User power consumption mode identification method and device considering multi-factor association and medium

Abstract

The invention relates to a user power consumption mode identification method and device considering multi-factor association and a medium. The method comprises the steps of obtaining historical load data and historical meteorological data and performing preprocessing; processing the historical load data and the historical meteorological data under different time resolutions, and constructing a load data set and a meteorological data set under different resolutions; constructing a load curve of each day based on the load data set, performing coarse clustering on the load curves by using a self-organizing mapping neural network, extracting a power utilization mode of the user, and identifying whether the user has an energy storage device or not; calculating the relevance between the meteorological factors and the load through correlation analysis, and selecting key factors; and based on the selected key factors, performing fine-grained analysis on the power consumption mode of the user by adopting a mode of combining a time convolutional network and a pyramid attention mechanism, and accurately identifying the power consumption mode. Compared with the prior art, the method provided by the invention can realize the identification of the power consumption mode of the user with coarse and fine granularity under the multi-resolution condition and multi-factor association.

Description

Technical Field

[0001] The present invention relates to the field of electricity usage behavior recognition, and in particular to a method, device and medium for recognizing a user's electricity usage pattern taking into account multi-factor associations. Background Art

[0002] With the transformation of the global energy landscape and the increase in the proportion of renewable energy, energy storage technology has gradually become a key link in the modern power system. Energy storage devices can not only balance the fluctuations in power supply, but also optimize users' electricity costs by shaving peaks and filling valleys, especially in the industrial and commercial fields. Users can charge during low periods and discharge during peak periods, thereby significantly reducing electricity costs and improving efficiency. However, the introduction of energy storage devices makes the electricity consumption pattern more complicated, and analyzing the electricity consumption behavior of users with energy storage devices has become an important issue in power dispatch optimization.

[0003] Traditional methods for analyzing electricity consumption patterns are mostly based on historical user load data, using static analysis with a single time resolution, and focusing mainly on the user's electricity consumption patterns in different time periods. This method works well for conventional users, but it is difficult to cope with its complexity in scenarios involving energy storage users. First, the charging and discharging behavior of energy storage devices causes the user load curve to change nonlinearly, making it difficult for traditional models to accurately predict. Second, user behavior is not only affected by fluctuations in electricity prices, but is also related to external meteorological factors such as weather, temperature, and humidity, which traditional methods often ignore. Therefore, the current recognition accuracy of user electricity consumption patterns is low. Summary of the invention

[0004] The purpose of the present invention is to provide a method, device and medium for identifying a user's electricity consumption pattern taking into account the association of multiple factors, so as to improve the recognition accuracy.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A method for identifying a user's electricity consumption pattern taking into account the association of multiple factors comprises the following steps:

[0007] S1, data acquisition: acquiring the historical load data of the electricity user within a preset period and the historical meteorological data of the corresponding period of the area where the user is located, and preprocessing the historical load data and historical meteorological data;

[0008] S2, data set construction: Considering the differences in electricity consumption of users in different seasons, the historical load data and historical meteorological data of each day are processed at different time resolutions, and the average value, peak and valley value and variance of the historical load data and historical meteorological data at different resolutions are calculated, and the load data set and meteorological data set at different resolutions are constructed respectively;

[0009] S3, coarse-grained identification: construct the load curve of each day based on the load data set, use the self-organizing map neural network to roughly cluster the load curve, roughly extract the user's power consumption pattern, and combine the clustering results to identify whether the user has an energy storage device;

[0010] S4, key factor selection: for load data sets and meteorological data sets at different resolutions, the correlation between meteorological factors and load is calculated through Pearson and Spearman correlation analysis, and key factors are selected;

[0011] S5, fine-grained recognition: Based on the selected key factors, a combination of temporal convolutional network and pyramid attention mechanism is used to perform fine-grained analysis of the user's power usage pattern, and accurately identify the power usage pattern at different time resolutions.

[0012] The preprocessing includes missing value filling and outlier detection and processing.

[0013] The self-organizing map neural network performs the following steps:

[0014] Data initialization process: Before training begins, each neuron in the self-organizing map neural network is assigned a random weight;

[0015] Competitive learning process: Each time, an input sample vector is randomly selected from the load curve set and input into the network. The self-organizing map neural network calculates the Euclidean distance between the input sample and the weight of each neuron, finds the neuron with the smallest distance, and records it as the winning neuron;

[0016] Collaborative learning process: With the winning neuron as the center, a winning neighborhood is defined, and the neurons around it are selected. The neurons in the winning neighborhood update their weights at the same time, and the neurons outside the winning neighborhood do not participate in the weight adjustment;

[0017] Weight adjustment process: update the weights of the winning neuron and the neurons in its winning neighborhood;

[0018] The above data initialization, competitive learning, collaborative learning and weight adjustment processes are repeated continuously, each time using a new sample vector to input the self-organizing map neural network for training until the network reaches a convergence state.

[0019] The formula for weight adjustment is:

[0020] W j k+1 =W j k +R k ·G ij ·(X i -W j )

[0021] Among them, R k is the learning rate of the kth iteration, X i is the i-th sample randomly selected, W j is the jth neuron weight that needs to be updated in the winning neighborhood, G ij is the update constraint of the jth neuron in the winning neighborhood corresponding to the i-th sample.

[0022] In step S4, based on the load data set and meteorological data set obtained in step S2, the user's daily electricity consumption feature matrix at different resolutions and the corresponding meteorological data matrix are extracted, and the correlation between different meteorological factors and load power is calculated based on the Person correlation coefficient. According to the calculation result of the correlation, the similarity correlation degree is defined as:

[0023]

[0024] Where r is the correlation calculated based on the Person correlation coefficient.

[0025] In step S4, based on the load data set and meteorological data set obtained in step S2, the user's daily electricity consumption characteristic matrix at different resolutions and the corresponding meteorological data matrix are extracted, and the monotonic relationship between meteorological factors and load power is calculated based on the Spearman correlation coefficient, and the monotonic correlation degree is defined as:

[0026]

[0027] Among them, r s is the monotonic relationship value calculated based on the Spearman correlation coefficient.

[0028] In step S4, based on the calculation results of the similarity correlation degree and the monotonicity correlation degree, the features of the meteorological data that are moderately and strongly correlated with the load are screened as key factors.

[0029] The step S5 specifically performs the following steps:

[0030] S51, the meteorological data and load data corresponding to the key factors are used as input sequences and input into the time convolution network. The long-range dependencies in the time series are captured through causal convolution and dilated convolution, and the long-term time series features of the input sequence are extracted.

[0031] S52, the output of the temporal convolutional network is input into the pyramid attention mechanism for processing, and the global context information is captured through feature maps of different scales, wherein the output of the temporal convolutional neural network is divided into multiple subsequences using a sliding window or hierarchical sampling, each subsequence represents a different time scale, and at each time scale, a self-attention mechanism is applied to capture the dependencies within the sequence, and the outputs of the self-attention at different time scales are fused through the pyramid structure;

[0032] S53, the outputs of the temporal convolutional network and the pyramid attention mechanism are fused by splicing to obtain fused features;

[0033] S54, inputting the fused features into a classifier to predict the electricity consumption pattern.

[0034] A device for identifying a user's electricity usage pattern taking into account multiple factors, comprising a memory, a processor, and a program stored in the memory, wherein the processor implements the method described above when executing the program.

[0035] A storage medium stores a program, which implements the method described above when executed.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] (1) Aiming at the processing problem of load data and meteorological data, the present invention provides a multi-resolution data processing method based on the peak, flat and valley time periods of industrial and commercial electricity consumption, that is, the 96-point load data of each day and the 24-point meteorological data of each day are processed according to the peak time periods, peak time periods, flat time periods and valley time periods in summer (July-September) and winter (January and December) and the peak time periods and flat time periods in other seasons, forming multiple resolution level features of daily load data. This method is based on the seasonal and time characteristics of electricity consumption, which can increase the richness of the data set and improve the efficiency of machine learning.

[0038] (2) For load data sets and meteorological data sets at different resolutions, the present invention provides a more comprehensive correlation analysis model from linear and nonlinear perspectives. The model first uses the Pearson correlation analysis method to test the linear correlation between meteorological factors and load. If this correlation is not significant, the Spearman correlation analysis method is further used to measure the monotonic relationship between meteorological factors and load from the perspective of rank or grade. It should be pointed out that the correlation analysis used in the present invention can comprehensively analyze the impact of various meteorological factors on load, which can not only provide a basis for the extraction of key meteorological factors, but also help to provide the execution time efficiency of algorithms such as machine learning.

[0039] (3) In response to the problem of load data analysis, the present invention provides a more comprehensive method for identifying electricity consumption behavior from both coarse-grained and fine-grained perspectives. The method first performs coarse clustering of load data through a self-organizing mapping neural network (SOM) based on the load curve of each day, and combines the fluctuation and change rules of the load curve to identify whether to install an energy storage device. Next, the impact of non-electric factors such as meteorology on the load is analyzed, and a time series convolutional network and a pyramid attention mechanism are used to capture the complexity of the load data in terms of time series, thereby achieving a fine-grained analysis of the user's electricity consumption pattern.

[0040] (4) Through accurate user pattern recognition, the present invention can not only provide more accurate load forecasts for power system operators, but also provide users with charging and discharging strategies for optimizing energy storage equipment, help to reasonably allocate energy storage resources, improve the operating efficiency of the energy storage system, and thus improve economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the method of the present invention;

[0042] Figure 2 It is a schematic diagram of the SOM network structure;

[0043] Figure 3 is a coarse-grained recognition result diagram in an embodiment;

[0044] Figure 4 This is the network diagram of the PAM-TCN model;

[0045] Figure 5 is a loss curve of a PAM-TCN model in one embodiment;

[0046] Figure 6 A diagram showing a fine-grained recognition result in one embodiment. DETAILED DESCRIPTION

[0047] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0048] Example 1

[0049] The present invention provides a method for identifying user power consumption patterns taking into account multi-factor associations, and constructs a multi-dimensional and multi-resolution analysis framework by acquiring historical power consumption and meteorological data of users. The specific steps include: first, collecting historical power consumption data of users with energy storage devices at 96 points per day for a period of one year, and obtaining corresponding meteorological data (such as temperature, humidity, wind speed, etc.). After preprocessing these data, according to the peak, peak, flat and valley periods of industrial and commercial electricity consumption, the average value, peak and valley value and variance under different time resolutions are calculated to construct a matrix set of power consumption and meteorological data. After the data processing is completed, the coarse-grained analysis based on the load curve is used to cluster the overall power consumption trend of the user, identify different power consumption patterns, and preliminarily analyze the power consumption behavior characteristics of the user. In addition, the correlation between meteorological data and power consumption data is analyzed to explore the influence of external factors on the power consumption behavior of users. In order to further accurately characterize the power consumption pattern of users with energy storage, this method proposes a multi-resolution recognition technology based on fine-grainedness, combined with coarse-grained results, and analyzes the charging and discharging behavior of energy storage equipment to describe the user's power consumption pattern in detail, accurately judge the power consumption pattern of new data, and provide a basis for power dispatching.

[0050] Specifically, Figure 1 As shown, the following steps are included:

[0051] S1, data acquisition: obtain the historical load data of the electricity user with a cycle of 1 year and 96 o'clock every day, and the historical meteorological data of the corresponding cycle and 24 o'clock every day in the area where the user is located, and perform pre-processing such as outlier detection, correction, missing value filling and normalization on the historical load data and historical meteorological data.

[0052] S2, data set construction: Taking into account the differences in power consumption characteristics of loads in different time periods in different seasons, the peak, flat and valley time periods of industrial and commercial electricity consumption in different seasons are used as intervals. The historical load data and historical meteorological data of each day are processed at different time resolutions, and the average, peak and valley values ​​and variance of the historical load data and historical meteorological data at different resolutions are calculated to obtain the corresponding data feature matrix, and the load data set and meteorological data set at different resolutions are constructed respectively.

[0053] S3, coarse-grained identification: construct a daily load curve based on the load data set, use the self-organizing map neural network (SOM) to coarsely cluster the load curve, roughly extract the user's power consumption pattern, and combine the clustering results to identify whether the user has an energy storage device.

[0054] The structure of the self-organizing map neural network is as follows Figure 2 As shown, it performs the following steps:

[0055] Data initialization process: Before training begins, each neuron in the self-organizing map neural network is assigned a random weight;

[0056] Competitive learning process: Each time, an input sample vector is randomly selected from the load curve set and input into the network. The self-organizing map neural network calculates the Euclidean distance between the input sample and the weight of each neuron, finds the neuron with the smallest distance, and records it as the winning neuron (Best Matching Unit, BMU);

[0057] Collaborative learning process: With the winning neuron as the center, a winning neighborhood is defined, and the neurons around it are selected. The neurons in the winning neighborhood update their weights at the same time, and the neurons outside the winning neighborhood do not participate in the weight adjustment;

[0058] Weight adjustment process: update the weights of the winning neuron and the neurons in its winning neighborhood;

[0059] The above data initialization, competitive learning, collaborative learning and weight adjustment processes are repeated continuously, each time using a new sample vector to input the self-organizing map neural network for training until the network reaches a convergence state.

[0060] In this embodiment, the weight adjustment formula is:

[0061] W j k+1 =W j k +R k ·G ij ·(X i -W j )

[0062] Among them, R k is the learning rate of the kth iteration, X i is the i-th sample randomly selected, W j is the jth neuron weight that needs to be updated in the winning neighborhood, G ij is the update constraint of the jth neuron in the winning neighborhood corresponding to the i-th sample.

[0063] In one embodiment, the coarse-grained recognition structure is as follows Figure 3 shown.

[0064] S4, key factor selection: For load data sets and meteorological data sets at different resolutions, the correlation between meteorological factors and load is calculated through Pearson and Spearman correlation analysis, and key factors are selected.

[0065] S41, Person correlation calculation

[0066] Based on the load data set and meteorological data set obtained in step S2, the user's daily electricity consumption characteristic matrix and the corresponding meteorological data matrix at different resolutions are extracted, and the correlation between different meteorological factors and load power is calculated based on the Person correlation coefficient. The calculation formula of the Person correlation coefficient is:

[0067]

[0068] Where r is the Pearson correlation coefficient, X and Y represent the variables in the meteorological data matrix and the load data matrix, respectively. and Represent the average values ​​of the variables in the meteorological data matrix and the load data matrix respectively.

[0069] According to the calculation results of the correlation, the similarity correlation degree is defined as:

[0070]

[0071] Where r is the correlation calculated based on the Person correlation coefficient.

[0072] S42, Spearman correlation calculation

[0073] Based on the load data set and meteorological data set obtained in step S2, the user's daily electricity consumption characteristic matrix at different resolutions and the corresponding meteorological data matrix are extracted, and the monotonic relationship between meteorological factors and load power is calculated based on the Spearman correlation coefficient. The calculation formula of the Spearman correlation coefficient is:

[0074]

[0075] Among them, r s is the Spearman correlation coefficient, R X and R Y are the ranks of the variables in the meteorological data matrix and the load data matrix, respectively.

[0076] The monotonicity correlation is defined as:

[0077]

[0078] Among them, r s is the monotonic relationship value calculated based on the Spearman correlation coefficient.

[0079] S43, feature selection

[0080] According to the calculation results of the similarity correlation degree and the monotonicity correlation degree, the features of the meteorological data showing medium and strong correlation with the load are screened as key factors and used as the input of step S5.

[0081] S5, fine-grained recognition: Based on the selected key factors, a combination of temporal convolutional network (TCN) and pyramid attention mechanism (Pyraformer) is used to perform fine-grained analysis of the user's power usage pattern and accurately identify the power usage pattern at different time resolutions.

[0082] In this embodiment, the network combining the temporal convolutional network (TCN) and the pyramid attention mechanism (Pyraformer) is denoted as the PAM-TCN model, and its structure is as follows: Figure 4 In one embodiment, the loss function is as follows: Figure 5 As shown, it can be seen that its convergence performance is good. Specifically, it performs the following steps:

[0083] S51, the meteorological data and load data corresponding to the key factors are taken as input sequences and input into the time convolution network TCN. The long-range dependencies in the time series are captured through causal convolution and dilated convolution, and the long-term time series features of the input sequence are extracted.

[0084] Assume that the input sequence is Where B is the batch size, C is the number of channels (feature dimension), and L is the sequence length. The output of TCN can be expressed as:

[0085] Y TCN =TCN(X)

[0086] Among them, each layer of TCN uses the convolution operation formula:

[0087]

[0088] Among them, Y l is the output of the lth layer, W l (k) is the convolution kernel of the lth layer, b l is the bias term, K is the number of convolution kernels, and * represents the convolution operation.

[0089] The output of TCN is calculated by the following formula:

[0090] Y TCN =TCN(X)=LayerNorm(Dropout(Y l ))

[0091] Among them, Dropout is used to prevent overfitting.

[0092] S52, the output of the temporal convolutional network is input into the pyramid attention mechanism (Pyraformer) for processing, capturing global context information through feature maps of different scales.

[0093] Assume that the output of TCN is Y TCN , and pass it into the pyramid attention mechanism as input X for processing, that is, first input data Divide into multiple subsequences, each subsequence represents a different time scale. This embodiment uses a sliding window or hierarchical sampling for division, and the obtained subsequences can be expressed as:

[0094]

[0095] Where i represents the i-th time scale, is the set of time indices associated with this scale.

[0096] At each time scale, Pyraformer applies a self-attention mechanism to capture dependencies within the sequence. The calculation formula for self-attention is:

[0097]

[0098] in, is the query matrix, is the key matrix, is the value matrix, d k is the dimension of the key. Through the application of the self-attention mechanism, each element in the sequence is assigned a different weight A, which means that important time points are emphasized.

[0099] Pyraformer uses a pyramid structure to fuse the attention outputs at different time scales. That is, for the attention output Y at N time scales, (i) , the output of each layer of the pyramid structure can be expressed as:

[0100] Y pyramid (j) = Aggregation(Y pyramid (j-1), Y (j) )

[0101] Among them, j represents the number of pyramid layers, and Aggregation is a function used to merge features from different time scales (such as weighted summation or concatenation). The final output pyramid feature is expressed as:

[0102] Y final =∑ j＝1 NW j Y pyramid (j)

[0103] Among them, W j are learnable weights.

[0104] In order to enable the model to understand the relative position relationship between elements in the sequence, relative position encoding is introduced. For each element x in the input sequencet , the relative position encoding can be expressed as:

[0105] PE(t)=Encode(t)

[0106] Among them, Encode is a function that converts the position t into an embedding vector. The relative position encoding is added to the input features so that the model can capture the sequential information in the time series data:

[0107] Z t =x t +PE(t)

[0108] After being processed by the multi-scale attention layer and pyramid structure, Pyraformer finally outputs a time series representation that combines multi-scale features, Y Pyraformer =Pyraformer(X).

[0109] S53, the output of the temporal convolutional network and the pyramid attention mechanism are fused by splicing to obtain the fused features:

[0110] Y final =concat(Y TCN ,Y Pyraformer )

[0111] S54, inputting the fused features into a classifier to predict the electricity consumption pattern.

[0112] This embodiment uses the Softmax function for processing and outputs the probability of each load category:

[0113] P(y|X)=Siftmax(Linear(Y final ))

[0114] Among them, P(y|X) represents the category probability distribution of a given input X.

[0115] Figure 6 The fine-grained recognition results are shown, showing that by combining TCN and the pyramid attention mechanism, this step successfully achieves fine-grained analysis of load data. This model can not only capture short-term dependencies in power consumption patterns, but also effectively handle long-range dependencies, thereby improving the recognition accuracy of power consumption patterns.

[0116] Example 2

[0117] The above is an introduction to a method embodiment. The following is a further explanation of the solution of the present invention through an apparatus embodiment.

[0118] In a preferred embodiment, a user power usage pattern recognition device taking into account multi-factor association includes:

[0119] Data acquisition module: obtains the historical load data of the electricity user within a preset period and the historical meteorological data of the corresponding period of the area where the user is located, and pre-processes the historical load data and historical meteorological data;

[0120] Dataset construction module: Considering the differences in electricity consumption of users in different seasons, the historical load data and historical meteorological data of each day are processed at different time resolutions, and the average value, peak and valley value and variance of the historical load data and historical meteorological data at different resolutions are calculated to construct load data sets and meteorological data sets at different resolutions respectively;

[0121] Coarse-grained identification module: constructs the load curve of each day based on the load data set, uses the self-organizing map neural network to roughly cluster the load curve, roughly extracts the user's power consumption pattern, and combines the clustering results to identify whether the user has an energy storage device;

[0122] Key factor selection module: for load data sets and meteorological data sets at different resolutions, the correlation between meteorological factors and load is calculated through Pearson and Spearman correlation analysis, and key factors are selected;

[0123] Fine-grained recognition module: Based on the selected key factors, a combination of temporal convolutional network and pyramid attention mechanism is used to perform fine-grained analysis of the user's power consumption pattern, and accurately identify the power consumption pattern at different time resolutions.

[0124] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0125] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0126] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for identifying user power consumption patterns taking into account multi-factor associations, characterized in that: The following steps are involved: S1, data acquisition: acquiring the historical load data of the electricity user within a preset period and the historical meteorological data of the corresponding period of the area where the user is located, and preprocessing the historical load data and historical meteorological data; S2, data set construction: Considering the differences in electricity consumption of users in different seasons, the historical load data and historical meteorological data of each day are processed at different time resolutions, and the average value, peak and valley value and variance of the historical load data and historical meteorological data at different resolutions are calculated, and the load data set and meteorological data set at different resolutions are constructed respectively; S3, coarse-grained identification: construct the load curve of each day based on the load data set, use the self-organizing map neural network to roughly cluster the load curve, roughly extract the user's power consumption pattern, and combine the clustering results to identify whether the user has an energy storage device; S4, key factor selection: for load data sets and meteorological data sets at different resolutions, the correlation between meteorological factors and load is calculated through Pearson and Spearman correlation analysis, and key factors are selected; S5, fine-grained recognition: Based on the selected key factors, a combination of temporal convolutional network and pyramid attention mechanism is used to perform fine-grained analysis of the user's power usage pattern, and accurately identify the power usage pattern at different time resolutions.

2. A method for identifying a user's electricity consumption pattern taking into account multiple factors according to claim 1, characterized in that: The preprocessing includes missing value filling and outlier detection and processing.

3. The method for identifying a user's electricity consumption pattern taking into account multiple factors according to claim 1, characterized in that: The self-organizing map neural network performs the following steps: Data initialization process: Before training begins, each neuron in the self-organizing map neural network is assigned a random weight; Competitive learning process: Each time, an input sample vector is randomly selected from the load curve set and input into the network. The self-organizing map neural network calculates the Euclidean distance between the input sample and the weight of each neuron, finds the neuron with the smallest distance, and records it as the winning neuron; Collaborative learning process: With the winning neuron as the center, a winning neighborhood is defined, and the neurons around it are selected. The neurons in the winning neighborhood update their weights at the same time, and the neurons outside the winning neighborhood do not participate in the weight adjustment; Weight adjustment process: update the weights of the winning neuron and the neurons in its winning neighborhood; The above data initialization, competitive learning, collaborative learning and weight adjustment processes are repeated continuously, each time using a new sample vector to input the self-organizing map neural network for training until the network reaches a convergence state.

4. A method for identifying a user's electricity consumption pattern taking into account multiple factors according to claim 3, characterized in that: The formula for weight adjustment is: Among them, R k is the learning rate of the kth iteration, X i is the i-th sample randomly selected, W j is the jth neuron weight that needs to be updated in the winning neighborhood, G ij is the update constraint of the jth neuron in the winning neighborhood corresponding to the i-th sample.

5. The method for identifying a user's electricity consumption pattern taking into account multiple factors according to claim 1, characterized in that: In step S4, based on the load data set and meteorological data set obtained in step S2, the user's daily electricity consumption feature matrix at different resolutions and the corresponding meteorological data matrix are extracted, and the correlation between different meteorological factors and load power is calculated based on the Person correlation coefficient. According to the calculation result of the correlation, the similarity correlation degree is defined as: Where r is the correlation calculated based on the Person correlation coefficient.

6. A method for identifying a user's electricity consumption pattern taking into account multiple factors according to claim 5, characterized in that: In step S4, based on the load data set and meteorological data set obtained in step S2, the user's daily electricity consumption characteristic matrix at different resolutions and the corresponding meteorological data matrix are extracted, and the monotonic relationship between meteorological factors and load power is calculated based on the Spearman correlation coefficient, and the monotonic correlation degree is defined as: Among them, r s is the monotonic relationship value calculated based on the Spearman correlation coefficient.

7. A method for identifying a user's electricity consumption pattern taking into account multiple factors according to claim 6, characterized in that: In step S4, based on the calculation results of the similarity correlation degree and the monotonicity correlation degree, the features of the meteorological data that are moderately and strongly correlated with the load are screened as key factors.

8. The method for identifying a user's electricity consumption pattern taking into account multiple factors according to claim 1, characterized in that: The step S5 specifically performs the following steps: S51, the meteorological data and load data corresponding to the key factors are used as input sequences and input into the time convolution network. The long-range dependencies in the time series are captured through causal convolution and dilated convolution, and the long-term time series features of the input sequence are extracted. S52, the output of the temporal convolutional network is input into the pyramid attention mechanism for processing, and the global context information is captured through feature maps of different scales, wherein the output of the temporal convolutional neural network is divided into multiple subsequences using a sliding window or hierarchical sampling, each subsequence represents a different time scale, and at each time scale, a self-attention mechanism is applied to capture the dependencies within the sequence, and the outputs of the self-attention at different time scales are fused through the pyramid structure; S53, the outputs of the temporal convolutional network and the pyramid attention mechanism are fused by splicing to obtain fused features; S54, inputting the fused features into a classifier to predict the electricity consumption pattern.

9. A user power usage pattern recognition device taking into account multi-factor association, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 8 is implemented.

Images