Dynamically updated multi-source heterogeneous power consumer portrait method and system

Abstract

The invention discloses a dynamically updated multi-source heterogeneous power consumer portrait method and system, and relates to the technical field of power consumer portrait generation, and the method comprises the steps: collecting multi-source data, carrying out the preprocessing of the collected data, carrying out the feature splicing and weight adjustment of the preprocessed multi-source data, and carrying out the feature splicing and weight adjustment of the preprocessed multi-source data; according to the method, features of different data sources are mapped to the same embedding space, dimensionality reduction and noise filtering are carried out on fused high-dimensional features, user features are extracted, time sequence modeling is carried out, a dynamic user portrait is generated based on core features after dimensionality reduction and a time sequence modeling result, and the user portrait is updated in real time. And user portrait feature parameters are optimized according to the user behavior feedback, and the change and the dynamic state of the user portrait are displayed. According to the method, the change and dynamic state of the user portrait are displayed through the chart, more personalized services are provided for an electric power company, and environmental protection and improvement of energy efficiency are promoted at the same time.

Description

Technical Field

[0001] The present invention relates to the technical field of power user portrait generation, and particularly to a method and system for dynamically updating multi-source heterogeneous power user portraits. Background Art

[0002] With the popularization of smart grids and the development of the power industry, power companies' in-depth analysis of users' power usage situations and generation of power user portraits have become important means for providing personalized services and precise demand response management. Traditional power user portrait generation methods usually rely on static single data sources, such as historical power consumption data and user basic information, and generate fixed user tags through simple data analysis. These static portraits cannot reflect users' behavioral changes in a timely manner and are difficult to meet the requirements of dynamic updating and precise services.

[0003] Existing power user portrait methods usually analyze historical data in an offline manner to generate fixed user tags and portraits. These methods mainly rely on historical power consumption records and basic user attribute data, ignoring the dynamic changes in users' power consumption behaviors. As users' behaviors are affected by multiple factors (such as weather changes, equipment usage situations, social events, etc.), existing methods cannot perform real-time updates, resulting in lagging user portraits and being difficult to reflect users' latest needs and behaviors.

[0004] Currently, power user portrait methods based on single data sources face the following main problems: static portraits, relying only on historical data, cannot reflect users' behavioral changes in a timely manner, affecting the service response speed and accuracy of power enterprises; the limitations of single data sources, lacking multi-dimensional information support, cannot comprehensively analyze users' behavioral characteristics, resulting in inaccurate portraits; lack of adaptability, the model parameters of existing methods are usually fixed and difficult to be adjusted in real time according to new data, thus affecting the accuracy and stability of portraits; the difficulty of integrating heterogeneous data, existing methods cannot effectively integrate data from different sources such as power, environment, and social, and are difficult to comprehensively grasp users' multi-dimensional needs.

[0005] Therefore, how to generate dynamic and real-time power user portraits based on multi-source heterogeneous data has become a technical problem urgently to be solved in the current power industry. In addition, existing research has less attention to the spatio-temporal characteristics of user-side carbon emissions and is difficult to accurately evaluate the contributions of carbon emissions at different time periods and different power-consuming devices. Power companies not only need to understand users' power consumption demands but also need to master the carbon emission situations on the user side and formulate precise low-carbon power consumption strategies; therefore, how to introduce carbon emission calculation and optimization into user portraits has become an important research direction. Summary of the Invention

[0006] In view of the above existing problems, the present invention provides a method and system for dynamically updating multi-source heterogeneous power user portraits to solve the problems in the prior art that the changes in user behavior cannot be reflected in a timely manner, there is a lack of multi-dimensional information support, the lack of adaptability, and it is difficult to comprehensively grasp the multi-dimensional needs of users.

[0007] To solve the above technical problems, a method for dynamically updating multi-source heterogeneous power user portraits is proposed, including

[0008] Collecting multi-source data, preprocessing the collected data, performing feature splicing and weight adjustment on the preprocessed multi-source data; using an embedding space fusion method to map the features of different data sources to the same embedding space, using a deep autoencoder to reduce the dimension and filter noise of the fused high-dimensional features, extracting user features, and using a long short-term memory network to perform time series modeling on user power consumption behavior and carbon emission data; based on the core features after dimension reduction and the time series modeling results, generating a dynamic user portrait, using a stream processing framework and a reinforcement learning feedback mechanism to update the user portrait in real time, optimizing the user portrait feature parameters according to user behavior feedback, and displaying the changes and dynamics of the user portrait.

[0009] As a preferred solution of the method for dynamically updating multi-source heterogeneous power user portraits according to the present invention, wherein: the collecting of multi-source data includes collecting power consumption data, environmental data, power carbon emission data, social feedback data, and equipment data;

[0010] The power consumption data includes power consumption load, voltage, current, peak-valley power consumption information; the environmental data includes temperature, humidity, wind speed; the social feedback data includes the concerns and dissatisfaction of users extracted from the opinions, comments, and suggestions of users on social platforms and power company feedback systems; the equipment data includes the usage frequency and usage time of electrical equipment.

[0011] As a preferred solution of the method for dynamically updating multi-source heterogeneous power user portraits according to the present invention, wherein: the preprocessing includes performing outlier cleaning, feature standardization, and time window alignment on the collected data;

[0012] The outlier cleaning includes using the box plot method to identify and remove outliers, wherein the box plot method includes calculating the first quartile, second quartile, and third quartile of the data, calculating the interquartile range, and determining the upper and lower boundaries of the outliers, identifying and removing data points below the lower boundary or above the upper boundary; the feature standardization includes using the mean-standard deviation normalization formula to perform feature data standardization, and the formula is expressed as:

[0013]

[0014] Among them, X is the original data, μ is the data mean, σ is the data standard deviation, and X norm is the normalized data;

[0015] The time window alignment includes aligning data with different time granularities to a fixed time window of 15 minutes through interpolation; the collected carbon emission data is standardized using the maximum-minimum normalization method, and the formula is expressed as:

[0016]

[0017] Among them, C is the original carbon emission data, C max is the maximum value of the carbon emission data, C min is the minimum value of the carbon emission data, C norm is the standardized carbon emission data.

[0018] As a preferred solution of the method for dynamically updating multi-source heterogeneous power user portraits described in the present invention, among them: the feature splicing and weight adjustment include splicing the features of each data source into a high-dimensional vector, assigning weights to different data sources, mapping the features of each data source to the same embedding space using a multi-layer perceptron, and performing weighted summation on the embedding representations to obtain a fused feature representation;

[0019] The formula for obtaining the fused feature representation is:

[0020]

[0021] Among them, w i is the weight of the features of the i-th data source, X fused is the fused feature vector, N is the total number of data sources, i is the variable index, and x i is the feature vector of the i-th data source;

[0022] The formula for the multi-layer perceptron is expressed as:

[0023] e i = MLP(x i )

[0024] Among them, x i is the feature vector of the i-th data source, e i is the embedded data, MLP is the multi-layer perceptron transformation, and i is the variable index;

[0025] The formula for obtaining the fused feature representation is:

[0026]

[0027] Among them, E is the obtained feature representation, w iis the weight of the i-th data source feature, e i is the embedded data, N is the total number of data sources, and i is the variable index.

[0028] As a preferred solution of the multi-source heterogeneous power user portrait method with dynamic update described in the present invention, wherein: the time series modeling of the user's electricity consumption behavior and carbon emission data includes using a deep autoencoder to reduce the dimension of the fused high-dimensional features, extracting user features, and using a long short-term memory network to perform time series modeling on the user's electricity consumption behavior and carbon emission data to predict the user's carbon emissions;

[0029] The long short-term memory network includes the calculations of an input gate, a forget gate, and an output gate, and the formula is expressed as:

[0030] f t = σ(W f [h t-1 , y t + b f )

[0031] i t = σ(W i [h t-1 , y t + b i )

[0032] o t = σ(W o [h t-1 , y t + b o )

[0033]

[0034] h t = o t · tanh(C t )

[0035] Among them, f t is the forget gate activation value, W f is the weight matrix of the forget gate, h t-1 is the hidden state of the previous time step, y t is the input of the current time step, b f is the bias vector of the forget gate, σ is the sigmoid activation function, i t is the input gate activation value, W i is the weight matrix of the input gate, b i is the bias vector of the input gate, o t is the output gate activation value, W o is the weight matrix of the output gate, b o is the bias vector of the output gate, Ct is the unit state at the current time step, C t-1 is the unit state at the previous time step is the candidate value of the unit state, h t is the generated dynamic time series feature, tanh is the hyperbolic tangent activation function, and t is the time

[0036] As a preferred solution of the dynamically updated multi-source heterogeneous power user portrait method described in the present invention, wherein: the real-time update of the user portrait includes using a stream processing framework to process data in real time, dynamically updating the user portrait according to a preset time window, and optimizing the user portrait feature parameters by using a reinforcement learning feedback mechanism to guide users to form low-carbon electricity consumption behavior habits

[0037] The dynamic update formula is expressed as

[0038]

[0039] where ω new is the updated model weight, ω old is the current weight, and η is the learning rate is the gradient of the loss function

[0040] The guiding users to form low-carbon electricity consumption behavior habits includes constructing a reward function, optimizing the reward function, using a reinforcement learning algorithm to adjust the feature parameters of the user portrait, encouraging users to use high-power devices during periods with lower carbon emissions, and adopting a reinforcement learning strategy based on user behavior feedback to optimize the Q function through a deep Q network

[0041] The construction of the reward function is expressed as

[0042] R t =-αD t +βU t

[0043] where D t is the carbon emission of the user at time t, U t is the user satisfaction, α and β are adjustable weight factors of the model, and R t is the reward

[0044] The optimization of the Q function is expressed as

[0045]

[0046] where S t is the current state, γ is the discount factor, R t is the reward, A t is the action taken, and is the next state S t+1Under this, the operation of taking the maximum Q value for all actions A is the expected return of taking action A in state S t state t action

[0047] As a preferred solution of the method for dynamically updating multi-source heterogeneous power user portraits according to the present invention, wherein: the display of the changes and dynamics of the user portrait includes extracting key features in the user portrait, converting the user portrait features into structured data, and displaying the changes and dynamics of the user portrait through charts;

[0048] The obtained final user portrait is expressed as:

[0049] UP = J(Z, h t )

[0050] where UP is the final user portrait, Z is the low-dimensional feature generated by the autoencoder, h t is the generated dynamic time series feature, and J is the dynamic change function; the generated user portrait includes electricity consumption demand features, demand preferences, and behavior patterns.

[0051] Another object of the present invention is to provide a system for dynamically updating multi-source heterogeneous power user portraits. The present invention promotes environmental protection and energy efficiency improvement by displaying the changes and dynamics of the user portrait through charts; the system of the present invention collects and preprocesses multi-source data, including electricity consumption, environment, carbon emissions, social feedback, and device usage data, and after feature splicing and weight adjustment of these data, uses a deep autoencoder and a long short-term memory network for feature fusion, dimensionality reduction, and time series modeling, thereby generating a dynamic user portrait including user behavior patterns, demand preferences, and electricity consumption habits.

[0052] As a preferred solution of a system for dynamically updating multi-source heterogeneous power user portraits according to the present invention, it includes a data collection and preprocessing module, a data collection and preprocessing module, a time series modeling and user portrait generation module, and a real-time update and optimization module.

[0053] The data collection and preprocessing module is used to collect heterogeneous data from different data sources and preprocess the collected data.

[0054] The feature fusion and dimensionality reduction module is used to splice and adjust the weights of the preprocessed multi-source data, map the features of different data sources to the same embedding space using the embedding space fusion method, and use a deep autoencoder to reduce the dimensionality and filter noise of the high-dimensional features, and extract the core features of the user.

[0055] The time series modeling and user profile generation module is used to perform time series modeling on user electricity consumption behavior and carbon emission data using a long short-term memory network, capture the dynamic behavior characteristics of users, and generate a dynamic user profile based on the core features after dimensionality reduction and the time series modeling results.

[0056] The real-time update and optimization module is used to perform real-time update of the user profile using a stream processing framework, optimize the feature parameters of the user profile through a reinforcement learning feedback mechanism according to user behavior feedback, guide users to form low-carbon electricity consumption behavior habits, and adjust the user profile by constructing and optimizing the reward function and the deep Q network to achieve a reduction in carbon emissions on the user side.

[0057] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method for a dynamically updated multi-source heterogeneous power user profile are implemented.

[0058] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for a dynamically updated multi-source heterogeneous power user profile are implemented.

[0059] Advantages of the present invention: By collecting and preprocessing multi-source data, including outlier cleaning, feature standardization, and time window alignment, the present invention ensures the multi-dimensionality and reliability of the data, laying a solid foundation for subsequent feature fusion and time series modeling; using embedding space fusion and deep autoencoder technology to extract the core features of users, and performing time series modeling on user electricity consumption behavior and carbon emission data using a long short-term memory network, improving the data processing efficiency and enhancing the prediction ability of the model; generating a dynamic user profile based on the core features after dimensionality reduction and the time series modeling results, and performing real-time update using a stream processing framework and a reinforcement learning feedback mechanism, realizing continuous optimization and personalization of the user profile, timely reflecting user behavior changes, guiding users to form low-carbon electricity consumption habits, and promoting environmental protection and energy conservation; and by converting the key features of the user profile into structured data and presenting them in the form of charts, the readability and intuitiveness of the user profile are improved, facilitating power companies and other stakeholders to understand and analyze user behavior, and formulating effective market strategies and energy management plans. Description of the Drawings

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0061] Figure 1The overall flowchart of a method for dynamically updating multi-source heterogeneous power user portraits provided by an embodiment of the present invention.

[0062] Figure 2 The system scheme flowchart of a system for dynamically updating multi-source heterogeneous power user portraits provided by an embodiment of the present invention. Detailed implementation manners

[0063] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following describes the detailed implementation manners of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0064] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0065] Secondly, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is mutually exclusive with other embodiments alone or selectively.

[0066] The present invention is described in detail in conjunction with schematic diagrams. When detailing the embodiments of the present invention, for the convenience of explanation, the cross-sectional views showing the device structure will be enlarged locally not in accordance with the general scale, and the schematic diagrams are only examples, which should not limit the protection scope of the present invention herein. In addition, in actual production, three-dimensional spatial dimensions of length, width, and depth should be included.

[0067] At the same time, in the description of the present invention, it should be noted that the orientation or positional relationships indicated by terms such as "upper, lower, inner, and outer" are based on the orientation or positional relationships shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0068] Unless otherwise clearly defined and limited in the present invention, the terms "installation, connection, and coupling" shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, or may be indirectly connected through an intermediate medium, or may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0069] Example 1. Refer to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a method for dynamically updating multi-source heterogeneous power user portraits, including:

[0070] S1: Collect multi-source data, preprocess the collected data, and perform feature splicing and weight adjustment on the preprocessed multi-source data;

[0071] The collection of multi-source data includes collecting power consumption data, environmental data, power carbon emission data, social feedback data, and device data;

[0072] The power consumption data includes power load, voltage, current, and peak-valley power consumption information; the environmental data includes temperature, humidity, and wind speed; the social feedback data includes extracting users' concerns and dissatisfaction through opinions, comments, and suggestions of users on social platforms and power company feedback systems; the device data includes the usage frequency and usage time of electrical equipment;

[0073] The carbon emission data includes power carbon emission factors, device-level carbon emission data, and time-series carbon emission data; the power carbon emission factors include calculating the carbon emissions per unit of electricity consumption based on the power grid energy structure (thermal power, wind power, photovoltaic); the device-level carbon emission data includes analyzing the carbon emission contributions of different household appliances (such as air conditioners, water heaters); the time-series carbon emission data includes the carbon emission characteristics in different time periods (such as night and day), and analyzing the peak-valley carbon emission pattern.

[0074] The preprocessing includes cleaning outliers, feature standardization, and time window alignment on the collected data;

[0075] Further, the outlier cleaning includes using the box plot method to identify and remove outliers. Among them, the box plot method includes calculating the first quartile, the second quartile, and the third quartile of the data, calculating the interquartile range, and determining the upper and lower boundaries of the outliers, and identifying and removing the data points below the lower boundary or above the upper boundary; the feature standardization includes using the mean-standard deviation normalization formula to standardize the feature data, and the formula is expressed as:

[0076]

[0077] Among them, X is the original data, μ is the data mean, σ is the data standard deviation, and X norm is the normalized data;

[0078] The above-mentioned time window alignment includes aligning data with different time granularities by interpolation according to a fixed time window, namely 15 minutes; the collected carbon emission data is standardized by the maximum-minimum normalization method, and the formula is expressed as:

[0079]

[0080] Among them, C is the original carbon emission data, and C max is the maximum value of the carbon emission data, and C min is the minimum value of the carbon emission data, and C norm is the standardized carbon emission data.

[0081] S2: Use the embedding space fusion method to map the features of different data sources to the same embedding space, adopt a deep autoencoder to reduce the dimension and filter noise of the fused high-dimensional features, extract user features, and use a long short-term memory network to perform time series modeling on user electricity consumption behavior and carbon emission data.

[0082] It should be noted that the above-mentioned feature splicing and weight adjustment include splicing the features of each data source into a high-dimensional vector, assigning weights to different data sources, using a multi-layer perceptron to map the features of each data source to the same embedding space, and performing weighted summation on the embedding representation to obtain the fused feature representation;

[0083] During the data fusion process, the present invention introduces a carbon emission calculation model to calculate the carbon emission characteristics on the user side, and the specific formula is as follows:

[0084] O t =P t ×EF t

[0085] Among them, O t is the carbon emission of the user at time t, P t is the electricity consumption power of the user at time t, EF t is the electricity carbon emission factor at this moment, and t is the time; the fused user portrait features not only include the user behavior pattern, but also include carbon emission influencing factors, providing data support for low-carbon electricity consumption optimization.

[0086] The formula for obtaining the fused feature representation is:

[0087]

[0088] Among them, w i is the weight of the feature of the i-th data source, and Xfused is the fused feature vector, N is the total number of data sources, i is the variable index, and x i is the feature vector of the i-th data source;

[0089] The formula of the multi-layer perceptron is expressed as:

[0090] e i = MLP(x i )

[0091] where x i is the feature vector of the i-th data source, e i is the embedded data, MLP is the multi-layer perceptron transformation, and i is the variable index;

[0092] The formula for obtaining the fused feature representation is:

[0093]

[0094] where E is the obtained feature representation, w i is the weight of the i-th data source feature, e i is the embedded data, N is the total number of data sources, and i is the variable index.

[0095] Furthermore, the time series modeling of the user's electricity consumption behavior and carbon emission data includes using a deep autoencoder to reduce the dimension of the fused high-dimensional features, extract user features, and using a long short-term memory network to perform time series modeling on the user's electricity consumption behavior and carbon emission data to predict the user's carbon emissions;

[0096] The long short-term memory network includes the calculations of an input gate, a forget gate, and an output gate, and the formula is expressed as:

[0097] f t = σ(W f [h t-1 , y t + b f )

[0098] i t = σ(W i [h t-1 , y t + b i )

[0099] o t = σ(W o [h t-1 , y t + b o )

[0100]

[0101] h t = o t ·tanh(C t )

[0102] where f t is the forget gate activation value, W f is the weight matrix of the forget gate, h t-1 is the hidden state at the previous time step, y t is the input at the current time step, b f is the bias vector of the forget gate, σ is the sigmoid activation function, i t is the input gate activation value, W i is the weight matrix of the input gate, b i is the bias vector of the input gate, o t is the output gate activation value, W o is the weight matrix of the output gate, b o is the bias vector of the output gate, C t is the cell state at the current time step, C t-1 is the cell state at the previous time step, is the candidate value of the cell state, h t is the generated dynamic time series feature, tanh is the hyperbolic tangent activation function, t is time.

[0103] S3: Based on the dimensionality-reduced core features and the time series modeling results, generate a dynamic user portrait, use a stream processing framework and a reinforcement learning feedback mechanism to update the user portrait in real time, optimize the user portrait feature parameters according to the user behavior feedback, and display the changes and dynamics of the user portrait.

[0104] Furthermore, the real-time update of the user portrait includes using a stream processing framework to process data in real time, dynamically updating the user portrait according to a preset time window, using a reinforcement learning feedback mechanism to optimize the user portrait feature parameters, and guiding users to form low-carbon electricity consumption behavior habits;

[0105] The dynamic update formula is expressed as:

[0106]

[0107] where ω new is the updated model weight, ω old is the current weight, η is the learning rate, is the gradient of the loss function;

[0108] The guidance for users to form low-carbon electricity consumption behavior habits includes constructing a reward function, optimizing the reward function, using a reinforcement learning algorithm to adjust the characteristic parameters of the user profile, encouraging users to use high-power devices during periods with lower carbon emissions, and adopting a reinforcement learning strategy based on user behavior feedback to optimize the Q function through a deep Q-network;

[0109] The construction of the reward function is expressed as:

[0110] R t =-αD t +βU t

[0111] where D t is the carbon emission of the user at time t, U t is the user satisfaction, α and β are weight factors adjustable by the model, and R t is the reward;

[0112] By optimizing the above reward function, the reinforcement learning algorithm will continuously adjust the characteristic parameters of the user profile, guiding users to be more inclined to use high-power devices during periods with lower carbon emissions, thereby achieving a reduction in carbon emissions on the user side.

[0113] The optimization of the Q function is expressed as:

[0114]

[0115] where S t is the current state, γ is the discount factor, R t is the reward, A t is the action taken, is the operation of taking the maximum Q value for all actions A in the next state S t+1 , and is the expected return of taking action A t in state S t .

[0116] Furthermore, the display of the changes and dynamics of the user profile includes extracting the key features in the user profile, converting the user profile features into structured data, and displaying the changes and dynamics of the user profile through charts;

[0117] The obtained final user profile is expressed as:

[0118] UP=J(Z,h t )

[0119] where UP is the final user profile, Z is the low-dimensional feature generated by the autoencoder, h t is the generated dynamic time-series feature, and J is the dynamic change function; the generated user profile includes electricity consumption demand features, demand preferences, and behavior patterns.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

[0121] Example 2. Referring to Figure 2 , which is the second embodiment of the present invention. This embodiment provides a dynamically updated multi-source heterogeneous power user portrait system, including a data collection and preprocessing module, a data collection and preprocessing module, a time series modeling and user portrait generation module, and a real-time update and optimization module.

[0122] The data collection and preprocessing module is used to collect heterogeneous data from different data sources and preprocess the collected data.

[0123] The feature fusion and dimensionality reduction module is used to splice and weight-adjust the features of the preprocessed multi-source data, map the features of different data sources to the same embedding space using the embedding space fusion method, and use a deep autoencoder to reduce the dimensionality and filter noise of the high-dimensional features to extract the core features of the user.

[0124] The time series modeling and user portrait generation module is used to perform time series modeling on the user's electricity consumption behavior and carbon emission data using a long short-term memory network, capture the dynamic behavior characteristics of the user, and generate a dynamic user portrait based on the dimensionality-reduced core features and the time series modeling results.

[0125] The real-time update and optimization module is used to perform real-time updates on the user portrait using a stream processing framework, optimize the feature parameters of the user portrait through a reinforcement learning feedback mechanism according to the user behavior feedback, guide the user to form low-carbon electricity consumption behavior habits, and adjust the user portrait by constructing and optimizing the reward function and the deep Q network to achieve a reduction in carbon emissions on the user side.

[0126] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

[0127] Example 3. The third embodiment of the present invention is different from the previous two embodiments in that:

[0128] If the above-mentioned functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0129] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0130] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), fiber optic devices, and portable compact disc read-only memories (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or processing it in other suitable ways if necessary, and then storing it in a computer memory.

[0131] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following techniques well known in the art or a combination thereof can be used: discrete logic circuits having logic gate circuits for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), and the like.

Claims

1. A method for dynamically updating multi-source heterogeneous power user portraits, characterized in that: Including Collecting multi-source data, preprocessing the collected data, and performing feature splicing and weight adjustment on the preprocessed multi-source data; Using the embedded space fusion method to map the features of different data sources to the same embedded space, using a deep autoencoder to reduce the dimension and filter noise of the fused high-dimensional features, extracting user features, and using a long short-term memory network to perform time series modeling on user electricity consumption behavior and carbon emission data; Based on the core features after dimension reduction and the time series modeling results, generating a dynamic user profile, using a stream processing framework and a reinforcement learning feedback mechanism to update the user profile in real time, optimizing the user profile feature parameters according to user behavior feedback, and displaying the changes and dynamics of the user profile.

2. The dynamic update multi-source heterogeneous power user profiling method according to claim 1, wherein: The collecting of multi-source data includes collecting electricity consumption data, environmental data, electricity carbon emission data, social feedback data, and device data; The electricity consumption data includes electricity load, voltage, current, and peak-valley electricity consumption information; the environmental data includes temperature, humidity, and wind speed; The social feedback data includes extracting users' concerns and dissatisfaction through opinions, comments, and suggestions of users in social platforms and the power company's feedback system; the device data includes the usage frequency and usage time of electrical equipment.

3. A method for dynamically updating multi-source heterogeneous power user portraits according to claim 2, characterized in that: The preprocessing includes cleaning outliers, feature standardization, and time window alignment on the collected data; The outlier cleaning includes using the box plot method to identify and remove outliers. Among them, the box plot method includes calculating the first quartile, second quartile, and third quartile of the data, calculating the interquartile range, and determining the upper and lower boundaries of the outliers, identifying and removing data points below the lower boundary or above the upper boundary; the feature standardization includes using the mean-standard deviation normalization formula to standardize the feature data, and the formula is expressed as: Among them, X is the original data, μ is the data mean, σ is the data standard deviation, and X norm is the data after normalization; The time window alignment includes aligning data with different time granularities to a fixed time window of 15 minutes by interpolation for different time granularity data; using the maximum-minimum normalization method to standardize the collected carbon emission data, and the formula is expressed as: Among them, C is the original carbon emission data, C max is the maximum value of the carbon emission data, C min is the minimum value of the carbon emission data, C norm is the carbon emission data after standardization.

4. The method for dynamically updating multi-source heterogeneous power user portraits according to claim 3, wherein: The performing of feature splicing and weight adjustment includes splicing the features of each data source into a high-dimensional vector, assigning weights to different data sources, using a multi-layer perceptron to map the features of each data source to the same embedded space, and performing weighted summation on the embedded representation to obtain the fused feature representation; The formula for obtaining the fused feature representation is: where, w i is the weight of the i-th data source feature, X fused is the fused feature vector, N is the total number of data sources, i is the variable index, and x i is the feature vector of the i-th data source; The formula of the multi-layer perceptron is expressed as: e i = MLP(x i ) where x i is the feature vector of the i-th data source, e i is the data after embedding, MLP is the multi-layer perceptron transformation, and i is the variable index; The formula for obtaining the fused feature representation is: Among them, E is the obtained feature representation, w i is the weight of the i-th data source feature, e i is the embedded data, N is the total number of data sources, and i is the variable index.

5. The method for dynamically updating multi-source heterogeneous power user portraits according to claim 4, wherein: The performing of time series modeling on user electricity consumption behavior and carbon emission data includes using a deep autoencoder to reduce the dimension of the fused high-dimensional features, extracting user features, and using a long short-term memory network to perform time series modeling on user electricity consumption behavior and carbon emission data to predict the user's carbon emissions; The long short-term memory network includes the calculations of an input gate, a forget gate, and an output gate, and the formula is expressed as: f t = σ(W f [h t-1 , y t + b f ) i t = σ(W i [h t-1 , y t + b i ) o t = σ(W o [h t-1 , y t + b o ) h t = o t ·tanh(C t ) Among them, f t is the forget gate activation value, W f is the weight matrix of the forget gate, h t-1 is the hidden state of the previous time step, y t is the input of the current time step, b f is the bias vector of the forget gate, σ is the sigmoid activation function, i t is the input gate activation value, W i is the weight matrix of the input gate, b i is the bias vector of the input gate, o t is the output gate activation value, W o is the weight matrix of the output gate, b o is the bias vector of the output gate, C t is the cell state of the current time step, C t-1 is the cell state of the previous time step, is the candidate value of the cell state, h t is the generated dynamic time series feature, tanh is the hyperbolic tangent activation function, t is the time.

6. The dynamic update multi-source heterogeneous power user profiling method according to claim 5, characterized in that: The real-time updating of the user profile includes using a stream processing framework to process data in real time, dynamically updating the user profile according to a preset time window, using a reinforcement learning feedback mechanism to optimize the user profile feature parameters, and guiding users to form low-carbon electricity consumption behavior habits; The dynamic update formula is expressed as: where ω new is the updated model weight, ω old is the current weight, η is the learning rate, is the gradient of the loss function; Guiding users to form low-carbon electricity consumption behavior habits includes constructing a reward function, optimizing the reward function, using a reinforcement learning algorithm to adjust the characteristic parameters of the user profile, motivating users to use high-power devices during periods with lower carbon emissions, and adopting a reinforcement learning strategy based on user behavior feedback to optimize the Q function through a deep Q-network; The construction of the reward function is expressed as: R t = -αD t + βU t Among them, D t is the carbon emission of the user at time t, U t is the user satisfaction, and α and β are weight factors adjustable in the model, R t is the reward; The optimization of the Q function is expressed as: Among them, S t is the current state, γ is the discount factor, R t is the reward, A t is the action taken, and is the operation of taking the maximum Q value for all actions A under the next state S t+1 , and is the expected return of taking the action A t under the state S t .

7. The method for dynamically updating multi-source heterogeneous power user portraits according to claim 6, wherein: Showing the changes and dynamics of the user profile includes extracting the key features in the user profile, transforming the user profile features into structured data, and presenting the changes and dynamics of the user profile through charts; The obtained final user profile is expressed as: UP = J(Z, h t ) Among them, UP is the end-user portrait, Z is the low-dimensional feature generated by the autoencoder, and h t is the generated dynamic time-series feature, and J is the dynamic change function; the generated user portrait includes electricity demand characteristics, demand preferences, and behavior patterns.

8. A system adopting a multi-source heterogeneous power user profiling method with dynamic update as described in any one of claims 1 to 7, characterized in that: It includes a data collection and preprocessing module, a feature fusion and dimensionality reduction module, a time series modeling and user profile generation module, and a real-time update and optimization module; The data collection and preprocessing module is used to collect heterogeneous data from different data sources and preprocess the collected data; The feature fusion and dimensionality reduction module is used to splice features and adjust weights for the preprocessed multi-source data, map the features of different data sources to the same embedding space using the embedding space fusion method, and use a deep autoencoder to reduce the dimensionality and filter noise of high-dimensional features to extract the core features of the user; The time series modeling and user profile generation module is used to perform time series modeling on the user's electricity consumption behavior and carbon emission data using a long short-term memory network to capture the dynamic behavior characteristics of the user, and generate a dynamic user profile based on the core features after dimensionality reduction and the time series modeling results; The real-time update and optimization module is used to perform real-time updates on the user profile using a stream processing framework, and optimize the characteristic parameters of the user profile through a reinforcement learning feedback mechanism according to user behavior feedback to guide users to form low-carbon electricity consumption behavior habits.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a method for dynamically updating a multi-source heterogeneous power user profile according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for dynamically updating a multi-source heterogeneous power user profile according to any one of claims 1 to 7.

Images