JP5215822B2 - Energy system control device and control method - Google Patents

Description

  The present invention relates to a control device and a control method for an energy system having an energy generation device, an energy storage device, and an energy load.

  As a method of performing low-cost operation control by effectively using the energy of the distributed energy system, there is a “distributed energy community system and its control method” described in Patent Document 1.

  In this system, the control center receives data on the power generation amount of the fuel cell, the energy storage amount of the storage battery, and the power consumption amount of the load from the control device of the distributed power system via the communication line, and generates power to each distributed energy system The power value and the received / transmitted power value are commanded, and the power supply / demand via the power line is complementarily controlled between the plurality of distributed energy systems having different daily load characteristics of the power demand.

  As a technique for predicting the energy demand necessary for controlling the energy system, there are techniques using regression analysis and a neural network.

  However, as in the energy system described in Patent Document 1, the demand for energy such as electric power / hot water supply for each household in a general household depends on the irregular living behavior of the consumer, and thus accurate prediction is difficult. . Hot water supply demand is more difficult to predict because there are many times when there is no demand at all, and demand peaks occur once.

  On the other hand, power generation systems that use natural energy such as solar and wind power use weather forecasts such as temperature and weather to forecast demand and power generation, so the accuracy of the forecast depends greatly on the accuracy of the weather forecast. To do.

Therefore, as described in Patent Document 2, a predicted deviation pattern and its occurrence probability are calculated, an operation plan of the energy system is corrected based on the calculation result, and an operation simulation is performed based on the corrected operation plan. There is a method of creating an optimal operation plan that can execute a control operation corresponding to the uncertainty of prediction or the deviation from the prediction.
JP 2002-44870 A JP 2006-304402 A

  In such an energy system, if the weather forecast and the demand / power generation forecast using it are not met, optimal operation control cannot be performed, and optimization corresponding to the deviation from the forecast has a huge calculation cost. There is a problem that control takes place in real time and takes too much time.

  The present invention has been made in view of the above-described problems, and is an energy system that combines a prediction and an optimal operation plan in the prior art, and calculates an operation plan close to the optimum regardless of the deviation of the prediction at high speed. It is an object to provide a control device and a control method.

In order to achieve the above object, an energy system control device according to the present invention can communicate with an energy system having an energy generation device, an energy storage device, and an energy load, and uses the weather forecast information to communicate the energy. An energy system controller for determining an operation plan of the generator and the energy storage device, and a result database unit for storing a result value of energy generation by the energy generator and a result value of energy demand by the energy load; Optimal scheduling unit that creates an optimal operation plan for the energy generation device and the energy storage device based on the actual value of energy generation and the actual value of energy demand in the actual database unit using a predetermined optimization method And the Enel A state determination function for determining a state evaluation function for evaluating the state of the generator and the energy storage device based on an optimal operation plan created by the optimal scheduling unit, and a state determined by the determination unit and the evaluation function, using, and the weather forecast information, see including the operation plan determination section, the determining the operation plan, the fixed portion, and the actual value of the weather forecast information, the actual value of the energy generation A learning database unit storing the actual value of the energy demand and the optimum operation plan, and when the data in the learning database unit satisfies a learning execution condition, a predetermined parameter of the state evaluation function Using a predetermined machine learning method, the actual value of the weather forecast information in the learning database unit, the actual value of energy generation, and the energy demand Including and actual value, and the optimum operation plan, and an evaluation function learning unit for determining based on.

An energy system control method according to the present invention can communicate with an energy system having an energy generation device, an energy storage device, and an energy load, and uses the weather forecast information to operate the energy generation device and the energy storage device. An energy system control method performed by an energy system control device for determining a plan, wherein an actual value of energy generation by the energy generator and an actual value of energy demand by the energy load are stored in an actual data database unit; An optimal schedule for creating an optimal operation plan for the energy generator and the energy storage device based on the actual value of energy generation and the actual value of energy demand in the actual database unit using a predetermined optimization method. A step of determining a state evaluation function for evaluating states of the energy generation device and the energy storage device based on the optimum operation plan, the state evaluation function, and the weather forecast information. with, viewed including the operation plan determining step, the determining the operation plan, the determination step includes the actual value of the weather forecast information, and actual value of the energy generation, the actual value of the energy demand, the A storage step for storing the optimum operation plan in the learning database unit, and a predetermined parameter value of the state evaluation function when the data in the learning database unit satisfies a learning execution condition. Using the learning method, the actual value of the weather forecast information in the learning database part, the actual value of energy generation and the energy demand Including between actual value, and the optimum operation plan, the evaluation function learning step of determining, based on the.

  The learning database unit includes, for each predetermined date and time, the optimum operation plan, the actual value of the weather forecast information, the actual value of the energy generation, the actual value of the energy demand, and the output state of the energy generator. It is desirable that the storage state of the energy storage device is stored in association with each other.

  The predetermined parameters include a first parameter for an evaluation value of an energy storage amount in the energy storage device, and a second parameter for an evaluation value of an output state of the energy generation device, and the first parameter and the first parameter The two parameters preferably have values according to the weather forecast information.

  The optimum scheduling unit uses the optimization method to optimize the operation plan so that the value of an objective function for evaluating the degree of achievement of the purpose in the operation plan is optimized in the optimum operation plan. Based on the actual value of the energy generation and the actual value of the energy demand, the evaluation function learning unit uses an operation evaluation function composed of the objective function and the state evaluation function as the machine learning method. Using a machine learning method, the value of the predetermined parameter is determined based on the actual value of weather forecast information, the actual value of energy generation and the actual value of energy demand in the learning database unit, and the optimal operation plan The operation plan determination unit preferably determines the operation plan using the operation evaluation function and the weather forecast information.

  The evaluation function learning unit creates a plurality of other operation plan candidates different from the optimal operation plan, and the value of each operation evaluation function of the operation plan candidates is greater than the value of the operation evaluation function of the optimal operation plan. It is desirable to determine the predetermined parameter value that minimizes the number to be improved.

  As described above, according to the present invention, it is possible to calculate an operation plan close to the optimum at high speed regardless of the deviation in prediction.

  For example, since the optimal scheduling result using the actual value of the input / output information of prediction in the prior art is learned as teacher data, the prediction model can be convoluted with the learned parameter, and calculation processing only for prediction is unnecessary.

  In addition, depending on the weather information (temperature and weather) that is the input information for prediction, the accumulated amount of the energy storage device and the operating state of the energy generator, which are factors that do not lead to long-term optimization in the optimization by time zone, Therefore, it is possible to obtain an output close to an operation obtained by prediction and optimization over a long period of time.

  Moreover, the data creation process for machine learning using the optimal scheduling over a long period of time that requires a calculation cost can be performed at an arbitrary time using the results database unit, and the operation plan can be determined, for example, for a short period of time. Since optimization calculation processing can be performed, optimal control can be performed at high speed with respect to fluctuations in demand and power generation amount.

  Next, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration of an energy system according to an embodiment of the present invention. The energy system 1 includes devices used by the consumers 1a to 1n. The energy system 1 is controlled by the control device 2.

  Each consumer 1a-1n uses the solar cell 11, the fuel cell 12, the storage battery 13, the heat storage apparatus 14, the electric power load 15, the thermal load 16, and the thermal piping 17.

  The solar cell 11 is used as an energy generating device that generates energy using natural energy. The fuel cell 12 is used as an energy generator. The storage battery 13 and the heat storage device 14 are used as energy storage devices. The electric power load 15 and the heat load 16 are used as energy loads.

  Solar cell 11, fuel cell 12, storage battery 13, and power load 15 are connected to system power (commercial power) 4 by power line 3. For this reason, electric power can be interchanged between a plurality of consumers (only two consumers are shown in FIG. 1). Further, the heat storage device 14 and the heat load 16 are connected to the fuel cell 12 by a heat pipe 17.

  The energy generation device and the energy storage device are not limited to those described above, and other types of energy generation devices and energy storage devices may be used for each consumer.

  The control device 2 measures the energy generation device, the energy storage device, the system power 4 and the energy load, and controls the power generation amount of the fuel cell 12 and the charge / discharge amount of the storage battery 13. The control device 2 can communicate with the energy system 1 and determines an operation plan of the energy generation device and the energy storage device using the weather forecast information.

  The control device 2 includes a communication unit 21, a performance database unit 22, an optimal scheduling unit 23, a learning database unit 24, an evaluation function learning unit 25, and an optimal operation calculation unit 26. The learning database unit 24 and the evaluation function learning unit 25 are included in the parameter determination unit 27.

  Here, an outline of the communication unit 21, the performance database unit 22, the optimum scheduling unit 23, the learning database unit 24, the evaluation function learning unit 25, the optimum operation calculation unit 26, and the parameter determination unit 27 will be described.

  The communication unit 21 can connect to the Internet or the like and receive information on weather forecast information, energy prices, and the like. These pieces of information are provided to the result database unit 22 and the optimum operation calculation unit 26.

  The performance database unit 22 stores, for example, a performance value of energy demand due to energy load and a performance value of energy generation by the energy generator.

  The optimum scheduling unit 23 uses the predetermined optimization method (hereinafter, simply referred to as “optimization method”) to generate the energy in the performance database unit 22 for the optimum operation plan of the energy generation device and the energy storage device. Based on the actual value and actual value of energy demand.

  For example, the optimum scheduling unit 23 uses the optimization method to optimize the operation plan so that the value of the objective function for evaluating the degree of achievement of the goal in the operation plan is optimized in the optimum operation plan. Created based on the actual value of occurrence and the actual value of energy demand.

  The parameter determination unit 27 is an example of a determination unit, and determines a state evaluation function for evaluating the states of the energy generation device and the energy storage device based on the optimum operation plan created by the optimum scheduling unit 23.

  In the present embodiment, the parameter determination unit 27 selects a predetermined parameter different from the operation plan from a plurality of parameters included in the state evaluation function for evaluating the state of the energy generation device and the energy storage device. A learning method (hereinafter simply referred to as “machine learning method”) is used to make a determination based on the optimum operation plan created by the optimum scheduling unit 23.

  The predetermined parameters include a first parameter for the evaluation value of the energy storage amount in the energy storage device and a second parameter for the evaluation value of the output state of the energy generation device. The first parameter and the second parameter have values corresponding to the weather forecast information.

  The learning database unit 24 stores, for example, the actual value of weather forecast information, the actual value of energy generation, the actual value of energy demand, and the optimum operation plan.

  In the present embodiment, the learning database unit 24 performs the optimum operation plan, the actual value of weather forecast information, the actual value of energy generation, the actual value of energy demand, and the output state of the energy generator for each predetermined date and time. And the storage state of the energy storage device are stored in association with each other.

  The evaluation function learning unit 25 uses the machine learning method to calculate the actual parameter value of the weather forecast information, the actual value of energy generation, the actual value of energy demand, and the optimum operation plan using the machine learning method. And based on the above.

  In the present embodiment, the evaluation function learning unit 25 uses a machine learning method using a driving evaluation function composed of an objective function and a state evaluation function as a machine learning method, and calculates a value of a predetermined parameter as a learning database unit. It is determined based on the actual value of the weather forecast information in 24, the actual value of energy generation, the actual value of energy demand, and the optimum operation plan.

  For example, the evaluation function learning unit 25 creates a plurality of other operation plan candidates different from the optimal operation plan, and the value of each operation evaluation function of the operation plan candidates is better than the value of the operation evaluation function of the optimal operation plan. The predetermined parameter value that minimizes the number is determined.

  The optimum operation calculation unit 26 is an example of an operation plan determination unit, and uses the state evaluation function in which the value of a predetermined parameter is determined by the parameter determination unit 27 and the weather forecast information, and the energy generator and the energy Determine the operation plan of the storage device.

  In the present embodiment, the optimum operation calculation unit 26 determines an operation plan for the energy generation device and the energy storage device using the operation evaluation function and the weather forecast information.

  FIG. 2 is a flowchart showing the flow of the evaluation function learning process in the control device 2 of the system shown in FIG.

  First, the performance database unit 22 records actual values corresponding to weather forecast information such as weather and temperature, actual values of energy demand at the power load 15, actual values of energy demand at the heat load 16, and the solar cell 11. When the communication unit 21 and the energy system 1 are received the actual value of power generation (the actual value of energy generation) (step 201), they are stored in association with the date / time information (step 202).

  Next, the process of the optimal scheduling unit 23 will be described.

  The optimal scheduling is to create an operation plan for a predetermined period for power generation of the fuel cell 12 that is a controllable energy generation device and charge / discharge of the storage battery 13 that is a controllable energy storage device. One day or one week is appropriate for the predetermined period.

  The optimal scheduling unit 23 does not perform the optimal scheduling before the date and time when the operation plan is to be executed, but stores the actual value of the energy demand at the power load 15 and the thermal load 16 stored in the actual database unit 22. The optimal scheduling is calculated after the date and time when the operation plan is to be executed, using the actual value of the energy demand at and the actual value of the power generation amount of the solar cell 11.

  The optimal scheduling unit 23 starts the optimal scheduling at the timing when the actual values are obtained (step 203).

  In addition, when the predetermined period is one week, it is desirable that the optimum scheduling unit 23 performs the optimum scheduling while updating the actual value data for one day every day.

  When calculating the optimal scheduling, the optimal scheduling unit 23 sets an objective function for evaluating the degree of achievement of the objective in the operation plan such as minimizing the cost or the carbon dioxide emission amount. An operation plan having the best function value (specifically, a combination of the operation plan of the fuel cell 12 and the operation plan of the storage battery 13) is searched.

  For this search, metaheuristic methods such as tabu search and genetic algorithm are suitable in that a practical approximate solution can be calculated relatively quickly even for large-scale optimization problems. However, mathematical programming such as linear programming can also be used.

  The meta-heuristic method is a unified methodology for using “rules of thumb” to efficiently obtain solutions to optimization problems. An accurate approximate solution can be obtained.

  Tabu Search (TS: Tabu <Contraindication> Search) is an optimization method that has an analogy in human memory processes, and relies on memory so that the same solution (driving schedule) is not circulated. If it gets worse, aim for the gentlest slope. In the tabu search, a memory for avoiding returning to the already passed solution is stored in a memory called a tabu list and used. In tabu search, backtracking to the same solution or moving to a similar solution is prohibited, and it is possible to escape from the local solution, so that the optimal solution can be searched globally.

  Genetic algorithm (GA) is an algorithm that imitates the process of evolution of living organisms, and is an engineering model of genetic laws. In the process of evolution of living organisms in the natural world, a set of several individuals is formed, and individuals with high adaptability to the environment are selected with high probability and survive.

  In the genetic algorithm, the fitness is calculated for a plurality of individuals (operation schedules) based on the objective function value, and then an operation called genetic operator consisting of selection, crossover, and mutation is performed to determine the fitness. High individuals survive and proliferate to the next generation. A series of these operation cycles is called a generation, and an individual having the maximum fitness among all generations is set as an optimal solution. Since the genetic algorithm is a multipoint (number of individuals) search, an optimal solution can be searched globally.

  When the optimal scheduling is performed, the optimal scheduling unit 23 receives the input information (the actual value of energy demand at the power load 15, the actual value of the heat load 16, and the actual value of the power generation amount of the solar cell 11), The optimum operation plan as an output result is stored in the learning database unit 24 (step 204).

  If the data in the learning database unit 24 satisfies the learning execution conditions such as being updated a predetermined number of times (step 205), the evaluation function learning unit 25 determines the parameters of the state evaluation function using a machine learning method. (Step 206).

  FIG. 3 is an explanatory diagram illustrating an example of a method for creating a learning database.

  As shown in FIG. 3, the optimal scheduling unit 23 divides the optimally scheduled operation plan for each predetermined time step, and stores it in the learning database unit 24 together with the input information.

  Input information includes weather information such as weather and temperature. As the weather information, it is desirable to include performance information on future weather close to the target time zone. For example, in the morning time zone, it is the afternoon weather performance and the highest temperature, and in the evening time zone, it is the next day weather performance and the lowest temperature. This is to reflect the near future situation that should be reflected in the optimum scheduling result also in the learned state evaluation function.

  Other input information includes actual values of power demand, heat demand, and photovoltaic power generation, which have been subject to forecasting in the past, and the above-mentioned near-term weather information is reflected in these results and reflected in the optimal operation plan. Should be.

  Information obtained from the operation plan includes the output value of each fuel cell 12 and storage battery 13 in the operation plan, as well as the storage amount and heat storage amount at the beginning of the time step that is an element of the state evaluation function.

  FIG. 4 is an explanatory diagram showing an example of a machine learning method for parameters of the state evaluation function in the evaluation function learning unit 25.

  First, the state evaluation function includes a parameter for an evaluation value related to starting and stopping of the fuel cell 12, a parameter for the evaluation value of the power generation state of the fuel cell 12 (second parameter), and a parameter for the evaluation value of the storage amount of the storage battery 13. (First parameter), and a parameter (first parameter) for an evaluation value of the heat storage amount of the heat storage device 14, and the parameters of the evaluation value depend on a time zone, weather forecast information, and the like.

  The fact that each evaluation value depends on the time zone, weather forecast information, etc., may be that each evaluation value has a value for each time zone and weather, or may be a function of the time zone and temperature.

  The evaluation value of the amount of stored electricity and the evaluation value of the stored amount of heat are not set as a unit price, but by depending on the amount of itself, for example, when the full charge approaches, the evaluation value of the additional amount of stored electricity decreases, It is learned to prevent overcharging.

  In addition, it is assumed that the evaluation value of the stored electricity amount and the evaluation value of the heat storage amount also depend on weather information in the near future, which is input information of the learning database unit 24.

  For example, the value of power storage should be higher when it is raining than when it is sunny in the afternoon, and in summer the higher maximum temperature is expected to increase the power load. Should have.

  Similarly, the starting value of the fuel cell 12 and the value of the amount of heat stored in the heat storage device 14 should depend on the prediction of not only the power load 15 but also the heat load 16, that is, weather information in the near future. It is.

  In this way, a large number of dependent elements are conceivable, and accordingly, the parameters become enormous and it is difficult to set an appropriate value for each parameter. Therefore, the evaluation function learning unit 25 determines a large number of parameters of the state evaluation function. It is indispensable to use machine learning as a technique to do this.

  Further, the evaluation function learning unit 25 uses, as a driving evaluation function, a state evaluation function added to an objective function in a time step period such as a cost used in optimal scheduling.

  In this example, the objective function is valuable for reducing the cost for a certain period, and the state evaluation function is valuable for accumulating a lot of energy. It consists of the difference between the objective function and the state evaluation function.

  As a learning procedure, the evaluation function learning unit 25 first creates a plurality of operation patterns (operation plan candidates) different from the optimal operation plan for each predetermined period calculated by the optimal scheduling unit 23.

  For example, the operation plan candidate is an operation pattern in which the output of the fuel cell 12 is increased or the discharge power of the storage battery 13 is decreased on the basis of the optimum operation plan, and various combinations are possible. Therefore, the evaluation function learning unit 25 narrows down the driving pattern candidates by selecting, for example, those having good driving evaluation functions from the driving pattern candidates, and creates a set of driving pattern candidates for each time step.

  Next, the evaluation function learning unit 25 uses, as the learning objective function, a sigmoid function of the difference between the driving evaluation function value of the driving pattern at each time step in the optimal driving plan and the driving evaluation function value of each driving pattern candidate. Use the summation function.

  The reason why the sigmoid function is used is to make the learning objective function smooth and to make the learning to make the optimum driving pattern as advantageous as possible with respect to the driving pattern candidate whose evaluation value is relatively close to the optimum driving pattern.

  Specifically, the evaluation function learning unit 25 makes the value of each operation evaluation function of the operation plan candidate better than the value of the operation evaluation function of the optimal operation plan so that the objective function for learning is minimized. The parameters of the state evaluation function (excluding the operation plan) are obtained by using the steepest descent method, the conjugate gradient method, or the like so that the number is minimized.

  Although the machine learning method for the parameters of the state evaluation function has been described above, the evaluation function learning unit 25 may implement a neural network, various regression analysis, and other methods in combination with the machine learning method.

  FIG. 5 is a flowchart showing a process flow of the optimum operation calculation unit 26 in the control device 2 shown in FIG.

  First, the optimum driving calculation unit 26 receives weather forecast information and the like from the communication unit 21 (step 501).

  Subsequently, the optimum operation calculation unit 26 prepares an initial operation plan to be evaluated first after arranging information necessary for the evaluation of the operation (step 502).

  However, this operation plan does not extend over a long period of time as calculated by the optimal scheduling unit 23, and is preferably 1 to several times the time step when the learning database is created.

  The candidate for the initial operation plan may be an operation plan that continues the current operation as it is.

  Next, the optimum operation calculation unit 26 calculates an objective function value for the initial operation plan (step 503).

  In this calculation, since the target period is short, the optimum operation calculation unit 26 obtains the power demand, the heat demand, the amount of photovoltaic power generation, and the like based on the current values and the past trends.

  By this calculation, the optimum operation calculation unit 26 also calculates the state such as the amount of stored electricity and the amount of stored heat after the end of the period. The optimum operation calculating unit 26 calculates the state evaluation function after the end of the period by using the latest learned parameters, using the state of the amount of stored electricity and the amount of heat storage after the end of the period (step 504).

  The optimal operation calculation unit 26 uses the latest received weather forecast information received as near-term weather information that is a dependent element of the state evaluation function, so that the demand prediction in the prior art is folded into the calculation of the state evaluation function. And reflected in the calculation result of the state evaluation function.

  The optimum operation calculation unit 26 evaluates the initial operation plan based on the operation evaluation function value calculated from the objective function value and the state evaluation function value thus calculated.

  The optimum operation calculation unit 26 performs this evaluation on various operation plans (steps 505 and 506), determines an operation plan having the best evaluation, and determines an operation command value based on the operation plan (step 507). ).

  The optimal operation calculation unit 26 may use the metaheuristic method used by the optimal scheduling unit 23 or other optimization methods for this optimal calculation, but the target period is short, so that it is relatively short. Can be optimized.

  The function of the control device 2 described above is executed by recording a program for realizing the function on a computer-readable recording medium, causing the computer to read the program recorded on the recording medium, and executing the program. It may be. The computer-readable recording medium refers to a recording medium such as a flexible disk, a magneto-optical disk, and a CD-ROM, and a storage device such as a hard disk device built in a computer system. Further, the computer-readable recording medium is a medium that dynamically holds the program for a short time (transmission medium or transmission wave) as in the case of transmitting the program via the Internet, and in the computer serving as a server in that case Such as a volatile memory that holds a program for a certain period of time.

It is the block diagram which showed the structure of the energy system by one Embodiment of this invention. It is a flowchart which shows the process of evaluation function learning. It is explanatory drawing which shows an example of the method of producing a learning database from the optimal scheduling result. It is explanatory drawing which showed an example of the machine learning method of the parameter of a state evaluation function. 3 is a flowchart showing a process flow of an optimum operation calculation unit 26.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Energy system 1a-1n Consumer 11 Solar cell 12 Fuel cell 13 Storage battery 14 Thermal storage device 15 Electric power load 16 Thermal load 17 Thermal piping 2 Controller 21 Communication part 22 Performance database part 23 Optimal scheduling part 24 Learning database part 25 Evaluation function learning Unit 26 Optimal operation calculation unit 27 Parameter determination unit 3 Power line 4 System power

Claims (10)

An energy system control device capable of communicating with an energy system having an energy generation device, an energy storage device, and an energy load, and determining an operation plan of the energy generation device and the energy storage device using weather forecast information There,
An actual value database for storing the actual value of energy generation by the energy generator and the actual value of energy demand by the energy load;
Optimal scheduling unit that creates an optimal operation plan for the energy generation device and the energy storage device based on the actual value of energy generation and the actual value of energy demand in the actual database unit using a predetermined optimization method When,
A determination unit for determining a state evaluation function for evaluating a state of the energy generation device and the energy storage device based on an optimal operation plan created by the optimal scheduling unit;
An operation plan determination unit that determines the operation plan using the state evaluation function determined by the determination unit and the weather forecast information;
Only including,
The determination unit is
A learning database unit for storing the actual value of the weather forecast information, the actual value of the energy generation, the actual value of the energy demand, and the optimum operation plan;
When the data in the learning database unit satisfies the learning implementation conditions, the value of the predetermined parameter of the state evaluation function is used to determine the actual weather forecast information in the learning database unit using a predetermined machine learning method. An energy system control device comprising: an evaluation function learning unit that is determined based on the value, the actual value of energy generation, the actual value of energy demand, and the optimum operation plan . The learning database unit includes, for each predetermined date and time, the optimum operation plan, the actual value of the weather forecast information, the actual value of the energy generation, the actual value of the energy demand, and the output state of the energy generator. When, a storage state of the energy storage device, and stores in association with each other, energy system control device according to claim 1. The predetermined parameters include a first parameter for an evaluation value of an energy storage amount in the energy storage device, and a second parameter for an evaluation value of an output state of the energy generation device,
The energy system control device according to claim 1 or 2 , wherein the first parameter and the second parameter have values corresponding to the weather forecast information. The optimum scheduling unit uses the optimization method to optimize the operation plan so that the value of an objective function for evaluating the degree of achievement of the purpose in the operation plan is optimized in the optimum operation plan. Create based on the actual value of the energy generation and the actual value of the energy demand,
The evaluation function learning unit uses a machine learning method using an operation evaluation function composed of the objective function and the state evaluation function as the machine learning method, and calculates the value of the predetermined parameter in the learning database unit. Determined based on the actual value of the weather forecast information, the actual value of energy generation and the actual value of energy demand, and the optimum operation plan,
The operation plan determination unit includes: the driving evaluation function, the using the weather forecast information, and to determine the operation plan, the energy system control apparatus according to any one of claims 1 to 3. The evaluation function learning unit creates a plurality of other operation plan candidates different from the optimal operation plan, and the value of each operation evaluation function of the operation plan candidates is greater than the value of the operation evaluation function of the optimal operation plan. The energy system controller according to claim 4 , wherein the predetermined parameter value that minimizes the number to be improved is determined. An energy system control device capable of communicating with an energy system having an energy generation device, an energy storage device, and an energy load, and determining an operation plan of the energy generation device and the energy storage device using weather forecast information Is an energy system control method performed by
A storage step of storing the actual value of energy generation by the energy generator and the actual value of energy demand due to the energy load in the actual database unit;
Optimal scheduling step of creating an optimal operation plan of the energy generator and the energy storage device based on the actual value of energy generation and the actual value of energy demand in the actual database unit using a predetermined optimization method When,
A determination step for determining a state evaluation function for evaluating a state of the energy generation device and the energy storage device based on the optimum operation plan;
An operation plan determination step for determining the operation plan using the state evaluation function and the weather forecast information;
Only including,
The confirmation step includes
A storage step of storing the actual value of the weather forecast information, the actual value of the energy generation, the actual value of the energy demand, and the optimum operation plan in a learning database unit;
When the data in the learning database unit satisfies the learning implementation conditions, the value of the predetermined parameter of the state evaluation function is used to determine the actual weather forecast information in the learning database unit using a predetermined machine learning method. An energy system control method comprising: an evaluation function learning step that is determined based on the value, the actual value of energy generation, the actual value of energy demand, and the optimum operation plan . In the storing step, for each predetermined date and time, the optimum operation plan, the actual value of the weather forecast information, the actual value of the energy generation, the actual value of the energy demand, and the output state of the energy generator, The energy system control method according to claim 6 , wherein the accumulated state of the energy storage device is stored in the learning database unit in association with each other. The predetermined parameters include a first parameter for an evaluation value of an energy storage amount in the energy storage device, and a second parameter for an evaluation value of an output state of the energy generation device,
The energy system control method according to claim 6 or 7 , wherein the first parameter and the second parameter have values corresponding to the weather forecast information. In the optimal scheduling step, the optimal operation plan is optimized using the optimization method so that the value of the objective function for evaluating the degree of achievement of the objective in the operation plan is optimal in the optimal operation plan, Create based on the actual value of the energy generation and the actual value of the energy demand,
In the evaluation function learning step, a machine learning method using an operation evaluation function composed of the objective function and the state evaluation function is used as the machine learning method, and the value of the predetermined parameter is stored in the learning database unit. Determined based on the actual value of the weather forecast information, the actual value of energy generation and the actual value of energy demand, and the optimum operation plan,
The energy system control method according to any one of claims 6 to 8 , wherein, in the operation plan determination step, the operation plan is determined using the operation evaluation function and the weather forecast information. In the evaluation function learning step, a plurality of other operation plan candidates different from the optimum operation plan are created, and the value of each operation evaluation function of the operation plan candidates is greater than the value of the operation evaluation function of the optimum operation plan. The energy system control method according to claim 9 , wherein the predetermined parameter value that minimizes the number to be improved is determined.

Images