Space load prediction method based on DBSCAN and improved ConvLSTM

Abstract

The invention provides a spatial load prediction method based on DBSCAN and improved ConvLSTM, relates to the technical field of a power distribution network in a power system, and solves the problem that potential interaction relationships among different spatial distribution cells are insufficiently mined. The adopted scheme is as follows: establishing a power geographic information system according to information collected in a planning area; obtaining a class I cellular layer and a class II cellular layer; obtaining an abnormal value of the I-class cellular load data, and removing the abnormal value to form a reasonable maximum value of each I-class cellular load in a specified time period; according to the reasonable maximum value of the I-type cellular load, generating II-type cellular-based spatial load historical data; and strengthening the spatial load data features of the class-II cells, constructing a spatial load prediction model, and obtaining prediction data. According to the method, abnormal values can be effectively identified and eliminated, the SLF model which is oriented to the type-II cells and synchronously considers the spatial-temporal correlation of the cells is constructed, and the precision and generalization ability of the prediction model are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network in a power system, and in particular to a spatial load prediction method based on DBSCAN and improved ConvLSTM. Background Art

[0002] Spatial Load Forecasting (SLF) is the prediction of the magnitude and spatial distribution of future power loads. Scientific SLF is the basis and guarantee for the correct planning of power systems and the foundation for the economic operation of power systems. Determining the future power load magnitude and spatial distribution of the planned area can more accurately plan and design the substation layout and line channels of the power grid, thereby establishing a reasonable and reliable grid structure;

[0003] The current spatial load forecasting method mainly predicts the spatial power load based on Class I cells. Since Class I cells are generated according to the power supply range of power supply equipment or power supply facilities, with the grid connection and commissioning of new equipment or new facilities, as well as the load transfer of the handle line, the shape and size of Class I cells will change, and their load law will also change, which will have a serious adverse impact on SLF.

[0004] In addition, most relevant literature focuses on the spatial relationship between adjacent cells, lacks a concise joint prediction model that takes into account the temporal and spatial characteristics of cellular loads, and the exploration of potential interaction relationships between cells with different spatial distributions is not sufficient and needs to be further addressed. Summary of the invention

[0005] The purpose of the present invention is to provide a solution to the problem that the shape and size of type I cells will change with the development and construction of the power grid, which will have an adverse effect on spatial load forecasting, and the existing relevant prediction models are not sufficient in exploring the potential interaction relationship between type II cells with different spatial distributions, and to establish a scientific, reasonable, highly applicable and highly accurate spatial load forecasting method.

[0006] The present invention provides a spatial load prediction method based on DBSCAN and improved ConvLSTM, comprising:

[0007] According to the information collected in the planning area, the power geographic information system is established to obtain the type I cell layer and the type II cell layer;

[0008] An outlier detection method is constructed using the DBSCAN algorithm to obtain outliers of the Class I cell load data, and the reasonable maximum values ​​of the Class I cell loads formed by the outliers within a specified period of time are eliminated;

[0009] According to the reasonable maximum value of the class I cell load, generating spatial load data based on the class II cell;

[0010] The ConvLSTM algorithm is improved by using the linear rectification function and the hyperbolic tangent function, the spatial load data characteristics of the type II cells are strengthened, and a spatial load prediction model is constructed;

[0011] According to the spatial load prediction model, prediction data is obtained.

[0012] Optionally, as one embodiment, the power geographic information system is established based on the information collected in the planning area to obtain a type I cell layer divided by the feeder power consumption range and a type II cell layer determined by latitude and longitude and spatial resolution, specifically including:

[0013] Collect and integrate the grid structure, land use information, 10kV feeder power supply range and corresponding historical load data within the planning area;

[0014] In the power GIS environment, build a registration base map of the planning area;

[0015] In the registered base map, the land use types of the predicted area are divided into eight categories according to the regional planning scheme, namely residential area, commercial area, cultural and entertainment area, industrial area, administrative office area, green space, municipal facilities area and special land, and a land use information layer is constructed;

[0016] The type I cell layer and the type II cell layer are established in turn according to the power supply range of the 10kV feeder and the gridding technology.

[0017] Optionally, as one embodiment, the outlier detection method is constructed by using the DBSCAN algorithm to obtain the outlier value of the class I cell load data, specifically including:

[0018] Step 1: Refer to the KNN algorithm to select an appropriate neighborhood radius, that is, the neighborhood range of the load value of any type I cell at any time point, denoted as ε; through clustering comparison experiments, determine the critical number of data that can define the core load of type I cells within the neighborhood radius, denoted as M;

[0019] Step 2: Randomly select an unlabeled type I cell load data, denoted as x, and check whether the ε neighborhood of its load value contains a sufficient number of load data. If the number of load data is greater than or equal to M, x is determined to be core load data, denoted as P1; otherwise, x is determined to be non-core load data, denoted as P2;

[0020] Step 3: Add each load data in the load value neighborhood of P1 and P1 to the normal data cluster;

[0021] Step 4: If P2 is not within the ε neighborhood of the load value of P1, add P2 to the abnormal data cluster, otherwise, add it to the normal data cluster;

[0022] Step 5: Repeat step 2 to step 4 until each of the type I cell load data is marked;

[0023] Step 6: After the above steps, a complete normal data cluster and an abnormal data cluster are obtained, and the data in the abnormal data cluster are all abnormal values;

[0024] Optionally, as one embodiment, the reasonable maximum value of each type I cell load formed after removing the abnormal value within a specified time period specifically includes:

[0025] The abnormal values ​​of all the Class I cellular load data in the abnormal data cluster are processed, and the maximum load value after eliminating the abnormal values ​​within a specified time period is the reasonable maximum value of the Class I cellular load within the specified time period.

[0026] Optionally, as one embodiment, generating the spatial load data based on the type II cells according to the reasonable maximum value of the type I cell load specifically includes:

[0027] In the electric power GIS environment, the type I cellular load data is converted into type II cellular load data by using spatial electric power load gridding technology;

[0028] The position information of each of the class II cells is used as the horizontal and vertical coordinates, and the cellular load data is used as the value of the cell, and a spatial load spatiotemporal information graph depicted by the class II cells is constructed through an Access database.

[0029] Optionally, as one embodiment, the improving the ConvLSTM algorithm by using a linear rectification function and a hyperbolic tangent function specifically includes:

[0030] The steps of the ConvLSTM algorithm to introduce convolution operations on the LSTM architecture to replace the matrix multiplication are as follows:

[0031] Input Gate t ), using formula (1):

[0032]

[0033] in: is the Hadamard product operation; * is the convolution operation; σ is the activation function; i t is the state retention rate of the input gate; X t is the input tensor at time t; H t-1 is the hidden state tensor at time t-1; Ct-1 is the storage unit tensor at time t-1; W Xi , W Hi , W ci is the weight matrix of the input gate, b i is bias;

[0034] Forget Gate t ), using formula (2):

[0035]

[0036] Where: f t The state retention probability of the forget gate; W Ci , W Hf , W Xf are the weight matrices of the input gate and the forget gate, respectively, f is bias;

[0037] Output Gate t ), using formula (3):

[0038]

[0039] In the formula: o t is the output probability of the output gate at time t; W Xo , W Ho , W co is the weight matrix of the output gate, b o is bias;

[0040] Cell State Update (Cell State Update, C t ), using formula (4):

[0041]

[0042] Where: C t is the storage unit tensor at time t; tanh is the hyperbolic tangent function; W Xc , W Hc is the weight matrix of the state gate, b c is bias;

[0043] At the output hidden state H t Previously, ReLU was used to improve the hidden state of the original algorithm, and the hidden state update (Hidden State Update, H t ), using formula (5)

[0044]

[0045] Where: Ht is the hidden state tensor at time t.

[0046] The technical solution of the present invention has at least the following beneficial technical effects: the present invention takes power system planning as the research background, and aims at the problem that Class I cells do not match the actual power supply range and the problem that the relevant prediction model does not simultaneously consider the spatiotemporal information of cells, and proposes a SLF method based on DBSCAN and improved ConvLSTM, which has the following advantages: 1) The present invention can effectively identify and eliminate outliers, so as to accurately and efficiently determine the reasonable maximum value of each Class I cell load, and reduce the prediction error caused by outliers; 2) The present invention uses Class II cells whose cell shapes are determined by longitude and latitude to characterize spatial power loads and perform SLF, avoiding the problem that the shape and size of cells do not match the actual power grid construction level, and is more in line with the actual requirements of power system planning; 3) The present invention constructs an SLF model that simultaneously considers the spatiotemporal correlation of cells, jointly models the spatiotemporal characteristics of cells, uses convolution to implement a gating mechanism, and uses activation functions to enhance the power load data characteristics of the predicted data, thereby effectively improving the accuracy and generalization ability of the prediction model to improve the planning benefits of urban power grids. The solution of the present invention effectively identifies and eliminates outliers, thereby accurately and efficiently determining the reasonable maximum value of each Class I cell load, and at the same time constructs an SLF model for Class II cells that synchronously considers the spatiotemporal correlation of cells, thereby avoiding the problem of mismatch between the shape and size of cells and the actual degree of grid construction and effectively improving the accuracy and generalization ability of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the conventional technology, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 It is a basic principle diagram of the prediction method provided by an embodiment of the present invention;

[0049] Figure 2 The embodiment of the present invention provides a KNN distance curve;

[0050] Figure 3 The embodiment of the present invention provides the influence of different minimum data numbers on clustering categories;

[0051] Figure 4 The embodiment of the present invention provides DBSCAN clustering results;

[0052] Figure 5 The embodiment of the present invention provides a prediction model structure diagram;

[0053] Figure 6 The embodiment of the present invention provides a ReLU-ConvLSTM unit module structure;

[0054] Figure 7 The embodiment of the present invention provides the target year's quasi-measured value and the predicted value of each method;

[0055] Figure 8 The embodiment of the present invention provides error distribution of prediction results. DETAILED DESCRIPTION

[0056] 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.

[0057] A spatial load forecasting method based on the density-based spatial clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) and the improved convolutional long short-term memory network (Convolutional Long Short-Term Memory Network, ConvLSTM); the spatial power load characterized by type II cells is used for prediction, which can well overcome the limitation of the inconsistent shape of type I cells when describing spatial loads, and can effectively reduce the error of prediction results; based on ConvLSTM, a spatial load prediction model characterized by type II cells is constructed, which simultaneously considers the time correlation and spatial correlation of the cellular load, effectively improving the accuracy and generalization ability of the prediction model, and can provide stable and reliable data support for power system planning.

[0058] The present invention provides a spatial load prediction method based on DBSCAN and improved ConvLSTM, which includes the following contents:

[0059] 1) Establishment of power geographic information system

[0060] ① Collect and integrate the grid structure, land use information, 10kV feeder power supply range and corresponding historical load data within the planning area;

[0061] ② In the electric power geographic information system (GIS) environment, a registration base map of the planning area is constructed; in the registered base map, the land use types of the predicted area are divided into eight categories, namely residential area, commercial area, cultural and entertainment area, industrial area, administrative office area, green space, municipal facilities area and special land, according to the regional planning scheme, and a land use information layer is constructed. Then, the Class I and Class II cell layers are established in accordance with the power supply range of the 10kV feeder and the gridding technology;

[0062] 2) Use DBSCAN algorithm to build an outlier detection method

[0063] Step 1: Parameter setting: Refer to the KNN algorithm to select an appropriate neighborhood radius, that is, the neighborhood range of the load value of any type I cell at any time point, denoted as ε; through clustering comparison experiments, determine the critical number of data that can define the core load of type I cells within the neighborhood radius, denoted as M;

[0064] Step 2: Classification of Class I cellular load data: Randomly select an unlabeled Class I cellular load data (denoted as x), and check whether the ε neighborhood of its load value contains a sufficient number of load data. If the number of load data is greater than or equal to M, then x is determined to be core load data, denoted as P1; otherwise, x is determined to be non-core load data, denoted as P2;

[0065] Step 3: Form a normal data cluster: add each load data in the load value neighborhood of P1 and P2 to the normal data cluster;

[0066] Step 4: Forming an abnormal data cluster: If P2 is not in the ε neighborhood of the load value of P1, then P2 is added to the abnormal data cluster, otherwise, it is added to the normal data cluster;

[0067] Step 5: Repeat steps (2) to (4) until each type I cell load data has been labeled;

[0068] Step 6: Determination of abnormal values ​​of type I cell load: After the above steps, complete normal data clusters and abnormal data clusters are obtained, and the data in the abnormal data clusters are all abnormal values;

[0069] 3) Generate spatial load spatiotemporal information graph based on type II cells

[0070] ① Process the abnormal values ​​of all Class I cellular load data in the abnormal data cluster. The maximum load value after removing the abnormal values ​​in the specified time period is the reasonable maximum value of Class I cellular load in the time period;

[0071] ② In the power GIS environment, the spatial power load gridding technology is used to transform the type I cellular load data into type II cellular load data;

[0072] ③Use the position information of each type II cell as the horizontal and vertical coordinates and the value of the cell load data cell, and construct a spatial load spatiotemporal information diagram depicted by type II cells through the Access database;

[0073] 4) Improve the ConvLSTM algorithm

[0074] The ConvLSTM algorithm introduces convolution operations on the LSTM architecture to replace the matrix multiplication. The calculation steps are as follows:

[0075] ① Input Gate t ) is shown in formula (1):

[0076]

[0077] Where: is the Hadamard product operation; * is the convolution operation; σ is the activation function; i t is the state retention rate of the input gate; X t is the input tensor at time t; H t-1 is the hidden state tensor at time t-1; C t-1 is the storage unit tensor at time t-1; W Xi , W Hi , W ci is the weight matrix of the input gate, b i is bias;

[0078] ②Forget Gate t ) is shown in formula (2):

[0079]

[0080] Where: f t The state retention probability of the forget gate; W Ci , W Hf , W Xf are the weight matrices of the input gate and the forget gate, respectively, f is bias;

[0081] ③ Output Gate t ) is shown in formula (3):

[0082]

[0083] In the formula: o t is the output probability of the output gate at time t; W Xo , W Ho , W co is the weight matrix of the output gate, b ois bias;

[0084] ④ Cell State Update (Cell State Update, C t ) is shown in formula (4):

[0085]

[0086] Where: C t is the storage unit tensor at time t; tanh is the hyperbolic tangent function; W Xc , W Hc is the weight matrix of the state gate, b c is bias;

[0087] ⑤In the output hidden state H t Previously, ReLU was used to improve the hidden state of the original algorithm, and the hidden state update (Hidden State Update, H t ) is shown in formula (5)

[0088]

[0089] Where: H t is the hidden state tensor at time t;

[0090] 5) Build a spatial load prediction model

[0091] Firstly, the DBSCAN algorithm is used to remove the abnormal data of type I cellular load, and the gridding technology is used to generate type II cellular load data. Secondly, the convolution operation is used to extract the spatial characteristics of type II cellular load. Then, the linear rectification function (ReLU) function is used to improve ConvLSTM to form a ReLU-ConvLSTM unit module to enhance the load data characteristics. The type II cellular load value is determined by combining the historical state of the multi-unit module with the current input. Finally, tanh is used to output the prediction results of each month of the target year and the annual maximum value of each type II cellular load is taken to form the prediction result of the target year, thereby realizing the joint prediction of the amplitude and spatial distribution of future spatial power loads.

[0092] The following uses Figure 1-Figure 8 And examples are provided to further illustrate the present invention;

[0093] The present invention is a spatial load prediction method based on DBSCAN and improved ConvLSTM. The basic principle diagram is as follows: Figure 1 As shown, the example contents of the method include: basic data integration, establishment of power GIS, generation of type II cell load data, construction of spatial load forecasting model, comparative evaluation;

[0094] According to the spatial load forecasting method based on DBSCAN and improved ConvLSTM proposed in the present invention, taking some administrative regions of a city in Northeast my country as an example, using its historical load data values ​​from 2009 to 2013, the Class II cell load value of the region in 2014 is predicted; in order to verify the accuracy of the method proposed in the present invention, the prediction results of the method of the present invention are compared and analyzed with the prediction results of the traditional grey theory method, linear regression method, exponential smoothing method, LSTM method, and CNN method. The specific contents are as follows:

[0095] 1) Establishing power GIS

[0096] Some administrative districts of a city in Northeast my country were selected as the forecast area, and the historical load data of the area with a sampling interval of 5 minutes from 2009 to 2013 were used to predict its spatial power load in 2014.

[0097] First, the GIS of the area to be studied is established, and the land use types of the area to be predicted are divided into eight categories according to the regional planning scheme, namely residential area, commercial area, cultural and entertainment area, industrial area, administrative office area, green space, municipal facilities area and special land. Then, according to the power supply range of 10kV feeder and grid technology, Class I and Class II cell layers are established. Then, on this basis, the spatial load prediction of the spatial power load data of the area in 2014 is carried out.

[0098] 2) Data preprocessing

[0099] The present invention adopts the monthly maximum value of the historical cellular load for SLF, first clusters the loads of each Class I cellular on a monthly basis, removes the abnormal clusters after detecting the clustering results, and finally calculates the reasonable monthly maximum value of each Class I cellular load after correction. Taking the Changchun Line in 2013 as an example, the reasonable values ​​of ε and M in the outlier detection method are determined.

[0100] When determining the value of ε, the original data is deduplicated and then plotted. Figure 2 The KNN distance curve is shown in Figure 1, and the elbow point of the curve is debugged with actual data. Finally, a reasonable value of ε is selected as 0.015.

[0101] When the neighborhood radius has been determined, the value of M is determined by empirical method. When M is set to 1, each data can become core data alone, and effective outlier detection cannot be achieved; when M is set to 2, an additional data and itself can constitute a cluster, and the algorithm will mistakenly regard most load data as core data, and the accuracy will decrease. Therefore, M should be selected to be greater than or equal to 3. At the same time, when M increases, if the clustering categories also increase, it means that some of the connected load data between the original clusters are regarded as outliers; if the clustering categories decrease at this time, it means that the number of clusters is shrinking, and some of the original small clusters are marked as outliers, because many small clusters are completely regarded as outliers. In view of the relatively uniform distribution of power load data, M is generally taken within 15. The experimental results of specific clustering categories with different M are shown as follows. Figure 3 shown.

[0102] Although the minimum value in the broken line will divide the load data into small clusters, it will greatly improve the accuracy of outlier detection. The maximum value in the broken line usually leads to more outliers and even misjudges normal values ​​as outliers. Since the goal of this clustering is to select the reasonable maximum load value of 12 months from a large amount of load data, the accuracy of outlier detection is required to be high, that is, M takes the minimum value of 5 in the experiment. The final outlier detection result is as follows: Figure 4 shown.

[0103] Then, in GIS, the monthly reasonable maximum load time series of each type I cell load from 2009 to 2013 were converted into a total of 378 type II cells according to square grids of equal size with a side length of 0.3 km, and the monthly quasi-measured values ​​of type II cell load from 2009 to 2013 were obtained.

[0104] 3) Constructing a spatial load prediction model

[0105] The model of the spatial load prediction method of the present invention is composed of a ReLU-ConvLSTM unit module structure. The specific principle and structure of the model are as follows: Figure 5-Figure 6 As shown in Table 1, in order to determine the specific structure of the prediction model, the parameters are manually adjusted based on the principle of priority of important parameters. First, the learning rate, which has the greatest impact on the model performance and training process, is optimized, then the model capacity is optimized, and finally the model details are adjusted. The parameters are from large to small. The specific experimental data are shown in Table 1. It should be pointed out that when determining the hidden layer dimension, start with the three layers with the widest applicability, and finally determine the parameters that give the best model performance at this time. Then, using this parameter as an example, a comparative experiment is conducted at the second and fourth layers to determine the specific hidden layer dimension and size.

[0106] Table 1 Comparison of specific parameters of prediction models

[0107]

[0108]

[0109] Finally, it was determined that the prediction network consists of three layers of ConvLSTM unit modules, and the dimensions of the hidden state unit modules in each layer are 64, 64, and 128 respectively. The prediction network is finally output through a layer of two-dimensional convolution 1×1. The specific experimental parameters of the training are shown in Table 2.

[0110] Table 2 Specific parameter settings of the prediction model

[0111]

[0112]

[0113] 4) Error evaluation analysis

[0114] The maximum value of each Class II cell load in the 12-step forecast is taken to form the SLF result of Class II cells in the target year. In order to show the accuracy of the SLF method of the present invention, the results of gray theory, linear regression, exponential smoothing, LSTM, convolutional neural network (CNN) and other methods are compared and analyzed. The mean absolute error and mean relative error are used to evaluate the error of the SLF results of each method, as shown in Table 3.

[0115] Table 3 Overall prediction error of type II cell load in the target year

[0116]

[0117] In Table 3, the prediction results of algorithms such as LSTM and CNN have smaller errors, showing the superiority of deep learning algorithms in prediction. The LSTM method only considers the time series characteristics of the load in SLF, while the CNN method simultaneously considers the time and space characteristics of the load. Its prediction accuracy is significantly improved compared with the LSTM method, which shows the importance of the spatial distribution characteristics of the load to the prediction accuracy of SLF. The method of the present invention combines the superiority of LSTM in time series characteristics with the function of CNN to consider spatial distribution characteristics to form a joint modeling of SLF time and space characteristics. Only one model is used to synchronously consider the time and space characteristics of the load, so the prediction accuracy is higher. Finally, the prediction results of all methods are visualized to generate a spatial load map and compared, as shown in the figure. Figure 7 shown.

[0118] The error distribution histograms of various methods are shown in Figure 8As shown. As can be seen from the figure, the prediction error of the SLF method based on deep learning is generally small. Among them, the smaller prediction errors of CNN and the method of the present invention show that the SLF that simultaneously considers the spatiotemporal correlation between type II cells is not only reasonable in theory, but also has smaller errors in actual predictions, proving the importance of synchronously considering the spatiotemporal correlation of cells for the SLF method. Compared with the other five SLF methods, the SLF method of the present invention has a more unified and scientific nature in the specific prediction process because it jointly models the spatiotemporal characteristics of cells, and conforms to the actual cellular load transformation law, so it performs best in terms of MSE and MAE.

[0119] The specific embodiments used in the present invention have been used to fully illustrate the contents of the present invention, but are not limited to the present embodiments. Any obvious changes made by those skilled in the art based on the teachings of the present invention are within the scope of protection of the present invention.

[0120] The specific embodiments used in the present invention have been used to fully illustrate the contents of the present invention, but are not limited to the present embodiments. Any obvious changes made by those skilled in the art based on the teachings of the present invention are within the scope of protection of the present invention.

[0121] Obviously, the described embodiments are only some embodiments of the present invention, but not all 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.

[0122] It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and do not constitute a limitation of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. In addition, the appended claims of the present invention are intended to cover all changes and modifications that fall within the scope and boundaries of the appended claims, or the equivalent forms of such scope and boundaries.

Claims

1. A spatial load forecasting method based on DBSCAN and improved ConvLSTM, characterized in that: include: According to the information collected in the planning area, an electric power geographic information system is established to obtain a type I cell layer divided by the feeder power supply range and a type II cell layer determined by latitude and longitude and spatial resolution, wherein the type I cell is a power supply area divided according to the power supply range of power supply equipment and power supply units in the electric power geographic information system environment, and the type II cell is a power supply area divided by equal-sized grids in the electric power geographic information system environment; An outlier detection method is constructed using the DBSCAN algorithm to obtain the outlier values ​​of the load data of the class I cells, and after removing the outliers, a reasonable maximum value of the load of each class I cell within a specified period of time is formed; According to the reasonable maximum value of the class I cell load, generate the spatial load history data based on the class II cell; The ConvLSTM algorithm is improved by using the linear rectification function and the hyperbolic tangent function, the spatial load data characteristics of the type II cells are strengthened, and a spatial load prediction model is constructed; According to the spatial load prediction model, prediction data is obtained.

2. The method according to claim 1, characterized in that According to the information collected in the planning area, a power geographic information system is established to obtain a type I cell layer divided by the feeder power supply range and a type II cell layer determined by latitude and longitude and spatial resolution, which specifically includes: Collect and integrate the grid structure, land use information, 10kV feeder power supply range and corresponding historical load data within the planning area; In the power GIS environment, build a registration base map of the planning area; In the registered base map, the land use types of the predicted area are divided into eight categories according to the regional planning scheme, namely residential area, commercial area, cultural and entertainment area, industrial area, administrative office area, green space, municipal facilities area and special land, and a land use information layer is constructed; The type I cell layer and the type II cell layer are established in turn according to the power supply range of the 10kV feeder and the gridding technology.

3. The method according to claim 1, characterized in that The outlier detection method is constructed by using the DBSCAN algorithm to obtain the outlier value of the class I cell load data, specifically including: Step 1: Refer to the KNN algorithm to select an appropriate neighborhood radius, that is, the neighborhood range of the load value of any type I cell at any time point, denoted as ε; through clustering comparison experiments, determine the critical number of data that can define the core load of type I cells within the neighborhood radius, denoted as M; Step 2: Randomly select an unlabeled type I cell load data, denoted as x, and check whether the ε neighborhood of its load value contains a sufficient number of load data. If the number of load data is greater than or equal to M, x is determined to be core load data, denoted as P1; otherwise, x is determined to be non-core load data, denoted as P2; Step 3: Add each load data in the load value neighborhood of P1 and P1 to the normal data cluster; Step 4: If P2 is not in the ε neighborhood of the load value of P1, add P2 to the abnormal data cluster, otherwise, add it to the normal data cluster; Step 5: Repeat step 2 to step 4 until each of the type I cell load data is marked; Step 6: After the above steps, a complete normal data cluster and an abnormal data cluster are obtained, and the data in the abnormal data cluster are all abnormal values.

4. The method according to claim 3, characterized in that The reasonable maximum value of each type I cell load formed after removing the abnormal value within the specified time period specifically includes: The abnormal values ​​of all the Class I cellular load data in the abnormal data cluster are processed, and the maximum load value after eliminating the abnormal values ​​within a specified time period is the reasonable maximum value of the Class I cellular load within the specified time period.

5. The method according to claim 1, characterized in that The generating of the spatial load history data based on the type II cells according to the reasonable maximum value of the type I cell load specifically includes: In the electric power GIS environment, the type I cellular load data is converted into type II cellular load data by using spatial electric power load gridding technology; The position information of each of the class II cells is used as the horizontal and vertical coordinates, and the cellular load data is used as the value of the cell, and a spatial load spatiotemporal information graph depicted by the class II cells is constructed through an Access database.

6. The method according to claim 1, characterized in that The improved ConvLSTM algorithm using the linear rectification function and the hyperbolic tangent function specifically includes: The steps of the ConvLSTM algorithm to introduce convolution operations on the LSTM architecture to replace the matrix multiplication are as follows: Input Gate t ), using formula (1): in: is the Hadamard product operation; * is the convolution operation; σ is the activation function; i t is the state retention rate of the input gate; X t is the input tensor at time t; H t-1 is the hidden state tensor at time t-1; C t-1 is the storage unit tensor at time t-1; W Xi , W Hi , W ci is the weight matrix of the input gate, b i is bias; Forget Gate t ), using formula (2): Where: f t The state retention probability of the forget gate; W Ci , W Hf , W Xf are the weight matrices of the input gate and the forget gate, respectively, f is bias; Output Gate t ), using formula (3): Where: t is the output probability of the output gate at time t; W Xo , W Ho , W co is the weight matrix of the output gate, b o is bias; Cell State Update (Cell State Update, C t ), using formula (4): Where: C t is the storage unit tensor at time t; tanh is the hyperbolic tangent function; W Xc , W Hc is the weight matrix of the state gate, b c is bias; At the output hidden state H t Previously, ReLU was used to improve the hidden state of the original algorithm, and the hidden state update (HiddenState Update, H t ), using formula (5) Where: H t is the hidden state tensor at time t.

Images