Short-term power load prediction method and system
Through the improved support vector machine model and particle swarm optimization algorithm, the problem of low power load prediction accuracy is solved, and higher prediction accuracy and model diversity are achieved.
Patent Information
- Application Number
- CN202311614429.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-29
- Publication Date
- 2025-05-30
AI Technical Summary
The existing power load prediction methods have low accuracy, especially artificial intelligence-based methods require a lot of data training and are difficult to interpret the prediction results. Support vector machine parameters are prone to errors, and particle swarm algorithms are prone to fall into local optimality.
The improved support vector machine (SVM) model is adopted to determine the penalty factor C and kernel parameter δ of the SVM model through the improved particle swarm optimization (PSO) algorithm, and the improved PSO algorithm avoids local optimization through variation conditions and inertial weight coefficient optimization, and improves the prediction accuracy of the model.
Through the improved SVM model and PSO algorithm, the accuracy of power load prediction is improved, local optimal problems are avoided, and the diversity and prediction capabilities of the model are enhanced.
Smart Images
Figure CN120073648A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric load forecasting, and particularly to a short-term electric load forecasting method and system. Background Art
[0002] Under the background of the "dual carbon" strategy, the importance of power system load forecasting has become increasingly prominent. Short-term electric load forecasting is an estimation of the future power demand within a certain period of time. Based on the load forecasting results, the power generation plan for the future period is reasonably arranged, which is the basis for the stable and economic operation of the power system. Therefore, accurately forecasting the electric load is of great significance for improving the utilization rate of power generation equipment and reducing the operation cost.
[0003] Currently, the existing forecasting methods mainly include: time series analysis, artificial neural network, statistical regression analysis method, support vector machine, wavelet analysis method, etc. Forecasting methods based on artificial intelligence such as neural network require a large amount of data for training and it is difficult to explain the reasons for the forecasting results, which also limits their popularization in practical applications. Support vector machine (SVM) is a machine learning method based on statistical learning theory that can achieve the VC dimension theory and the principle of minimum structural risk under the condition of limited samples. It can use less data to train and predict the model, but its parameters are usually determined mainly by experience, so it is easy to cause errors. Using the traditional particle swarm optimization (PSO) algorithm to optimize its parameters, although it overcomes the disadvantage of determining parameters by experience, due to the inherent characteristics of the particle swarm algorithm, it is easy to fall into the local optimum situation, affecting the accuracy of the SVM model through parameters, and thus affecting the accuracy of electric load forecasting. Summary of the Invention
[0004] The purpose of the present invention is to provide a short-term electric load forecasting method and system to solve the problem of low accuracy of electric load forecasting.
[0005] To solve the above technical problems, the present invention provides a short-term electric load forecasting method, including the following steps,
[0006] 1) Obtain the data required for electric load forecasting and perform corresponding preprocessing. The data required for electric load forecasting includes weather data;
[0007] 2) Input the preprocessed data into the trained power load forecasting model to obtain the power load forecasting result; among them, the power load forecasting model is an improved SVM model. The improvement of the improved SVM model lies in that the penalty factor C and kernel parameter δ of the SVM model are determined by an improved PSO algorithm. The improvement of the improved PSO algorithm includes judging whether the mutation condition is satisfied after updating the particle position, velocity, and optimal value. If the mutation condition is satisfied, randomly expand the number of the current particle swarm, and use the randomized particle swarm for the next iteration calculation. If the mutation condition is not satisfied, directly perform the next iteration calculation on the current particle swarm without any processing.
[0008] The beneficial effects of the above technical solution are as follows: The power load forecasting model selects an improved SVM model. Selecting the SVM model can perform forecasting under limited data. By determining the penalty factor C and kernel parameter δ of the SVM model through the improved PSO algorithm, the disadvantage of determining the SVM model parameters using traditional experience can be overcome. Using the mutation condition to improve the traditional PSO algorithm can prevent the algorithm from falling into the local optimal situation, enabling the particles to jump out of the previously searched local optimal position, ensuring the diversity of the population, and better determining the penalty factor C and kernel parameter δ of the SVM model.
[0009] Further, the mutation condition specifically is to judge whether the random number probability p m is greater than B + A*rand(), where B is a constant, A is a random function coefficient, and rand() represents a random function.
[0010] The beneficial effects of the above technical solution are as follows: The mutation condition of judging whether the random number probability p m is greater than B + A*rand() can be used to judge whether to randomly expand the number of the particle swarm to prevent the algorithm from falling into the local optimal situation.
[0011] Further, the improvement of the improved PSO algorithm also lies in that the inertia weight coefficient ρ is determined by the following formula:
[0012]
[0013]
[0014] In the formula, ρ(k) is the inertia weight coefficient, rand is a random function, ρ max is the maximum value of the inertia weight coefficient, ρ min is the minimum value of the inertia weight coefficient, t is the current iteration number, k max is the maximum iteration number, is the parameter variable of the above formula.
[0015] The beneficial effects of the above technical solution are as follows: The inertial weight coefficient ρ is determined by a formula rather than a fixed value. After the algorithm is optimized, the inertial weight coefficient has a relatively large value and a slow change rate in the early stage of the search, greatly improving the probability of finding the global optimum. In the later stage of the algorithm, the value is relatively small but the change rate is fast, enhancing the optimization ability of the algorithm in the later stage and improving the prediction accuracy of the model.
[0016] Furthermore, the improvement of the improved PSO algorithm also lies in that the velocity update formula is determined by the following formula:
[0017]
[0018] In the formula: is the new optimized velocity, ρ is the inertial weight coefficient, c 1 is the cognitive information learning factor, c 2 is the social information learning factor, rand 1 、rand 2 are random functions, is the d-th dimension component of the velocity vector of particle i in the (k - 1)-th iteration, p id is the best position experienced by particle i individually, p gd is the best position experienced by the population, is the d-th dimension component of the position vector of i in the (k - 1)-th iteration, λ is the compression factor, and
[0019]
[0020] In the formula: c is the sum of c 1 and c 2 .
[0021] The beneficial effects of the above technical solution are as follows: Optimize the velocity update formula of the PSO algorithm, introduce a compression factor to limit the parameters of the learning factor, and control the cognitive information learning factor c 1 and the social information learning factor c 2 within a reasonable range, enhancing the ability of the particle to search for the global optimal solution.
[0022] Furthermore, the weather data includes the highest temperature, the lowest temperature, the average temperature, the wind force level, the humidity, and the meteorological conditions.
[0023] The beneficial effects of the above technical solution are as follows: Select multiple different data as weather data, enriching the weather data and making the prediction results more accurate.
[0024] Furthermore, the preprocessing methods for the highest temperature, the lowest temperature, and the average temperature are all:
[0025]
[0026] In the formula, x is the data before preprocessing, and x min is the minimum value of x, and x max is the maximum value of x, and y is the data after preprocessing.
[0027] The beneficial effect of the above technical solution is: performing a preprocessing operation on the temperature to remove the influence of the dimension on the prediction result.
[0028] Further, the method for preprocessing the meteorological situation is: classifying the weather conditions into six situations: sunny, cloudy, overcast, light rain, moderate rain, and heavy rain, defining the corresponding values for each situation, and the values are in the range of [0, 1]; performing the following inverse normalization processing on the obtained meteorological situation to achieve preprocessing:
[0029] x = x min +(x max - x min )y.
[0030] In the formula, x is the value corresponding to the meteorological situation, and x min is the minimum value among all the values, and x max is the maximum value among all the values, and y is the data of the weather situation after preprocessing.
[0031] The beneficial effect of the above technical solution is: performing a preprocessing operation on the meteorological situation to remove the influence of the dimension on the prediction result.
[0032] Further, the data required for the electric load prediction further includes the daily type value, and the daily type value is specifically a [0, 1] numerical value selected based on the periodicity of the load to characterize the response of the load to different daily types of weekdays and weekends.
[0033] The beneficial effect of the above technical solution is: selecting the daily type value for the electric load prediction, considering the influence of weekdays and weekends on the electric load, and making the prediction result more accurate.
[0034] Further, the data required for the electric load prediction further includes the power consumption data per unit time.
[0035] The beneficial effect of the above technical solution is: selecting the power consumption data per unit time for the electric load prediction and making the prediction result more accurate.
[0036] To solve the above technical problems, the present invention also provides a short-term electric load prediction system, which includes a memory and a processor, and a computer program stored on the memory and running on the processor. The processor is used to execute the computer program instructions stored in the memory to implement the short-term electric load prediction method.
[0037] The beneficial effects of the above technical solution are as follows: In the short-term power load forecasting system, the improved SVM model is selected as the power load forecasting model. The SVM model can be used for forecasting under limited data. By determining the penalty factor C and kernel parameter δ of the SVM model through the improved PSO algorithm, the disadvantage of determining the parameters of the SVM model using traditional experience can be overcome. The traditional PSO algorithm is improved using mutation conditions to prevent the algorithm from falling into the local optimum, enabling the particles to jump out of the previously searched local optimum position, ensuring the diversity of the population, and better determining the penalty factor C and kernel parameter δ of the SVM model. Description of the Drawings
[0038] Figure 1 It is a flowchart of the optimization algorithm of a short-term power load forecasting optimization method according to an embodiment of the method of the present invention;
[0039] Figure 2 It is a functional diagram of the modules of a short-term power load forecasting system according to an embodiment of the method of the present invention;
[0040] Figure 3 It is a comparison diagram of the predicted value and the actual value of the optimization model in a short-term power load forecasting optimization method according to an embodiment of the method of the present invention;
[0041] Figure 4 It is a comparison diagram of the predicted values and the actual values of three methods in a short-term power load forecasting optimization method according to an embodiment of the method of the present invention. Detailed Embodiments
[0042] A short-term electric load forecasting method includes the following steps. First, obtain the data required for electric load forecasting and perform corresponding preprocessing. The data required for electric load forecasting includes weather data. Obtain the required data and perform corresponding processing on it. By preprocessing, eliminate the influence of different dimensions on forecasting and improve the forecasting accuracy. Second, input the preprocessed data into the trained electric load forecasting model to obtain the electric load forecasting result. Among them, the electric load forecasting model is an improved SVM model. The improvement of the improved SVM model lies in that the penalty factor C and kernel parameter δ of the SVM model are determined by an improved PSO algorithm. And the improvement of the improved PSO algorithm includes, after updating the particle position, velocity and optimal value, determining whether the mutation condition is satisfied. If the mutation condition is satisfied, randomly expand the number of the current particle swarm, and use the randomized particle swarm for the next iterative calculation. If the mutation condition is not satisfied, directly perform the next iterative calculation on the current particle swarm without any treatment. Selecting the SVM model can perform forecasting on data in the case of limited data. Determining the penalty factor C and kernel parameter δ of the SVM model by the improved PSO algorithm can overcome the defect of determining parameters by traditional experience. Using the mutation condition to improve the traditional PSO algorithm can prevent the algorithm from falling into the local optimal situation, enabling the particles to jump out of the previously searched local optimal position and ensuring the diversity of the population.
[0043] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0044] Method embodiment:
[0045] A short-term electric load forecasting method includes the following steps:
[0046] (1) Collect the data required for forecasting and perform corresponding preprocessing on the data.
[0047] Data collection is a key step in electric load forecasting. Not all of the collected data is suitable for sample forecasting and needs to be preprocessed. The size of the data volume should be appropriate. At the same time, situations such as data loss and data anomalies caused by power equipment failures or other human factors should be considered. Problematic data cannot be directly used for electric load forecasting, otherwise it will lead to large deviation forecasting. Therefore, it is necessary to fill or correct the problematic data.
[0048] For the processing of missing data, if the time interval between the front and back is not large, the linear interpolation method is adopted. If the interval is large and the linear interpolation method has poor effect, the load data of adjacent days is used for substitution. Compare the load at a certain moment with the adjacent load values before and after. If the difference is greater than a certain threshold, the horizontal processing method is adopted. Compare the load at a certain moment with the loads at the same moment of the previous day and the day before yesterday respectively. If the difference is greater than a certain threshold, the vertical processing method is adopted.
[0049] To eliminate the influence of different dimensions on the data, the input sample data is normalized. Each input sample contains 8 characteristic indicators: day type value, maximum temperature, minimum temperature, average temperature, wind force level, humidity, meteorological conditions, and power consumption per unit time, and the output is the load value of the prediction point.
[0050] ① Normalize the load data. The load data is logarithmically processed as follows:
[0051] y' ii = lg(y ij )
[0052] where y ij is the original load data and y' ij is the normalized load.
[0053] ② Normalize the maximum temperature, minimum temperature, and average temperature data:
[0054]
[0055] In the formula: x is the data before preprocessing, x min is the minimum value of x, x max is the maximum value of x, and y is the data after preprocessing.
[0056] ③ Division and normalization of day types:
[0057] Generally, the load patterns on weekdays and weekends are similar but also have obvious differences. During weekdays, the load is usually in a stable state, while during weekends, the proportion of industrial load drops significantly. Therefore, the load on weekends is at a lower level compared to weekdays. Based on the periodicity of the load, values in the range [0, 1] are selected to represent the response of the load to different day types on weekdays and weekends. In the present invention, Monday to Thursday are taken as 0.9, Friday is taken as 0.6, and Saturday to Sunday are taken as 0.4.
[0058] ④ Meteorological condition factors:
[0059] When processing rainfall load data, the weather conditions are divided into clear, cloudy, overcast, light rain, moderate rain, and heavy rain, and the corresponding quantization sets are {0, 0.25, 0.5, 0.65, 0.75, 1}. In this way, each sample data is transformed into the value range of [0, 1]. Finally, the inverse normalization process is performed on the output result as follows:
[0060] x = x min + (x max - x min )y
[0061] In the formula, x is the value corresponding to the meteorological condition, xmin is the minimum value among all the values, x max is the maximum value among all the values, and y is the preprocessed weather condition data.
[0062] (2) Construct a PSO-SVM model.
[0063] The specific steps to construct the PSO-SVM model are as follows, and the process is as Figure 1 shown.
[0064] Step 1: Initialize the population. Initialize the velocity and position of each particle, and set the maximum number of iterations k max , mutation threshold ε, maximum value P of the inertia weight coefficient max , minimum value P of the inertia weight coefficient min , mutation probability P m and learning factor c 1 , c 2 and other parameter values.
[0065] Step 2: Calculate the fitness value f(x i ) of each particle in the population according to the fitness function.
[0066] Step 3: Solve the individual best fitness value. For each particle, compare the fitness value of the current position with its historical best position p b . If the current position is higher, update it to the historical best position; otherwise, do not change it.
[0067] Step 4: Solve the best fitness value of the population. Compare the fitness value of each particle's current position with the global optimal value p g . If the current fitness value is higher, update the current value to the global optimal value; otherwise, do not change it.
[0068] Step 5: Optimize the inertia weight coefficient ρ of the PSO algorithm. The inertia weight coefficient ρ is determined by the following formula:
[0069]
[0070]
[0071] In the formula, ρ(k) is the inertia weight coefficient, rand is the random function, ρ max is the maximum value of the inertia weight coefficient, ρ min is the minimum value of the inertia weight coefficient, t is the current iteration number, k max is the maximum number of iterations, is the parameter variable in the above formula.
[0072] Step 6: Improve the velocity update formula of the PSO algorithm. The velocity update formula is determined by the following formula:
[0073] The optimized new speed calculation strategy is as follows:
[0074]
[0075] In the formula: is the optimized new speed, ρ is the inertia weight coefficient, c 1 is the cognitive information learning factor, c 2 is the social information learning factor, rand 1 and rand 2 are random functions, is the d-th dimensional component of the flight speed vector of particle i at the (k - 1)-th iteration, p id is the best position experienced by particle i individually, p gd is the best position experienced by the population, is the d-th dimensional component of the position vector of i at the (k - 1)-th iteration, λ is the compression factor, and
[0076]
[0077] In the formula: c is the sum of c 1 and c 2
[0078] Update the positions and speeds of each particle to obtain a new population Z(t).
[0079] Step Seven: Mutation operation. Introduce a mutation strategy to optimize the PSO algorithm. Specifically, after each particle update during the iteration, use the random probability mutation strategy to re-initialize the particles, enabling the particles to jump out of the previously searched local optimal position. The random probability mutation strategy is specifically to judge whether the random number probability p m is greater than B + A * rand(), where B is a constant, A is the random function coefficient, rand() represents the random function, and B and A are set according to actual needs. Here, B is set to 0.9 and A is set to 0.2; if it is greater, the population is randomized, and if it is less than or equal, the population numerical size remains unchanged.
[0080] Step Eight: Determine whether the optimization condition is satisfied according to k max and the accuracy threshold. If the condition is met, end the optimization; otherwise, go to Step Two.
[0081] Step Nine: Output the optimal solution and determine the penalty factor C and the kernel parameter δ.
[0082] Step Ten: Establish the optimized combined prediction model PSO - SVM.
[0083] (3) Train the PSO-SVM model and use the trained model to predict the results, and evaluate the prediction results at the same time.
[0084] Use the SVM, traditional PSO-SVM, and optimized PSO-SVM models to train and use the trained models to predict the results, and compare them with the actual values. Specifically, as Figure 3 and Figure 4 shown.
[0085] To evaluate the prediction results, the root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE) are introduced to evaluate the prediction results of the model. The results are shown in Table 1, and the expressions are as follows:
[0086]
[0087]
[0088]
[0089] In the formula, N is the size of the sample data; y(i) and represent the actual value and the predicted value of the prediction model respectively; RMSE is the root mean square difference between the true value and the predicted value; MAPE is the average of the ratios between the errors of all prediction points and the actual value; MAE is the average of the absolute errors between all actual values and predicted values.
[0090] Table 1
[0091]
[0092]
[0093] The implementation of the short-term power load forecasting method requires the following modules to cooperate. The relationship between the modules is as Figure 2 shown;
[0094] Data acquisition and preprocessing module: This module is used to collect historical power load data and preprocess the data, such as removing outliers, filling missing values, etc.
[0095] Feature extraction module: This module is used to extract useful features from the original load data, such as seasonality, trend, periodicity, etc., and convert the original load data into feature vectors for use in model training and prediction.
[0096] Model training module: This module is used to train the model, that is, to optimize the parameters of the SVM model using the particle swarm optimization algorithm. In this module, the parameters of the SVM model, the parameters of the PSO algorithm, etc. need to be set.
[0097] Prediction Module: This module is used to predict future power loads using a trained model. In this module, the feature vector of the prediction period needs to be input, and then the PSO-SVM model is used for prediction.
[0098] Result Evaluation Module: This module is used to evaluate the accuracy and precision of the prediction results, that is, to calculate indicators such as root mean square error, mean absolute error, and mean absolute percentage error. The evaluation results can be used to optimize the model and improve the prediction accuracy.
[0099] Visualization Display Module: This module is used to visually display the prediction results so that users can better understand and analyze the prediction results, that is, to display the comparison between the prediction results and the actual data using charts, as well as the trend changes of the prediction results, etc.
[0100] Parameter Tuning Module: This module is used to adjust the parameters of the model to improve the prediction accuracy and robustness, such as adjusting parameters such as the number of iterations and population size of the PSO algorithm, as well as the kernel function type and penalty coefficient of the SVM model.
[0101] System Management Module: This module is used to manage and maintain the entire power load prediction system, including user management, data backup, system upgrade, etc. In addition, the system performance can also be monitored and optimized through this module to ensure the stability and reliability of the system.
[0102] Data Analysis and Mining Module: This module is used to analyze and mine historical load data to discover the laws and trends of load changes and provide references for optimizing the prediction model.
[0103] Real-time Monitoring and Warning Module: This module is used to monitor and warn the power load in real time. Once an abnormal situation or a situation beyond the prediction range occurs, the system will send a warning message in time to help users take corresponding measures to avoid losses and risks.
[0104] Multi-source Data Fusion Module: This module is used to fuse the data from multiple data sources to improve the accuracy and robustness of the prediction model. For example, the data from multiple data sources such as meteorological data, economic data, and population data can be fused to more comprehensively consider the factors affecting load changes.
[0105] Human-computer Interaction Interface Module: This module is used to interact with users and provide a friendly graphical interface for users to perform operations such as data query, prediction operation, and result display.
[0106] Risk Management Module: This module is used to evaluate the risks and errors of load prediction and propose corresponding risk management strategies to reduce the impact of prediction errors on the power system.
[0107] Decision Support Module: This module is used to provide decision support based on the load forecasting results, such as power dispatching, energy planning, market operation, etc., providing reference and support for the operation and management of the power system.
[0108] Data Security Module: This module is used to ensure the security and privacy of data, including measures such as data encryption, backup, and restoration, to ensure the integrity and security of data.
[0109] System Integration Module: This module is used to integrate each module to ensure the collaborative work and interoperability of the system, so as to improve the overall performance and stability of the system.
[0110] System Optimization Module: This module is used to optimize and adjust the system to improve the performance and efficiency of the system, such as optimizing algorithms and adjusting parameters.
[0111] System Monitoring Module: This module is used to monitor and manage the system in real time, including system performance, operating status, abnormal conditions, etc., to ensure the stability and reliability of the system.
[0112] Embodiment of the Short-Term Power Load Forecasting System:
[0113] The short-term power load forecasting system includes a memory and a processor, as well as a computer program stored on the memory and running on the processor. The processor is used to execute the computer program instructions stored in the memory to implement the short-term power load forecasting method. The specific process has been described in detail in the embodiment of the short-term power load forecasting method and will not be repeated here. Among them, the processor can select processing devices such as a microprocessor MCU and a field programmable gate array FPGA, and the memory can select storage devices such as a mobile hard disk, a read-only memory (ROM), and a random access memory (RAM).
[0114] The specific implementation manners are given above, but the present invention is not limited to the described implementation manners. The basic idea of the present invention lies in the above basic solution. For those of ordinary skill in the art, according to the teachings of the present invention, it does not require creative labor to design various deformed models, formulas, and parameters. Changes, modifications, substitutions, and variations made to the implementation manners without departing from the principle and spirit of the present invention still fall within the protection scope of the present invention.
Claims
1. A short-term electric load forecasting method, characterized in that, it includes the following steps, 1) Obtain the data required for electric load forecasting and perform corresponding preprocessing. The data required for electric load forecasting includes weather data; 2) Input the preprocessed data into the trained electric load forecasting model to obtain the electric load forecasting result; among them, the electric load forecasting model is an improved SVM model. The improvement of the improved SVM model lies in that the penalty factor C and kernel parameter δ of the SVM model are determined by an improved PSO algorithm, and the improvement of the improved PSO algorithm includes judging whether the mutation condition is satisfied after updating the particle position, velocity and optimal value. If the mutation condition is satisfied, randomly expand the number of the current particle swarm, and use the randomized particle swarm for the next iteration calculation. If the mutation condition is not satisfied, directly perform the next iteration calculation on the current particle swarm without any processing.
2. The short-term electric load forecasting method according to claim 1, characterized in that, The mutation condition specifically is to determine whether the random number probability p m is greater than B + A*rand(), where B is a constant, A is the coefficient of the random function, and rand() represents the random function.
3. The short-term electric load forecasting method according to claim 1, characterized in that, The improvement of the improved PSO algorithm also lies in that the inertia weight coefficient ρ is determined by the following formula: Where ρ(k) is the inertia weight coefficient, rand is the random function, ρ max is the maximum value of the inertia weight coefficient, ρ min is the minimum value of the inertia weight coefficient, t is the current iteration number, k max is the maximum iteration number, is the parameter variable of the above formula.
4. The short-term electric load forecasting method according to claim 1, characterized in that, The improvement of the improved PSO algorithm also lies in that the velocity update formula is determined by the following formula: Wherein: is the d-th dimensional component of the flight velocity vector of particle i at the k-th iteration, ρ is the inertia weight coefficient, c 1 is the cognitive information learning factor, c 2 is the social information learning factor, rand 1 、rand 2 is a random function, is the d-th dimensional component of the flight velocity vector of particle i at the (k - 1)-th iteration, p id is the best position experienced by particle i individually, p gd is the best position experienced by the population, is the d-th dimensional component of the position vector of i at the (k - 1)-th iteration, λ is the compression factor, and Where: c is c 1 and c 2 The sum of.
5. The short-term electric load forecasting method according to claim 1, characterized in that, The weather data includes the highest temperature, the lowest temperature, the average temperature, the wind force level, the humidity and the meteorological condition.
6. The short-term electric load forecasting method according to claim 5, characterized in that, The preprocessing methods for the highest temperature, the lowest temperature and the average temperature are all: Where x is the data before preprocessing, x min is the minimum value of x, and x max is the maximum value of x, and y is the data after preprocessing.
7. The short-term electric load forecasting method according to claim 5, characterized in that, The preprocessing method for the meteorological condition is: Divide the weather condition into six situations: sunny, cloudy, overcast, light rain, moderate rain and heavy rain, define the corresponding values for each situation, and the values are in the range of [0,1]; perform the following anti-normalization processing on the obtained meteorological condition to achieve preprocessing: x = x min +(x max -x min )y Where x is the value corresponding to the meteorological situation, and x min is the minimum value among all the values, and x max is the maximum value among all the values, and y is the preprocessed weather situation data.
8. The short-term electric load forecasting method according to claim 1, characterized in that, The data required for electric load forecasting also includes the daily type value. The daily type value is specifically based on the periodicity of the load, and a [0,1] numerical value is selected to represent the response of the load to different daily types of weekdays and weekends.
9. The short-term electric load forecasting method according to claim 1, characterized in that, The data required for electric load forecasting also includes the power consumption data per unit time.
10. A short-term electric load forecasting system, characterized in that, This system includes a memory and a processor, as well as a computer program stored on the memory and running on the processor. The processor is used to execute the computer program instructions stored in the memory to implement the short-term electric load forecasting method according to any one of claims 1 to 9.