CN118520428A - Power load prediction method and system based on artificial intelligence - Google Patents

Abstract

The invention relates to the field of power load prediction, in particular to an artificial intelligence-based power load prediction method and system, wherein the method comprises the following steps: and fitting the load data in the obtained historical power load data segments, determining extreme points of a fitting curve, calculating the initial noise degree of the historical power load data segments, correcting the initial noise degree according to the initial noise degree of the corresponding historical power grid frequency data segments and the correlation between the historical power load data segments and the corresponding historical power grid frequency data segments to obtain final noise degrees of the historical power load data segments, and weighting predicted values determined by the corresponding historical power load data segments by utilizing the final noise degree of each historical power load data segment to obtain final predicted values. The method and the device can reduce the influence of noise data on the prediction result and improve the accuracy of power load prediction.

Description

A method and system for predicting power load based on artificial intelligence

Technical Field

The present invention relates to the field of power load forecasting, and more specifically, to a power load forecasting method and system based on artificial intelligence.

Background Art

With the development of economy and the acceleration of urbanization, the demand for electricity is growing. At the same time, the demand for intelligence and digitalization is also increasing. It is hoped that technical means can be used to improve energy efficiency and management level. The power load forecasting method and system based on artificial intelligence are in line with the development trend of intelligence and digitalization, and can provide accurate and efficient load forecasting services. The existing algorithm for predicting power load is the ARIMA model, which has the advantages of simplicity, flexibility, and strong interpretability. It is a commonly used and effective time series forecasting method.

In the related technology, for example, the application document with publication number CN116995668A discloses a medium- and long-term power load forecasting method and system based on an improved seasonal ARIMA model, the method comprising: decomposing the periodically extracted historical power load data according to the Fourier series expansion to obtain the mean, periodic component and non-periodic component; trend fitting, calculating the central moving average of the non-periodic component, and fitting the trend component using the least squares method; residual simulation, constructing an ARIMA autoregressive difference moving average model for the residual component obtained by subtracting the trend component from the non-periodic component; and then using the constructed ARIMA to predict the prediction results of each component for the next period.

However, the main improvement in the above scheme is the method of determining the parameters when constructing the ARIMA model, especially the method of determining the residual component, and does not take into account the impact of noise data in the collected historical power load data on the prediction results. As a result, when the pre-constructed ARIMA model predicts the corresponding components, it may be affected by the noise data, which reduces the accuracy of the prediction results to a certain extent.

Summary of the invention

In order to solve the problem of low accuracy of prediction results due to the influence of noise data, the present invention provides an artificial intelligence-based power load prediction method and system.

According to a first aspect of an embodiment of the present invention, there is provided a method for predicting power load based on artificial intelligence, comprising:

Acquire multiple historical power load data segments and corresponding historical power grid frequency data segments;

Fit the load data in the historical power load data segment, obtain the extreme value points of the fitting curve of the historical power load data segment, and calculate the initial noise level of the historical power load data segment. The initial noise level is positively correlated with the discreteness of the load data in the historical power load data segment and the number of extreme value points, and negatively correlated with the average interval time of the extreme value points.

Calculating the final noise level of the historical load data segment, the final noise level is negatively correlated with the initial noise level of the corresponding historical power grid frequency data segment, and the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment;

Determine the predicted value of the power load at the future time for the historical load data segment, use the final noise level of the historical load data to weight the predicted value, and obtain the weighted value of the predicted value corresponding to the historical load data segment, and then obtain the weighted value of the predicted value corresponding to each historical load data segment, and use the sum of the weighted values as the final predicted value of the power load at the future time.

The present invention can ensure the accuracy of the possibility of noise data existing in each determined historical power load data segment, and use the final noise level of each historical load data segment to weight the predicted value of the power load at a future moment in each historical load data segment. The predicted values of multiple historical load data segments can be comprehensively considered to ensure the accuracy of the determined final predicted value.

In an exemplary embodiment of the present invention, the final predicted value of the power load at a future time satisfies the following relationship:

;

In the formula, Indicates future time The final forecast value of the power load; Indicates the number of historical power load data segments; Indicates The historical power load data segment is used for the future time The predicted value of the power load; Indicates the future moment When predicting the power load, The final noise level of each historical power load data segment; Indicated by natural constant An exponential function with base .

By calculating the ratio of the final noise level of each historical power load data segment to the cumulative sum of the final noise levels of all historical power load data segments, the present invention can ensure that the sum of the weights of the prediction values corresponding to each historical power load data segment is 1, which can reduce the difficulty of weighting each prediction value to obtain the final prediction value.

In an exemplary embodiment of the present invention, the final noise level of the historical power load data segment satisfies the following relationship:

;

In the formula, Indicates the future moment When predicting the power load, The final noise level of each historical power load data segment; Indicates the future moment When predicting the power load, The initial noise level of a historical power load data segment; Indicates the future moment When predicting the power load, The initial noise level of the historical power grid frequency data segment corresponding to the historical power load data segment; Indicates the future moment When predicting the power load, The correlation between the load data in a historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment.

Since the load data and the frequency data are negatively correlated, adding 1 to the correlation between the load data and the frequency data can more clearly reflect the negative correlation between the final performance of each historical load data segment and the correlation.

In an exemplary embodiment of the present invention, a method for obtaining the correlation between load data in a historical power load data segment and frequency data in a corresponding historical power grid frequency data segment includes:

Calculate the Pearson correlation coefficient between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment, and use the Pearson correlation coefficient as the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment.

Since the Pearson correlation coefficient can reflect the change of the data in each data set in a monotonic direction, the present invention uses the Pearson correlation coefficient to measure the correlation between the load data and the frequency data, which can clearly characterize the negative correlation between the load data and the frequency data.

In an exemplary embodiment of the present invention, the initial noise level of the historical power load data segment satisfies the following relationship:

;

In the formula, Indicates the future moment When predicting the power load, The initial noise level of a historical power load data segment; Indicates the future moment When predicting the power load, The number of extreme value points in the fitting curve of a historical power load data segment; Represents the first intervals; and Respectively represent the first The acquisition time of the extreme value points on the right side and the acquisition time of the extreme value points on the left side corresponding to the interval; Indicates the future moment When predicting the power load, The discrete degree of load data in a historical power load data segment.

The present invention can consider the load data itself in each historical power load data segment from multiple aspects, and to a certain extent ensure the accuracy of the possibility of noise data existing in each historical power load data segment.

In an exemplary embodiment of the present invention, the discrete degree of load data in the historical power load data segment satisfies the following relationship:

;

In the formula, Indicates the future moment When predicting the power load, The discrete degree of load data in a historical power load data segment; and Respectively for the future moment When predicting the power load, The upper quartile and lower quartile of the load data in a historical power load data segment; Represents preset hyperparameters.

In an exemplary embodiment of the present invention, determining a predicted value of a power load at a future time by a historical load data segment includes:

When predicting the power load at a future moment, all the historical load data in the historical load data segment are input into the pre-trained ARIMA model to obtain the predicted value of the power load at a future moment by the historical load data segment, and then obtain the predicted value corresponding to each historical load data segment.

According to a second aspect of an embodiment of the present invention, there is provided an artificial intelligence-based power load forecasting system, which includes a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the steps of the first aspect of the embodiment of the present invention.

The present invention has the following effects:

The present invention corrects the initial noise level of the historical power load data segment through the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid data segment, thereby ensuring the accuracy of the possibility of the existence of noise data in the determined historical power load data segment; and uses the final noise level of the historical load data segment to weight the predicted value of the power load at a determined future moment, thereby considering the possibility of the existence of noise data in each historical load data segment, adjusting the proportion of the corresponding predicted value in the final predicted value, and ensuring the accuracy of the determined final predicted value.

BRIEF DESCRIPTION OF THE DRAWINGS

By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

FIG1 is a schematic diagram of a flow chart of steps of an artificial intelligence-based power load forecasting method according to an embodiment of the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

Referring to FIG1 , a method for predicting power load based on artificial intelligence includes steps S1-S4, which are as follows:

S1: Acquire multiple historical power load data segments and corresponding historical power grid frequency data segments.

Specifically, smart meters can be used to collect load data at various times, and frequency meters can be used to collect frequency data at various times in the power grid. When predicting the power load at a certain moment, the time before that moment can be used as the collection end time, and the time thirty minutes back from the predicted moment can be used as the collection start time. The load data and frequency data in the power grid can be continuously collected at a collection frequency of 1 time/second, thereby obtaining historical load data and historical frequency data.

Furthermore, multiple time periods can be selected from the collected historical load data, such as data segments corresponding to 6 time periods, to obtain multiple historical power load data segments, and then data segments corresponding to the time periods can be selected from the historical frequency data to obtain corresponding historical power grid frequency data segments, and the last moment of the multiple selected time periods is the previous moment of the moment when the power load needs to be predicted. This embodiment does not specifically limit the number of selected time periods.

Optionally, the length of the selected historical power load data segment and the grid frequency data segment can be 100, 150, 200, 250, 300 and 350. Of course, the appropriate length can also be selected according to the specific situation. This embodiment does not specifically limit the length of the historical power load data segment and the historical grid frequency data segment.

S2: Fit the load data in the historical power load data segment, obtain the extreme points of the fitting curve of the historical power load data segment, and calculate the initial noise level of the historical power load data segment. The initial noise level is positively correlated with the discreteness of the load data in the historical power load data segment and the number of extreme points, and negatively correlated with the average interval time of the extreme points.

It should be noted that noise data may affect the accuracy of power load prediction. Therefore, it is necessary to measure the presence of noise data in each selected historical power load data segment, so that the effectiveness of the prediction of power load at future times can be determined differently according to the situation of noise data in each historical power load data segment.

It should be further explained that the more noise data there is in a historical power load data segment, the more drastic and frequent the overall change of all load data in the historical power load data segment will be, and the more discrete the corresponding numerical distribution will be.

In an exemplary embodiment of the present invention, the initial noise level refers to an indicator used to measure the possibility of noise data existing in each historical power load data segment. Specifically, if the initial noise level of a historical power load data segment is higher, the possibility of noise data existing in the historical power load data segment is greater, and conversely, if the initial noise level of a historical power load data segment is lower, the possibility of noise data existing in the historical power load data segment is smaller.

Optionally, the load data in the historical power load data segments are fitted to determine the extreme points in the fitting curve of the historical power load data segments. The change of each load data in each historical power load data segment can be judged according to the number of extreme points. If the number of extreme points in the fitting curve corresponding to a historical power load data segment is large, it means that the load data in the historical power load data segment changes more frequently, and when the time interval corresponding to two extreme points is shorter, it can be said that the corresponding load data changes more drastically. Therefore, the frequency and severity of the change of the load data in the corresponding historical power load data segment can be evaluated according to the number of extreme points in the fitting curve corresponding to each historical power load data segment and the time interval between the corresponding extreme points.

Optionally, the degree of dispersion of the load data in each historical power load data segment can be determined using quartiles. Specifically, the load data in a historical power load data segment can be arranged in ascending order, and then the difference between the corresponding upper quartile and the lower quartile is calculated to measure the dispersion of the middle 50% of the load data arranged in ascending order. If the calculated difference is larger, the difference in the values of the load data in the historical power load data segment is larger, and the degree of dispersion of the load data in the historical power load data segment is higher, the possibility of noise data existing in the corresponding historical power load data segment is greater. Of course, the degree of dispersion of the load data in each historical power load data segment can also be determined in an appropriate manner according to the specific situation. This embodiment does not specifically limit the method for determining the degree of dispersion.

Specifically, the discrete degree of load data in the historical power load data segment satisfies the following relationship:

;

In the formula, Indicates the future moment When predicting the power load, The discrete degree of load data in a historical power load data segment; and Respectively for the future moment When predicting the power load, The upper quartile and lower quartile of the load data in a historical power load data segment; represents a preset hyperparameter. In this embodiment, =0.01. It should be noted that setting the hyperparameter can prevent the situation where the upper quartile and the lower quartile of the load data in the historical power load data segment have the same value, thereby improving the stability of the calculation formula.

Furthermore, after determining the extreme points in the fitting curve of each historical load data segment and the discrete degree of the load data in each historical power load data segment, the initial noise degree of each historical load data segment can be calculated. Specifically, the initial noise degree of the historical load data segment satisfies the following relationship:

;

In the formula, Indicates the future moment When predicting the power load, The initial noise level of a historical power load data segment; Indicates the future moment When predicting the power load, The number of extreme value points in the fitting curve of a historical power load data segment. It should be noted that the larger the value, the more frequent the changes in the load data in the historical power load data segment; Represents the first intervals; and Respectively represent the first The acquisition time of the extreme value points on the right side and the acquisition time of the extreme value points on the left side corresponding to the interval; Indicates the future moment When predicting the power load, The discrete degree of load data in a historical power load data segment.

in, Reflects the future moment When predicting the power load, In the fitting curve of the historical power load data segment, the first The time interval between the two extreme value points corresponding to the interval, the smaller the value, the more drastic the change of the load data in the historical power load data segment; It reflects the average interval time between all extreme points. The smaller the value is, the faster the load data in the historical power load data segment changes, and the greater the possibility of noise data existing in the corresponding historical power load data segment, and the higher the initial noise level of the historical power load data segment.

S3: Calculate the final noise level of the historical load data segment. The final noise level is negatively correlated with the initial noise level of the corresponding historical power grid frequency data segment, and the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment.

It should be noted that since the initial noise level of the historical load data segment is obtained based on the analysis of its own load data, there are certain limitations in judging the possibility of the existence of noise data. For a stably operating power grid system, the supply and demand of electricity is relatively balanced to a certain extent. When the power load increases, the power system needs to provide more electricity to meet the power supply demand, which may lead to a decrease in the frequency of the power system. Therefore, the initial noise level of the historical load data segment can be corrected in combination with the correlation between load data and frequency data, so as to more accurately evaluate the possibility of the existence of noise data in each historical load data segment.

In an exemplary embodiment of the present invention, the final noise level refers to an indicator that can accurately measure the possibility of noise data existing in each historical load data segment by combining the correlation between load data and frequency data, and is used to characterize the possibility of noise data existing in the corresponding historical load data segment. Specifically, if the final noise level of a historical load data segment is high, the possibility of noise data existing in the historical load data segment is high. On the contrary, if the final noise level of a historical load data segment is low, the possibility of noise data existing in the historical load data segment is low, so that the possibility of noise existing in each historical load data segment can be accurately measured according to the final noise level of each historical load data segment.

Optionally, the method for determining the initial noise level of each historical power grid frequency data segment is the same as the method for determining the initial noise level of a historical load data segment, which is as follows: first, fit the frequency data in any historical power grid frequency data segment, and then determine the extreme points of the fitting curve of the historical power grid frequency data segment, and use the calculation formula for calculating the initial noise level of the historical load data segment to adjust the corresponding parameters and calculate the initial noise level of the historical power grid frequency data segment. Therefore, when the initial noise level of the historical load data segment is determined, the initial noise level of the corresponding historical power grid frequency data segment can be determined.

In an exemplary embodiment of the present invention, the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment can be determined by the following steps:

Calculate the Pearson correlation coefficient between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment, and use the Pearson correlation coefficient as the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment.

Specifically, since there is a certain negative correlation between the load data and the frequency data collected in the same time period, and the Pearson correlation coefficient can well measure the monotonic relationship between the data in each data set, the Pearson correlation coefficient between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment can be calculated to measure the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment. Calculating the Pearson correlation coefficient between the data in each data set is a prior art and will not be described in detail here.

Further, after determining the initial noise level of the historical power grid frequency data segment corresponding to each historical power load data segment, and the correlation between the load data in each historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment, the final noise level of the historical power load data segment can be calculated. Specifically, the final noise level of the historical power load data segment satisfies the following relationship:

;

In the formula, Indicates the future moment When predicting the power load, The final noise level of each historical power load data segment; Indicates the future moment When predicting the power load, The initial noise level of a historical power load data segment; Indicates the future moment When predicting the power load, The initial noise level of the historical power grid frequency data segment corresponding to the historical power load data segment; Indicates the future moment When predicting the power load, The correlation between the load data in a historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment, The value range is [-1, 0].

in, Reflects the The difference between the initial noise level of a historical power load data segment and the initial noise level of the corresponding historical power grid frequency data segment. The larger the value, the lower the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment, and the greater the possibility that noise data exists in the historical power load data segment. Reflects the The sum of the Pearson correlation coefficient between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment and 1. Since the load data and the frequency data are negatively correlated, the larger the value, the more significant the correlation is. The worse the negative correlation between the load data in a historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment, the greater the possibility that noise data exists in the historical power load data segment.

S4: Determine the predicted value of the power load at the future time for the historical load data segment, and use the final noise level of the historical load data to weight the predicted value to obtain the weighted value of the predicted value corresponding to the historical load data segment, and then obtain the weighted value of the predicted value corresponding to each historical load data segment, and use the sum of each weighted value as the final predicted value of the power load at the future time.

In an exemplary embodiment of the present invention, the following steps may be used to determine the predicted value of the power load at a future time using the historical load data segment:

When predicting the power load at a future moment, all the historical load data in the historical load data segment are input into the pre-trained ARIMA model to obtain the predicted value of the power load at a future moment by the historical load data segment, and then obtain the predicted value corresponding to each historical load data segment.

It should be noted that the ARIMA model (Autoregressive Integrated Moving Average model) is a model that uses the number of autoregressive terms, the number of moving average terms, and the number of differences made to make the processed data segment a stationary series to predict the corresponding situation at future times. It can predict the power load at future times based on each historical load data segment.

Specifically, a historical load data segment arranged in the order of collection time can be arbitrarily intercepted from all load data within thirty minutes before the future moment of collection, and corresponding labels can be set. Then, the historical load data segment and the corresponding label can be used as a sample set, and the sample set can be divided into a training set and a test set according to a certain ratio, such as 8:2. The training set is used to train the model, and the trained model is evaluated through the test set to complete the parameter tuning of the model. When the minimization loss function of the model tends to converge, the training of the model is completed.

Furthermore, when the model training is completed, any historical load data segment can be input into the trained model. The model can analyze the historical load data segment based on the learning experience, and output the predicted value of the power load at the future time for the historical load data segment, thereby obtaining the predicted value corresponding to each historical load data segment.

Furthermore, the final noise level of any historical load data segment is used to weight the predicted value determined by the historical load data segment. When the final noise level of the historical load data segment is large, the proportion of the predicted value corresponding to the historical load data segment in the final predicted value can be reduced, thereby ensuring the accuracy of the final predicted value of the power load at a determined future time.

Specifically, the final predicted value of the power load at the future moment satisfies the following relationship:

;

In the formula, Indicates future time The final forecast value of the power load; Indicates the number of historical power load data segments; Indicates The historical power load data segment is used for the future time The predicted value of the power load; Indicates the future moment When predicting the power load, The final noise level of each historical power load data segment; Indicated by natural constant An exponential function with base .

in, Reflects the importance of determining the power load at a future time. The weight of the predicted value corresponding to each historical power load data segment; the larger the value, the greater the proportion of the predicted value corresponding to the historical power load data segment in the final predicted value, so that the predicted value corresponding to each historical power load data segment can be determined differently, and the proportion in the final predicted value can be guaranteed to ensure the accuracy of the predicted power load at the future moment.

The present invention also provides an artificial intelligence-based power load forecasting system, the system includes a memory and a processor, and a computer program is stored in the memory, the computer program integrates the function of an artificial intelligence-based power load forecasting method, when the computer program is executed, an artificial intelligence-based power load forecasting method can reduce the impact of noise data on the forecasting results, thereby improving the accuracy of power load forecasting.

In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims (8)

1. An artificial intelligence based power load prediction method, comprising:

acquiring a plurality of historical power load data segments and corresponding historical power grid frequency data segments;

Fitting the load data in the historical power load data segment, obtaining extreme points of a fitting curve of the historical power load data segment, and calculating initial noise degree of the historical power load data segment, wherein the initial noise degree is positively correlated with the discrete degree of the load data in the historical power load data segment and the number of the extreme points, and is negatively correlated with the average interval time of the extreme points;

Calculating the final noise degree of the historical load data segment, wherein the final noise degree is inversely related to the initial noise degree of the corresponding historical power grid frequency data segment and the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment;

and determining a predicted value of the historical load data segment on the electric load quantity at the future time, weighting the predicted value by utilizing the final noise degree of the historical load data to obtain a weighted value of the predicted value corresponding to the historical load data segment, further obtaining a weighted value of the predicted value corresponding to each historical load data segment, and taking the sum of the weighted values as the final predicted value of the electric load quantity at the future time.

2. An artificial intelligence based power load prediction method according to claim 1, wherein the final predicted value of the power load amount at the future time satisfies the following relation:

；

In the method, in the process of the invention, Indicating future time of dayA final predicted value of the electric power load amount of (2); Representing a number of historical power load data segments; Represent the first Historical power load data segment versus future timeA predicted value of the electric power load amount of (a); Indicating at a future time When predicting the power load of (a) a first timeFinal noise level of each historical power load data segment; Expressed in natural constant An exponential function of the base.

3. An artificial intelligence based power load prediction method according to claim 2, wherein the final noise level of the historical power load data segment satisfies the following relation:

；

In the method, in the process of the invention, Indicating at a future timeWhen predicting the power load of (a) a first timeFinal noise level of each historical power load data segment; Indicating at a future time When predicting the power load of (a) a first timeInitial noise levels of the historical power load data segments; Indicating at a future time When predicting the power load of (a) a first timeInitial noise degrees of historical grid frequency data segments corresponding to the historical power load data segments; Indicating at a future time When predicting the power load of (a) a first timeCorrelation between load data in each historical power load data segment and frequency data in the corresponding historical grid frequency data segment.

4. A method of artificial intelligence based power load prediction according to claim 3, wherein the method of obtaining a correlation between load data in the historical power load data segment and frequency data in the corresponding historical grid frequency data segment comprises:

And calculating a pearson correlation coefficient between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment, and taking the pearson correlation coefficient as the correlation between the load data in the historical power load data segment and the frequency data in the corresponding historical power grid frequency data segment.

5. An artificial intelligence based power load prediction method according to claim 3 wherein the initial noise level of the historical power load data segment satisfies the following relationship:

；

In the method, in the process of the invention, Indicating at a future timeWhen predicting the power load of (a) a first timeInitial noise levels of the historical power load data segments; Indicating at a future time When predicting the power load of (a) a first timeThe number of extreme points in the fitting curve of the historical power load data segments; representing the first between adjacent extreme points A plurality of intervals; And (3) with Respectively represent the first extreme points between adjacent extreme pointsThe acquisition time of the right extreme point and the acquisition time of the left extreme point corresponding to the intervals; Indicating at a future time When predicting the power load of (a) a first timeThe degree of dispersion of the load data in the historical power load data segment.

6. The artificial intelligence based power load prediction method of claim 5, wherein the degree of dispersion of load data in the historical power load data segment satisfies the following relationship:

；

In the method, in the process of the invention, Indicating at a future timeWhen predicting the power load of (a) a first timeThe degree of dispersion of the load data in the historical power load data segments; And (3) with Respectively indicate to future timeWhen predicting the power load of (a) a first timeThe upper quartile and the lower quartile of the load data in the historical power load data segments; indicating a preset super parameter.

7. An artificial intelligence based power load prediction method according to claim 1, wherein said determining a predicted value of said historical load data segment for an amount of power load at a future time comprises:

When the electric power load quantity at the future moment is predicted, all the historical load data in the historical load data segments are input into a pre-trained ARIMA model to obtain predicted values of the historical load data segments on the electric power load quantity at the future moment, and further the predicted values corresponding to each historical load data segment are obtained.

8. An artificial intelligence based power load prediction system comprising a memory having a computer program stored thereon and a processor executing the computer program to perform the steps of the method of any of claims 1-7.