Electrical load prediction method based on geographic information and neural network

Abstract

The invention discloses an electrical load prediction method based on geographic information and a neural network, and belongs to the technical field of electrical load prediction. According to the method, the problems that the existing prediction method does not consider the dominant electricity utilization industry in the coverage range of the geographic position to be predicted, so that the prediction accuracy is low, and the existing method is low in prediction efficiency and high in model complexity are solved. According to the method, the total power consumption of each industry in each sub-region is counted according to the geographical location information covered by each sub-region, then the power consumption prediction dominant industry in the sub-region is determined, then the One-hot coding vector is obtained according to the power consumption prediction dominant industry, the One-hot coding vector is fused into the feature vector of the sub-region, and finally the power consumption prediction dominant industry in the sub-region is obtained. And carrying out clustering processing on the sub-regions according to the feature vectors. And after a final clustering center corresponding to each cluster is calculated, an electrical load prediction result of each sub-region in the cluster can be calculated according to the clustering center through the output of the prediction model. The method can be applied to electrical load prediction.

Description

Technical Field

[0001] This invention belongs to the field of electricity load forecasting technology, specifically relating to an electricity load forecasting method based on geographic information and neural networks. Background Technology

[0002] With the continuous development of my country's economy across all sectors, electricity has become increasingly important. The stability of all industries is inseparable from electricity, therefore, the safe and stable operation of the power grid directly affects the stable development of all industries nationwide. The gradual progress of society, the continuous development of the economy, and the ever-increasing demands of people for a higher quality of life are all inextricably linked to the continuous, stable, and safe operation of electricity. Load forecasting, the process of predicting future load demand based on historical load data and other external factors, plays a crucial role in power system planning and dispatching.

[0003] With the rapid development of deep learning technology, it has been widely applied in the field of electricity load forecasting. However, existing deep learning-based forecasting methods do not consider the dominant electricity-consuming industries within the geographical area to be forecasted, resulting in low accuracy of electricity load forecasting. Moreover, when it is necessary to forecast the electricity consumption of each region separately, existing forecasting methods need to input the information of each sub-region into the model one by one, resulting in low forecasting efficiency and high complexity of existing forecasting models. Summary of the Invention

[0004] This invention addresses the problems of low prediction accuracy, low prediction efficiency, and high model complexity in existing prediction methods, which fail to consider the dominant electricity-consuming industries within the geographical area covered by the predicted location. To solve these problems, this invention proposes an electricity load prediction method based on geographic information and neural networks.

[0005] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a method for predicting electricity load based on geographic information and neural networks, the method specifically including the following steps:

[0006] Step 1: Treat the power supply area of ​​each local power supply system in the area to be predicted as a sub-region, and obtain the geographic coordinates of each local power supply system from the geographic information system. Use the obtained geographic coordinates as the geographic index of the corresponding sub-region.

[0007] And obtain the historical electricity consumption data for each sub-region separately, for the first The sub-region will be the first Each sub-region in history Annual electricity consumption is denoted as , will the The sequence of electricity consumption data for each sub-region in historical years is denoted as follows: ;

[0008] in, Indicates the first Electricity consumption in each sub-region in its first year of history Indicates the first Electricity consumption in each sub-region in the second year in history Indicates the first Each sub-region in history Annual electricity consumption;

[0009] Step 2: Process the historical electricity consumption data sequence for each sub-region to obtain the processed historical electricity consumption sequence for each sub-region, and construct the initial feature vector for each sub-region based on the geographical location information of the sub-region and the processed historical electricity consumption sequence.

[0010] Step 3: Perform initial clustering on the sub-regions based on the initial feature vectors of each sub-region to obtain the initial clustering results of the sub-regions. The set of geographic location indexes of the sub-regions contained in the initial clusters Recorded as:

[0011] ;

[0012] in:

[0013] Indicates the first The geographic location index of the first sub-region in the initial cluster. Indicates responsibility for the first The geographical longitude of the local power supply system for the first sub-region in the initial cluster. Indicates responsibility for the first The geographical latitude of the local power supply system for the first sub-region in the initial cluster;

[0014] Indicates the first The geographic location index of the second sub-region in the initial cluster. Indicates responsibility for the first The geographical longitude of the local power supply system for the second sub-region in the initial cluster. Indicates responsibility for the first The geographical latitude of the local power supply system for the second sub-region in the initial cluster;

[0015] Indicates the first In the initial clustering, the th Geographical index of each sub-region Indicates responsibility for the first In the initial clustering, the th The geographical longitude of the local power supply system for each sub-region, Indicates responsibility for the first In the initial clustering, the th The geographical latitude of the local power supply system for each sub-region;

[0016] Indicates the first The number of subregions contained in each initial cluster. , Indicates the total number of categories;

[0017] Step 4: Generate a final cluster center for each group of initial clustering results, obtain the final clustering results based on each final cluster center, output the sub-regions contained in each final cluster, and record the geographic location index of all sub-regions contained in each final cluster.

[0018] Step 5: Predict the rate of change of electricity load for each sub-region in the final cluster;

[0019] For the The final cluster is extracted. The historical electricity consumption data of each sub-region corresponding to the geographic location index included in the final cluster, based on the... The predicted electricity load rate of change for each sub-region is obtained by combining the predicted results of the final clustering with the extracted historical electricity consumption data.

[0020] Furthermore, the specific process of step two is as follows:

[0021] Step 21, place the first The processed historical electricity consumption sequence corresponding to each sub-region is denoted as follows: , They represent the first The sub-regions in the 2nd, 3rd, ..., 1st year of history Annual electricity consumption change rate;

[0022] Step 22: Calculate the first... The sub-regions in the 2nd, 3rd, ..., 1st year of history The average annual rate of change in electricity consumption:

[0023]

[0024] in, Indicates the first Average rate of change in electricity consumption in each sub-region;

[0025] Steps two and three: Calculate the first... The sub-regions in the 2nd, 3rd, ..., 1st year of history The maximum value of the annual electricity consumption change rate:

[0026]

[0027] in, Indicates the first The maximum electricity consumption change rate corresponding to each sub-region;

[0028] Step Two Four: Calculate the... Skewness of electricity consumption change rate corresponding to each sub-region:

[0029]

[0030] in, Represents the processed historical electricity consumption sequence The standard deviation of all data in the middle. Indicates the first Skewness of electricity consumption change rate corresponding to each sub-region;

[0031] Step 25: Determine the location information. The one-hot encoded vector of the nth sub-region, using the nth... The processed historical electricity consumption sequence corresponding to each sub-region , , and the The one-hot encoded vectors of the sub-regions constitute the first Feature vectors of each sub-region .

[0032] Furthermore, the specific process of step two-one is as follows:

[0033]

[0034] in:

[0035] Indicates the first Electricity consumption in each sub-region in the third year in history;

[0036] Indicates the first Each sub-region in history Annual electricity consumption.

[0037] Furthermore, the initial clustering method is as follows:

[0038] Step 3. Initialization Each gray wolf contains one individual gray wolf. There are 1 initial cluster center, and each initial cluster center is a vector with the same dimension as the feature vector of the sub-region;

[0039] Step 3.2: Initialize the number of iterations ;

[0040] Step 3: Calculate the number of... The objective function value corresponding to each individual gray wolf in each iteration;

[0041] For the The first of the individual gray wolves Cluster centers Calculate the first one respectively Feature vectors of each sub-region and The Euclidean distance is then the first... Each subregion belongs to the cluster center corresponding to the maximum Euclidean distance;

[0042] After processing the feature vectors of each sub-region in turn, the first... Objective function value for each individual gray wolf:

[0043]

[0044] in, Indicates the first Of the individual gray wolves, the one belonging to the first... Cluster centers The set of subregions; Indicates the first The objective function value of each individual gray wolf. express and Euclidean distance;

[0045] Steps three and four, from the first In the next iteration, the top three gray wolf individuals with the largest objective function values ​​are selected, and the others... Each individual gray wolf is updated based on the three selected gray wolf individuals;

[0046] The three selected gray wolf individuals and the updated The first gray wolf individual as the first Each individual gray wolf in the next iteration;

[0047] Step 35: Determine if the conditions are met. , The maximum number of iterations is set.

[0048] If satisfied Then from the first In the next iteration, select the gray wolf with the largest objective function value from the gray wolf individuals, and then execute step three six.

[0049] If not satisfied Then let Return to step three;

[0050] Step 36: Select the gray wolf individuals from Step 35. Using the first cluster center as the initial cluster center, and then according to... The initial clustering results are obtained from the initial cluster centers, and the set of geographic location indexes of the sub-regions contained in each cluster is output.

[0051] Furthermore, the specific process of step four is as follows:

[0052] For the obtained first Initial clustering results for the groups:

[0053] Step 41: Construct the first The first group in the initial clustering results The feature vector corresponding to the geographic location index of each sub-region eigenvectors , ;

[0054] Initialize the number of iterations ;

[0055] Step 42: Extract separately The first in The extracted elements are then sorted in ascending order to obtain the sorted result sequence. ;

[0056] Using a fixed-length window from the sequence The starting point begins in the sequence Swipe up to calculate the importance index value of the data in each window; for the sequence Select the sequence The window position corresponding to the most important indicator value is selected, and the average value of all data within the selected window is calculated. The calculated average value is denoted as... ;

[0057] Step 43: Determine if the conditions are met. ;

[0058] If satisfied Then proceed to step four;

[0059] If not satisfied Then let Return to step four two;

[0060] Steps four and five: As the first The final cluster center.

[0061] Furthermore, the method for calculating the importance index value of the data within each window is as follows:

[0062] Step 421: Divide the range of electricity consumption change rate;

[0063] Step 422: Calculate the importance index values ​​of the data in the current window:

[0064]

[0065] in, This indicates the importance index value of the data within the current window; Indicates the number of elements in the current window. One data point; Indicates in sequence In, with The number of data points belonging to the same range of electricity consumption change rates. This indicates the number of data items in the current window.

[0066] Furthermore, the specific process of step five is as follows:

[0067] Step 51, place the first The final cluster center is used as input to the electricity load change rate prediction model to obtain the result for the _th Prediction results of electricity load change rate for each cluster;

[0068] Step 52: Record the predicted electricity load change rate obtained in Step 51 as... ,in, They represent the first year, the second year, ..., the third year, respectively. Annual electricity load change rate;

[0069] For the The th final cluster in the th The sub-region corresponding to the geographic location index is calculated. The sub-regions in the future Annual electricity load forecast results;

[0070]

[0071] in, Indicates the first The sub-regions in the future Annual electricity load forecast results Indicates the first The sub-regions in the future Annual electricity load forecast results;

[0072] Similarly, calculate the first... The electricity load prediction results for the sub-regions corresponding to each geographic location index in the final cluster.

[0073] Furthermore, the method also includes step six, which involves determining whether the power supply capacity of the power supply system corresponding to the geographical location index of each sub-region can meet the requirements when predicting the predicted power load value for each sub-region.

[0074] Furthermore, the working process of the electricity load change rate prediction model is as follows:

[0075] The first The final cluster centers serve as inputs to two parallel branches, where the first branch includes a TCN network and the second branch includes a BiLSTM network.

[0076] The outputs of the two branches are fused along the channel dimension, and the fused result is used as the input of a fully connected neural network. The fully connected neural network then outputs the predicted result of the electricity load change rate.

[0077] The beneficial effects of this invention are:

[0078] This invention calculates the total electricity consumption of each industry within a sub-region based on the geographical location information it covers, thereby identifying the dominant industry for electricity consumption forecasting within that sub-region. A one-hot encoded vector is then obtained based on this dominant industry and integrated into the sub-region's feature vector. The sub-regions are then clustered based on this feature vector. After calculating the final cluster center for each cluster, the electricity load forecast for each sub-region within that cluster can be calculated using the output of the prediction model. This approach ensures both accuracy and efficiency in prediction. Based on the electricity load forecast results for each sub-region, it can be determined whether the power supply system corresponding to the geographical location index of that sub-region meets the power supply requirements. Furthermore, compared to existing methods, this invention reduces model complexity while maintaining prediction accuracy. Attached Figure Description

[0079] Figure 1 This is a flowchart of an electricity load prediction method based on geographic information and neural networks according to the present invention;

[0080] Figure 2 This is a graph showing the percentage of electricity consumption within a specific sub-region. Detailed Implementation

[0081] With social development and the continuous improvement of productivity, urban electricity consumption is also constantly increasing. To ensure the stability of power supply in various local power supply systems within a city, it is necessary to predict the electricity consumption of the areas served by each local power supply system. This invention, from a geographic information perspective, treats areas geographically belonging to the same local power supply system as sub-regions and predicts the future electricity load within each sub-region. When a local power supply system is insufficient to meet the power supply requirements of a sub-region, corresponding modification measures need to be taken in advance based on the prediction results to meet the power demand. To facilitate the prediction of electricity load in the area to be predicted, this invention clusters the sub-regions based on historical electricity consumption data of each sub-region within the area to be predicted and the main electricity consumption types within the sub-regions obtained from geographic information. Finally, electricity load prediction is performed for each type of sub-region.

[0082] Specific implementation method one: Combining Figure 1 This embodiment describes a method for predicting electricity load based on geographic information and neural networks. The method specifically includes the following steps:

[0083] Step 1: Treat the power supply area of ​​each local power supply system in the area to be predicted as a sub-region, and obtain the geographic coordinates of each local power supply system from the geographic information system. Use the obtained geographic coordinates as the geographic index of the corresponding sub-region.

[0084] And obtain the historical electricity consumption data for each sub-region separately, for the first The sub-region will be the first Each sub-region in history Annual electricity consumption is denoted as , will the The sequence of electricity consumption data for each sub-region in historical years is denoted as follows: ;

[0085] in, They are arranged in chronological order from front to back. Indicates the first Electricity consumption in each sub-region in its first year of history Indicates the first Electricity consumption in each sub-region in the second year in history Indicates the first Each sub-region in history Annual electricity consumption;

[0086] To reduce the computational load during model prediction while ensuring the accuracy of electricity load prediction for each sub-region, this invention processes the historical electricity consumption data of each sub-region separately. The processed electricity consumption data sequence can reflect the historical electricity consumption trend of each sub-region, so as to cluster sub-regions with similar electricity consumption trends.

[0087] Step 2: Process the historical electricity consumption data sequence for each sub-region to obtain the processed historical electricity consumption sequence for each sub-region, and construct the initial feature vector for each sub-region based on the geographical location information of the sub-region and the processed historical electricity consumption sequence.

[0088] Specifically,

[0089] Step 21, place the first The processed historical electricity consumption sequence corresponding to each sub-region is denoted as follows: , They represent the first The sub-regions in the 2nd, 3rd, ..., 1st year of history Annual electricity consumption change rate;

[0090]

[0091] in:

[0092] Indicates the first Electricity consumption in each sub-region in the third year in history;

[0093] Indicates the first Each sub-region in history Annual electricity consumption;

[0094] Step 22: Calculate the first... The sub-regions in the 2nd, 3rd, ..., 1st year of history The average annual rate of change in electricity consumption:

[0095]

[0096] in, Indicates the first Average rate of change in electricity consumption in each sub-region;

[0097] Steps two and three: Calculate the first... The sub-regions in the 2nd, 3rd, ..., 1st year of history The maximum value of the annual electricity consumption change rate:

[0098]

[0099] in, Indicates the first The maximum electricity consumption change rate corresponding to each sub-region;

[0100] Step Two Four: Calculate the... Skewness of electricity consumption change rate corresponding to each sub-region:

[0101]

[0102] in, Represents the processed historical electricity consumption sequence The standard deviation of all data in the middle. Indicates the first Skewness of electricity consumption change rate corresponding to each sub-region;

[0103] Step 25: Determine the location information. The one-hot encoded vector of the nth sub-region, using the nth... The processed historical electricity consumption sequence corresponding to each sub-region , , and the The one-hot encoded vectors of the sub-regions constitute the first Feature vectors of each sub-region ;

[0104] Specifically, no. The one-hot encoded vector for each sub-region is a four-dimensional vector. To reflect industry relevance in the electricity load forecasting process, this invention can determine the types of electricity consumption within a sub-region based on the geographical location information it covers. This invention categorizes electricity consumption types into residential electricity consumption areas, industrial electricity consumption areas, commercial electricity consumption areas, and other types of electricity consumption areas (mainly covering public services and special electricity users, including schools, hospitals, transportation, etc.). The residential electricity consumption area, industrial electricity consumption area, commercial electricity consumption area, and other types of electricity consumption areas are defined in the [missing information - likely a specific section or document]. In the One-hot encoded vectors of each sub-region, the first, second, third, and fourth elements are respectively assigned a value of 1 for the element corresponding to the electricity consumption type with the highest electrical load in the sub-region, and all other element values ​​are set to 0. Figure 2 The figure shows the percentage of electricity consumption for each type within a certain sub-region.

[0105] By introducing a one-hot encoded vector into the feature vector of a sub-region, the electricity consumption type of the sub-region can be fully considered in the subsequent clustering process, because the electricity load change trends of two sub-regions dominated by the same electricity consumption type are more likely to be similar.

[0106] Step 3: Perform initial clustering on the sub-regions based on the initial feature vectors of each sub-region to obtain the initial clustering results of the sub-regions. The set of geographic location indexes of the sub-regions contained in the initial clusters Recorded as:

[0107] ;

[0108] in:

[0109] Indicates the first The geographic location index of the first sub-region in the initial cluster. Indicates responsibility for the first The geographical longitude of the local power supply system for the first sub-region in the initial cluster. Indicates responsibility for the first The geographical latitude of the local power supply system for the first sub-region in the initial cluster;

[0110] Indicates the first The geographic location index of the second sub-region in the initial cluster. Indicates responsibility for the first The geographical longitude of the local power supply system for the second sub-region in the initial cluster. Indicates responsibility for the first The geographical latitude of the local power supply system for the second sub-region in the initial cluster;

[0111] Indicates the first In the initial clustering, the th Geographical index of each sub-region Indicates responsibility for the first In the initial clustering, the th The geographical longitude of the local power supply system for each sub-region, Indicates responsibility for the first In the initial clustering, the th The geographical latitude of the local power supply system for each sub-region;

[0112] Indicates the first The number of subregions contained in each initial cluster. , Indicates the total number of categories;

[0113] Specifically, the initial clustering method is as follows:

[0114] Step 3. Initialization Each gray wolf contains one individual gray wolf. There are 1 initial cluster center, and each initial cluster center is a vector with the same dimension as the feature vector of the sub-region;

[0115] Step 3.2: Initialize the number of iterations ;

[0116] Step 3: Calculate the number of... The objective function value corresponding to each individual gray wolf in each iteration;

[0117] For the The first of the individual gray wolves Cluster centers Calculate the first one respectively Feature vectors of each sub-region and The Euclidean distance is then the first... Each subregion belongs to the cluster center corresponding to the maximum Euclidean distance;

[0118] After processing the feature vectors of each sub-region in turn, the first... Objective function value for each individual gray wolf:

[0119]

[0120] in, Indicates the first Of the individual gray wolves, the one belonging to the first... Cluster centers The set of subregions; Indicates the first The objective function value of each individual gray wolf. express and Euclidean distance;

[0121] Steps three and four, from the first In the next iteration, the top three gray wolf individuals with the largest objective function values ​​are selected, and the others... Each individual gray wolf is updated based on the three selected gray wolf individuals;

[0122] The three selected gray wolf individuals and the updated The first gray wolf individual as the first Each individual gray wolf in the next iteration;

[0123] Step 35: Determine if the conditions are met. , The maximum number of iterations is set.

[0124] If satisfied Then from the first In the next iteration, select the gray wolf with the largest objective function value (i.e., directly obtain it from the calculation result of step 33), and then execute step 36;

[0125] If not satisfied Then let Return to step three;

[0126] Step 36: Select the gray wolf individuals from Step 35. Using the first cluster center as the initial cluster center, and then according to... The initial clustering results are obtained from the initial cluster centers, and the set of geographic location indexes of the sub-regions contained in each cluster is output.

[0127] Step 4: Generate a final cluster center for each group of initial clustering results, obtain the final clustering results based on each final cluster center, output the sub-regions contained in each final cluster, and record the geographic location index of all sub-regions contained in each final cluster.

[0128] Specifically, for the obtained first Initial clustering results for the groups:

[0129] Step 41: Construct the first The first group in the initial clustering results The feature vector corresponding to the geographic location index of each sub-region eigenvectors , ;

[0130] Initialize the number of iterations ;

[0131] Step 42: Extract separately The first in The extracted elements are then sorted in ascending order to obtain the sorted result sequence. ;

[0132] Using a fixed-length window from the sequence The starting point begins in the sequence By sliding upwards, the importance index value of the data within each window is calculated. In this invention, the window length is set to... The window sliding step size is The window length and sliding step size can both be set according to the actual situation; for sequences Select the sequence The window position corresponding to the most important indicator value is selected, and the average value of all data within the selected window is calculated. The calculated average value is denoted as... ;

[0133] It should be noted that the importance index value of the data within each window is calculated as follows:

[0134] Step 421: Divide the electricity consumption change rate range, for example, by... As the first interval, As the second interval, As the third interval, As the fourth interval, As the fifth interval, and so on, more intervals can be set according to the actual situation of the area to be predicted;

[0135] Step 422: Calculate the importance index values ​​of the data in the current window:

[0136]

[0137] in, This indicates the importance index value of the data within the current window; Indicates the number of elements in the current window. One data point; Indicates in sequence In, with The number of data points belonging to the same range of electricity consumption change rates. Indicates the number of data items in the current window;

[0138] Step 43: Determine if the conditions are met. (i.e., for the eigenvector) The front of the middle (The elements are processed using the same method).

[0139] If satisfied Then proceed to step four;

[0140] If not satisfied Then let Return to step four two;

[0141] Steps four and five: As the first The final cluster center.

[0142] After determining the final cluster centers, for any sub-region, the corresponding feature vector... We can calculate the eigenvectors separately. The feature vector of each sub-region is determined by its Euclidean distance from the final cluster center. If a sub-region's feature vector is the smallest Euclidean distance from the final cluster center, then that sub-region belongs to the final cluster center with the smallest Euclidean distance. Then, the final classification result can be determined.

[0143] The above process of the present invention can ensure that sub-regions with similar electricity load change trends are divided into a cluster as much as possible. We record the geographic location index set of each sub-region belonging to the final cluster center. Sub-regions within the same geographic location index set can be predicted based on the same final cluster center, which greatly reduces the amount of computation required for the prediction process.

[0144] Step 5: Predict the rate of change of electricity load for each sub-region in the final cluster;

[0145] For the The final cluster is extracted. The historical electricity consumption data of each sub-region corresponding to the geographic location index included in the final cluster, based on the... The predicted electricity load rate of change for each sub-region is obtained by combining the predicted results of the final clustering with the extracted historical electricity consumption data.

[0146] Specifically, for the first The final clusters:

[0147] Step 51, place the first The final cluster center is used as input to the electricity load change rate prediction model to obtain the result for the _th Prediction results of electricity load change rate for each cluster;

[0148] The working process of the electricity load change rate prediction model is as follows:

[0149] The first The final cluster centers serve as inputs to two parallel branches, where the first branch includes a TCN network and the second branch includes a BiLSTM network.

[0150] The outputs of the two branches are fused along the channel dimension, and the fused result is used as the input of a fully connected neural network. The fully connected neural network then outputs the predicted result of the electricity load change rate.

[0151] Step 52: Record the predicted electricity load change rate obtained in Step 51 as... ,in, They represent the first year, the second year, ..., the third year, respectively. Annual electricity load change rate;

[0152] For the The th final cluster in the th The sub-region corresponding to the geographic location index is calculated. The sub-regions in the future Annual electricity load forecast results;

[0153]

[0154] in, Indicates the first The sub-regions in the future Annual electricity load forecast results Indicates the first The sub-regions in the future Annual electricity load forecast results;

[0155] Similarly, calculate the first... The electricity load prediction results for the sub-regions corresponding to each geographic location index in the final cluster.

[0156] When predicting the power load of each sub-region, the power supply system corresponding to the geographical location index of the sub-region is determined, and then it is judged whether the power supply capacity of the power supply system can meet the requirements. If it does not meet the requirements, the power supply system needs to be upgraded.

[0157] Experimental Section

[0158] The electricity load forecast for each sub-region within region A was performed using the method of this invention. The forecast results are shown in Table 1.

[0159] Table 1

[0160]

[0161] In Table 1, the data from 1999 to 2005 are actual data, while the data for 2006 and 2007 are predicted data. The unit of the data in Table 1 is 100 million kilowatt-hours. Comparing the predicted results with the actual values ​​shows that the method of the present invention can effectively predict electricity load. Moreover, different numbers of sub-regions can be assigned to different areas to be measured.

[0162] The above examples of the present invention are merely illustrative of the computational model and process of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is impossible to exhaustively list all possible implementations here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting electricity load based on geographic information and neural networks, characterized in that, The method specifically includes the following steps: Step 1: Treat the power supply area of ​​each local power supply system in the area to be predicted as a sub-region, and obtain the geographic coordinates of each local power supply system from the geographic information system. Use the obtained geographic coordinates as the geographic index of the corresponding sub-region. And obtain the historical electricity consumption data for each sub-region separately, for the first The sub-region will be the first Each sub-region in history Annual electricity consumption is denoted as , will the The sequence of electricity consumption data for each sub-region in historical years is denoted as: ; in, Indicates the first Electricity consumption in each sub-region in its first year of history Indicates the first Electricity consumption in each sub-region in the second year in history Indicates the first Each sub-region in history Annual electricity consumption; Step 2: Process the historical electricity consumption data sequence for each sub-region to obtain the processed historical electricity consumption sequence for each sub-region, and construct the initial feature vector for each sub-region based on the geographical location information of the sub-region and the processed historical electricity consumption sequence. Step 3: Perform initial clustering on the sub-regions based on the initial feature vectors of each sub-region to obtain the initial clustering results of the sub-regions. The set of geographic location indexes of the sub-regions contained in the initial clusters Recorded as: ； in: Indicates the first The geographic location index of the first sub-region in the initial cluster. Indicates responsibility for the first The geographical longitude of the local power supply system for the first sub-region in the initial cluster. Indicates responsibility for the first The geographical latitude of the local power supply system for the first sub-region in the initial cluster; Indicates the first The geographic location index of the second sub-region in the initial cluster. Indicates responsibility for the first The geographical longitude of the local power supply system for the second sub-region in the initial cluster. Indicates responsibility for the first The geographical latitude of the local power supply system for the second sub-region in the initial cluster; Indicates the first In the initial clustering, the th Geographical index of each sub-region Indicates responsibility for the first In the initial clustering, the th The geographical longitude of the local power supply system for each sub-region, Indicates responsibility for the first In the initial clustering, the th The geographical latitude of the local power supply system for each sub-region; Indicates the first The number of subregions contained in each initial cluster. , Indicates the total number of categories; Step 4: Generate a final cluster center for each group of initial clustering results, obtain the final clustering results based on each final cluster center, output the sub-regions contained in each final cluster, and record the geographic location index of all sub-regions contained in each final cluster. Step 5: Predict the rate of change of electricity load for each sub-region in the final cluster; For the The final cluster is extracted. The historical electricity consumption data of each sub-region corresponding to the geographic location index included in the final cluster, based on the... The predicted electricity load change rate of the final cluster and the extracted historical electricity consumption data are used to obtain the predicted electricity load of each sub-region.

2. The electricity load forecasting method based on geographic information and neural networks according to claim 1, characterized in that, The specific process of step two is as follows: Step 21, place the first The processed historical electricity consumption sequence corresponding to each sub-region is denoted as follows: , They represent the first The sub-regions in the 2nd, 3rd, ..., 1st year of history Annual electricity consumption change rate; Step 22: Calculate the first... The sub-regions in the 2nd, 3rd, ..., 1st year of history The average annual rate of change in electricity consumption: in, Indicates the first Average rate of change in electricity consumption in each sub-region; Steps two and three: Calculate the first... The sub-regions in the 2nd, 3rd, ..., 1st year of history The maximum value of the annual electricity consumption change rate: in, Indicates the first The maximum electricity consumption change rate corresponding to each sub-region; Step Two Four: Calculate the... Skewness of electricity consumption change rate corresponding to each sub-region: in, Represents the processed historical electricity consumption sequence The standard deviation of all data in the middle. Indicates the first Skewness of electricity consumption change rate corresponding to each sub-region; Step 25: Determine the location information. The one-hot encoded vector of the nth sub-region, using the nth... The processed historical electricity consumption sequence corresponding to each sub-region , , and the The one-hot encoded vectors of the sub-regions constitute the first Feature vectors of each sub-region .

3. The electricity load forecasting method based on geographic information and neural networks according to claim 2, characterized in that, The specific process of step 21 is as follows: in: Indicates the first Electricity consumption in each sub-region in the third year in history; Indicates the first Each sub-region in history Annual electricity consumption.

4. The electricity load forecasting method based on geographic information and neural networks according to claim 3, characterized in that, The initial clustering method is as follows: Step 3. Initialization Each gray wolf contains one individual gray wolf. There are 1 initial cluster center, and each initial cluster center is a vector with the same dimension as the feature vector of the sub-region; Step 3.2: Initialize the number of iterations ; Step 3: Calculate the number of... The objective function value corresponding to each individual gray wolf in each iteration; For the The first of the gray wolf individuals Cluster centers Calculate the first one respectively Feature vectors of each sub-region and The Euclidean distance is then the first... Each subregion belongs to the cluster center corresponding to the maximum Euclidean distance; After processing the feature vectors of each sub-region in turn, the first... Objective function value for each individual gray wolf: in, Indicates the first Of the individual gray wolves, the one belonging to the first... Cluster centers The set of subregions; Indicates the first The objective function value of each individual gray wolf. express and Euclidean distance; Steps three and four, from the first In the next iteration, the top three gray wolf individuals with the largest objective function values ​​are selected, and the others... Each individual gray wolf is updated based on the three selected gray wolf individuals; The three selected gray wolf individuals and the updated The first gray wolf individual as the first Each individual gray wolf in the next iteration; Step 35: Determine if the conditions are met. , The maximum number of iterations is set. If satisfied Then from the first In the next iteration, select the gray wolf with the largest objective function value from the gray wolf individuals, and then execute step three six. If not satisfied Then let Return to step three; Step 36: Select the gray wolf individuals from Step 35. Using the first cluster center as the initial cluster center, and then according to... The initial clustering results are obtained from the initial cluster centers, and the set of geographic location indexes of the sub-regions contained in each cluster is output.

5. The electricity load forecasting method based on geographic information and neural networks according to claim 4, characterized in that, The specific process of step four is as follows: For the obtained first Initial clustering results for the groups: Step 41: Construct the first The first group in the initial clustering results The feature vector corresponding to the geographic location index of each sub-region eigenvectors , ; Initialize the number of iterations ; Step 42: Extract separately The first in The extracted elements are then sorted in ascending order to obtain the sorted result sequence. ; Using a fixed-length window from the sequence The starting point begins in the sequence Swipe up to calculate the importance index value of the data in each window; for the sequence Select the sequence The window position corresponding to the most important indicator value is selected, and the average value of all data within the selected window is calculated. The calculated average value is denoted as... ; Step 43: Determine if the condition is met. ; If satisfied Then proceed to step four; If not satisfied Then let Return to step four two; Steps four and five: As the first The final cluster center.

6. The electricity load forecasting method based on geographic information and neural networks according to claim 5, characterized in that, The method for calculating the importance index value of data within each window is as follows: Step 421: Divide the range of electricity consumption change rate; Step 422: Calculate the importance index values ​​of the data in the current window: in, This indicates the importance index value of the data within the current window; Indicates the number of elements in the current window. One data point; Indicates in sequence In, with The number of data points belonging to the same range of electricity consumption change rates. This indicates the number of data items in the current window.

7. The electricity load forecasting method based on geographic information and neural networks according to claim 6, characterized in that, The specific process of step five is as follows: Step 51, place the first The final cluster center is used as input to the electricity load change rate prediction model to obtain the result for the 1st cluster. Prediction results of electricity load change rate for each cluster; Step 52: Record the predicted electricity load change rate obtained in Step 51 as... ,in, They represent the first year, the second year, ..., the third year, respectively. Annual electricity load change rate; For the The th final cluster in the th The sub-region corresponding to the _ geographic location index, calculate the _ ... The sub-regions in the future Annual electricity load forecast results; in, Indicates the first The sub-regions in the future Annual electricity load forecast results Indicates the first The sub-regions in the future Annual electricity load forecast results; Similarly, calculate the first... The electricity load prediction results for the sub-regions corresponding to each geographic location index in the final cluster.

8. The electricity load forecasting method based on geographic information and neural networks according to claim 7, characterized in that, The method further includes a sixth step: when predicting the predicted power load value for each sub-region, it is then determined whether the power supply capacity of the power supply system corresponding to the geographical location index of the sub-region can meet the requirements.

9. The electricity load forecasting method based on geographic information and neural networks according to claim 8, characterized in that, The working process of the electricity load change rate prediction model is as follows: The first The final cluster centers serve as inputs to two parallel branches, where the first branch includes a TCN network and the second branch includes a BiLSTM network. The outputs of the two branches are fused along the channel dimension, and the fused result is used as the input of a fully connected neural network. The fully connected neural network then outputs the predicted result of the electricity load change rate.

Images