Power grid regional load prediction system based on big data

Abstract

The invention is suitable for the technical field of load prediction, and particularly relates to a power grid regional load prediction system based on big data, and the system comprises a search device, a generation device, an extraction device, and an input device. The searching device is used for dividing a power grid load prediction area, marking the power grid load prediction area as a total area, collecting historical load data of the total area, and configuring a plurality of influence factors which at least comprise weather and temperature, selecting test time, and defining first power consumption and a first factor. And finding out the same time of the previous year of the test time. According to the method, the global adversarial network is established, the weight value can be converged, the accuracy of the weight value is further improved, the local adversarial network is established, the weight value can be corrected under the condition of limited sample data, noise interference is reduced, the prediction precision of the adversarial network is greatly improved, and the prediction accuracy of the adversarial network is improved. And safe and stable operation of a power grid is effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of load forecasting, and particularly to a power grid regional load forecasting system based on big data. Background Art

[0002] Since electric energy cannot be stored in large quantities, during the operation of power plants and the dispatching of electric energy, it is required that the power output of power plants maintain a dynamic balance with the fluctuations of the power grid load. By using big data technology to predict the power load, the safe and economic operation of the power grid can be effectively guaranteed.

[0003] If the total regional power grid is called the large power grid, then in the sub-regional power grid (small power grid) relative to it, due to the small number of samples, the historical load data may be insufficient or incomplete, resulting in difficult data training and greatly affecting the prediction accuracy of big data. Therefore, "how to perform pre-training on the large power grid and apply it to the small power grid" is the technical problem to be solved by the present invention. Summary of the Invention

[0004] The purpose of the present invention is to provide a power grid regional load forecasting system based on big data to solve the problem of "how to perform pre-training on the large power grid and apply it to the small power grid" proposed in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A power grid regional load forecasting system based on big data, the system includes:

[0007] A search device, a generation device, an extraction device, and an input device;

[0008] The search device is used to divide the power grid load forecasting area, mark it as the total area, collect the historical load data of the total area, configure a number of influencing factors, where the influencing factors at least include: weather and temperature, select the test time, and define the first power consumption and the first factor, find the same period of the previous year of the test time, and define the second power consumption and the second factor;

[0009] The generation device is used to form a global adversarial network composed of a generator and a discriminator, upload the first factor, the second factor, and the second power consumption to the generator, synchronize the first power consumption to the discriminator, and generate a global adversarial network;

[0010] The extraction device is used to create a weight value calculation formula, construct a loss function, iterate the loss function, and converge to obtain the weight values of each influencing factor. The weight values are input into the global adversarial network, and the global adversarial network is trained using the historical load data, the weight values are adjusted, and feature data is extracted from the trained global adversarial network.

[0011] The input device is used to integrate and generate a local adversarial network using the feature data, select a target sub-region from the total region, create a training set, train the local adversarial network, correct the weight values, collect the real-time values of the influencing factors of the target sub-region, and input the real-time values into the local adversarial network to output the predicted value of the grid load.

[0012] Further, the search device includes:

[0013] A division module for dividing the grid load prediction area, marking it as the total region, collecting the historical load data of the total region, and configuring several influencing factors, where the influencing factors at least include: weather and temperature;

[0014] A data preparation module for selecting the test time, defining the first electricity consumption and the first factor, finding the same period of the previous year of the test time, and defining the second electricity consumption and the second factor.

[0015] Further, the generation device includes:

[0016] An upload module for forming a global adversarial network composed of a generator and a discriminator, and uploading the first factor, the second factor, and the second electricity consumption to the generator;

[0017] A synchronization module for synchronizing the first electricity consumption to the discriminator to generate a global adversarial network.

[0018] Further, the extraction device includes:

[0019] An input module for creating a weight value calculation formula, constructing a loss function, iterating the loss function, and converging to obtain the weight values of each influencing factor, and inputting the weight values into the global adversarial network;

[0020] An extraction module for training the global adversarial network using the historical load data, adjusting the weight values, and extracting feature data from the trained global adversarial network.

[0021] Further, the input device includes:

[0022] A calibration module, which is used to utilize the feature data to integrally generate a local adversarial network, select a target sub-region from the total region, create a training set, train the local adversarial network, and calibrate the weight value;

[0023] A prediction module, which is used to collect real-time values of influencing factors of the target sub-region, input the real-time values into the local adversarial network, and obtain a predicted value of the power grid load.

[0024] Furthermore, the partitioning module includes:

[0025] A traversal unit, which is used to traverse additional factors, where the additional factors at least include: holidays, events, and population flow;

[0026] A linking unit, which is used to link the additional factors to the global adversarial network, calculate corresponding weight values, and create an enabling mechanism.

[0027] Furthermore, the data preparation module includes:

[0028] A comparison unit, which is used to establish a mapping between the test time and the weight value, and construct a comparison table;

[0029] A plotting unit, which is used to use the test time as the abscissa and the weight value as the ordinate to plot several weight trend graphs and configure a correlation, where the correlation includes: positive, zero, and negative.

[0030] Furthermore, the synchronization module includes:

[0031] A slicing unit, which is used to integrate global constraints into the global adversarial network, slice the global constraints into several sub-items, and mark the priority corresponding to each sub-item, where the priority consists of high, medium, and low;

[0032] An insertion unit, which is used to push the sub-items into the generator in order from high to low according to the priority.

[0033] Furthermore, the input module includes:

[0034] A quantization unit, which is used to quantize the weather, calculate the weather difference between the first factor and the second factor, denoted as W, and calculate the temperature difference between the first factor and the second factor, denoted as T;

[0035] A calculation unit, which is used to construct a weight value calculation formula, and the weight value calculation formula is:

[0036] ;

[0037] where is the predicted value of the first electricity consumption, y is the second electricity consumption, α is the weight value corresponding to the weather, and b is the weight value corresponding to the temperature;

[0038] Construct a loss function, and the loss function is:

[0039] ;

[0040] where E is the deviation, is the first electricity consumption;

[0041] Take partial derivatives of α and b respectively, and perform several iterative updates to obtain the weight values of each of the influencing factors.

[0042] Further, the prediction module includes:

[0043] A corresponding unit for generating a label using the weight value and inserting the label into the corresponding target sub-region;

[0044] A writing unit for recording the true value of the grid load, calculating the difference between the true value and the predicted value, and writing the difference into the label.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] By determining the total area, it is possible to optimize power dispatching, improve the reliability of the power grid. At the same time, it greatly expands the source of historical load data, providing sufficient training data for the adversarial network. By determining the test time and the same period of the previous year, it is convenient to calculate the weight value and quantify the influence degree of weather and temperature on the grid load, greatly improving the accuracy and reliability of load prediction. By forming a global adversarial network, it is possible to converge the weight value, thereby further improving the accuracy of the weight value. By constructing a local adversarial network, it is possible to correct the weight value in the case of limited sample data, thereby reducing noise interference and greatly improving the prediction accuracy of the adversarial network, effectively ensuring the safe and stable operation of the power grid. Description of the Drawings

[0047] Figure 1 is the block diagram of the composition of the power grid area load prediction system based on big data provided by the embodiment of the present invention;

[0048] Figure 2 is the block diagram of the composition of the search device in the power grid area load prediction system based on big data provided by the embodiment of the present invention;

[0049] Figure 3 is the block diagram of the composition of the generation device in the power grid area load prediction system based on big data provided by the embodiment of the present invention;

[0050] Figure 4Block diagram of the extraction device in the power grid area load forecasting system based on big data provided by the embodiments of the present invention;

[0051] Figure 5 Block diagram of the input device in the power grid area load forecasting system based on big data provided by the embodiments of the present invention;

[0052] Figure 6 Block diagram of the division module in the power grid area load forecasting system based on big data provided by the embodiments of the present invention;

[0053] Figure 7 Block diagram of the data preparation module in the power grid area load forecasting system based on big data provided by the embodiments of the present invention;

[0054] Figure 8 Block diagram of the synchronization module in the power grid area load forecasting system based on big data provided by the embodiments of the present invention;

[0055] Figure 9 Block diagram of the input module in the power grid area load forecasting system based on big data provided by the embodiments of the present invention;

[0056] Figure 10 Block diagram of the forecasting module in the power grid area load forecasting system based on big data provided by the embodiments of the present invention. Detailed implementation manners

[0057] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] Figure 1 The block diagram of the composition structure of the power grid area load forecasting system based on big data provided by the embodiments of the present invention is shown. The power grid area load forecasting system 1 based on big data includes: a search device 11, a generation device 12, an extraction device 13, and an input device 14;

[0059] The search device 11 is used to divide the power grid load forecasting area, marked as the total area, collect the historical load data of the total area, configure a number of influencing factors, where the influencing factors at least include: weather and temperature, select the test time, and define the first electricity consumption and the first factor, find the same period time of the previous year of the test time, and define the second electricity consumption and the second factor.

[0060] According to the power grid area division, determine the power grid load forecasting area, which is marked as the total area. The power grid load forecasting area should have sufficient sample size. For example, the total area can be a city-level power grid, or a local power grid such as a community, campus, hospital, and commercial center. Collect the historical load data of the total area from the power supply department or public data. The historical load data is the historical power consumption. Determine the influencing factors of the historical load data, where the influencing factors include weather and temperature. In this application, the power grid load situation under extreme weather is not considered. Select the test time, such as June 1, 2024. Define the power consumption on this day as the first power consumption, collect the temperature and weather data on this day, and define it as the first factor. Find the same period of the previous year, that is, June 1, 2023, and use the same method to determine the second power consumption and the second factor. It should be noted that if the temperature difference between the first factor and the second factor is greater than the threshold, the same period of the previous year should be reselected until the temperature difference is less than or equal to the threshold. The specific reselection method is to postpone the same period of the previous year. Take June 2, 2023 as the same period of the previous year.

[0061] In this application, for the convenience of explaining the specific calculation process, only weather and temperature are taken as examples of influencing factors. In the actual prediction process, more influencing factors should be considered, but the calculation method is the same.

[0062] The generating device 12 is used to form a global adversarial network composed of a generator and a discriminator, upload the first factor, the second factor, and the second power consumption to the generator, synchronize the first power consumption to the discriminator, and generate a global adversarial network.

[0063] Upload the first factor, the second factor, and the second power consumption mentioned above to the generator, and upload the first power consumption to the discriminator. The role of the discriminator is to evaluate the authenticity of the generation result by comparing the load forecast output by the generator with the real historical load data. The generator continuously adjusts the weight value of the influencing factors to make the prediction result closer to the real load data. Use the game between the generator and the discriminator to construct a global adversarial network. The advantage of doing this is that it can accurately capture the complex factors and potential change laws affecting the power grid load, thereby improving the accuracy and reliability of load forecasting.

[0064] Continuing with the above example in detail, in the generator, by using the electricity consumption on June 1, 2023 and the weather and temperature data on that day, combined with the weather and temperature data on June 1, 2024, the electricity consumption on June 1, 2024 is deduced, and the deduced electricity consumption is determined as the predicted value; the predicted value is sent to the discriminator, and in the discriminator, the predicted value is compared with the actual electricity consumption to determine whether the deviation between the two is within the control range, where the control range is pre - determined by professionals. If it is, the weight value in the discriminator is output as the result. If not, the generator is used to adjust the weight value in the discriminator until the deviation is within the control range.

[0065] However, the above process is only used to initially determine the weight value, and it is also necessary to converge the weight value and train the adversarial network to improve the accuracy of the weight value and the predicted data.

[0066] The extraction device 13 is used to create a weight value calculation formula, construct a loss function, iterate the loss function, and converge to obtain the weight value of each influencing factor. The weight value is input into the global adversarial network, and the global adversarial network is trained using the historical load data, the weight value is adjusted, and the feature data is extracted from the trained global adversarial network.

[0067] Create a weight value calculation formula, that is , substitute the first factor, the second factor, and the first electricity consumption into the weight value calculation formula to calculate the predicted value of the electricity consumption at the test time; define the difference between the predicted value and the second electricity consumption as the deviation, calculate the mean square value of the deviation, construct a loss function, take the partial derivatives of α and b in it, and perform multiple iterations and convergence to obtain the weight value; after the weight value calculation is completed, continue to select other test times to train the global adversarial network and adjust the weight value; after the training is completed, define the weight value as the feature data.

[0068] Continuing with the above example in detail, assume that the electricity consumption in a certain sub - region on June 1, 2023 is 1000 MW, the temperature on that day is 10 °C, and the weather is sunny (quantify sunny as 1), and the electricity consumption on June 1, 2024 is 1200 MW, the temperature on that day is 12 °C, and the weather is rainy (quantify rainy as 2); in other words, the first electricity consumption is 1200 MW, the first factor is 12 °C and 2, the second electricity consumption is 1000 MW, and the second factor is 10 °C and 1. Substitute the data into the weight value calculation formula to get

[0069]

[0070] where α is the weight value corresponding to the temperature, b is the weight value corresponding to the weather, and the deviation is , denoted as E, calculate the mean square value of the deviation, and construct the following loss function:

[0071] ;

[0072] where ;

[0073] The partial derivative with respect to α is:

[0074]

[0075] The partial derivative with respect to b is:

[0076] ;

[0077] Set the learning step size to 0.001 (this step size is pre - determined by professionals), denoted as , and set the initial value of the weight value to 0, that is , , substitute the initial value into the deviation, and get ;

[0078] Perform gradient calculation on the above partial derivatives to get:

[0079]

[0080] ;

[0081] Using the learning step size, update and to get

[0082]

[0083]

[0084] Substitute the calculated and into E, and continue to update α and b along the direction where the gradient ratio of temperature and weather is 2:1 until is infinitely close to 1200MW, then define the α and b at this time as the final weight values.

[0085] The input device 14 is used to utilize the feature data to integrate and generate a local adversarial network, select a target sub - region from the total region, create a training set, train the local adversarial network, correct the weight values, collect the real - time values of the influencing factors of the target sub - region, and input the real - time values into the local adversarial network to output the predicted value of the power grid load.

[0086] Extract feature data from the adversarial network, where the feature data is the weight value, and use this weight value to construct a local adversarial network; divide the total area into several sub-areas, where the sub-areas can be a building, a community, etc. The total area and the sub-areas are only relative concepts without specific division criteria; select the areas that need to be predicted for the power grid load from the sub-areas, that is, the target sub-areas; read the power consumption load data of the target sub-areas from the power supply department or public data, and train the local adversarial network so that it can learn the load change rules of the target sub-areas and adjust the weight values of the influencing factors within the target sub-areas.

[0087] Collect the real-time values of the influencing factors within the target sub-areas, where the real-time values are the predicted data of the weather and temperature on a future day, and the real-time values can be obtained from meteorological data providers; input the real-time values into the generator in the local adversarial network, and output the predicted value of the power grid load on a future day through the discriminator.

[0088] Figure 2 The block diagram of the composition structure of the power grid area load prediction system based on big data provided by the embodiment of the present invention is shown. The searching device 11 includes:

[0089] The dividing module 111 is used to divide the power grid load prediction area, marked as the total area, collect the historical load data of the total area, and configure several influencing factors, where the influencing factors at least include: weather and temperature.

[0090] Determine the boundary of the area that needs to be predicted for the power grid load, integrate the boundary, and divide the total area, where the total area consists of several sub-areas; read the historical load data of each day within the total area from the power supply department or public data, and determine the influencing factors of each day at the same time.

[0091] The data preparation module 112 is used to select the test time, define the first power consumption and the first factor, find the same period time of the previous year of the test time, and define the second power consumption and the second factor.

[0092] The number of test times is not limited. Define the power consumption load during the test time as the first power consumption, and the influencing factors during the test time as the first factor; similarly, in the same period time of the previous year, determine the second power consumption and the second factor.

[0093] Figure 3 The block diagram of the composition structure of the power grid area load prediction system based on big data provided by the embodiment of the present invention is shown. The generating device 12 includes:

[0094] The uploading module 121 is used to form a global adversarial network composed of a generator and a discriminator, and upload the first factor, the second factor, and the second power consumption to the generator.

[0095] An adversarial network is formed using a generator and a discriminator, where the adversarial network is mainly used to calculate and correct weight values.

[0096] The synchronization module 122 is used to synchronize the first power consumption to the discriminator to generate a global adversarial network.

[0097] Upload the first factor, the second factor, and the second power consumption to the generator, and upload the first power consumption to the discriminator. Use the discriminator to compare the predicted data in the generator and determine whether the deviation is within the control range.

[0098] Figure 4 The block diagram of the composition structure of the power grid area load forecasting system based on big data provided by the embodiment of the present invention is shown. The extraction device 13 includes:

[0099] The input module 131 is used to create a weight value calculation formula, construct a loss function, iterate the loss function, and converge to obtain the weight value of each influencing factor, and input the weight value into the global adversarial network.

[0100] Construct a weight value calculation formula, and on this basis, construct a loss function, calculate the weight value, and transfer this weight value into the global adversarial network.

[0101] The extraction module 132 is used to train the global adversarial network using the historical load data, adjust the weight value, and extract feature data from the trained global adversarial network.

[0102] Use the historical load data of the total area to train the global adversarial network to improve the accuracy of the weight value, and after the training is completed, define the weight value in the global adversarial network as feature data.

[0103] Figure 5 The block diagram of the composition structure of the power grid area load forecasting system based on big data provided by the embodiment of the present invention is shown. The input device 14 includes:

[0104] The correction module 141 is used to use the feature data to integrate and generate a local adversarial network, select a target sub-region from the total area, create a training set, and train the local adversarial network to correct the weight value.

[0105] Use the feature data to form a local adversarial network, select a target sub-region from several sub-regions; use the power grid load data of the target sub-region to create a training set, train the local adversarial network, and gradually adjust the weight value during the training process.

[0106] A prediction module 142, configured to collect real-time values of influencing factors of the target sub-region, input the real-time values into a local adversarial network, and obtain a predicted value of the grid load through the input.

[0107] Determine the influencing factors of the target sub-region, collect the real-time values of the influencing factors on a certain future day, input the real-time values into the local adversarial network, and obtain a predicted value of the grid load through the input.

[0108] Figure 6 The block diagram of the composition structure of the grid area load prediction system based on big data provided by the embodiment of the present invention is shown. The division module 111 includes:

[0109] A traversal unit 1111, configured to traverse additional factors, where the additional factors at least include: holidays, events, and population mobility.

[0110] In this embodiment, the influencing factors may include holidays, events, and population mobility in addition to weather and temperature; however, in the specific data processing process, the additional factors need to be quantified.

[0111] A linking unit 1112, configured to link the additional factors to the global adversarial network, calculate the corresponding weight values, and create an enabling mechanism.

[0112] After determining the additional factors, link the additional factors in the global adversarial network, and calculate the weight values corresponding to each additional factor according to the weight value calculation method mentioned in the extraction device 13; however, in the process of predicting the grid load at different times, it is not necessary to introduce all the additional factors, that is, according to the enabling mechanism, introduce the additional factors; where the enabling mechanism may be: when the time for predicting the grid load is a holiday, then introduce the weight value corresponding to the holiday.

[0113] Figure 7 The block diagram of the composition structure of the grid area load prediction system based on big data provided by the embodiment of the present invention is shown. The data preparation module 112 includes:

[0114] A comparison unit 1121, configured to establish a mapping between the test time and the weight value, and construct a comparison table.

[0115] The test time is different, and the corresponding weight values are also different. Integrate the test time and the corresponding weight values to construct a comparison table; the advantage of doing this is that the weight value at each test time can be quickly determined.

[0116] A plotting unit 1122, configured to use the test time as the abscissa and the weight value as the ordinate to plot a plurality of weight trend graphs and configure the correlation, where the correlation includes: positive, zero, and negative.

[0117] Find several adjacent test times, determine them as the abscissa, and successively use the weight values of the influencing factors corresponding to the test times as the ordinate to plot a weight trend graph; judge whether there is a correlation between the weather and the weight values, where the correlation can be a positive correlation, a negative correlation, or no correlation.

[0118] For example, record the daily temperature from June 1, 2024 to July 30, 2024, calculate the corresponding weight values, use time as the abscissa and temperature as the ordinate to plot a temperature curve, and then use time as the abscissa and the weight value of the temperature as the ordinate to plot a weight trend graph, and judge whether the temperature curve and the weight trend graph show a correlation.

[0119] Figure 8 The composition structure block diagram of the power grid area load forecasting system based on big data provided by an embodiment of the present invention is shown. The synchronization module 122 includes:

[0120] A segmentation unit 1221, configured to integrate global constraints into the global adversarial network, segment the global constraints into several sub-items, and mark the priority corresponding to each sub-item, where the priority consists of high, medium, and low.

[0121] In order to converge the weight values faster and reduce invalid calculations, global constraints are inserted into the global adversarial network. The global constraints are a set of constraint conditions for data; for example, a certain constraint condition is: if the temperature difference between the test time and the same period of the previous year is greater than the threshold, then reselect the same period of the previous year; another constraint condition is: if the temperature corresponding to a certain test time exceeds the fluctuation range, reselect the test time; segment the global constraints into several sub-items, where the sub-items are the constraint conditions, and determine the priority of each constraint condition, where the priority can be high, medium, and low.

[0122] An insertion unit 1222, configured to push the sub-items into the generator in order from high to low according to the priority.

[0123] Enable the constraint conditions in order from high to low according to the priority.

[0124] Figure 9 The composition structure block diagram of the power grid area load forecasting system based on big data provided by an embodiment of the present invention is shown. The input module 131 includes:

[0125] A quantization unit 1311, configured to quantize the weather, calculate the weather difference between the first factor and the second factor, denoted as W, and calculate the temperature difference between the first factor and the second factor, denoted as T;

[0126] A calculation unit 1312 for constructing a weight value calculation formula, and the weight value calculation formula is:

[0127]

[0128] where is the predicted value of the first electricity consumption, y is the second electricity consumption, α is the weight value corresponding to the weather, and b is the weight value corresponding to the temperature;

[0129] Construct a loss function, and the loss function is:

[0130]

[0131] where E is the deviation, is the first electricity consumption;

[0132] Respectively take partial derivatives of α and b, and perform several iterations of update to obtain the weight values of each of the influencing factors.

[0133] Calculate the weight values of each influencing factor according to the steps shown in the extraction device 13.

[0134] Figure 10 The block diagram of the composition structure of the power grid area load prediction system provided by the embodiment of the present invention is shown. The prediction module 142 includes:

[0135] A corresponding unit 1421 for generating a label by using the weight value and inserting the label into the corresponding target sub-region.

[0136] Insert a label generated by the corresponding weight value into the target sub-region; the advantage of doing this is that the corresponding weight value can be quickly found and the power grid load can be predicted.

[0137] A writing unit 1422 for recording the true value of the power grid load, calculating the difference between the true value and the predicted value, and writing the difference into the label.

[0138] The difference between the true value and the predicted value, that is, the error. For different target sub-regions, the magnitude and distribution characteristics of the error may be different; by writing the error into the label, it can help the generator adopt different strategies for different regions during the training process; for example, in regions with larger errors, strengthen the adjustment of the weight values of key influencing factors (such as temperature and human flow, etc.), and in regions with smaller errors, optimize the generalization ability of the model; in addition, by inserting the label, the discrimination accuracy of the discriminator can be intuitively displayed, and it can quickly identify which sub-regions have larger errors, so as to optimize the stability of the adversarial training.

[0139] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as falling within the scope described in this specification.

[0140] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention shall be subject to the appended claims.

[0141] The foregoing is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A power grid regional load forecasting system based on big data, characterized in that: The system comprises: a search device, a generation device, an extraction device and an input device; The search device is used to divide the power grid load forecast area, mark it as a total area, collect historical load data of the total area, configure several influencing factors, wherein the influencing factors at least include: weather and temperature, select the test time, and define the first power consumption and the first factor, find the time of the same period of the previous year of the test time, and define the second power consumption and the second factor; The generating device is used to form a global adversarial network composed of a generator and a discriminator, upload the first factor, the second factor, and the second power consumption to the generator, synchronize the first power consumption to the discriminator, and generate a global adversarial network; The extraction device is used to create a weight value calculation formula, construct a loss function, iterate the loss function, and converge to obtain a weight value of each of the influencing factors, input the weight value into a global adversarial network, use the historical load data to train the global adversarial network, adjust the weight value, and extract feature data from the trained global adversarial network; The input device is used to utilize the feature data to integrate and generate a local adversarial network, select a target sub-area from the total area, create a training set, train the local adversarial network, correct the weight value, collect the real-time value of the influencing factors of the target sub-area, input the real-time value into the local adversarial network, and output the predicted value of the power grid load.

2. The power grid regional load forecasting system based on big data according to claim 1 is characterized in that: The search device comprises: A partitioning module is used to partition a power grid load forecasting area, mark it as a total area, collect historical load data of the total area, and configure a number of influencing factors, wherein the influencing factors at least include: weather and temperature; The data preparation module is used to select the test time, define the first power consumption and the first factor, find out the time of the same period in previous years of the test time, and define the second power consumption and the second factor.

3. The power grid regional load forecasting system based on big data according to claim 1 is characterized in that: The generating device comprises: An uploading module, used to form a global adversarial network consisting of a generator and a discriminator, and upload the first factor, the second factor, and the second power consumption to the generator; A synchronization module is used to synchronize the first power consumption to the discriminator to generate a global adversarial network.

4. The power grid regional load forecasting system based on big data according to claim 3 is characterized in that: The extraction device comprises: An input module is used to create a weight value calculation formula, construct a loss function, iterate the loss function, and converge to obtain a weight value of each influencing factor, and input the weight value into the global adversarial network; The extraction module is used to train the global adversarial network using the historical load data, adjust the weight value, and extract feature data from the trained global adversarial network.

5. The power grid regional load forecasting system based on big data according to claim 1, characterized in that: The input device comprises: A correction module, used to integrate and generate a local adversarial network using the feature data, select a target sub-region from the total region, create a training set, train the local adversarial network, and correct the weight value; The prediction module is used to collect the real-time values ​​of the influencing factors of the target sub-area, input the real-time values ​​into the local adversarial network, and obtain the predicted value of the power grid load.

6. The power grid regional load forecasting system based on big data according to claim 2, characterized in that: The division module comprises: A traversal unit, used for traversing additional factors, wherein the additional factors at least include: holidays, activities and population mobility; A linking unit is used to link the additional factors to the global adversarial network, calculate corresponding weight values, and create an enabling mechanism.

7. The power grid regional load forecasting system based on big data according to claim 2, characterized in that: The data preparation module includes: A comparison unit, used to establish a mapping between the test time and the weight value, and construct a comparison table; The drawing unit is used to draw a plurality of weight trend diagrams with the test time as the horizontal coordinate and the weight value as the vertical coordinate, and configure the correlation relationship, wherein the correlation relationship includes: positive, zero and negative.

8. The power grid regional load forecasting system based on big data according to claim 3 is characterized in that: The synchronization module comprises: A segmentation unit, used for integrating a global constraint into the global adversarial network, segmenting the global constraint into a plurality of sub-items, and marking a priority corresponding to each sub-item, wherein the priority consists of high, medium and low; The insertion unit is used to push the sub-items into the generator in descending order of priority.

9. The power grid regional load forecasting system based on big data according to claim 4, characterized in that: The input module comprises: A quantification unit, used to quantify the weather, calculate the weather difference between the first factor and the second factor, recorded as W, and calculate the temperature difference between the first factor and the second factor, recorded as T; The calculation unit is used to construct a weight value calculation formula, and the weight value calculation formula is: ； in is the predicted value of the first power consumption, y is the second power consumption, α is the weight value corresponding to the weather, and b is the weight value corresponding to the temperature; Construct a loss function, which is: ； Where E is the deviation, It is the first electricity consumption; Partial derivatives are performed on α and b respectively, and several iterations are performed to update the weight value of each influencing factor.

10. The power grid regional load forecasting system based on big data according to claim 5, characterized in that: The prediction module comprises: A corresponding unit, used to generate a label using the weight value, and insert the label into the corresponding target sub-region; The writing unit is used to record the real value of the power grid load, calculate the difference between the real value and the predicted value, and write the difference into the tag.

Images