Distributed user baseline power prediction method

Abstract

The invention discloses a distributed user baseline power prediction method in the field of electric energy metering data processing, and aims to solve the problems of low accuracy and difficulty in acquiring time positioning information of data in distributed user baseline power prediction. The method comprises the following steps of: screening an original distributed user data sequence by using a Szely-Rizzo distance correlation coefficient, and removing an attribute sequence which is not highly correlated with the distributed user data sequence from the original distributed user data sequence, so that the accuracy of predicting a future baseline power sequence by using the screened distributed user data sequence is improved; according to the method, the extracted time information is utilized in advance to carry out fusion time positioning coding operation on the screened distributed user data sequence, so that the problem that a Transform model cannot capture the data time information contained in the baseline power sequence is solved, the Transform model has time correlation when predicting the baseline power sequence, and more accurate prediction is realized.

Description

Technical Field

[0001] The invention relates to a method for predicting decentralized user baseline power, and belongs to the field of electric energy metering data processing. Background Art

[0002] With the rapid advancement of the construction of new power systems, massive distributed resources are involved in spot transactions, which also puts higher requirements on the prediction of baseline power of distributed users.

[0003] Distributed user baseline power refers to the power consumed by multiple independent users or devices under normal operating conditions without external influence or interference. This baseline power is usually used to evaluate and analyze the user's regular power consumption pattern, and as a reference for predicting future power demand and formulating power dispatch plans. The prediction of distributed user baseline power is to use massive user power consumption data to predict future power demand through intelligent algorithm models, thereby optimizing the management and dispatch of power systems. In recent years, the use of deep learning technology for baseline power prediction has made the understanding of distributed user behavior more in-depth and accurate, realized real-time monitoring and prediction of user power consumption behavior, efficiently integrated massive distributed resources to participate in spot transactions, promoted the efficient and safe consumption of renewable energy, greatly improved the flexibility of the power system, and optimized its strategy for participating in spot transactions, which is of great significance to the construction of a new power system.

[0004] However, when traditional methods of predicting distributed user baseline power use multiple attributes to predict distributed user baseline power, some attributes that are not strongly correlated with the distributed user baseline power data will interfere with the prediction, affecting the accuracy of the distributed user baseline power prediction; in addition, traditional methods of predicting distributed user baseline power are difficult to capture the time correlation of distributed user baseline power. Therefore, the accuracy of these methods in predicting distributed user baseline power still needs to be improved.

[0005] Due to the diversity of user types and the complexity of baseline power data, traditional machine learning algorithms have problems with low accuracy and difficulty in obtaining the time location information of data for decentralized user baseline power prediction. Summary of the invention

[0006] The purpose of the present application is to overcome the deficiencies in the prior art and to provide a method for predicting the baseline power of distributed users with high prediction accuracy and time correlation.

[0007] In order to achieve the above objectives, this application is implemented by adopting the following technical solutions:

[0008] In a first aspect, the present application provides a method for predicting a decentralized user baseline power, comprising:

[0009] Acquire an original distributed user data sequence, wherein the original distributed user data sequence includes a baseline power sequence and a plurality of attribute sequences;

[0010] Obtaining a Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence, and obtaining a filtered distributed user data sequence according to a comparison result between a preset Székely-Rizzo distance correlation coefficient threshold and each Székely-Rizzo distance correlation coefficient;

[0011] Extracting time information from the filtered distributed user data sequence, and performing a fusion time positioning coding operation on the filtered distributed user data sequence according to the time information;

[0012] The distributed user data sequence after fusion time positioning coding is input into the pre-built and trained Transformer model for prediction to obtain the prediction result of the baseline power sequence.

[0013] In some embodiments of the first aspect, the obtaining of the Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence, and obtaining the filtered distributed user data sequence according to a comparison result between a preset Székely-Rizzo distance correlation coefficient threshold and each Székely-Rizzo distance correlation coefficient, include:

[0014] A Székely-Rizzo distance correlation coefficient threshold is set. In response to the Székely-Rizzo distance correlation coefficient between any attribute sequence and the baseline power sequence being greater than the Székely-Rizzo distance correlation coefficient threshold, the attribute sequence is placed into the filtered distributed user data sequence; the baseline power sequence is placed into the filtered distributed user data sequence; the filtered distributed user data sequence is obtained as follows:

[0015] ,

[0016] In the formula, is the filtered distributed user data sequence, is the original distributed user data sequence, In order to use the Székely-Rizzo distance correlation coefficient threshold to filter all attribute sequences in the original distributed user data sequence, is the Székely-Rizzo distance correlation coefficient threshold, For all attribute sequences, is the attribute sequence after filtering, is the baseline power series.

[0017] In some embodiments of the first aspect, a central distance matrix of each attribute sequence and a central distance matrix of the baseline power sequence are obtained, a Székely-Rizzo distance covariance between each attribute sequence and the baseline power sequence is obtained according to the central distance matrix of each attribute sequence and the baseline power sequence, and a Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence is obtained according to the Székely-Rizzo distance covariance between each attribute sequence and the baseline power sequence;

[0018] The Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence is obtained as follows:

[0019] ,

[0020] In the formula, For the Székely-Rizzo distance correlation coefficient calculation operation, For the The Székely-Rizzo distance correlation coefficient between the attribute series and the baseline power series, is the number of the attribute sequence, is the distance covariance calculation operation, For the The distance covariance between the attribute sequence and the baseline power sequence, For the The distance covariance between an attribute sequence and itself, is the distance covariance between the baseline power sequence and itself, For the A sequence of attributes, is the number of elements in the attribute sequence, For the The center distance matrix of the attribute sequence Line The elements of the column, is the center distance matrix of the baseline power sequence Line The elements of the column, is the counting number of the element, The counting number of the element.

[0021] In some embodiments of the first aspect, a distance matrix of each attribute sequence is obtained, and a corresponding central distance matrix is ​​obtained according to the distance matrix of each attribute sequence; a distance matrix of the baseline power sequence is obtained, and a corresponding central distance matrix is ​​obtained according to the distance matrix of the baseline power sequence;

[0022] The central distance matrix of each attribute sequence is obtained as follows:

[0023] ,

[0024] In the formula, For the The first elements and The distance of the first element, The distance matrix of the attribute sequence Line The elements of the column, For the The distance matrix of the attribute sequence The average value of the row elements, For the The distance matrix of the attribute sequence The average value of the column elements, For the The mean of all elements in the distance matrix of the attribute sequence;

[0025] The central distance matrix of the baseline power sequence is obtained as follows:

[0026] ,

[0027] In the formula, is the first elements and The distance of the element in the distance matrix of the baseline power sequence Line The elements of the column, is the distance matrix of the baseline power sequence The mean of the row elements, is the distance matrix of the baseline power sequence The mean of the column elements, is the mean of all elements in the distance matrix of the baseline power sequence.

[0028] In some embodiments of the first aspect, the distance matrix of each attribute sequence is obtained by an acquisition formula, and the acquisition formula of the distance matrix of each attribute sequence is as follows:

[0029] ,

[0030] In the formula, is an element in the attribute sequence, and Respectively The first elements and elements;

[0031] The distance matrix of the baseline power sequence is obtained by an acquisition formula, and the acquisition formula of the distance matrix of the baseline power sequence is as follows:

[0032] ,

[0033] In the formula, is the element in the baseline power sequence, and are the first elements and elements.

[0034] In some embodiments of the first aspect, extracting time information from the filtered distributed user data sequence, and performing a fusion time positioning coding operation on the filtered distributed user data sequence according to the time information, includes:

[0035] Performing local position coding on the filtered distributed user data sequence to obtain local position coding information, finding the position information of the time series data from the local position coding information, and extracting accurate time information from the filtered distributed user data sequence according to the position information of the time series data;

[0036] Using the accurate time information, globally time-code the filtered distributed user data sequence to obtain global time code information;

[0037] The global time coding information and the local position coding information are fused to obtain fused time positioning coding information. The fused time positioning coding information is obtained as follows:

[0038] ,

[0039] In the formula, The local position encoding information of the filtered distributed user data sequence, is the global time coding information of the filtered distributed user data sequence, The fused time positioning coding information of the filtered distributed user data sequence is It is the filtered distributed user data sequence.

[0040] In some embodiments of the first aspect, performing local position coding on the filtered distributed user data sequence to obtain local position coding information includes:

[0041] A sine local position coding operation is performed on the even dimensions of the filtered distributed user data sequence, and a cosine local position coding operation is performed on the odd dimensions. The sine code and the cosine code are combined into the local position coding information. The sine code and the cosine code are obtained as follows:

[0042] ,

[0043] In the formula, is the local position encoding operation, is the position in the filtered distributed user data sequence, is the dimension counting number of the filtered distributed user data sequence, , is the input dimension of the filtered distributed user data sequence, is a sinusoidal local position encoding operation for even dimensions, is the cosine local position encoding operation for odd dimensions, is sinusoidal coding, Cosine encoding.

[0044] In a second aspect, the present application also provides a computer device, comprising a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for predicting distributed user baseline power as described in any embodiment of the first aspect are performed.

[0045] In a third aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for predicting distributed user baseline power as described in any embodiment of the first aspect.

[0046] In a fourth aspect, the present application also provides a computer program product, comprising a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the method for predicting distributed user baseline power described in any embodiment of the first aspect are implemented.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The method for predicting the baseline power of distributed users provided in the present application uses the Székely-Rizzo distance correlation coefficient to screen the original distributed user data sequence, and eliminates the attribute sequence with low correlation with the distributed user data sequence, thereby improving the accuracy of predicting the future baseline power sequence using the screened distributed user data sequence; the extracted time information is pre-used to perform a fusion time positioning coding operation on the screened distributed user data sequence, thereby overcoming the problem that the Transformer model cannot capture the time information contained in the baseline power sequence, so that the Transformer model has time correlation when predicting the baseline power sequence, and achieves more accurate prediction; by improving the accuracy of the prediction of the baseline power sequence of distributed users, technical support is provided for distributed users to participate in spot transactions, so that users can choose a suitable energy market service model according to their own energy production and consumption needs and market environment, and promote the power system to achieve the optimal configuration and utilization of distributed user resources according to the changes in time and space conditions and elements. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 is an algorithm framework diagram of the Transformer model in the decentralized user baseline power prediction method provided in this embodiment;

[0051] Figure 2 is a deployment and implementation flow chart of the method for predicting decentralized user baseline power provided by this embodiment;

[0052] Figure 3 is a flowchart of the steps of the method for predicting the distributed user baseline power provided by this embodiment;

[0053] Figure 4 It is a schematic block diagram of the principles of the computer device provided in this embodiment. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0055] The term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0056] Embodiment 1:

[0057] Figure 3 This is a flow chart of a method for predicting a distributed user baseline power in the first embodiment of the present invention. This flow chart only shows the logical sequence of the method described in this embodiment. In other possible embodiments of the present invention, different methods may be used without conflict. Figure 1 The steps shown or described are accomplished in the order shown.

[0058] See also Figure 3 The method for predicting the distributed user baseline power of this implementation specifically includes the following steps:

[0059] Acquire an original distributed user data sequence, the original distributed user data sequence including a baseline power sequence and a plurality of attribute sequences;

[0060] Obtain the Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence (reference: Gábor J. Székely, Maria L. Rizzo, Nail K. Bakirov. Measuring and testingdependence by correlation of distances[J], Annals of Statistics, 2008, 35(6):2769-2794.). The Székely-Rizzo distance correlation coefficient can capture the nonlinear relationship between the attribute sequence and the baseline power sequence, and shows good performance in a variety of statistical tests; obtain the filtered distributed user data sequence based on the comparison result between the preset correlation coefficient threshold and each Székely-Rizzo distance correlation coefficient, and eliminate the attribute sequence with poor correlation with the baseline power sequence to prevent the attribute sequence with poor correlation with the baseline power sequence from interfering with the prediction of the baseline power sequence;

[0061] Since the subsequent Transformer model has the problem that the local position coding cannot obtain the data time information, the time information is extracted from the filtered distributed user data sequence, and the filtered distributed user data sequence is fused with the time positioning coding operation according to the time information;

[0062] The distributed user data sequence after fusion of time positioning coding is input into the Transformer model for prediction. The distributed user data sequence after fusion of time positioning coding already has time information. On the basis of eliminating the attribute sequence with low correlation, the Transformer model performs prediction and obtains the prediction result of the baseline power sequence, which exceeds the existing prediction method in accuracy.

[0063] In addition, the main challenge of predicting the baseline power of distributed users lies in the uncertainty, clustering and predictability of users' electricity consumption behavior. The randomness, fuzziness and roughness of users' electricity consumption behavior require the model or method to be able to adapt to time-varying characteristics, accurately define user behavior categories, and overcome the limitations of measurement accuracy. Existing prediction methods may have limitations in these aspects and it is difficult to fully capture the complexity and variability of electricity consumption behavior.

[0064] In contrast, the method for predicting the baseline power of a decentralized user provided in this embodiment calculates the Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence, and performs threshold comparison, which can screen out the attribute sequence most relevant to the baseline power, help reduce the impact of irrelevant or redundant information, and make the prediction process more focused on features useful for the prediction task. Extracting and fusing time information is one of the key steps. Since the user's electricity consumption behavior is often closely related to time (for example, weekdays and weekends, seasonal changes, etc.), incorporating time information into the baseline power sequence can significantly improve the accuracy and adaptability of the prediction. The Transformer model is suitable for processing sequence data and can capture long-term dependencies in the data, which is crucial for predicting baseline power with time-varying characteristics. A trained Transformer model can learn the complex patterns and variability of user electricity consumption behavior. This embodiment also aims to build a prediction model that can adapt to time-varying characteristics and accurately define user behavior categories. By training on a large-scale and diverse data set, the model can learn the electricity consumption behavior patterns of different users and have a certain generalization ability, thereby overcoming the limitations of measurement accuracy.

[0065] Secondly, the electricity consumption behavior of users shows certain group characteristics, and a model or method is required to identify and analyze the electricity consumption characteristics of different user groups and effectively distinguish between different groups. In the method for predicting the baseline power of distributed users provided in this embodiment, time information is crucial for identifying the electricity consumption characteristics of user groups, because different user groups may show different electricity consumption behaviors in different time periods. By integrating time information, the model can better capture the time-varying laws of these group characteristics and distinguish different groups.

[0066] Furthermore, although predictability is an important advantage of user behavior prediction, its limitations cannot be ignored. The inherent laws of user behavior may be affected by individual differences and changes in the external environment, making it difficult for the prediction model to achieve completely accurate predictions in practical applications. The method for predicting the decentralized user baseline power provided in this embodiment overcomes the limitations of the predictability of user behavior and achieves relatively accurate predictions in practical applications by capturing individual characteristics, integrating time information, utilizing the flexibility of the Transformer model, performing model tuning and verification, and continuous learning and updating. The combination of these methods helps to improve the adaptability and generalization ability of the model, and better cope with the impact of individual differences and changes in the external environment on prediction accuracy.

[0067] The Székely-Rizzo distance correlation coefficient is a new dependency measure for testing the joint independence of random vectors. Its definition is based on the Euclidean distance between sample elements, rather than on sample moments. Specifically, the Székely-Rizzo distance correlation coefficient satisfies 0 <= R <= 1, and R = 0 if and only if the random vectors are independent. In the case of normal distribution, the Székely-Rizzo distance correlation coefficient is directly related to the traditional correlation. The Székely-Rizzo distance correlation coefficient can be applied to random vectors of any dimension, not just two-dimensional data. In addition, the Székely-Rizzo distance correlation coefficient provides a new tool to test the independence of random vectors. It performs well in the case of non-normal distribution, non-regular dependency structure, and non-monotonic dependency structure, is easy to calculate, and has statistical consistency. These characteristics make the Székely-Rizzo distance correlation coefficient a valuable supplement to traditional methods.

[0068] Therefore, the distributed user baseline power prediction method provided in this embodiment integrates time information into the prediction of the baseline power sequence, which can significantly improve the accuracy and adaptability of the prediction. By integrating time information, the model can better capture the time-varying laws of power consumption behavior of different user groups, effectively distinguish different user groups, and make targeted predictions. By calculating the Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence, and performing threshold comparison, this method can screen out the attribute sequence that is most relevant to the baseline power, effectively reducing the impact of irrelevant or redundant information, and allowing the model to focus more on features that are useful for prediction tasks.

[0069] Embodiment 2:

[0070] This embodiment provides a method for predicting a distributed user baseline power. This embodiment is optimized on the basis of the first embodiment to improve the technical effect and refine the technical solution. For details of the contents not fully described in this embodiment, please refer to the first embodiment.

[0071] Embodiment 1 mentions using the comparison between the Székely-Rizzo distance correlation coefficient and the Székely-Rizzo distance correlation coefficient threshold to screen the attribute sequence. Optionally, in a further but non-limiting embodiment of the present application, the comparison strategy is:

[0072] The Székely-Rizzo distance correlation coefficient threshold is set. In response to the Székely-Rizzo distance correlation coefficient between any attribute sequence and the baseline power sequence being greater than the Székely-Rizzo distance correlation coefficient threshold, the attribute sequence is placed in the filtered distributed user data sequence, in order to filter and eliminate the attribute sequence that is not highly correlated with the baseline power sequence; the baseline power sequence is placed in the filtered distributed user data sequence; the filtered distributed user data sequence is obtained as follows

[0073] ,

[0074] In the formula, is the filtered distributed user data sequence, , is the original distributed user data sequence, , In order to use the Székely-Rizzo distance correlation coefficient threshold to filter the attribute sequence in the original distributed user data sequence, is the Székely-Rizzo distance correlation coefficient threshold, For all attribute sequences, is the attribute sequence after filtering, is the baseline power series.

[0075] As one embodiment, a method for obtaining a Székely-Rizzo distance correlation coefficient threshold is as follows:

[0076] Obtain the central distance matrix of each attribute sequence and the central distance matrix of the baseline power sequence, obtain the Székely-Rizzo distance covariance between each attribute sequence and the baseline power sequence according to the central distance matrix of each attribute sequence and the baseline power sequence, and obtain the Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence according to the Székely-Rizzo distance covariance between each attribute sequence and the baseline power sequence;

[0077] The Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence is obtained as follows:

[0078] ,

[0079] In the formula, For the Székely-Rizzo distance correlation coefficient calculation operation, For the The Székely-Rizzo distance correlation coefficient between the attribute series and the baseline power series, is the number of the attribute sequence, , is the distance covariance calculation operation, For the The distance covariance between the attribute sequence and the baseline power sequence, For the The distance covariance between an attribute sequence and itself, is the distance covariance between the baseline power sequence and itself, For the attribute sequences, that is, all attribute sequences One of them, is the number of elements in the attribute sequence, and is also used to represent the length of the attribute sequence. For the The center distance matrix of the attribute sequence Line The elements of the column, is the center distance matrix of the baseline power sequence Line The elements of the column, is the counting number of the element, The counting number of the element.

[0080] The Székely-Rizzo distance correlation coefficient between each attribute series and the baseline power series is calculated using the above formula.

[0081] As one embodiment, this embodiment also provides a method for obtaining a center distance matrix:

[0082] Obtain the distance matrix of each attribute sequence, and obtain the corresponding central distance matrix according to the distance matrix of each attribute sequence; obtain the distance matrix of the baseline power sequence, and obtain the corresponding central distance matrix according to the distance matrix of the baseline power sequence;

[0083] The central distance matrix of each attribute sequence is obtained as follows:

[0084] ,

[0085] In the formula, For the The first elements and The distance between elements. In this embodiment, the attribute sequence is converted into the corresponding distance matrix for easy understanding. As the The distance matrix of the attribute sequence Line The elements of the column, For the The distance matrix of the attribute sequence The average value of the row elements, For the The distance matrix of the attribute sequence The average value of the column elements, For the The mean of all elements in the distance matrix of the attribute sequence;

[0086] The central distance matrix of the baseline power sequence is obtained as follows:

[0087] ,

[0088] In the formula, is the first elements and The distance of each element. In this embodiment, for ease of understanding, the baseline power sequence is converted into a corresponding distance matrix. That is, the distance matrix of the baseline power sequence Line The elements of the column, is the distance matrix of the baseline power sequence The mean of the row elements, is the distance matrix of the baseline power sequence The mean of the column elements, is the mean of all elements in the distance matrix of the baseline power sequence.

[0089] As one embodiment, this embodiment also provides an acquisition formula for the distance matrix:

[0090] ,

[0091] In the formula, is an element in the attribute sequence, and Respectively The first elements and elements;

[0092] The distance matrix of the baseline power sequence is obtained by the acquisition formula. The acquisition formula of the distance matrix of the baseline power sequence is as follows:

[0093] ,

[0094] In the formula, is the element in the baseline power sequence, and are the first elements and elements.

[0095] In addition, the first embodiment mentions extracting time information from the filtered distributed user data sequence, and performing a fusion time positioning coding operation on the filtered distributed user data sequence according to the time information, as one embodiment, including:

[0096] Performing local position coding on the filtered distributed user data sequence, obtaining local position coding information, finding the position information of the time series data from the local position coding information, and extracting accurate time information from the filtered distributed user data sequence according to the position information of the time series data;

[0097] Using accurate time information, globally time-code the filtered distributed user data sequence to obtain global time-code information;

[0098] The global time coding information and the local position coding information are fused to obtain the fused time positioning coding information. The fusion time positioning coding information is obtained as follows:

[0099] ,

[0100] In the formula, The local position encoding information of the filtered distributed user data sequence, is the global time coding information of the filtered distributed user data sequence, The fused time positioning coding information is provided for the filtered distributed user data sequence.

[0101] Considering the local position encoding of the filtered distributed user data sequence, different input dimensions should use different encoding methods.

[0102] The even dimensions of the filtered distributed user data sequence are subjected to sine local position coding operations, and the odd dimensions are subjected to cosine local position coding operations. The sine coding and cosine coding are combined into local position coding information. The sine coding and cosine coding are obtained as follows:

[0103] ,

[0104] In the formula, is the local position encoding operation, is the position of the filtered distributed user data sequence, is the dimension counting number of the filtered distributed user data sequence, , is the input dimension of the filtered distributed user data sequence, is a sinusoidal local position encoding operation for even dimensions, is the cosine local position encoding operation for odd dimensions, is sinusoidal coding, Cosine encoding.

[0105] Then, the distributed user data sequence after fusion of time positioning coding is input into the pre-built and trained Transformer model for prediction to obtain the prediction result of the baseline power sequence.

[0106] refer to Figure 1 As one embodiment, the original distributed user data sequence As input, after calculating the Székely-Rizzo distance correlation coefficient (i.e., the SR distance correlation coefficient in the figure), the Székely-Rizzo distance correlation coefficient is used to filter the distributed user data sequence, linearize the filtered distributed user data sequence, and perform fusion time positioning coding operation on the linearized filtered distributed user data sequence. The distributed user data sequence after fusion time positioning coding is input into the Transformer model for prediction. The Transformer model includes an encoder and a decoder. The encoder includes multi-head attention mechanism, residual and layer normalization, and feedforward network operations. The distributed user data sequence input into the encoder is first processed by the multi-head attention mechanism.

[0107] The distributed user data sequence after fusion time location coding is input into the matrix The multi-head attention mechanism is input in the form of With three weight matrices respectively , , Multiply to get the query ,key Sum , then calculate the query and key The dot product of To zoom, is the dimension of the key, through The activation function is used to obtain the weight of the value, and the final weight of the value is equal to the value Multiply to get the result of attention calculation, the specific calculation formula is as follows:

[0108] ,

[0109] In the formula, Score for attention, is the transpose symbol.

[0110] The multi-head attention mechanism will query ,key Sum Projected multiple times to , and Dimensions, and are the dimensions of query and value respectively, assuming the number of attention heads is , and The calculation formula is as follows:

[0111] .

[0112] In the query ,key Sum At each projection level of , the attention function is executed in parallel to generate dimensional output values, concatenate them and project them again to get the final attention score.

[0113] The specific formula for connecting and re-projecting attention heads at different levels is as follows:

[0114] ,

[0115] in, is the multi-head attention calculation operation, , is the counting number of the attention head, , , They are respectively The learning parameters of the attention head, is the field of real numbers, For the connection operation, is a learnable parameter when connecting projections.

[0116] Behind the multi-head attention mechanism and the feedforward network, there is a residual and layer normalization operation. The residual mechanism is implemented by adding the input and output of the previous layer.

[0117] Each encoder and decoder contains a fully connected feed-forward network, which consists of two linear transformations with different parameters.

[0118] The decoder contains three sub-layers, the masked multi-head attention layer, the multi-head attention layer, and the feedforward network layer. Similar to the encoder, residual connections are used around each sub-layer, followed by layer normalization. The only difference in the decoder is the masked multi-head attention mechanism, which masks the true value of future data during training to ensure that the current time prediction can only depend on the data before the current time. The data output after the decoder is decoded at the decoder end is linearized to obtain the predicted baseline power sequence .

[0119] refer to Figure 2 The method for predicting the distributed user baseline power provided in this embodiment can be regarded as a prediction model. This embodiment also provides a corresponding training method.

[0120] First, data acquisition is performed to obtain the electricity consumption data of distributed users. The key attributes are extracted using a clustering method (K-means clustering can be used) to obtain the original distributed user data sequence. The original distributed user data sequence is divided into two parts, one for the training set and the other for the test set, which are used for model migration and training. The trained model is packaged for deployment.

[0121] The method for predicting the distributed user baseline power provided in this embodiment improves the technical effect and refines the technical solution compared to the first embodiment.

[0122] Embodiment three:

[0123] This embodiment provides a computer device, including a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for predicting the baseline power of distributed users provided in Embodiment 1 or 2 are performed.

[0124] The computer device may be a server or an electronic terminal. As one embodiment, refer to Figure 4, the computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data obtained and generated in the method for predicting the baseline power of distributed users. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the method for predicting the baseline power of distributed users provided in embodiment one or two is implemented.

[0125] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0126] The computer device provided in this embodiment has the same technical effects as those in Embodiment 1 or 2, which will not be described in detail here.

[0127] Embodiment 4:

[0128] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method for predicting the decentralized user baseline power provided in the first or second embodiment are implemented.

[0129] The computer-readable storage medium provided in this embodiment has the same technical effects as those in Embodiment 1 or 2, and will not be described in detail here.

[0130] Embodiment five:

[0131] This embodiment provides a computer program product, on which a computer program is stored, and when the program is executed by a processor, the steps of the method for predicting the distributed user baseline power provided in Embodiment 1 or Embodiment 2 are implemented. The computer program product provided in this embodiment can be transmitted, distributed and downloaded in the form of a signal via the Internet.

[0132] The computer program product provided in this embodiment has the same technical effects as those in Embodiment 1 or 2, and will not be described in detail here.

[0133] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0134] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0135] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0137] In the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0138] For those skilled in the art, the specific meanings of the above terms in the present invention can be understood through specific circumstances.

[0139] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting decentralized user baseline power, characterized in that: include, Acquire an original distributed user data sequence, wherein the original distributed user data sequence includes a baseline power sequence and a plurality of attribute sequences; Obtaining a Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence, and obtaining a filtered distributed user data sequence according to a comparison result between a preset Székely-Rizzo distance correlation coefficient threshold and each Székely-Rizzo distance correlation coefficient; Extracting time information from the filtered distributed user data sequence, and performing a fusion time positioning coding operation on the filtered distributed user data sequence according to the time information; The distributed user data sequence after fusion time positioning coding is input into the pre-built and trained Transformer model for prediction to obtain the prediction result of the baseline power sequence.

2. The method for predicting the distributed user baseline power according to claim 1, characterized in that: The step of obtaining the Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence, and obtaining the filtered distributed user data sequence according to the comparison result between the preset Székely-Rizzo distance correlation coefficient threshold and each Székely-Rizzo distance correlation coefficient, includes: A Székely-Rizzo distance correlation coefficient threshold is set. In response to the Székely-Rizzo distance correlation coefficient between any attribute sequence and the baseline power sequence being greater than the Székely-Rizzo distance correlation coefficient threshold, the attribute sequence is placed into the filtered distributed user data sequence; the baseline power sequence is placed into the filtered distributed user data sequence; the filtered distributed user data sequence is obtained as follows: ， In the formula, is the filtered distributed user data sequence, is the original distributed user data sequence, In order to use the Székely-Rizzo distance correlation coefficient threshold to filter all attribute sequences in the original distributed user data sequence, is the Székely-Rizzo distance correlation coefficient threshold, For all attribute sequences, is the attribute sequence after filtering, is the baseline power series.

3. The method for predicting the distributed user baseline power according to claim 2, characterized in that: Obtaining the central distance matrix of each attribute sequence and the central distance matrix of the baseline power sequence, obtaining the Székely-Rizzo distance covariance between each attribute sequence and the baseline power sequence according to the central distance matrix of each attribute sequence and the baseline power sequence, and obtaining the Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence according to the Székely-Rizzo distance covariance between each attribute sequence and the baseline power sequence; The Székely-Rizzo distance correlation coefficient between each attribute sequence and the baseline power sequence is obtained as follows: ， In the formula, For the Székely-Rizzo distance correlation coefficient calculation operation, For the The Székely-Rizzo distance correlation coefficient between the attribute series and the baseline power series, is the number of the attribute sequence, is the distance covariance calculation operation, For the The distance covariance between the attribute sequence and the baseline power sequence, For the The distance covariance between an attribute sequence and itself, is the distance covariance between the baseline power sequence and itself, For the A sequence of attributes, is the number of elements in the attribute sequence, For the The center distance matrix of the attribute sequence Line The elements of the column, is the center distance matrix of the baseline power sequence Line The elements of the column, is the counting number of the element, The counting number of the element.

4. The method for predicting the distributed user baseline power according to claim 3, characterized in that: Obtaining the distance matrix of each attribute sequence, and obtaining the corresponding central distance matrix according to the distance matrix of each attribute sequence; obtaining the distance matrix of the baseline power sequence, and obtaining the corresponding central distance matrix according to the distance matrix of the baseline power sequence; The central distance matrix of each attribute sequence is obtained as follows: ， In the formula, For the The first elements and The distance of the first element, The distance matrix of the attribute sequence Line The elements of the column, For the The distance matrix of the attribute sequence The average value of the row elements, For the The distance matrix of the attribute sequence The average value of the column elements, For the The mean of all elements in the distance matrix of the attribute sequence; The central distance matrix of the baseline power sequence is obtained as follows: ， In the formula, is the first elements and The distance of the element in the distance matrix of the baseline power sequence Line The elements of the column, is the distance matrix of the baseline power sequence The mean of the row elements, is the distance matrix of the baseline power sequence The mean of the column elements, is the mean of all elements in the distance matrix of the baseline power sequence.

5. The method for predicting the distributed user baseline power according to claim 4, characterized in that: The distance matrix of each attribute sequence is obtained by the acquisition formula. The acquisition formula of the distance matrix of each attribute sequence is as follows: ， In the formula, is an element in the attribute sequence, and Respectively The first elements and elements; The distance matrix of the baseline power sequence is obtained by an acquisition formula, and the acquisition formula of the distance matrix of the baseline power sequence is as follows: ， In the formula, is the element in the baseline power sequence, and are the first elements and elements.

6. The method for predicting the distributed user baseline power according to claim 1, characterized in that: Extracting time information from the filtered distributed user data sequence, and performing a fusion time positioning coding operation on the filtered distributed user data sequence according to the time information, including: Performing local position coding on the filtered distributed user data sequence to obtain local position coding information, finding the position information of the time series data from the local position coding information, and extracting accurate time information from the filtered distributed user data sequence according to the position information of the time series data; Using the accurate time information, globally time-code the filtered distributed user data sequence to obtain global time code information; The global time coding information and the local position coding information are fused to obtain fused time positioning coding information. The fused time positioning coding information is obtained as follows: ， In the formula, The local position encoding information of the filtered distributed user data sequence, is the global time coding information of the filtered distributed user data sequence, The fused time positioning coding information of the filtered distributed user data sequence is It is the filtered distributed user data sequence.

7. The method for predicting the distributed user baseline power according to claim 6, characterized in that: The step of performing local position coding on the filtered distributed user data sequence to obtain local position coding information includes: A sine local position coding operation is performed on the even dimensions of the filtered distributed user data sequence, and a cosine local position coding operation is performed on the odd dimensions. The sine code and the cosine code are combined into the local position coding information. The sine code and the cosine code are obtained as follows: ， In the formula, is the local position encoding operation, is the position in the filtered distributed user data sequence, is the dimension counting number of the filtered distributed user data sequence, , is the input dimension of the filtered distributed user data sequence, is a sinusoidal local position encoding operation for even dimensions, is the cosine local position encoding operation for odd dimensions, is sinusoidal coding, Cosine encoding.

8. A computer device, characterized in that: The method comprises a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for predicting the baseline power of distributed users as claimed in any one of claims 1 to 7 are executed.

9. 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 for predicting the decentralized user baseline power according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the method for predicting the baseline power of distributed users as claimed in any one of claims 1 to 7 are implemented.

Images