Multi-zone-area load probability prediction method and system considering time-space characteristics

Abstract

The invention belongs to the technical field of power load prediction, and discloses a multi-transformer-area load probability prediction method and system considering time-space characteristics, and the method comprises the steps: collecting the geographic position data, historical load data and load-related external data of transformer areas, and carrying out the preprocessing; constructing a graph structure by taking the transformer areas as nodes and taking electrical connection between the transformer areas as edges; performing feature extraction on the graph structure to obtain spatial features among the plurality of courts, and performing operation on the spatial features to obtain time features; carrying out dynamic fusion on the space and time features to obtain fused features, and carrying out load probability prediction based on the fused features; and defining a loss function, optimizing model parameters, and obtaining a trained load probability prediction model. According to the method, the actual physical space information and the time characteristic information are considered to predict the load of the multiple transformer areas, the uncertainty of the load is reflected in a probability prediction mode, and more comprehensive and reliable load information is provided for dispatching, planning and operation management of a power system.

Description

Technical Field

[0001] The present invention relates to the technical field of electric load prediction, and in particular to a method and system for predicting the load probability of multiple substations taking into account time and space characteristics. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous expansion of the power system and the widespread popularity of electric vehicles, the load characteristics of multi-area power grids have become increasingly complex. There are electrical connections and power interactions between multiple areas, and their loads are not only affected by the electrical equipment in their own area, but also by the correlation of other areas.

[0004] At the same time, load data has obvious time series characteristics, including periodicity, trend and uncertainty. Traditional load forecasting methods, such as those based on statistical models, are difficult to effectively capture the complex spatial correlation and time series characteristics between multiple substations. Although machine learning methods can process complex data to a certain extent, they are not deep enough in mining the spatial structure of multi-station data. Recurrent neural networks (RNNs) and their variants LSTMs in deep learning perform well in processing time series data and can learn the long-term dependencies of load data, but they do not consider the spatial relationships between substations in multi-station data enough. Graph convolutional networks (GCNs) are good at processing data with graph structures and can mine the actual spatial associations between multiple substations, but most of them do not fully integrate actual spatial information, resulting in the model's inaccurate capture of spatial correlations.

[0005] At present, some load forecasting methods based on deep learning have been proposed. However, some methods cannot fully integrate the data of multiple substations and the actual spatial and temporal characteristics in the model structure, resulting in limited prediction accuracy, and less consideration of the uncertainty of load forecasting, which cannot provide sufficient risk information for power system decision-making. Therefore, a probabilistic forecasting method is needed that can comprehensively consider the actual spatial structure of multiple substations, the differences in load characteristics in different regions, and the uncertainty of load. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a multi-station load probability prediction method and system considering the temporal and spatial characteristics, integrating the advantages of multiple models, considering the actual physical space information and time characteristic information to realize the prediction of multi-station load, reflecting the uncertainty of load through probabilistic prediction, and providing more comprehensive and reliable load information for the dispatching, planning and operation management of the power system.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for predicting load probability of multiple substations considering spatiotemporal characteristics, comprising the following steps: Collect geographical location data, historical load data and external data related to load for each substation, and pre-process the collected data; Based on the preprocessed data, each substation is used as a node, and the electrical connections between substations are used as edges to construct a graph structure; Extract features from graph structure data to obtain spatial features between multiple stations, and input the spatial features into the LSTM layer in chronological order to obtain temporal features; Dynamically fuse spatial features and temporal features to obtain fused features, and perform load probability prediction based on the fused features; Define the loss function, optimize the model parameters, and obtain the trained multi-zone load probability prediction model.

[0008] As an optional implementation, the collected data is preprocessed, specifically: The collected data is cleaned, outliers and missing values ​​are removed, and the data is normalized. The normalized data is divided into multiple time windows in chronological order, and the data in each time window is taken as a sample.

[0009] As an optional implementation, feature extraction is performed on the graph structure data, specifically: multiple stacked GCN layers are used to perform linear transformation and nonlinear activation function ReLU processing on the node features and adjacency matrix of the graph structure to obtain the spatial features between multiple areas.

[0010] As an optional implementation, the adjacency matrix is ​​constructed according to the physical connection relationship and power transmission characteristics between the substations, and the element values ​​in the adjacency matrix are set according to the line impedance and power transmission capacity between the substations.

[0011] As an optional implementation, the node features of the graph structure are the geographical location information, load and related data of each substation.

[0012] As an optional implementation, a cross-gating block is used to dynamically fuse spatial features and temporal features. The cross-gating block consists of two gating units, which are used to fuse the spatial features extracted by GCN and the time series features learned by LSTM respectively. The gating signal is generated by calculating the weighted sum of the input features, and the spatial and temporal features are dynamically fused according to the gating signal.

[0013] In a second aspect, the present invention provides a multi-station load probability prediction system considering time and space characteristics, comprising: The data collection and preprocessing module is configured to: collect geographical location data, historical load data and external data related to the load of each substation, and preprocess the collected data; The graph structure building module is configured to: construct a graph structure based on the preprocessed data, taking each station area as a node and the electrical connections between the stations as edges; The feature extraction module is configured to: extract features from the graph structure data to obtain spatial features between multiple zones, and input the spatial features into the LSTM layer in time sequence to obtain time features; The load probability prediction module is configured to: dynamically fuse the spatial features and the temporal features to obtain the fused features, and perform load probability prediction based on the fused features; The model training module is configured to: define the loss function, optimize the model parameters, and obtain the trained multi-station load probability prediction model.

[0014] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0016] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a multi-station load probability prediction method and system considering spatiotemporal characteristics, constructs a CG-GCN-LSTM model, gives full play to the advantages of GCN mining multi-station spatial correlation, LSTM learning time series features and cross-gating block dynamic fusion of spatiotemporal features, can more comprehensively and accurately capture the complex characteristics of the actual space and load data of multiple stations, and significantly improves the accuracy of load prediction compared with a single model or a simple combination model. By constructing reasonable graph structure data and adjacency matrix, the physical connection and power transmission relationship of the multi-station power grid are effectively reflected, so that the model can better learn the actual spatial dependence and data spatial dependence between stations, and provide more accurate spatial information support for load prediction. The introduction of cross-gating blocks realizes the adaptive fusion of spatial and temporal features, can flexibly adjust the feature fusion strategy according to different load data characteristics and prediction tasks, improves the adaptability and robustness of the model, and can still maintain good prediction performance in the face of data noise and distribution changes. The model finally performs probabilistic prediction, which can stably and reliably consider different load differences and load uncertainty in practical applications to perform multi-station load prediction, providing a strong guarantee for the scientific dispatch and efficient operation of the power system.

[0018] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0020] Figure 1 A flow chart of a method for predicting load probability of multiple substations considering time and space characteristics provided in Example 1 of the present invention; Figure 2 It is a schematic diagram of the spatiotemporal model structure of charging piles in multiple areas; Figure 3 This is a structural diagram of the CG-GCN-LSTM model. DETAILED DESCRIPTION

[0021] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0023] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0025] Example 1 like Figure 1 As shown, this embodiment provides a method for predicting the load probability of multiple substations considering the temporal and spatial characteristics, including the following steps: S1. Collect geographical location data, historical load data and external data related to load of each substation, and pre-process the collected data; S2. Based on the preprocessed data, each substation is used as a node and the electrical connections between substations are used as edges to construct a graph structure; S3, extract features from the graph structure data to obtain spatial features between multiple stations, and input the spatial features into the LSTM layer in chronological order to obtain temporal features; S4, dynamically fuse the spatial features and the temporal features to obtain fused features, and perform load probability prediction based on the fused features; S5. Define the loss function, optimize the model parameters, and obtain the trained multi-station load probability prediction model.

[0026] Collect the geographical location (latitude and longitude) information of each substation, and use smart meters, sensors and other equipment to collect historical load data of each substation, including active power, reactive power, voltage and other information of each substation. At the same time, collect external data related to load, such as meteorological data (temperature, humidity, wind speed), date type (working day, holiday). Clean the collected data to remove outliers and missing values. For outliers, use statistical methods (3σ principle) to identify and correct them; for missing values, use linear interpolation method to fill them according to the time series characteristics and correlation of the data. After that, normalize the data and map all data to the [0, 1] interval.

[0027] Use the normalization method to normalize all data: ; Among them, x is the original data, and are the minimum and maximum values ​​of the data feature, respectively. is the normalized data.

[0028] The normalized data is divided into multiple time windows in chronological order. The data in each time window is taken as a sample. Each sample contains the load data of multiple stations in the time window and related external data. These samples are constructed into graph structure data as the input of subsequent models. The graph structure is constructed as follows: The multi-area power grid is regarded as a graph structure, in which each area is a node of the graph and the electrical connection between areas is an edge. According to the physical connection relationship and power transmission characteristics between areas, an adjacency matrix is ​​constructed to describe the spatial correlation strength between areas. The element values ​​in the adjacency matrix can be set according to factors such as line impedance and power transmission capacity between areas. If there is a direct electrical connection between areas i and j and the connection is tight, the value of the corresponding element in the adjacency matrix is ​​relatively large; if the connection is weak or there is no direct connection, the value is small or 0.

[0029] Node information of graph structure: Each substation is regarded as a node of the graph. In addition to the traditional load and related data (such as active power, reactive power, voltage, etc.) as node features, the node information also incorporates the geographical location information of the substation and the longitude and latitude coordinates. This information is combined into a feature vector to represent each node. The eigenvector of [load, longitude, latitude].

[0030] Edge information and adjacency matrix: The construction of edges not only considers the electrical connection relationship between substations, but also combines their geographical distance. For two substations with electrical connection, the closer the geographical distance is, the greater the weight of the edge; otherwise, the smaller the weight is.

[0031] Adjacency Matrix It is used to describe the connection relationship and weight between nodes. and There is an electrical connection between them and the geographical distance is , we can define the adjacency matrix elements for: ; in and is an adjustable parameter that controls the influence of geographic distance on edge weights.

[0032] GCN is used to process the constructed graph structure data. GCN can automatically learn the spatial feature representation between multiple substations by performing convolution operations on the graph. The input of GCN is the node features of the graph (the load and related data of each substation) and the adjacency matrix. After stacking multiple GCN layers, higher-level spatial features are gradually extracted. In each layer of GCN, the output features of the layer are obtained by performing linear transformation and nonlinear activation function ReLU processing on the node features and adjacency matrix. These output features integrate the information of the substation itself and the associated information of adjacent substations, thereby effectively capturing the spatial correlation between multiple substations.

[0033] The input of the GCN layer includes the node feature matrix , adjacency matrix And the node location information matrix (Consists of the longitude and latitude information of the node). The first-layer GCN calculation is extended on the basis of traditional GCN, taking into account the actual spatial location information of the node. The calculation of the first-layer GCN can be expressed as: ; in, ; ; In the formula, is the normalized adjacency matrix, is the degree matrix (diagonal matrix), is the learnable weight matrix of the first layer, is the activation function.

[0034] The node feature matrix and node location information matrix Adding them together enables the model to consider both the load characteristics and spatial location characteristics of the nodes when performing feature propagation. In order to extract more advanced spatial features, multiple layers of GCN are stacked. The calculation of layer GCN is: ; in is the output of the previous layer, It is The learnable weight matrix of the layer. By stacking multiple layers of GCN, the model can gradually capture the synergy of load changes between multiple stations and the load distribution pattern in space.

[0035] In order to enhance the model's ability to capture spatial features, a spatial attention mechanism is introduced. The attention weight is calculated for each node Defining the query vector , the key vector .

[0036] ; The dot product similarity is: ; node The update characteristics are: ; in, and The nodes are and The initial eigenvector of and is a learnable weight matrix.

[0037] The spatial features output by the GCN layer are input to the LSTM layer in chronological order. LSTM has memory units and gating mechanisms, which can effectively learn the long-term temporal dependencies of load data. The LSTM layer processes the input sequence data step by step, and through the synergy of the forget gate, input gate, and output gate, it decides which information to retain, which information to update, and which information to output. During the processing, LSTM can remember the trend, periodicity and other characteristics of the load data, thereby effectively predicting future load changes. By stacking multiple layers of LSTM, the model's ability to learn complex time series features can be further enhanced.

[0038] The spatial features output by the GCN layer Arranged in chronological order, the input is fed into a network consisting of three LSTM layers. The number of hidden units in each LSTM layer is set to 64. In the first LSTM layer, the input sequence Passing the Gate of Oblivion , Input Gate , output gate and memory unit The calculation results in the output : ; ; ; ; ; in, is the weight matrix, is the bias vector, is the Sigmoid function, Indicates element-wise multiplication. The second and third LSTM layers process the output of the previous layer in the same way to further learn time series features.

[0039] After the GCN layer and the LSTM layer, the cross-gating block is introduced. This module consists of two gating units, which are used to fuse the spatial features extracted by GCN and the time series features learned by LSTM. One gating unit is responsible for controlling the information flow of spatial features, and the other gating unit is responsible for controlling the information flow of temporal features. The gating unit generates a gating signal by calculating the weighted sum of the input features, and dynamically fuses the spatial and temporal features according to the gating signal. The gating signal is obtained by performing linear transformation, nonlinear activation, and weighted summation on the spatial and temporal features. In this way, the cross-gating block can adaptively adjust the fusion ratio of spatial and temporal features according to different input data and prediction task requirements, so as to better play the advantages of the two features and improve the prediction accuracy.

[0040] The cross-gating block contains two gating units, which respectively And the temporal features of LSTM output (Output of the third layer LSTM) is processed. For the spatial feature gating unit, the gating signal is calculated : ; The fused spatial features: ; For the temporal feature gated unit, calculate the gating signal : ; Time characteristics after fusion: ; Features of the final fusion: ; After the feature output is fused by the cross-gating block, the deep-AR base is used to perform probability prediction. The probability prediction layer uses the Gaussian mixture model (GMM) method to map the fused features to the probability distribution of the load. For the Gaussian mixture model, the negative log-likelihood loss function is used, and the root mean square error (RMSE) is reported. Quantile loss normalization and ( ), and the continuous ranked probability score (CRPS) to measure the difference between the probability distribution predicted by the model and the true load value.

[0041] By inputting a Gaussian distribution and using the negative log-likelihood as the loss function: ; A DeepAR infrastructure with 3 layers of LSTM is used. To evaluate the performance, the root mean square error (RMSE), Quantile loss normalization and ( ), and the continuous ranked probability score (CRPS) (lower scores indicate better performance). The calculation formula is as follows: ; in, ， Indicates the predicted Percentile value.

[0042] For the predicted Gaussian distribution , The score is defined as: ; in, They are the probability density function (PDF) and cumulative distribution function (CDF) of the Gaussian distribution. , the CRPS score is defined as The average of the scores is: .

[0043] Example 2 This embodiment provides a multi-station area load probability prediction system considering time and space characteristics, including: The data collection and preprocessing module is configured to: collect geographical location data, historical load data and external data related to the load of each substation, and preprocess the collected data; The graph structure building module is configured to: construct a graph structure based on the preprocessed data, taking each station area as a node and the electrical connections between the stations as edges; The feature extraction module is configured to: extract features from the graph structure data to obtain spatial features between multiple zones, and input the spatial features into the LSTM layer in time sequence to obtain time features; The load probability prediction module is configured to: dynamically fuse the spatial features and the temporal features to obtain the fused features, and perform load probability prediction based on the fused features; The model training module is configured to: define the loss function, optimize the model parameters, and obtain the trained multi-station load probability prediction model.

[0044] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0045] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0046] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0047] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0048] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0049] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0050] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0051] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0052] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0053] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, and the like. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, and the like.

[0054] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0055] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A multi-station load probability forecasting method considering spatiotemporal characteristics, characterized in that: The following steps are involved: Collect geographical location data, historical load data and external data related to load for each substation, and pre-process the collected data; Based on the preprocessed data, each substation is used as a node, and the electrical connections between substations are used as edges to construct a graph structure; Extract features from graph structure data to obtain spatial features between multiple stations, and input the spatial features into the LSTM layer in chronological order to obtain temporal features; Dynamically fuse spatial features and temporal features to obtain fused features, and perform load probability prediction based on the fused features; Define the loss function, optimize the model parameters, and obtain the trained multi-zone load probability prediction model.

2. The method for predicting load probability of multiple substations considering time and space characteristics as claimed in claim 1, characterized in that: Preprocess the collected data, specifically: The collected data is cleaned, outliers and missing values ​​are removed, and the data is normalized. The normalized data is divided into multiple time windows in chronological order, and the data in each time window is taken as a sample.

3. The method for predicting load probability of multiple substations considering time and space characteristics as claimed in claim 1, characterized in that: Feature extraction is performed on graph structure data, specifically: multiple stacked GCN layers are used to perform linear transformation and nonlinear activation function ReLU processing on the node features and adjacency matrix of the graph structure to obtain the spatial features between multiple areas.

4. The method for predicting load probability of multiple substations considering time and space characteristics as claimed in claim 3, characterized in that: The adjacency matrix is ​​constructed according to the physical connection relationship and power transmission characteristics between the substations, and the element values ​​in the adjacency matrix are set according to the line impedance and power transmission capacity between the substations.

5. The method for predicting load probability of multiple substations considering time and space characteristics as claimed in claim 3, characterized in that: The node features of the graph structure are the geographical location information, load and related data of each substation.

6. The method for predicting load probability of multiple substations considering time and space characteristics as claimed in claim 1, characterized in that: The spatial features and temporal features are dynamically fused using a cross-gating block. The cross-gating block consists of two gating units, which are used to fuse the spatial features extracted by GCN and the time series features learned by LSTM respectively. The gating signal is generated by calculating the weighted sum of the input features, and the spatial and temporal features are dynamically fused according to the gating signal.

7. A multi-station load probability forecasting system considering time and space characteristics, characterized in that: include: The data collection and preprocessing module is configured to: collect geographical location data, historical load data and external data related to the load of each substation, and preprocess the collected data; The graph structure building module is configured to: construct a graph structure based on the preprocessed data, taking each station area as a node and the electrical connections between the stations as edges; The feature extraction module is configured to: extract features from the graph structure data to obtain spatial features between multiple zones, and input the spatial features into the LSTM layer in time sequence to obtain time features; The load probability prediction module is configured to: dynamically fuse the spatial features and the temporal features to obtain the fused features, and perform load probability prediction based on the fused features; The model training module is configured to: define the loss function, optimize the model parameters, and obtain the trained multi-station load probability prediction model.

8. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images