Power grid model predictive error correction method and system based on machine learning

Abstract

The invention discloses a power grid model predictive error correction method and system based on machine learning, and the method comprises the steps: building a training set and a verification set based on an obtained historical load time sequence in a region and a corresponding influence factor sequence, and carrying out the training through a neural network model, and obtaining a load prediction power grid model; predicting the consumption of the power grid load in the region to obtain a load prediction sequence; whether the load prediction power grid model is suitable for the current power grid operation condition is judged based on the load trend characteristics and the load seasonal characteristics in the load prediction sequence, and if yes, the load prediction power grid model continues to be used for load prediction; and if the load prediction power grid model is not suitable for the current power grid operation condition, reconstructing a historical load time sequence, and retraining and optimizing the load prediction power grid model conforming to the current condition by taking the reconstructed historical load sequence as a verification set again. It can be ensured that the load prediction power grid model and the actual operation condition of the power grid are kept synchronous.

Description

Technical Field

[0001] The present invention relates to a predictive error correction method and system for a power grid model based on machine learning, which belongs to the technical field of power system automation. Background Art

[0002] As modern power grids become increasingly large and complex, their coverage spans vast areas, integrating various energy production, transmission and distribution facilities, forming a highly interconnected and dynamically changing system. This growth has not only improved the supply capacity of electric energy, but also significantly increased the uncertainty and potential risks in the operation of the power grid. The frequent occurrence of extreme weather events caused by climate change, the increase in equipment aging and failure rates, and the fluctuating loads caused by the large-scale grid connection of renewable energy have all brought unprecedented challenges to the safe and stable operation of the power grid.

[0003] In this context, the use of advanced machine learning, big data analysis, artificial intelligence and other technical means to conduct comprehensive and in-depth monitoring and analysis of the operating status of the power grid has become an important means of power processing. Existing power grid models are usually modeled using offline static methods, and their parameters are determined during the modeling stage and are only applicable to specific power grid operating states. As the scale of the power grid continues to expand, the operating state becomes more complex and dynamic, and the prediction accuracy of existing power grid models will significantly decrease when dealing with new operating states.

[0004] For example, when a certain operating state is added or removed in the power grid, such as a generator, load or line, the parameters of the power grid model need to be recalibrated to maintain the accuracy of the prediction. However, this process usually requires a lot of time and historical data, and cannot respond to rapid changes in the power grid in a timely manner. In addition, most existing power grid models can only predict potential problems or risks in power grid operation, and lack the function of generating effective response strategies. Summary of the invention

[0005] The purpose of the present invention is to provide a machine learning power grid model predictive error correction method and system, which optimizes and adjusts the load prediction power grid model to a state suitable for the current power grid operation through the predicted load.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A predictive error correction method for a power grid model based on machine learning, the steps of which include:

[0008] Obtain the historical load time series in the region and its corresponding influencing factor series; the influencing factors include external environmental factors and all load operation states in the region; the external environmental factors include temperature, precipitation, and seasonal factors;

[0009] Construct training and validation sets based on the historical load time series in the region and its corresponding influencing factor series;

[0010] Based on the training set and the validation set, a neural network model is used to train a load forecasting power grid model, and the load forecasting power grid model is used to forecast the power grid load in the region to obtain a load forecasting sequence;

[0011] Extract load trend characteristics and load seasonal characteristics in load forecasting sequence;

[0012] Based on the load trend characteristics and load seasonal characteristics in the load forecast sequence, determine whether the load forecasting power grid model is suitable for the current power grid operation status. If it is suitable, continue to use the load forecasting power grid model for load forecasting;

[0013] If the load forecasting power grid model is not suitable for the current power grid operation status, the historical load time series is reconstructed, and the reconstructed historical load series is used as a validation set to retrain and optimize the load forecasting power grid model to suit the current situation.

[0014] The characteristics of power grid load are trend and seasonal. The trend of power grid load is mainly reflected in its long-term growth trend with economic growth, social progress and population increase. With the development of economy, the demand for electricity in industrial production, commercial activities and residents' lives is increasing, which leads to the continuous increase of the total power grid load. The seasonal variation of power grid load is mainly caused by factors such as climate conditions, holiday arrangements and the periodicity of production activities. For example, in summer, due to the large-scale use of refrigeration equipment such as air conditioners caused by high temperature weather, the power grid load will increase significantly, forming the so-called "summer peak". In winter, although the use of heating equipment will also increase the load, it is usually not as obvious as the summer peak. Therefore, the factors affecting the load can be the operating status of all loads in the region, temperature, precipitation, seasonal factors, etc.

[0015] The operating status of all loads in the area refers to the operating status of all loads (i.e., power users and equipment) in the power system at a specific time point, which can be represented by a matrix.

[0016] The present invention is further achieved by the following technical solutions:

[0017] In the above-mentioned predictive error correction method of the power grid model based on machine learning, the verification set uses the mean square error as an evaluation index to adjust and optimize the load forecasting power grid model. The Nash coefficient NSE, Kling-Gupta efficiency coefficient KGE and root mean square error, RMSE, can also be used as model evaluation indicators to adjust and optimize the load forecasting power grid model.

[0018] In the aforementioned predictive error correction method for power grid models based on machine learning, the load forecasting sequence uses a hybrid autoregressive expert with one-dimensional causal convolution to extract the load trend representation and then uses a time domain contrast loss function to learn to obtain the final load trend characteristics;

[0019] The load forecast sequence uses a learnable Fourier transform in the frequency domain to obtain a frequency domain seasonal trend representation, then uses a frequency domain contrast loss function to learn to obtain a frequency domain seasonal trend representation, and converts the frequency domain seasonal trend representation back to the time domain to obtain the load seasonal characteristics.

[0020] The aforementioned predictive error correction method for the power grid model based on machine learning, the method for judging whether the load prediction power grid model is applicable to the current power grid operation status is:

[0021] Determine the time period T of the load forecast sequence in the annual cycle;

[0022] Compare the load trend characteristic data in the load forecast sequence with the upper limit value and the lower limit value in the load trend standard of the current load forecast power grid model in time period T;

[0023] Compare the load seasonality characteristics in the load forecast sequence with the upper limit value and the lower limit value in the load seasonality standard of the current load forecast power grid model in time period T;

[0024] If the corresponding load seasonal characteristic data in the load forecast sequence exceeding α% are all within the range of the upper and lower limits of the load seasonal standard of the current load forecast power grid model in time period T, and the corresponding load trend characteristic data in the load forecast sequence are within the range of the upper and lower limits of the load trend standard of the current load forecast power grid model in time period T, then the load forecast power grid model is judged to be applicable to the current power grid operation status; otherwise, the load forecast power grid model is judged to be not applicable to the current power grid operation status.

[0025] Because the load is seasonal and has different stages such as peak and trough in a day, when determining the time period T in which the load forecast sequence is located in an annual cycle, it can be expressed by the stage division of seasonal load trends and the stage division of daily load trends, such as seasonal peak period-daily peak period, seasonal trough period-daily peak period, seasonal trough period-daily peak period, seasonal trough period-daily trough period, etc.

[0026] Obtain the load seasonality characteristics and load trend characteristics of the historical load time series within the time period T, and use statistical analysis methods to calculate the load seasonality characteristics of the historical load time series within the time period T. The confidence interval or tolerance interval can be used to calculate the load seasonality standard of the current load forecasting power grid model in the time period T.

[0027] Similarly, the confidence interval or tolerance interval can be used to calculate the load trend standard of the current load forecasting power grid model in time period T.

[0028] Because changes in the operating conditions of all loads in the power system (i.e., power users and equipment) at a specific point in time can directly affect the trend and seasonality of the grid load in the area, when the load forecasting grid model is not applicable to the current grid operating conditions, it is necessary to immediately check whether the operating conditions of the internal loads in the grid have changed.

[0029] The aforementioned predictive error correction method for the power grid model based on machine learning, when the reason why the load forecasting power grid model is not applicable is that the operating status of all loads in the area has changed, then the time period in which the load forecasting sequence is located with a year as the cycle is confirmed, and the historical load time series of the corresponding time period is adjusted according to the changed load operating status, and the adjusted historical load time series is used as a verification set to optimize the load forecasting power grid model;

[0030] If external environmental factors are the reason why the load forecasting power grid model is not applicable, then confirm the time period of the load forecasting sequence in the annual cycle;

[0031] Filter out the historical load time series M in the same time period as the load forecast sequence, calculate the ratio a of the forecast trend correlation of the load forecast sequence and the trend correlation of the historical load time series M, and use the ratio a as the weight of the load trend characteristics of the historical load time series in the same time period;

[0032] Calculate the ratio b of the seasonal correlation of the load forecast sequence and the seasonal correlation of the historical load time series M, and use the ratio b as the weight of the seasonal trend characteristics of the historical load time series M;

[0033] Based on the ratio a and the ratio b, the corresponding load trend characteristics and load seasonal characteristics in the historical load time series M are connected using the total loss function to obtain the reconstructed historical load time series.

[0034] The aforementioned predictive error correction method for power grid model based on machine learning, wherein the reconstructed load in the reconstructed historical load time series is:

[0035] Q=aQ t +bQ s ;

[0036] Where Q represents the reconstructed load, a is the ratio of the predicted trend correlation of the load forecast sequence to the trend correlation of the historical load time series M, b is the predicted seasonal correlation of the load forecast sequence to the seasonal correlation of the historical load time series M, Q t To reconstruct the load trend characteristics of the historical load corresponding to load Q, Q sThe seasonality of the load corresponding to the historical load of the reconstructed load Q.

[0037] The above-mentioned predictive error correction method for power grid model based on machine learning, the total loss function is:

[0038]

[0039] Among them, t represents the time domain contrast loss function, ζ a represents the frequency domain amplitude contrast loss function, ζ p represents the frequency domain phase contrast loss function, and a is a parameter that balances the trade-off between trend and seasonal factors.

[0040] The present invention also includes another technical solution, a predictive error correction system for a power grid model based on machine learning, comprising:

[0041] A data acquisition module is used to acquire the historical load time series in the region and its corresponding influencing factor series; the influencing factors include external environmental factors and all load operation states in the region; the external environmental factors include temperature, precipitation, and seasonal factors;

[0042] The load forecasting module builds a training set and a validation set based on the historical load time series in the region and its corresponding influencing factor series. Based on the training set and the validation set, a neural network model is used to train a load forecasting power grid model. The load forecasting power grid model is used to predict the power grid load usage in the region to obtain a load forecasting sequence.

[0043] The power grid model judgment module is used to extract the load trend characteristics and load seasonal characteristics in the load forecast sequence, and judge whether the load forecast power grid model is suitable for the current power grid operation status based on the load trend characteristics and load seasonal characteristics in the load forecast sequence. If it is suitable, the load forecast power grid model is continued to be used for load forecasting;

[0044] The power grid model re-optimization module is used to reconstruct the historical load time series and retrain and optimize the load forecasting power grid model that conforms to the current situation based on the reconstructed historical load series as a validation set.

[0045] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: by extracting the load trend characteristics and load seasonal characteristics in the load forecasting sequence, the model can capture the changing rules of the power grid load in a more detailed manner, thereby more accurately predicting the future load. The load trend characteristics and load seasonal characteristics of the predicted load are used to evaluate whether the load forecasting power grid model is suitable for the current power grid operation status, and whether to retrain or optimize the load forecasting power grid model is determined based on the evaluation results, thereby ensuring that the load forecasting power grid model is always synchronized with the actual operation status of the power grid. This dynamic adjustment mechanism greatly improves the adaptability and robustness of the model. By continuously optimizing the prediction model, the risks caused by prediction errors can be reduced to ensure the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0047] Figure 1 It is a flow chart of the steps of the predictive error correction method of the power grid model based on machine learning of the present invention;

[0048] Figure 2 It is a structural schematic diagram of the power grid model predictive error correction system based on machine learning of the present invention. DETAILED DESCRIPTION

[0049] 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 only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Example 1

[0051] Predictive error correction methods for power grid models based on machine learning, such as Figure 1 As shown, the steps include:

[0052] S1. Obtain the historical load time series in the region and its corresponding influencing factor series; the influencing factors include external environmental factors and all load operating states in the region; the external environmental factors include temperature, precipitation, and seasonal factors.

[0053] S2. Construct a training set and a validation set based on the historical load time series in the region and its corresponding influencing factor series. Specifically, use a neural network model to train the load prediction power grid model based on the training set, and use the mean square error as an evaluation index based on the validation set to adjust and optimize the load prediction power grid model.

[0054] S3. Based on the training set and the validation set, a neural network model is used to train a load prediction power grid model, and the load prediction power grid model is used to predict the power grid load usage in the region to obtain a load prediction sequence.

[0055] S4, extracting the load trend characteristics and load seasonal characteristics in the load forecast sequence, the specific execution steps include: extracting the load trend representation based on the load forecast sequence using a hybrid autoregressive expert with one-dimensional causal convolution, and then learning using a time domain contrast loss function to obtain the final load trend characteristics;

[0056] Based on the load forecast sequence, a frequency domain seasonal trend representation is obtained by using a learnable Fourier transform in the frequency domain, and then the frequency domain contrast loss function is used to learn the frequency domain seasonal trend representation, and the frequency domain seasonal trend representation is converted back to the time domain to obtain the load seasonal characteristics.

[0057] S5. Based on the load trend characteristics and load seasonal characteristics in the load forecast sequence, determine whether the load forecasting power grid model is suitable for the current power grid operation status. If it is suitable, continue to use the load forecasting power grid model to perform load forecasting.

[0058] The specific execution steps for determining whether the load forecasting power grid model is applicable to the current power grid operation status include:

[0059] Determine the time period T of the load forecast sequence in the annual cycle;

[0060] Compare the load trend characteristic data in the load forecast sequence with the upper limit value and the lower limit value in the load trend standard of the current load forecast power grid model in time period T;

[0061] Compare the load seasonality characteristics in the load forecast sequence with the upper limit value and the lower limit value in the load seasonality standard of the current load forecast power grid model in time period T;

[0062] If the corresponding load seasonal characteristic data in the load forecast sequence exceeding α% are within the range of the upper and lower limits of the load seasonal standard of the current load forecast power grid model in time period T, and the corresponding load trend characteristic data in the load forecast sequence are within the range of the upper and lower limits of the load trend standard of the current load forecast power grid model in time period T, then it is proved that the load forecast power grid model is suitable for the current power grid operation status; otherwise, it is proved that the load forecast power grid model is not suitable for the current power grid operation status.

[0063] S6. If the load forecasting grid model is not suitable for the current grid operation status, the historical load time series is reconstructed, and the load forecasting grid model that meets the current situation is retrained and optimized based on the reconstructed historical load series as a validation set. The specific execution steps include:

[0064] When the load forecasting grid model is not applicable to the current grid operation status, the first step is to check whether the operation status of the load inside the grid has changed. If the reason why the load forecasting grid model is not applicable is that the operation status of all loads in the area has changed, the corresponding historical load time series is adjusted according to the changed load operation status, and the adjusted historical load time series is used as a validation set to optimize the load forecasting grid model. For example, if the initial grid operation status matrix is ​​[1,1,1,1,1] and the current grid operation status matrix is ​​[1,0,1,1,0], it means that the grid operation status has changed, where 1 indicates that the load is in operation and 0 indicates that the load is stopped or paused. Then the adjusted historical load time series is the load data in the load time series when the initial grid operation status matrix is ​​[1,1,1,1,1] minus the load data corresponding to the load being stopped or paused in the current grid operation status matrix [1,0,1,1,0].

[0065] If the operation status of the internal load of the power grid has not changed, the external environmental factors are considered as the reason for the inapplicability of the load forecasting power grid model. The grey correlation analysis method is used to calculate the correlation between the load trend characteristics of the load forecasting sequence and the forecast trend of each external environmental factor data, as well as the correlation between the load seasonal characteristics of the load forecasting sequence and the forecast season of each external environmental factor data;

[0066] Filter out the historical load time series M in the same time period as the load forecast sequence, calculate the ratio a of the forecast trend correlation of the load forecast sequence and the trend correlation of the historical load time series M, and use the ratio a as the weight of the load trend characteristics of the historical load time series in the same time period;

[0067] Calculate the ratio b of the seasonal correlation of the load forecast sequence and the seasonal correlation of the historical load time series M, and use the ratio b as the weight of the seasonal trend characteristics of the historical load time series M;

[0068] Based on the ratio a and the ratio b, the corresponding load trend characteristics and load seasonal characteristics in the historical load time series M are connected using the total loss function to obtain the reconstructed historical load time series.

[0069] Reconstruct loads from historical load time series:

[0070] Q=aQ t +bQ s ;

[0071] Where Q represents the reconstructed load, a is the ratio of the predicted trend correlation of the load forecast sequence to the trend correlation of the historical load time series M, b is the predicted seasonal correlation of the load forecast sequence to the seasonal correlation of the historical load time series M, Q t To reconstruct the load trend characteristics of the historical load corresponding to load Q, Q s The seasonality of the load corresponding to the historical load of the reconstructed load Q.

[0072] Total loss function:

[0073]

[0074] Among them, t represents the time domain contrast loss function, ζ a represents the frequency domain amplitude contrast loss function, ζ p represents the frequency domain phase contrast loss function, and a is a parameter that balances the trade-off between trend and seasonal factors.

[0075] Example 2

[0076] The predictive error correction system of power grid model based on machine learning, such as Figure 2 As shown, it includes: a data acquisition module, which is used to obtain the historical load time series in the region and its corresponding influencing factor series; the influencing factors include external environmental factors and all load operation states in the region; the external environmental factors include temperature, precipitation, and seasonal factors;

[0077] The load forecasting module builds a training set and a validation set based on the historical load time series in the region and its corresponding influencing factor series. Based on the training set and the validation set, a neural network model is used to train a load forecasting power grid model. The load forecasting power grid model is used to predict the power grid load usage in the region to obtain a load forecasting sequence.

[0078] The power grid model judgment module is used to extract the load trend characteristics and load seasonal characteristics in the load forecast sequence, and judge whether the load forecast power grid model is suitable for the current power grid operation status based on the load trend characteristics and load seasonal characteristics in the load forecast sequence. If it is suitable, the load forecast power grid model is continued to be used for load forecasting;

[0079] The power grid model re-optimization module is used to reconstruct the historical load time series and retrain and optimize the load forecasting power grid model that conforms to the current situation based on the reconstructed historical load series as a validation set.

[0080] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0081] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A predictive error correction method for a power grid model based on machine learning, characterized in that: The steps include: Obtain the historical load time series in the region and its corresponding influencing factor series; the influencing factors include external environmental factors and all load operation states in the region; the external environmental factors include temperature, precipitation, and seasonal factors; Construct training and validation sets based on the historical load time series in the region and its corresponding influencing factor series; Based on the training set and the validation set, a neural network model is used to train a load forecasting power grid model, and the load forecasting power grid model is used to forecast the load usage of the power grid in the region to obtain a load forecasting sequence; Extract load trend characteristics and load seasonal characteristics in load forecasting sequence; Based on the load trend characteristics and load seasonal characteristics in the load forecast sequence, it is judged whether the load forecasting power grid model is suitable for the current power grid operation status. If it is suitable, the load forecasting power grid model is continued to be used for load forecasting. If the load forecasting power grid model is not suitable for the current power grid operation status, the historical load time series is reconstructed, and the load forecasting power grid model that conforms to the current situation is retrained and optimized based on the reconstructed historical load series as a validation set.

2. The predictive error correction method for power grid model based on machine learning according to claim 1, characterized in that: The validation set uses mean square error as an evaluation index to adjust and optimize the load forecasting power grid model.

3. The predictive error correction method for power grid model based on machine learning according to claim 1, characterized in that: The load forecasting sequence uses a hybrid autoregressive expert of one-dimensional causal convolution to extract the load trend representation and then uses the time domain contrast loss function to learn to obtain the final load trend characteristics; The load forecast sequence uses a learnable Fourier transform in the frequency domain to obtain a frequency domain seasonal trend representation, then uses a frequency domain contrast loss function to learn to obtain a frequency domain seasonal trend representation, and converts the frequency domain seasonal trend representation back to the time domain to obtain the load seasonal characteristics.

4. The predictive error correction method for power grid model based on machine learning according to claim 1, characterized in that: The method to determine whether the load forecasting power grid model is applicable to the current power grid operation status is: Determine the time period T of the load forecast sequence in the annual cycle; Compare the load trend characteristic data in the load forecast sequence with the upper limit value and the lower limit value in the load trend standard of the current load forecast power grid model in time period T; Compare the load seasonality characteristics in the load forecast sequence with the upper limit value and the lower limit value in the load seasonality standard of the current load forecast power grid model in time period T; If the corresponding load seasonal characteristic data in the load forecast sequence exceeding α% are all within the range of the upper and lower limits of the load seasonal standard of the current load forecast power grid model in time period T, and the corresponding load trend characteristic data in the load forecast sequence are within the range of the upper and lower limits of the load trend standard of the current load forecast power grid model in time period T, then the load forecast power grid model is judged to be applicable to the current power grid operation status; otherwise, the load forecast power grid model is judged to be not applicable to the current power grid operation status.

5. The predictive error correction method for power grid model based on machine learning according to claim 1, characterized in that: The step of reconstructing the historical load time series includes: When the reason why the load forecasting power grid model is not applicable is that the operating status of all loads in the region has changed, the corresponding historical load time series is adjusted according to the changed load operating status, and the adjusted historical load time series is used as the reconstructed historical load time series; If the external environmental factors are the reasons for the inapplicability of the load forecasting power grid model, the grey correlation analysis method is used to calculate the correlation between the load trend characteristics of the load forecasting sequence and the forecast trend of each external environmental factor data, as well as the correlation between the load seasonal characteristics of the load forecasting sequence and the forecast season of each external environmental factor data; Filter out the historical load time series M in the same time period as the load forecast sequence, calculate the ratio a of the forecast trend correlation of the load forecast sequence and the trend correlation of the historical load time series M, and use the ratio a as the weight of the load trend characteristics of the historical load time series in the same time period; Calculate the ratio b of the seasonal correlation of the load forecast sequence and the seasonal correlation of the historical load time series M, and use the ratio b as the weight of the seasonal trend characteristics of the historical load time series M; Based on the ratio a and the ratio b, the corresponding load trend characteristics and load seasonal characteristics in the historical load time series M are connected using the total loss function to obtain the reconstructed historical load time series.

6. The predictive error correction method for power grid model based on machine learning according to claim 5, characterized in that: The reconstructed load in the reconstructed historical load time series is: Q=aQ t +bQ s ; Where Q represents the reconstructed load, a is the ratio of the predicted trend correlation of the load forecast sequence to the trend correlation of the historical load time series M, b is the predicted seasonal correlation of the load forecast sequence to the seasonal correlation of the historical load time series M, Q t To reconstruct the load trend characteristics of the historical load corresponding to load Q, Q s The seasonality of the load corresponding to the historical load of the reconstructed load Q.

7. The predictive error correction method for power grid model based on machine learning according to claim 5, characterized in that: The total loss function is: Among them, t represents the time domain contrast loss function, ζ a represents the frequency domain amplitude contrast loss function, ζ p represents the frequency domain phase contrast loss function, and a is a parameter that balances the trade-off between trend and seasonal factors.

8. A predictive error correction system for power grid models based on machine learning, characterized in that: include: A data acquisition module is used to obtain the historical load time series in the region and its corresponding influencing factor series; The influencing factors include external environmental factors and all load operating conditions in the area; the external environmental factors include temperature, precipitation, and seasonal factors; The load forecasting module builds a training set and a validation set based on the historical load time series in the region and its corresponding influencing factor series. Based on the training set and the validation set, a neural network model is used to train a load forecasting power grid model. The load forecasting power grid model is used to predict the power grid load usage in the region to obtain a load forecasting sequence. The power grid model judgment module is used to extract the load trend characteristics and load seasonal characteristics in the load forecast sequence, and judge whether the load forecast power grid model is suitable for the current power grid operation status based on the load trend characteristics and load seasonal characteristics in the load forecast sequence. If it is suitable, the load forecast power grid model is continued to be used for load forecasting; The power grid model re-optimization module is used to reconstruct the historical load time series and retrain and optimize the load forecasting power grid model that conforms to the current situation based on the reconstructed historical load series as a validation set.

Images