Power load forecasting method and system based on hybrid attention mechanism

Abstract

The present invention proposes a hybrid attention mechanism power load forecasting method and system, including: obtaining data related to power load changes; analyzing the obtained archival information and power load sequence data characterizing power users to obtain customer grouping data with different power consumption behavior patterns; extracting features with a correlation degree greater than a preset correlation threshold with the power load operation changes of customer groups from the potential influencing factor sequence data, and extracting power consumption context features related to the power load changes of customers from the extracted features; constructing a power load forecasting model; and outputting the predicted value of the power load of the customer to be predicted. The present invention realizes the collaborative modeling of complex correlation relationships such as the mutual influence between features and the influence of time factors within features through the attention mechanism, so as to ensure the accuracy of power load operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of power information processing, and in particular to a power load forecasting method and system based on a hybrid attention mechanism. 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 advancement of power grid construction, large-scale grid disturbances always pose a threat to grid security. The successive proposals of elastic and resilient grids have put forward higher requirements for grid load control. Power companies need to have a deeper understanding of grid load characteristics in order to carry out load control and other tasks in a targeted manner.

[0004] Accurate prediction of power load has a very important impact and role in load control. However, the change of power load is affected by a variety of factors, and how to achieve accurate load prediction is a challenging task.

[0005] There are differences in electricity consumption patterns among different types of customers. For example, there are differences between residential and industrial electricity consumption patterns, there are also certain differences between residential electricity consumption in different communities, and there are also differences in electricity consumption in different industries. How to effectively distinguish these different types of customers is the premise and basis for accurate prediction of power load.

[0006] There is a correlation between the electricity consumption patterns of different types of customers. The electricity consumption patterns reflect the living and working conditions of residents within a certain range to a certain extent. The living and working conditions of residents are related, which is reflected in the electricity consumption of residents and industrial electricity. Their electricity consumption patterns are also related and mutually influenced to a certain extent. How to effectively extract the correlation between the electricity consumption patterns of these different types of customers is also an important issue in achieving accurate load forecasting.

[0007] Customers' electricity usage behavior is affected by various external factors, such as weather, seasons, holidays, etc. How to evaluate the impact of these factors on customers' electricity usage behavior is also an important issue that needs to be considered in power load forecasting. Summary of the invention

[0008] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a power load forecasting method and system based on a hybrid attention mechanism to ensure the accuracy of power load forecasting.

[0009] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0010] In a first aspect, a hybrid attention mechanism power load forecasting method is disclosed, comprising:

[0011] Obtaining data related to power load changes, including archival information characterizing power users, power load sequence data, and sequence data of potential factors affecting power load changes;

[0012] Analyze the archival information and power load sequence data of electricity users to obtain customer grouping data with different electricity consumption behavior patterns;

[0013] Extracting features whose correlation with the power load operation change of the customer group is greater than a preset correlation threshold from the potential influencing factor sequence data, and extracting power consumption context features related to the power load change of the customer from the extracted features;

[0014] According to different types of electricity users, the attention mechanism is used to analyze the mutual influence of different electricity consumption context features and build an electricity load prediction model.

[0015] The power load of the customer to be predicted, the power consumption information of other customers and the power consumption context information are input into the power load prediction model, and the predicted value of the power load of the customer to be predicted is obtained as output.

[0016] In a further technical solution, the sequence data of potential influencing factors affecting the change of power load includes temperature, wind force, haze, air pressure and holidays;

[0017] The potential influencing factor sequence data is vectorized and converted into a corresponding power consumption environment context vector. The power consumption environment context change within a time interval can constitute a context sequence.

[0018] In a further technical solution, the power load sequence data is a power load state change vector of the power customer formed within a time interval;

[0019] The power load state change vector and the power consumption environment context vector are normalized.

[0020] A further technical solution is to analyze the acquired data to obtain customer grouping data with different electricity consumption behavior patterns, specifically:

[0021] Obtain the power load vectors of different customers and set the number of groups;

[0022] A clustering algorithm is used to obtain multiple customer groups, and all customer power load vectors in each group are averaged to obtain an average power load change vector corresponding to each customer group.

[0023] A further technical solution is to extract the power usage context features related to the customer's power load changes, specifically:

[0024] For the average power load change vector of multiple customer groups within a time interval, obtain the change of each potential influencing factor;

[0025] Given a set containing multiple groups, for each sample, calculate its correlation with different influencing factors in the electricity consumption context features;

[0026] Set a threshold, extract influencing factors whose Pearson correlation coefficient exceeds the threshold, and use the influencing factors whose correlation statistics are greater than the threshold as screening features.

[0027] A further technical solution is to construct a power load forecasting model as follows:

[0028] The average power load vector of the customer group, the power load vector of the specific customer, and the power usage context load vector are respectively embedded and represented to obtain corresponding vector representations;

[0029] Perform hidden state learning on the various learned embeddings to obtain the hidden state representation of each embedded representation;

[0030] Concatenate the latent states of each customer group embedding representation into a group joint representation;

[0031] Through the self-attention mechanism, learn the changing rules of the customer's power load sequence data itself and its weight impact; learn the impact of all customer grouping power consumption patterns on the customer's power load; learn the impact of power consumption context on customer x's power load;

[0032] Comprehensively consider the changing patterns of power load of power users, the mutual influence between users, and the impact of power consumption context on customer power consumption, and make predictions through the full link layer to obtain prediction results;

[0033] The predicted results are compared with the actual results, and the iteration is continued until the model converges or the maximum number of training times is reached to obtain the power load prediction model.

[0034] In a second aspect, a hybrid attention mechanism power load forecasting system is disclosed, comprising:

[0035] The customer grouping module is configured to: obtain data related to power load changes, including archive information characterizing power users, power load sequence data, and sequence data of potential influencing factors affecting power load changes;

[0036] Analyze the archival information and power load sequence data of electricity users to obtain customer grouping data with different electricity consumption behavior patterns;

[0037] The feature extraction module is configured to: extract features whose correlation with the operation change of the power load of the customer group is greater than a preset correlation threshold from the potential influencing factor sequence data, and extract power consumption context features related to the change of the power load of the customer from the extracted features;

[0038] The power load forecasting training module integrating the attention mechanism is configured as follows: for different types of power customer groups, the attention mechanism is used to analyze the mutual influence of different power consumption context features and build a power load forecasting model;

[0039] The customer power load prediction module is configured to: input the power load of the customer to be predicted, the power consumption information of other customers and the power consumption context information into the power load prediction model, and output the predicted value of the power load of the customer to be predicted.

[0040] One or more of the above technical solutions have the following beneficial effects:

[0041] The present invention automatically extracts features related to customer power load changes through an attention-based power load forecasting method, and through the attention mechanism, realizes collaborative modeling of complex correlations such as the mutual influence between features and the influence of time factors within features, thereby ensuring the accuracy of power load operation.

[0042] The present invention analyzes the changes in customer power load, extracts features related to the changes in customer power load operation, and uses them as influencing factors of power load operation to better characterize the actual situation of customer power load changes.

[0043] The present invention integrates the attention mechanism to explore the complex correlations such as the mutual influence between power load changes and the influence of factors within features, thereby improving the accuracy of power load prediction.

[0044] 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

[0045] 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.

[0046] Figure 1 A schematic diagram of a power load forecasting method based on a hybrid attention mechanism provided in an embodiment of the present disclosure;

[0047] Figure 2 An overall flow chart of a power load forecasting method based on a hybrid attention mechanism provided in an embodiment of the present disclosure;

[0048] Figure 3 A flow chart for extracting features associated with power load changes provided in an embodiment of the present disclosure;

[0049] Figure 4 A diagram of a power load forecasting model based on a hybrid attention mechanism provided for an example of the present disclosure;

[0050] Figure 5 A schematic diagram of the structure of an electric load prediction device based on a hybrid attention mechanism provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0051] 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.

[0052] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0053] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0054] Embodiment 1

[0055] See attached Figure 1 , 2 As shown, this embodiment discloses a method for predicting power load based on a hybrid attention mechanism, firstly obtaining archive data of power customers, power load sequence data of user customers, and sequence data of potential influencing factors affecting power load changes, and processing them;

[0056] Analyze the acquired user profile data and user power load sequence data to obtain customer groups with different power consumption behavior patterns;

[0057] Extract features whose correlation with the power load operation change of the customer group is greater than a preset correlation threshold, and extract power consumption context features related to the power load change of the customer;

[0058] We group different types of electricity users and use the attention mechanism to analyze their mutual influence in different electricity consumption contexts, and build a customer electricity load prediction model.

[0059] The power load of the customer to be predicted, the power consumption information of other customers and the power consumption context information are input into the power load prediction model, and the predicted value of the power load of the target customer is obtained as output.

[0060] Based on the archival data of electricity users and combined with the power load sequence data of users, users are grouped to obtain customer groups with different electricity consumption behavior patterns. Based on the grouping and combined with the attention mechanism, the similarity of power consumption patterns in different groups and the correlation of power consumption patterns in different groups are examined to make a more accurate prediction of the power load of electricity users.

[0061] The specific steps include:

[0062] Step A: Obtain relevant data on customer power load changes, including customer power load sequence data, customer profile data, power usage environment data, etc., and pre-process the acquired data.

[0063] Based on the acquired customer electricity consumption data and the electricity consumption data series data that potentially affects the customer's electricity load, data preprocessing is first performed, and then data normalization is performed.

[0064] The general process of data preprocessing is as follows:

[0065] A1. Analyze various factors that affect the change of power load. Based on experience, collect sequence data of potential factors affecting customers' electricity consumption from public data services such as weather information network, including but not limited to: temperature (Temp), wind (Wind), haze (Haze), air pressure (Air), humidity, holidays (Holiday), etc.

[0066] The collected data of the above-mentioned influencing factors are processed and converted into a vector of potential influencing factors of customer electricity consumption. The changes within a time interval can constitute a trajectory data.

[0067] Specifically, the present invention collects data on the influencing factors, vectorizes them, and converts them into corresponding power environment context vectors. i The power usage environment context vector can be identified as

[0068] Context ti =(Temp ti ,Wind ti ,Haze ti ,Air ti ,Holiday ti )

[0069] In a time interval [t i ,t j The power environment context changes within ] can constitute a context sequence, marked as Context ti～tj .

[0070] A2. Collection and preprocessing of customer power load change data.

[0071] When collecting customer electricity consumption data, we first obtain the customer's corresponding electricity load at a certain moment, and then construct the customer electricity load series data of the corresponding customer based on the customer's power load changes within a time interval.

[0072] For simplicity, for a certain electricity customer i, assume that at a certain time t i The generated power load is x i,ti , whose changes in a time interval can constitute the power load state change vector of the electricity customer:

[0073] x i,ti～tj ={x i,ti ,x i,ti+1 ,…,x i,tj}

[0074] The invention is to predict the power customer i at time t j+1 Power load x i,tj+1 .

[0075] The customer power load data vector and power consumption context vector are normalized to eliminate the deviation caused by inconsistent dimensions. The min-max normalization method is used for normalization, and the feature data value is [0,1].

[0076] Step B: Analyze the acquired data to obtain customer groups with different electricity consumption behavior patterns.

[0077] The general process of customer grouping is:

[0078] Analyze the electricity customer profile data and power load series data, and embed them into electricity customer representation vectors;

[0079] The electricity customer characterization vectors are clustered to obtain electricity customer groups.

[0080] Specifically:

[0081] B1. Obtain the power load vector x of different customers i i,ti～tj , the number of groups k is set based on experience or manually.

[0082] B2. Use the k-means clustering algorithm to obtain K customer groups, where the number of customers in each customer group is n i ,1≤i≤K, all customer power load vectors in each group are averaged, and the average power load change vector corresponding to each customer group is obtained as follows:

[0083] X i,ti～tj ={X i,ti ,X i,ti+1 ,…,X i,tj}

[0084] Among them, 1≤i≤K, K is the number of customer groups.

[0085] Step C: extract relevant features to calculate relevant statistics, obtain screening features based on the calculated statistics, and obtain power consumption context features related to the customer's power load changes based on the screening features.

[0086] Features whose correlation with the operation change of the power load of the customer group is greater than a preset correlation threshold are extracted, and power consumption context features related to the change of the customer's power load are extracted.

[0087] For the customer power load sequence data within a given time interval and the sequence data of various potential influencing factors of the power consumption environment, the features that affect the changes in customer power load are selected based on Pearson correlation, the redundancy of features is eliminated and features with a high correlation with customer power load changes are extracted. During the customer's power consumption process, the correlation statistics between the selected features and the customer's power load change with time.

[0088] like Figure 3 As shown, the specific process of feature extraction is: normalizing the customer power load time series data;

[0089] For a set of power load sequence samples containing N customers, the correlation between each sample and different influencing factors in the context features is calculated based on Pearson correlation;

[0090] A threshold is set, and an influencing factor whose Pearson correlation coefficient exceeds the threshold is extracted, which is used as a screening feature. Based on the screening feature data, the extracted feature is obtained.

[0091] Specifically, the general process of extracting the features that affect the customer's power load changes is:

[0092] C1. For the time interval [t i ,t j The average power load change vector X of K customer groups in i,ti～tj (1≤i≤K), the changes of each potential influencing factor are Context ti～tj .

[0093] C2. Given a set of K groups {X 1,ti～tj ,X 2,ti～tj ,…, K,ti～tj}, for each sample X i,ti～tj , based on Pearson correlation calculation and electricity usage context feature Context ti～tj The correlation of different influencing factors in . The calculation formula is as follows:

[0094]

[0095] in, C3. Set the threshold γ, extract the influencing factors whose Pearson correlation coefficient exceeds the threshold γ, and use the influencing factors whose correlation statistics are greater than the threshold γ as the screening features, denoted as Cont ti ～tj .

[0096] Step D: Group different types of electricity users and use the hybrid attention mechanism to analyze their mutual influence in different electricity consumption contexts and build a customer electricity load prediction model, such as Figure 4 shown.

[0097] The process of building and training a customer power load forecasting model using the attention mechanism is:

[0098] According to the electricity customer grouping results, the embedded representation of each customer group is obtained, and the embedded representations of all customer groups are concatenated as the representation of the electricity customer group;

[0099] Embedding representation of electricity usage context features;

[0100] Embedding representation of customer power load data;

[0101] The attention mechanism is used to model the relationship between customer power load data and power usage context representation, power customer group representation, and customer group representation of the power customer, and then concatenate them.

[0102] The concatenation results obtained by the above-mentioned multiple attention mechanisms are input into the fully connected layer to obtain the predicted output;

[0103] Through continuous iterations, until the model converges or the maximum number of training times is reached.

[0104] The general process of building a customer power load forecasting model based on the hybrid attention mechanism is:

[0105] D1.Data embedding and processing.

[0106] D1.1. Data embedding. The Embedding layer is used to map feature data into index representations, and a feature-index mapping matrix A is constructed. Through the Embedding mapping, the average power load vector of the customer group, the power load vector of the specific customer, and the power consumption context load vector are embedded and represented to obtain the corresponding vector representations.

[0107] The embedding representation of the average power load vector of the kth (1≤i≤K) customer group is:

[0108] X′ k=(X′ k,ti ,X′ k,ti+1 ,…,X′ k,tj )=A T (X k,ti ,X k,ti+1 ,…,X k,tj )

[0109] The embedded representation of the power load vector of a specific customer i is:

[0110] x i ′=(x′ i,ti ,x′ i,ti+1 ,…,x′ i,tj )=A T (x i,ti ,x i,ti+1 ,…,x i,tj )

[0111] Embedded representation of power consumption context load vector

[0112] C=(C ti ,C ti+1 ,…,C tj )=A T {c ti ,c ti+1 ,…,c tj}

[0113] in,

[0114]

[0115] K is the dimension of the embedding layer, A∈R |C|·K Represents the feature mapping matrix, C is Cont ti～tj The number of relevant influencing factors.

[0116] D1.2 Hidden state learning: Based on Bi-GRU, hidden state learning is performed on various learned embeddings.

[0117] The hidden state representation of each embedded representation is:

[0118] The hidden state of the k-th (1≤i≤K) customer group embedding representation is:

[0119]

[0120] The hidden state of the power load embedding representation of specific customer i is:

[0121]

[0122] The hidden state of the power consumption context load vector embedding representation is:

[0123]

[0124] Among them, BiGRU is the hidden state generation function of bidirectional GRU.

[0125] D1.3 Grouped customer joint representation. The hidden states of each customer group embedding representation are concatenated into a group joint representation:

[0126]

[0127] Among them, K is the number of customer groups.

[0128] D2. Attention mechanism for power load sequence data. Through the self-attention mechanism, we learn the change law of customer x's power load sequence data itself and its weight influence, and get:

[0129]

[0130] D3. Attention mechanism for all groups. Through the Attention mechanism, we learn the impact of the electricity consumption patterns of all customer groups on the electricity load of customer i, and get:

[0131]

[0132] D4. Attention mechanism for power consumption context. The influence of power consumption context on the power load of customer x is learned through the Attention mechanism, and the following is obtained:

[0133]

[0134] D5. Global attention mechanism. The attention mechanism comprehensively considers the power load change law of power customer i, the mutual influence between users, and the impact of power consumption context on customer power consumption, and obtains the following:

[0135]

[0136] D6. Full-link layer prediction. Prediction is performed through the full-link layer, and the prediction result is

[0137]

[0138] D7. Compare the predicted results with the actual results and continue iterating until the model converges or the maximum number of training times MaxNum is reached.

[0139] Step E: Input the power load of the customer to be predicted, the power consumption information of other customers and the power consumption context information into the power load prediction model, and output the predicted value of the power consumption of the target customer.

[0140] The general process of power forecasting is:

[0141] E1. For the customer y to be predicted, obtain its power load vector x y,ti～tj and context sequence Cont ti～tj , from step B2, we get the group electricity consumption grouping vector set

[0142] {X 1,ti～tj ,X 2,ti～tj ,…,X K,ti～tj}(1≤i≤K, K is the number of customer groups).

[0143] E2. Input the data obtained in E1 into the power load forecasting model based on the hybrid attention mechanism to obtain customer y at time t j+1 The power load forecast value Embodiment 2

[0144] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0145] Embodiment 3

[0146] The purpose of this embodiment is to provide a computer-readable storage medium.

[0147] A computer-readable storage medium stores a computer program, which executes the steps of the above method when executed by a processor.

[0148] Embodiment 4

[0149] The purpose of this embodiment is to provide Figure 5 As shown, this embodiment provides a power load forecasting system evaluation device based on a hybrid attention mechanism, including:

[0150] The customer grouping module is configured to: perform cluster analysis based on customer profile data and electricity load conditions to obtain customer groups with similar electricity usage behaviors.

[0151] The electricity usage context extraction module is configured to crawl context variables related to customer power load changes through public services.

[0152] The feature extraction module is configured to: perform feature extraction based on the acquired data, and extract features whose correlation with the customer's power load change is greater than a preset correlation threshold.

[0153] The power load prediction training module integrating the attention mechanism is configured to mine the power load of power customers at the next moment based on customer grouping, power load changes, and power usage context.

[0154] The customer power load prediction module is configured to output the power load of a specific customer at the next moment by inputting power usage context, customer power load change status, etc.

[0155] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0156] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0157] 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 hybrid attention mechanism power load forecasting method, characterized in that: include: Obtain data related to power load changes, including archival information characterizing power customers, power load sequence data, and sequence data of potential factors affecting power load changes; The potential influencing factor sequence data affecting the change of power load includes temperature, wind force, haze, air pressure and holidays; Analyze the archival information and power load sequence data of electricity users to obtain customer grouping data with different electricity consumption behavior patterns; Extracting features whose correlation with the power load operation change of the customer group is greater than a preset correlation threshold from the potential influencing factor sequence data, and extracting power consumption context features related to the power load change of the customer from the extracted features; According to different types of electricity users, the attention mechanism is used to analyze the mutual influence of different electricity consumption context features and build an electricity load prediction model. Input the power load of the customer to be predicted, the power consumption information of other customers and the power consumption context information into the power load prediction model, and output the predicted value of the power load of the customer to be predicted; The acquired data is analyzed to obtain customer grouping data with different electricity consumption behavior patterns: Obtain the power load vectors of different customers and set the number of groups; Using a clustering algorithm, multiple customer groups are obtained, and all customer power load vectors in each group are averaged to obtain an average power load change vector corresponding to each customer group; Extract electricity usage context features related to customer power load changes: For the average power load change vector of multiple customer groups within a time interval, obtain the change of each potential influencing factor; Given a set containing multiple groups, for each sample, calculate its correlation with different influencing factors in the electricity consumption context features; Set a threshold, extract the influencing factors whose Pearson correlation coefficient exceeds the threshold, and use the influencing factors whose correlation statistics are greater than the threshold as screening features; The process of building a power load forecasting model is as follows: The average power load vector of the customer group, the power load vector of the specific customer, and the power usage context load vector are respectively embedded and represented to obtain corresponding vector representations; Perform hidden state learning on the various learned embeddings to obtain the hidden state representation of each embedded representation; Concatenate the latent states of each customer group embedding representation into a group joint representation; Through the self-attention mechanism, we learn the changing rules of the customer's power load sequence data and its weight impact; learn the impact of all customer grouping power consumption patterns on the customer's power load; learn the impact of power consumption context on the customer The impact of electricity load; Comprehensively consider the changing patterns of power load of power users, the mutual influence between users, and the impact of power consumption context on customer power consumption, and make predictions through the full link layer to obtain prediction results; The predicted results are compared with the actual results, and the iteration is continued until the model converges or the maximum number of training times is reached to obtain the power load prediction model.

2. The hybrid attention mechanism power load forecasting method according to claim 1, characterized in that: The potential influencing factor sequence data is vectorized and converted into a corresponding power consumption environment context vector. The power consumption environment context change within a time interval can constitute a context sequence.

3. A hybrid attention mechanism power load forecasting method as claimed in claim 2, characterized in that: The power load sequence data is a power load state change vector of an electricity customer formed within a time interval.

4. The hybrid attention mechanism power load forecasting method according to claim 3, characterized in that: The power load state change vector and the power consumption environment context vector are normalized.

5. A hybrid attention mechanism power load forecasting system, characterized in that: include: The customer grouping module is configured to: obtain data related to power load changes, including archive information characterizing power users, power load sequence data, and sequence data of potential influencing factors affecting power load changes; The potential influencing factor sequence data affecting the change of power load includes temperature, wind force, haze, air pressure and holidays; Analyze the archival information and power load sequence data of electricity users to obtain customer grouping data with different electricity consumption behavior patterns; The feature extraction module is configured to: extract features whose correlation with the operation change of the power load of the customer group is greater than a preset correlation threshold from the potential influencing factor sequence data, and extract power consumption context features related to the change of the power load of the customer from the extracted features; The power load forecasting training module integrating the attention mechanism is configured as follows: for different types of power customer groups, the attention mechanism is used to analyze the mutual influence of different power consumption context features and build a power load forecasting model; The customer power load prediction module is configured to: input the power load of the customer to be predicted, the power consumption information of other customers and the power consumption context information into the power load prediction model, and output the predicted value of the power load of the customer to be predicted; The acquired data is analyzed to obtain customer grouping data with different electricity consumption behavior patterns: Obtain the power load vectors of different customers and set the number of groups; Using a clustering algorithm, multiple customer groups are obtained, and all customer power load vectors in each group are averaged to obtain an average power load change vector corresponding to each customer group; Extract electricity usage context features related to customer power load changes: For the average power load change vector of multiple customer groups within a time interval, obtain the change of each potential influencing factor; Given a set containing multiple groups, for each sample, calculate its correlation with different influencing factors in the electricity consumption context features; Set a threshold, extract the influencing factors whose Pearson correlation coefficient exceeds the threshold, and use the influencing factors whose correlation statistics are greater than the threshold as screening features; The process of building a power load forecasting model is as follows: The average power load vector of the customer group, the power load vector of the specific customer, and the power usage context load vector are respectively embedded and represented to obtain corresponding vector representations; Perform hidden state learning on the various learned embeddings to obtain the hidden state representation of each embedded representation; Concatenate the latent states of each customer group embedding representation into a group joint representation; Through the self-attention mechanism, we learn the changing rules of the customer's power load sequence data and its weight impact; learn the impact of all customer grouping power consumption patterns on the customer's power load; learn the impact of power consumption context on the customer The impact of electricity load; Comprehensively consider the changing patterns of power load of power users, the mutual influence between users, and the impact of power consumption context on customer power consumption, and make predictions through the full link layer to obtain prediction results; The predicted results are compared with the actual results, and the iteration is continued until the model converges or the maximum number of training times is reached to obtain the power load prediction model.

6. A hybrid attention mechanism power load forecasting system as claimed in claim 5, characterized in that: The potential influencing factor sequence data is vectorized and converted into a corresponding power consumption environment context vector. The power consumption environment context change within a time interval can constitute a context sequence.

7. A hybrid attention mechanism power load forecasting system as claimed in claim 6, characterized in that: The power load sequence data is a power load state change vector of an electricity customer formed within a time interval.

8. A hybrid attention mechanism power load forecasting system as claimed in claim 7, characterized in that: The power load state change vector and the power consumption environment context vector are normalized.

9. A computing device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method described in any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 4 are performed.

Images