CN104268638A - Photovoltaic power generation system power predicting method of elman-based neural network - Google Patents

Abstract

The invention discloses a method for predicting the power of a photovoltaic power generation system based on an elman neural network. Determine the optimal number of neurons in the hidden layer, thereby establishing a neural network based on elman; normalize the historical data of power generation and historical weather parameter information, and then use the normalized data to establish a neural network Conduct training to control the prediction error of the neural network based on elman within the preset range; take the historical data of power generation one week before the forecast day and the weather parameter data of the forecast day as input, and use the trained neural network to predict the forecasting day. power generation forecast. The invention has the advantages of stability, ability to adapt to time-varying characteristics and high prediction accuracy, and can be widely used in the field of photovoltaic power generation.

Description

A Power Prediction Method of Photovoltaic Power Generation System Based on Elman Neural Network

technical field

The invention relates to the field of photovoltaic power generation, in particular to a method for predicting the power of a photovoltaic power generation system based on an elman neural network.

Background technique

Renewable energy power generation is a relatively efficient and clean renewable energy utilization method, and it is also one of the most mature, large-scale development conditions and commercial development prospects among current renewable energy utilization technologies. Photovoltaic power generation is the main way of utilizing renewable energy and is the main component of smart grid. The prediction of short-term power generation is the key to the successful promotion of photovoltaic power generation. It is also the basis for the power dispatching department to formulate power dispatching plans, and it is also an important guarantee for the benefits of self-built photovoltaic power generation systems such as households or enterprises.

At present, all short-term solar photovoltaic power generation prediction methods are based on the same idea: first, use mathematical and physical theories and related data to establish a prediction formula or model, and then use the prediction formula or model to predict the power generation of photovoltaic power plants. According to the mathematical and physical theory adopted and its predicted output, photovoltaic power generation prediction methods can be divided into two categories: (1) direct prediction method (also called statistical method) to directly predict the output power of photovoltaic system; Radiation is predicted, and then the indirect prediction method (also called physical method) of photoelectric output power is obtained according to the photoelectric conversion efficiency.

Forecasting methods based on statistical methods include probability method, time series method and artificial intelligence method. The advantage is that the procedure is simple and there is no requirement for the location of the photovoltaic power station and power conversion parameters; the disadvantage is that it does not consider the environmental factors that affect photovoltaic power generation and requires a lot The historical operating data of the photovoltaic power station is used to ensure the accuracy of the forecast results, and it is easy to cause excessive fluctuations in the forecast accuracy due to changes in the environment. The prediction method based on physics method is mainly based on the physical power generation principle of photovoltaic power generation system. Its advantage is that it does not require historical operating data, and it can be directly predicted after the photovoltaic power station is built; the disadvantage is that it needs data such as detailed topographic maps of the photovoltaic power station, coordinates of the power station, power curve of the photovoltaic power station, and other related photoelectric conversion parameters.

At present, the prediction method based on BP neural network (a kind of artificial intelligence method) is widely used in the industry. However, the prediction method based on BP neural network still has the following defects:

(1) There is only feedforward without feedback, and the sensitivity to historical data is too poor, which will easily lead to the disappearance of the information of the learned learning mode, which is not stable enough;

(2) The ability to process dynamic information is too weak to directly and dynamically reflect the characteristics of the photovoltaic power generation system in the dynamic process, and it does not have the ability to adapt to time-varying characteristics, and the prediction accuracy fluctuates greatly.

Contents of the invention

In order to solve the above-mentioned technical problems, the object of the present invention is to provide a method for predicting the power of a photovoltaic power generation system based on an elman neural network, which is stable, capable of adapting to time-varying characteristics, and high in prediction accuracy.

The technical solution adopted by the present invention to solve its technical problems is:

A method for predicting the power of a photovoltaic power generation system based on an elman neural network, comprising:

A. Obtain historical power generation data of photovoltaic power generation equipment in relevant areas and corresponding historical weather parameter information;

B, determine the input and output data of neural network, and determine optimal hidden layer neuron number, thereby set up the neural network based on elman, described neural network based on elman comprises input layer, hidden layer, carrying layer and An output layer, the receiving layer is used to memorize the output value of the hidden layer at the previous moment and return the output value to the input of the hidden layer;

C. Perform normalization processing on the historical data of power generation and historical weather parameter information, and then train the established neural network according to the normalized data, so as to control the prediction error of the neural network based on elman at the preset value within the scope;

D. Taking the historical power generation data of the week before the forecast day and the weather parameter data on the forecast day as input, use the trained neural network to predict the power generation on the forecast day.

Further, the preset range is 5%-10%.

Further, the historical power generation data includes hourly power generation and effective power generation time period, and the historical weather parameter information includes air temperature, air pressure, wind direction, wind speed, cloud cover, rainfall, sunshine time and weather type.

Further, the step B includes:

B1. Statistically obtained historical power generation data and historical weather parameter information, the actual power generation power of a day is used as the output data of the neural network, and the hourly power generation W and the weather of the day in the effective time period f of the previous week are used The parameter data is used as the input data of the neural network;

B2. Initialize the elman neural network, determine the input node unit vector, the hidden layer node unit vector, the feedback state vector and the output node vector according to the input and output sequences, thereby establishing a neural network training model based on elman, wherein, The number of nodes in the hidden layer is obtained by incremental trial and error.

Further, the nonlinear state space expression of the elman neural network is:

,

Among them, y is the m-dimensional output node vector; l is the n -dimensional hidden layer node unit vector; x is the u-dimensional input vector; c is the n-dimensional feedback state vector; w ³ is the connection weight from the hidden layer to the output layer ; w ² is the connection weight from the input layer to the hidden layer; w ¹ is the connection weight from the receiving layer to the hidden layer; g (*) is the transfer function of the output neuron; f (*) is the hidden layer neuron transfer function.

Further, said step C, which includes:

C1. Use the maximum and minimum method to normalize the historical data of power generation and historical weather parameter information. The formula for the normalization process is:

,

Among them, x _max is the maximum number in the data sequence, and x _min is the minimum number in the data sequence;

C2. Perform error calculation, weight update and threshold update on the established neural network according to the normalized data, so as to control the prediction error of the elman-based neural network within the range of 5%-10%.

Further, the elman neural network uses the BP algorithm to correct and update the weights, and uses the error sum of squares function to learn the indicator function, and the formula for learning the indicator function E (w) is:

,

in, Enter a vector for the target.

The beneficial effects of the present invention are: through the construction of Elman neural network, combined with historical weather parameter information corresponding to the historical data of power generation of photovoltaic power generation equipment in relevant areas, the power generation power of the forecast day is obtained, wherein, Elman neural network includes input layer, Hidden layer, output layer, and the input receiving layer used to remember the output value of the hidden layer at the previous moment and return the output value to the hidden layer, which increases feedback, is more sensitive to historical data, and is more stable; when training It needs error control, can approach any nonlinear mapping with arbitrary precision, and does not consider the influence of external noise on the system, so that the system has high precision, and has the ability to adapt to time-varying characteristics, and can directly and dynamically reflect the dynamic process system characteristics, reducing the volatility of prediction accuracy.

Description of drawings

The present invention will be further described below in conjunction with drawings and embodiments.

Fig. 1 is the overall flowchart of a kind of photovoltaic power generation system power prediction method based on elman neural network of the present invention;

Fig. 2 is the structural representation of the neural network based on elman of the present invention;

Fig. 3 is the flowchart of step B of the present invention;

Fig. 4 is a flowchart of step C of the present invention.

Detailed ways

Referring to Figure 1 and Figure 2, a photovoltaic power generation system power prediction method based on elman neural network, including:

A. Obtain the historical power generation data of photovoltaic power generation equipment in relevant areas and the corresponding historical weather parameter information;

B, determine the input and output data of neural network, and determine optimal hidden layer neuron number, thereby set up the neural network based on elman, described neural network based on elman comprises input layer, hidden layer, carrying layer and An output layer, the receiving layer is used to memorize the output value of the hidden layer at the previous moment and return the output value to the input of the hidden layer;

C. Perform normalization processing on the historical data of power generation and historical weather parameter information, and then train the established neural network according to the normalized data, so as to control the prediction error of the neural network based on elman at the preset value within the scope;

D. Taking the historical power generation data of the week before the forecast day and the weather parameter data on the forecast day as input, use the trained neural network to predict the power generation on the forecast day.

Further as a preferred embodiment, the preset range is 5%-10%.

As a further preferred embodiment, the historical power generation data includes hourly power generation and effective power generation time periods, and the historical weather parameter information includes air temperature, air pressure, wind direction, wind speed, cloud cover, rainfall, sunshine time and weather type.

Referring to Fig. 3, further as a preferred embodiment, the step B includes:

B1. Statistically obtained historical power generation data and historical weather parameter information, the actual power generation power of a day is used as the output data of the neural network, and the hourly power generation W and the weather of the day in the effective time period f of the previous week are used The parameter data is used as the input data of the neural network;

B2. Initialize the elman neural network, determine the input node unit vector, the hidden layer node unit vector, the feedback state vector and the output node vector according to the input and output sequences, thereby establishing a neural network training model based on elman, wherein, The number of nodes in the hidden layer is obtained by incremental trial and error.

Further as a preferred embodiment, the nonlinear state space expression of the elman neural network is:

,

Among them, y is the m-dimensional output node vector; l is the n -dimensional hidden layer node unit vector; x is the u-dimensional input vector; c is the n-dimensional feedback state vector; w ³ is the connection weight from the hidden layer to the output layer ; w ² is the connection weight from the input layer to the hidden layer; w ¹ is the connection weight from the receiving layer to the hidden layer; g (*) is the transfer function of the output neuron; f (*) is the hidden layer neuron transfer function.

Referring to Fig. 4, further as a preferred embodiment, the step C includes:

C1. Use the maximum and minimum method to normalize the historical data of power generation and historical weather parameter information. The formula for the normalization process is:

,

Among them, x _max is the maximum number in the data sequence, and x _min is the minimum number in the data sequence;

C2. Perform error calculation, weight update and threshold update on the established neural network according to the normalized data, so as to control the prediction error of the elman-based neural network within the range of 5%-10%.

Further as a preferred embodiment, the elman neural network adopts the BP algorithm to correct and update the weight, and adopts the error sum of squares function to carry out the index function learning, and the formula of the index function E (w) learning is:

,

in, Enter a vector for the target.

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments.

Embodiment one

Referring to Fig. 2, the first embodiment of the present invention:

The present invention adopts the prediction model based on Elman neural network to realize the short-term prediction of photovoltaic system power generation, and the structure of the Elman neural network model is shown in FIG. 2 . Among them, X ₁ , X ₂ ···Xu are nodes in the input layer, corresponding to the input forecast weather parameters and last week’s photovoltaic system power generation; Y ₁ is a node in the output layer, corresponding to the output forecast day system power generation; l ₁ , l ₂ ···l _N is the node of the hidden layer, and the number of nodes in the hidden layer n (that is, the optimal number of neurons in the hidden layer) is determined by gradually increasing the trial method; C ₁ , C ₂ ··· _CN is the node of the receiving layer, which is used to memorize the output value of the hidden layer unit at the previous moment and return it to the input of the hidden layer.

The nonlinear state space expression of this Elman neural network is:

Among them, g(*) is the transfer function of the output neuron, which is a linear combination of the output of the hidden layer; f(*) is the transfer function of the hidden layer neuron, and the S function is often used.

Embodiment two

With reference to Fig. 1-4, the second embodiment of the present invention:

The main implementation steps of the photovoltaic power generation system power prediction method of the present invention are as follows:

Step 1, obtain the historical data of the power generation of photovoltaic power generation equipment in the relevant area, including the hourly power generation W and the effective power generation time period f, so as to obtain the effective power generation time period of the predicted power generation;

Step 2, obtain corresponding historical weather parameter information, including but not limited to temperature T, air pressure P, wind direction WD, wind speed WS, cloud cover C, rainfall R, sunshine time t and weather type P;

Step 3: Statistically obtain the historical data of generated power and historical weather parameter information, take the actual generated power of a day as the output data of the neural network, and use the hourly generated power W in the effective time period f of the previous week and the predicted day’s Weather parameter data as input data;

Step 4, network initialization, according to the input and output sequence (X, Y), determine u-dimensional input node unit vector x, n-dimensional hidden layer node unit vector l, n-dimensional feedback state vector c, m-dimensional output node vector y , where the number of hidden layer nodes n is determined by gradually increasing the number of hidden layer neurons, until the network performance reaches the set threshold or stops when it is optimal, so as to establish the neural network training based on Elman Model;

Step 5, use the maximum and minimum method to normalize the historical data of power generation and historical information of weather parameters, and then use it to train the neural network (including the process of error calculation, weight update and threshold update), and the network's The prediction error is controlled within 5% to 10%;

Step 6, after the training is completed, the neural network can be used to predict the power generation on the forecast day, so as to obtain the forecast result.

Compared with the prior art, the present invention establishes a photovoltaic power generation system neural network power prediction model through a photovoltaic power generation system power prediction method based on Elman neural network, which can approach any nonlinear mapping with arbitrary precision, regardless of external The influence of noise on the system makes the system have high precision and the ability to adapt to time-varying characteristics, which can directly and dynamically reflect the characteristics of the dynamic process system.

The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements without violating the spirit of the present invention. , these equivalent modifications or replacements are all within the scope defined by the claims of the present application.

Claims (7)

1. A photovoltaic power generation system power prediction method based on elman neural network, characterized in that: comprising: A. Obtain historical power generation data of photovoltaic power generation equipment in relevant areas and corresponding historical weather parameter information; B, determine the input and output data of neural network, and determine optimal hidden layer neuron number, thereby set up the neural network based on elman, described neural network based on elman comprises input layer, hidden layer, carrying layer and An output layer, the receiving layer is used to memorize the output value of the hidden layer at the previous moment and return the output value to the input of the hidden layer; C. Perform normalization processing on the historical data of power generation and historical weather parameter information, and then train the established neural network according to the normalized data, so as to control the prediction error of the neural network based on elman at the preset value within the scope; D. Taking the historical power generation data of the week before the forecast day and the weather parameter data on the forecast day as input, use the trained neural network to predict the power generation on the forecast day. 2. A kind of photovoltaic power generation system power prediction method based on elman neural network according to claim 1, is characterized in that: described preset range is 5%-10%. 3. a kind of photovoltaic power generation system power prediction method based on elman neural network according to claim 2, is characterized in that: described generation power historical data comprises hourly generation power and effective generation time period, and described historical weather parameter information Including temperature, air pressure, wind direction, wind speed, cloud cover, rainfall, sunshine time and weather type. 4. a kind of photovoltaic power generation system power prediction method based on elman neural network according to claim 3, is characterized in that: described step B, it comprises: B1. Statistically obtained historical power generation data and historical weather parameter information, the actual power generation power of a day is used as the output data of the neural network, and the hourly power generation W and the weather of the day in the effective time period f of the previous week are used The parameter data is used as the input data of the neural network; B2. Initialize the elman neural network, determine the input node unit vector, the hidden layer node unit vector, the feedback state vector and the output node vector according to the input and output sequences, thereby establishing a neural network training model based on elman, wherein, The number of nodes in the hidden layer is obtained by incremental trial and error. 5. a kind of photovoltaic power generation system power prediction method based on elman neural network according to claim 4, is characterized in that: the nonlinear state space expression of described elman neural network is: , Among them, y is the m-dimensional output node vector; l is the n -dimensional hidden layer node unit vector; x is the u-dimensional input vector; c is the n-dimensional feedback state vector; w ³ is the connection weight from the hidden layer to the output layer ; w ² is the connection weight from the input layer to the hidden layer; w ¹ is the connection weight from the receiving layer to the hidden layer; g (*) is the transfer function of the output neuron; f (*) is the hidden layer neuron transfer function. 6. A kind of photovoltaic power generation system power prediction method based on elman neural network according to claim 4, is characterized in that: described step C, it comprises: C1, using the maximum and minimum method to normalize the historical data of power generation and historical weather parameter information, the formula for the normalization process is: , Among them, x _max is the maximum number in the data sequence, and x _min is the minimum number in the data sequence; C2. Perform error calculation, weight update and threshold update on the established neural network according to the normalized data, so as to control the prediction error of the elman-based neural network within the range of 5%-10%. 7. a kind of photovoltaic power generation system power prediction method based on elman neural network according to claim 6, is characterized in that: described elman neural network adopts BP algorithm to carry out weight correction update, and adopts error square sum function to carry out indicator function Learning, the formula of described indicator function E (w) learning is: , in, Enter a vector for the target.