Resident energy consumption behavior analysis method and system based on multi-modal data

Abstract

The invention belongs to the technical field of energy consumption analysis, and discloses a resident energy consumption behavior analysis method and system based on multi-modal data. The method comprises the following steps: constructing a regional user portrait generation model, a multi-modal feature fusion model, a resident energy consumption behavior analysis model and a regional electric energy scheduling model based on a cloud data center; real-time resident information and real-time multi-mode monitoring data are collected and preprocessed; performing regional user portrait generation by using a regional user portrait generation model; performing multi-modal feature fusion by using a multi-modal feature fusion model; performing resident energy consumption behavior analysis by using the resident energy consumption behavior analysis model; and performing regional electric energy scheduling by using the regional electric energy scheduling model. According to the invention, the problems of insufficient accuracy, insufficient data processing capability and lack of response mechanism in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the field of energy consumption analysis, and in particular relates to a method and system for analyzing residents' energy consumption behavior based on multimodal data. Background Art

[0002] With the development of the global economy and population growth, energy demand continues to rise. Residential energy demand has a significant impact on energy supply and distribution. Therefore, understanding and analyzing residential energy behavior is crucial for energy planning and policy making.

[0003] The existing technology for analyzing residents’ energy consumption behavior has the following defects:

[0004] 1) Insufficient accuracy: Existing energy consumption behavior analysis methods often rely on a single type of data source, such as only considering residential energy consumption data, while ignoring other relevant multimodal data, such as environmental data and user information data. The limitations of this single data source may result in incomplete and inaccurate analysis results;

[0005] 2) Insufficient data processing capabilities: Existing technologies lack sufficient data processing capabilities when faced with large amounts of data, resulting in low analysis efficiency and difficulty in meeting the needs of real-time analysis;

[0006] 3) Lack of response mechanism: The existing technology only analyzes the energy consumption behavior of residents, but lacks a regional power dispatch mechanism to respond to the analysis results, and the degree of automation and intelligence is low. Summary of the invention

[0007] In order to solve the problems of insufficient accuracy, insufficient data processing capability and lack of response mechanism in the prior art, the present invention aims to provide a method and system for analyzing residents' energy consumption behavior based on multimodal data.

[0008] The technical solution adopted by the present invention is:

[0009] A method for analyzing residents' energy consumption behavior based on multimodal data includes the following steps:

[0010] Based on the cloud data center, using machine learning algorithms, we build regional user portrait generation models, multimodal feature fusion models, resident energy consumption behavior analysis models, and regional power dispatch models.

[0011] Collecting real-time resident information and corresponding real-time multimodal monitoring data of the target area, and preprocessing them to obtain a number of preprocessed real-time resident information and a number of preprocessed real-time multimodal monitoring data;

[0012] Based on some pre-processed real-time resident information, a regional user portrait generation model is used to generate a regional user portrait, and a real-time regional user portrait of the target area is obtained;

[0013] According to the real-time regional user portrait and some pre-processed real-time multimodal monitoring data, a multimodal feature fusion model is used to perform multimodal feature fusion to obtain the corresponding real-time multimodal fusion features;

[0014] According to the real-time multimodal fusion characteristics, the resident energy consumption behavior analysis model is used to analyze the resident energy consumption behavior and obtain the real-time resident energy consumption behavior analysis results in the target area;

[0015] According to the real-time analysis results of residents' energy consumption behavior, regional power dispatching model is used to carry out regional power dispatching and obtain the real-time regional power dispatching plan for the target area.

[0016] Furthermore, the real-time multimodal monitoring data includes real-time resident energy consumption data, real-time regional energy configuration data, and real-time regional environmental data in the target area.

[0017] Furthermore, based on the cloud data center, a regional user portrait generation model, a multimodal feature fusion model, a resident energy consumption behavior analysis model, and a regional power dispatch model are constructed using a machine learning algorithm, including the following steps:

[0018] Based on the cloud data center, a number of historical resident information and corresponding historical multimodal monitoring data of different regions are collected and preprocessed to obtain a number of preprocessed historical resident information and a number of preprocessed historical multimodal monitoring data;

[0019] Based on the pre-processed historical resident information of different regions, a regional user portrait generation model is constructed using clustering and deep learning fusion algorithms, and several historical regional user portraits are generated;

[0020] According to the historical regional user portraits of different regions and the corresponding pre-processed historical multimodal monitoring data, a multimodal feature fusion algorithm is used to build a multimodal feature fusion model and generate several historical multimodal fusion features;

[0021] Based on several historical multimodal fusion features, a deep learning algorithm is used to build a resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results;

[0022] Based on the analysis results of several historical residents' energy consumption behaviors, a regional power dispatch model was constructed using an enhanced reinforcement learning algorithm.

[0023] Furthermore, the regional user portrait generation model is constructed based on the FCM-RF algorithm, and the regional user portrait generation model includes a group user information clustering module constructed based on the FCM clustering algorithm and a regional user portrait generation module constructed based on the RF algorithm, which are connected in sequence;

[0024] The multimodal feature fusion model is constructed based on the GCN-LSTM-Attention algorithm, and the multimodal feature fusion model includes a graph feature extraction module constructed based on the GCN algorithm, a data feature extraction module constructed based on the LSTM algorithm, and a weighted fusion module constructed based on the Attention mechanism. Both the graph feature extraction module and the data feature extraction module are connected to the weighted fusion module;

[0025] The resident energy consumption behavior analysis model is constructed based on the CFWA-MLP algorithm, and the resident energy consumption behavior analysis model includes a network parameter optimization module constructed based on the CFWA algorithm and a resident energy consumption behavior analysis module constructed based on the MLP algorithm;

[0026] The regional electric energy dispatching model is constructed based on the ML-DQN algorithm, and the regional electric energy dispatching model includes a meta-learning module constructed based on the ML algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is respectively connected to the deep Q network, the experience replay pool and the meta-learning module, and the meta-learning module is connected to the experience replay pool.

[0027] Furthermore, based on several historical multimodal fusion features, a deep learning algorithm is used to construct a resident energy consumption behavior analysis model, and several historical resident energy consumption behavior analysis results are generated, including the following steps:

[0028] Using the CFWA-MLP algorithm, an initial resident energy consumption behavior analysis model is constructed, which includes a network parameter optimization module and an initial resident energy consumption behavior analysis module;

[0029] Use the network parameter optimization module to iteratively optimize the initial network parameters of the initial resident energy consumption behavior analysis module to obtain the optimal initial network parameters;

[0030] According to the optimal initial network parameters, the initial resident energy consumption behavior analysis module is optimized to obtain an optimized resident energy consumption behavior analysis module;

[0031] According to several historical multimodal fusion features, the optimized resident energy consumption behavior analysis module is optimized and trained to obtain the final resident energy consumption behavior analysis module;

[0032] The network parameter optimization module and the final resident energy consumption behavior analysis module are integrated to obtain the final resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results.

[0033] Furthermore, based on the analysis results of several historical residents’ energy consumption behaviors, an enhanced reinforcement learning algorithm is used to construct a regional power dispatch model, which includes the following steps:

[0034] The regional power dispatch problem is used as the simulation environment of the DQN algorithm, and the regional power dispatch problem is divided into several subtasks;

[0035] Define meta-feature parameters in the historical analysis results of residents' energy consumption behavior, and extract several historical meta-features of the historical analysis results of residents' energy consumption behavior based on the meta-feature parameters;

[0036] According to several historical meta-features of the analysis results of different historical residents’ energy consumption behaviors, ML algorithms are used for training under several subtasks to obtain the initial meta-learning module and several historical meta-learning experiences;

[0037] Use the initial meta-learning module to initialize the deep Q network of the DQN algorithm, and update the initial meta-learning module to obtain an updated meta-learning module;

[0038] Set the environment state, action space, and reward function of the DQN algorithm, and build the agent and experience replay pool to obtain the initial reinforcement learning module;

[0039] Based on the analysis results of several historical residents’ energy consumption behaviors, the initial reinforcement learning module is trained and optimized to obtain the final reinforcement learning module and several historical regional power dispatch experiences;

[0040] Several historical meta-learning experiences and several historical regional power dispatching experiences are stored in the experience replay pool, and combined with the final reinforcement learning module and the updated meta-learning module to obtain the regional power dispatching model.

[0041] Furthermore, according to some pre-processed real-time resident information, a regional user portrait generation model is used to generate a regional user portrait to obtain a real-time regional user portrait of the target area, including the following steps:

[0042] Input some pre-processed real-time resident information into the regional user portrait generation model;

[0043] Using the group user information clustering module, clustering is performed on a number of pre-processed real-time resident information to obtain a number of real-time clustering centers;

[0044] According to the pre-processed real-time resident information corresponding to several real-time clustering centers, the regional user portrait generation module is used to generate the regional user portrait to obtain the real-time regional user portrait of the target area.

[0045] Furthermore, according to the real-time regional user portrait and a number of pre-processed real-time multimodal monitoring data, a multimodal feature fusion model is used to perform multimodal feature fusion to obtain corresponding real-time multimodal fusion features, including the following steps:

[0046] Input the real-time regional user portrait and several pre-processed real-time multimodal monitoring data into the multimodal feature fusion model;

[0047] Use the graph feature extraction module to extract the real-time graph features of the real-time regional user portrait, and use the data feature extraction module to extract the real-time data features of several pre-processed real-time multimodal monitoring data;

[0048] According to the preset attention weights of the weighted fusion module, the real-time graph features and real-time data features are weightedly fused to obtain the corresponding real-time multimodal fusion features.

[0049] Furthermore, according to the real-time analysis results of residents' energy consumption behavior, regional power dispatching is performed using a regional power dispatching model to obtain a real-time regional power dispatching plan for the target area, including the following steps:

[0050] The real-time analysis results of residents' energy consumption behavior are input into the regional electric energy dispatch model, and several real-time meta-features of the real-time analysis results of residents' energy consumption behavior are extracted according to the meta-feature parameters;

[0051] According to a number of real-time meta-features, the corresponding real-time meta-learning experience is matched in the experience replay pool, and the regional power dispatch model is updated according to the real-time meta-learning experience to obtain an updated regional power dispatch model;

[0052] According to the real-time analysis results of residents' energy consumption behavior, the corresponding real-time regional power dispatch experience is matched in the experience playback pool;

[0053] According to the real-time analysis results of residents' energy consumption behavior and the real-time regional power dispatching experience, the updated regional power dispatching model is used to carry out regional power dispatching and obtain the real-time regional power dispatching plan for the target area.

[0054] A resident energy consumption behavior analysis system based on multimodal data is used to implement a resident energy consumption behavior analysis method. The system includes a cloud data center and several data acquisition devices. The cloud data center is communicated with the several data acquisition devices respectively, and the cloud data center includes a model construction unit, a regional user portrait generation unit, a multimodal fusion unit, a resident energy consumption behavior analysis unit and a regional power dispatching unit which are connected in sequence.

[0055] The beneficial effects of the present invention are:

[0056] The present invention discloses a resident energy consumption behavior analysis method and system based on multimodal data, which comprehensively considers the importance of multimodal data for the analysis of resident energy consumption behavior, breaks the limitation of a single data source, and can more comprehensively analyze the resident energy consumption behavior, thereby improving the accuracy of the analysis result; adopts multimodal feature fusion technology to effectively integrate the features of different data sources, improves data processing capabilities, and thus improves analysis efficiency, meeting the demand for real-time analysis; constructs a regional user portrait generation model, which can dynamically update the regional user portrait according to real-time data, while reducing the difficulty of analysis, improves the accuracy and real-time performance of the regional user portrait, better reflects the energy consumption habits and needs of regional users, and thus improves the accuracy of the analysis of resident energy consumption behavior, is suitable for the energy consumption characteristics of different regions or different user groups, and improves the adaptability of the resident energy consumption behavior analysis model; constructs a regional electric energy dispatching model, realizes a response mechanism, and performs automatic and intelligent regional electric energy dispatching according to the analysis results of the resident energy consumption behavior, reduces the demand for manual intervention, optimizes energy utilization efficiency, and achieves energy conservation and emission reduction.

[0057] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flowchart of the method for analyzing residents' energy consumption behavior based on multimodal data in the present invention.

[0059] Figure 2 It is a structural block diagram of the resident energy consumption behavior analysis system based on multimodal data in the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0061] Embodiment 1:

[0062] like Figure 1 As shown, this embodiment provides a method for analyzing residents' energy consumption behavior based on multimodal data, comprising the following steps:

[0063] S1: Based on the cloud data center, using machine learning algorithms, build a regional user portrait generation model, a multimodal feature fusion model, a resident energy consumption behavior analysis model, and a regional power scheduling model, including the following steps:

[0064] S1-1: Based on the cloud data center, collect some historical resident information and corresponding historical multimodal monitoring data of different regions, and perform preprocessing to obtain some preprocessed historical resident information and some preprocessed historical multimodal monitoring data;

[0065] S1-2: Based on the pre-processed historical resident information of different regions, a clustering and deep learning fusion algorithm is used to build a regional user portrait generation model and generate several historical regional user portraits;

[0066] The regional user portrait generation model is constructed based on the fuzzy C-mean (FCM)-random forest (RF) algorithm, and the regional user portrait generation model includes a group user information clustering module constructed based on the FCM clustering algorithm and a regional user portrait generation module constructed based on the RF algorithm, which are connected in sequence;

[0067] The FCM clustering algorithm clusters data based on similarities, divides user information with the same characteristics, and obtains information clusters of residents belonging to the same group in the region. Through clustering, the complexity and computational complexity of resident information processing are reduced, and the efficiency of user portrait generation is improved; the RF algorithm screens the features of user information through the internal Classification and Regression Tree (CART), which can process a large number of features and select the most stable and distinguishing key features as the basis for predicting user portrait labels, including group commonality, energy usage preferences, energy usage rate, and energy configuration requirements.

[0068] S1-3: Based on the historical regional user portraits of different regions and the corresponding pre-processed historical multimodal monitoring data, a multimodal feature fusion algorithm is used to build a multimodal feature fusion model and generate several historical multimodal fusion features;

[0069] The multimodal feature fusion model is constructed based on the graph convolutional network (GCN)-long short-term memory network (LSTM)-Attention algorithm, and the multimodal feature fusion model includes a graph feature extraction module constructed based on the GCN algorithm, a data feature extraction module constructed based on the LSTM algorithm, and a weighted fusion module constructed based on the Attention mechanism. The graph feature extraction module and the data feature extraction module are both connected to the weighted fusion module;

[0070] The LSTM algorithm can capture long-distance dependencies and keep the memory of previous information in the sequence, which can help the model better understand the temporal relationship in the sequence and extract deep features in multimodal monitoring data. The GCN algorithm propagates features on regional user portraits through operations similar to convolution, extracts node features and edge features between entities, and constructs graph features of regional user portraits. The weighted fusion module uses preset attention weights to achieve weighted fusion of deep data features and graph features, integrate scattered features, and enhance the influence of features on label prediction.

[0071] S1-4: Based on several historical multimodal fusion features, a deep learning algorithm is used to construct a resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results;

[0072] The resident energy consumption behavior analysis model is constructed based on the Convergent Fireworks Optimization Algorithm (CFWA)-Multilayer Perceptron (MLP) algorithm, and the resident energy consumption behavior analysis model includes a network parameter optimization module constructed based on the CFWA algorithm and a resident energy consumption behavior analysis module constructed based on the MLP algorithm;

[0073] The MLP network is a general feedforward neural network that is suitable for classification and prediction of various types of fusion data, which improves the accuracy of analysis of residents' energy consumption behavior. The CFWA algorithm optimizes the initial network parameters of the MLP network, preventing the MLP network from falling into a local optimal value and being unable to jump out, reducing the sensitivity to the initial value of the network, and improving the robustness of the model, the efficiency and accuracy of feature extraction.

[0074] Based on several historical multimodal fusion features, a deep learning algorithm is used to build a resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results, including the following steps:

[0075] S1-4-1: Using the CFWA-MLP algorithm, construct an initial resident energy consumption behavior analysis model including a network parameter optimization module and an initial resident energy consumption behavior analysis module;

[0076] S1-4-2: Use the network parameter optimization module to iteratively optimize the initial network parameters of the initial resident energy consumption behavior analysis module to obtain the optimal initial network parameters, including the following steps:

[0077] S1-4-2-1: Taking the minimization of mean square error as the optimization objective, the initial network parameters of the residents’ energy consumption behavior analysis module are encoded into individual vectors of the CFWA algorithm;

[0078] S1-4-2-2: According to the optimization goal, set the fitness function of the CFWA algorithm, and set the CFWA population parameters and the maximum number of iterations;

[0079] The formula is:

[0080]

[0081] Where fit is the fitness function; MSE is the mean square error function; Y n' is the true output value of the n'th sample; is the ideal output value of the n'th sample; n' is the sample indicator; N is the total number of samples;

[0082] S1-4-2-3: Initialize according to the CFWA population parameters and individual vectors to generate several initial CFWA individuals of the initial CFWA population;

[0083] The formula is:

[0084]

[0085] In the formula, q l' is the initial CFWA individual of the Circle chaos map; q' l' is the randomly generated initial CFWA individual; l' is the CFWA individual indicator;

[0086] S1-4-2-4: According to the fitness function, the explosion radius, number of sparks and fitness value of the initial CFWA individual are obtained;

[0087] The formula is:

[0088]

[0089] In the formula, S l' is the initial CFWA individual q l' The number of sparks; M' is the number constant; f max is the maximum fitness value in the initialized CFWA population; f(q l' ) is the initial CFWA individual q l' The fitness value; τ is an infinitesimal constant; a is the convergence factor; σ is a positive real number that is not 0;

[0090]

[0091] In the formula, R l' is the initial CFWA individual q l' Explosion radius; f is the explosion radius adjustment constant; min is the minimum fitness value in the initialized CFWA population;

[0092]

[0093] Where a is the convergence factor; tanh(.) is the hyperbolic tangent function; t, t max are the current number of iterations and the maximum number of iterations respectively; a max 、a min are the maximum and minimum values ​​of the convergence factor, respectively; λ is the decreasing rate parameter, k is the decreasing period parameter, λ = -2π, k = π;

[0094] The number of sparks determines the number of sub-fireworks produced after each firework explodes, and the explosion radius determines the distribution range of the sparks produced after the firework explodes in the solution space. In the early iteration, the value of a is larger, the number of sparks of the CFWA individual is smaller, and the explosion radius is larger, which helps to reduce the computational burden and is more widely distributed, which helps to explore more solution spaces. In the later iteration, a smaller explosion radius helps to perform fine search in local areas, and a larger number of sparks helps to increase the diversity of the search.

[0095] S1-4-2-5: Perform fireworks explosion according to the explosion radius, the number of sparks and the fitness value to obtain a number of updated CFWA individuals of the updated CFWA population;

[0096] The formula is:

[0097] q' l' =q l' +S l' ×rand(-1,1)

[0098] In the formula, q' l' is the updated CFWA individual; rand(-1,1) is a random number between -1 and 1;

[0099] S1-4-2-6: Use the Gaussian mutation algorithm to perform Gaussian mutation on the initialized CFWA population to generate several Gaussian mutated CFWA individuals of the Gaussian mutated CFWA population;

[0100] The formula is:

[0101] q" l' =q l' +S l' ×G(1,1)

[0102] In the formula, q" l' is a CFWA individual with Gaussian variation; G(1,1) is a random number with Gaussian distribution whose mean and variance are both 1;

[0103] S1-4-2-7: Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the initialized CFWA population to obtain several reverse CFWA individuals of the reverse CFWA population;

[0104] The formula is:

[0105] q"' l' =γ(L max +L min )-q l'

[0106] In the formula, q"' l' is the reverse CFWA individual; γ is the decreasing inertia coefficient; L max , L min are the maximum and minimum values ​​of the vector space respectively;

[0107] S1-4-2-8: Obtain the fitness value of each updated CFWA individual, Gaussian mutated CFWA individual and reverse CFWA individual, and take the CFWA individual with the minimum fitness value as the optimal individual;

[0108] S1-4-2-9: If the number of iterations of iterative optimization reaches the maximum number of iterations or the fitness value of the optimal individual meets the requirements, the optimal individual will be output;

[0109] S1-4-2-10: Decode the individual vector of the optimal individual to obtain the optimal initial network parameters of the initial resident energy consumption behavior analysis module; the initial network parameters include the initial weights of neurons in the MLP network, the initial threshold, the bias between hidden layers, the bias from the hidden layer to the output layer, the learning rate, etc.;

[0110] S1-4-3: Optimizing the initial resident energy consumption behavior analysis module according to the optimal initial network parameters to obtain an optimized resident energy consumption behavior analysis module;

[0111] S1-4-4: Based on several historical multimodal fusion features, the optimized resident energy consumption behavior analysis module is optimized and trained to obtain the final resident energy consumption behavior analysis module;

[0112] S1-4-5: Integrate the network parameter optimization module and the final resident energy consumption behavior analysis module to obtain the final resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results;

[0113] S1-5: Based on the analysis results of several historical residents' energy consumption behaviors, a regional power dispatch model is constructed using an enhanced reinforcement learning algorithm;

[0114] The regional electric energy dispatching model is constructed based on the Meta-Learning (ML)-Deep Q Network (DQN) algorithm, and the regional electric energy dispatching model includes a meta-learning module constructed based on the ML algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is connected to the deep Q network, the experience replay pool and the meta-learning module respectively, and the meta-learning module is connected to the experience replay pool.

[0115] The meta-learning module enables the model to learn quickly on a small number of samples. This allows the model to quickly adapt to new working environments and improve the generalization ability of the model for different resident energy behavior analysis results in different regions. Even under unseen resident energy behavior analysis results, the deep Q network can be updated based on previous learning experience. Through meta-learning, the DQN algorithm can converge faster, reducing the amount of data and time required for training, which can significantly improve learning efficiency for regional power scheduling tasks.

[0116] Based on the analysis results of several historical residents' energy consumption behaviors, an enhanced reinforcement learning algorithm is used to build a regional power dispatch model, which includes the following steps:

[0117] S1-5-1: Use the regional power dispatch problem as the simulation environment of the DQN algorithm and divide the regional power dispatch problem into several subtasks;

[0118] S1-5-2: Define meta-feature parameters in the historical analysis results of residents’ energy consumption behavior, and extract several historical meta-features of the historical analysis results of residents’ energy consumption behavior based on the meta-feature parameters;

[0119] S1-5-3: Based on several historical meta-features of the analysis results of different historical residents’ energy consumption behaviors, ML algorithms are used for training under several subtasks to obtain the initial meta-learning module and several historical meta-learning experiences;

[0120] S1-5-4: Use the initial meta-learning module to initialize the deep Q network of the DQN algorithm, and update the initial meta-learning module to obtain an updated meta-learning module;

[0121] S1-5-5: Set the environment state, action space, and reward function of the DQN algorithm, and build the agent and experience replay pool to obtain the initial reinforcement learning module;

[0122] S1-5-6: Based on the analysis results of several historical residents’ energy consumption behaviors, the initial reinforcement learning module is trained and optimized to obtain the final reinforcement learning module and several historical regional power dispatch experiences;

[0123] S1-5-7: Store several historical meta-learning experiences and several historical regional power dispatching experiences into the experience replay pool, and combine the final reinforcement learning module and the updated meta-learning module to obtain the regional power dispatching model;

[0124] S2: collecting real-time resident information and corresponding real-time multimodal monitoring data of the target area, and preprocessing them to obtain a number of preprocessed real-time resident information and a number of preprocessed real-time multimodal monitoring data;

[0125] Real-time multimodal monitoring data include real-time residential energy consumption data in the target area (residential energy consumption-related data, including electricity, water, gas and other energy consumption data), real-time regional energy configuration data (including equipment and information for the production, transmission and distribution of electricity, water, gas and other energy sources in the region), and real-time regional environmental data (including regional climate, weather, ambient temperature and humidity data); resident information includes basic information such as family demographic information, living habits, preferences, and historical energy consumption data;

[0126] Preprocessing includes data cleaning, format conversion, and normalization. While eliminating duplicate and erroneous data, it also performs standardized conversion, eliminates magnitude differences and noise, and converts the data into a format that can be recognized by subsequent models.

[0127] S3: Based on some pre-processed real-time resident information, a regional user profile generation model is used to generate a regional user profile to obtain a real-time regional user profile of the target area, including the following steps:

[0128] S3-1: Input some pre-processed real-time resident information into the regional user portrait generation model;

[0129] S3-2: using the group user information clustering module, clustering a number of pre-processed real-time resident information to obtain a number of real-time clustering centers;

[0130] S3-3: Based on the pre-processed real-time resident information corresponding to a number of real-time clustering centers, a regional user portrait generation module is used to generate a regional user portrait to obtain a real-time regional user portrait of the target area;

[0131] S4: Based on the real-time regional user portrait and a number of pre-processed real-time multimodal monitoring data, a multimodal feature fusion model is used to perform multimodal feature fusion to obtain corresponding real-time multimodal fusion features, including the following steps:

[0132] S4-1: Input the real-time regional user portrait and several pre-processed real-time multimodal monitoring data into the multimodal feature fusion model;

[0133] S4-2: Use the graph feature extraction module to extract the real-time graph features of the real-time regional user portrait, and use the data feature extraction module to extract the real-time data features of several pre-processed real-time multimodal monitoring data;

[0134] S4-3: According to the attention weights preset by the weighted fusion module, the real-time graph features and the real-time data features are weightedly fused to obtain the corresponding real-time multimodal fusion features;

[0135] S5: Based on the real-time multimodal fusion features, the resident energy consumption behavior analysis model is used to analyze the resident energy consumption behavior and obtain the real-time resident energy consumption behavior analysis results in the target area;

[0136] The results of the analysis of residents’ energy consumption behavior include the overall regional energy consumption overview, energy consumption trend forecast, peak energy consumption period, energy-consuming equipment analysis, energy consumption structure, energy-saving potential analysis, environmental impact assessment, and regional energy consumption recommendations;

[0137] S6: Based on the real-time analysis results of residents' energy consumption behavior, regional power dispatching is performed using a regional power dispatching model to obtain a real-time regional power dispatching plan for the target area, including the following steps:

[0138] S6-1: input the real-time resident energy consumption behavior analysis results into the regional power dispatch model, and extract several real-time meta-features of the real-time resident energy consumption behavior analysis results according to the meta-feature parameters;

[0139] S6-2: According to a number of real-time meta-features, the corresponding real-time meta-learning experience is matched in the experience replay pool, and the regional power dispatch model is updated according to the real-time meta-learning experience to obtain an updated regional power dispatch model, including the following steps:

[0140] S6-2-1: According to several real-time meta-features, several historical meta-learning experiences in the experience replay pool are matched, and the historical meta-learning experience with the highest similarity is used as the real-time meta-learning experience;

[0141] S6-2-2: Based on the real-time meta-learning experience, use the meta-learning module to generate the optimal network parameters of the reinforcement learning module in the regional power scheduling model, including the number of neurons, neuron weights, and learning rate of the deep Q network;

[0142] S6-2-3: updating the reinforcement learning module according to the optimal network parameters to obtain an updated reinforcement learning module, and updating the meta-learning module to obtain an updated meta-learning module;

[0143] S6-2-4: Integrate the updated reinforcement learning module and the updated meta-learning module to obtain an updated regional power dispatch model;

[0144] S6-3: According to the real-time analysis results of residents’ energy consumption behavior, the corresponding real-time regional power dispatch experience is matched in the experience playback pool;

[0145] S6-4: Based on the real-time analysis results of residents' energy consumption behavior and the real-time regional power dispatching experience, the updated regional power dispatching model is used to perform regional power dispatching to obtain a real-time regional power dispatching plan for the target area, including the following steps:

[0146] S6-4-1: According to the real-time regional power dispatch experience, the action space of the updated regional power dispatch model is adjusted to obtain the updated action space A'=[a'1,...,a' j" ,...,a' I ], where a' j" is the updated j-th "action value", j" is the action indicator, and I is the total number of action space dimensions;

[0147] S6-4-2: According to the real-time analysis results of residents' energy consumption behavior, the state space of the updated regional power dispatch model is adjusted to obtain the updated state space S'=[s'1,...,s' i" ,...,s' I' ], where s' i" is the updated i'th state value, i' is the state indicator, and I' is the total number of state space dimensions;

[0148] The status includes the overall regional energy consumption overview, energy consumption trend forecast, peak energy consumption period, energy consumption equipment analysis, energy consumption structure, energy saving potential analysis, environmental impact assessment and regional energy consumption recommendations, etc.

[0149] Actions include adjusting power generation plans, implementing demand response, and controlling energy storage systems;

[0150] S6-4-3: input the updated state space into the deep Q network of the updated regional power dispatch model, and generate the Q value of each possible action in the updated action space using the deep Q network based on the updated state space;

[0151] S6-4-4: Use the agent to obtain the reward value of each possible action in the updated action space according to the reward function, and update the Q value of the possible action according to the reward value to obtain the updated Q value of each possible action;

[0152] The formula is:

[0153] Q(s' p' ,a' p' )=(1-α")·Q(s p' ,a p' )+α"·(R(s p' ,ap' ,s' p' )+γ·Q max (s p' ,a p' ))

[0154] In the formula, Q(s' p' ,a' p' ) is the updated state value s' p' and the updated action value a' p' The corresponding updated Q value; Q(s p' ,a p' ) is the state value s p' and action value a p' The corresponding predicted Q value; α" is the learning rate; Q max (s p' ,a p' ) is the highest predicted Q value; p' is the comprehensive indicator; γ is the update parameter;

[0155] S6-4-5: Repeat the above Q-value updating steps until the number of iterations reaches the iteration number threshold, and use the greedy strategy to select the possible action with the highest updated Q-value as the execution action;

[0156] S6-4-6: Based on the execution action, obtain the real-time regional power dispatch plan.

[0157] Embodiment 2:

[0158] like Figure 2 As shown, this embodiment provides a resident energy behavior analysis system based on multimodal data, which is used to implement a resident energy behavior analysis method. The system includes a cloud data center and several data acquisition devices. The cloud data center is respectively connected to the several data acquisition devices for communication, and the cloud data center includes a model building unit, a regional user portrait generation unit, a multimodal fusion unit, a resident energy behavior analysis unit, and a regional power scheduling unit connected in sequence;

[0159] A data collection device, used to collect real-time resident information and corresponding real-time multimodal monitoring data of the target area, and send the real-time resident information and corresponding real-time multimodal monitoring data to a cloud data center;

[0160] A model building unit, which is used to use machine learning algorithms to build a regional user portrait generation model, a multimodal feature fusion model, a resident energy consumption behavior analysis model, and a regional power dispatch model;

[0161] A regional user portrait generation unit is used to perform preprocessing to obtain a number of preprocessed real-time resident information and a number of preprocessed real-time multimodal monitoring data; based on the preprocessed real-time resident information, a regional user portrait generation model is used to generate a regional user portrait to obtain a real-time regional user portrait of the target area;

[0162] The multimodal fusion unit is used to perform multimodal feature fusion based on the real-time regional user portrait and a number of pre-processed real-time multimodal monitoring data, using a multimodal feature fusion model to obtain corresponding real-time multimodal fusion features;

[0163] The resident energy consumption behavior analysis unit is used to analyze the resident energy consumption behavior according to the real-time multimodal fusion characteristics and the resident energy consumption behavior analysis model, and obtain the real-time resident energy consumption behavior analysis results of the target area;

[0164] The regional power dispatching unit is used to perform regional power dispatching according to the real-time analysis results of residents' energy consumption behavior and use the regional power dispatching model to obtain a real-time regional power dispatching plan for the target area.

[0165] The present invention discloses a resident energy consumption behavior analysis method and system based on multimodal data, which comprehensively considers the importance of multimodal data for the analysis of resident energy consumption behavior, breaks the limitation of a single data source, and can more comprehensively analyze the resident energy consumption behavior, thereby improving the accuracy of the analysis result; adopts multimodal feature fusion technology to effectively integrate the features of different data sources, improves data processing capabilities, and thus improves analysis efficiency, meeting the demand for real-time analysis; constructs a regional user portrait generation model, which can dynamically update the regional user portrait according to real-time data, while reducing the difficulty of analysis, improves the accuracy and real-time performance of the regional user portrait, better reflects the energy consumption habits and needs of regional users, and thus improves the accuracy of the analysis of resident energy consumption behavior, is suitable for the energy consumption characteristics of different regions or different user groups, and improves the adaptability of the resident energy consumption behavior analysis model; constructs a regional electric energy dispatching model, realizes a response mechanism, and performs automatic and intelligent regional electric energy dispatching according to the analysis results of the resident energy consumption behavior, reduces the demand for manual intervention, optimizes energy utilization efficiency, and achieves energy conservation and emission reduction.

[0166] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing residents' energy consumption behavior based on multimodal data, characterized in that: The steps include: Based on the cloud data center, using machine learning algorithms, we build regional user portrait generation models, multimodal feature fusion models, resident energy consumption behavior analysis models, and regional power dispatch models. Collecting real-time resident information and corresponding real-time multimodal monitoring data of the target area, and preprocessing them to obtain a number of preprocessed real-time resident information and a number of preprocessed real-time multimodal monitoring data; Based on some pre-processed real-time resident information, a regional user portrait generation model is used to generate a regional user portrait, and a real-time regional user portrait of the target area is obtained; According to the real-time regional user portrait and some pre-processed real-time multimodal monitoring data, a multimodal feature fusion model is used to perform multimodal feature fusion to obtain the corresponding real-time multimodal fusion features; According to the real-time multimodal fusion characteristics, the resident energy consumption behavior analysis model is used to analyze the resident energy consumption behavior and obtain the real-time resident energy consumption behavior analysis results in the target area; According to the real-time analysis results of residents' energy consumption behavior, regional power dispatching model is used to carry out regional power dispatching and obtain the real-time regional power dispatching plan for the target area.

2. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 1, characterized in that: The real-time multimodal monitoring data includes real-time resident energy consumption data, real-time regional energy configuration data and real-time regional environmental data in the target area.

3. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 1, characterized in that: Based on the cloud data center, using machine learning algorithms, we build a regional user portrait generation model, a multimodal feature fusion model, a resident energy consumption behavior analysis model, and a regional power dispatch model, including the following steps: Based on the cloud data center, a number of historical resident information and corresponding historical multimodal monitoring data of different regions are collected and preprocessed to obtain a number of preprocessed historical resident information and a number of preprocessed historical multimodal monitoring data; Based on the pre-processed historical resident information of different regions, a regional user portrait generation model is constructed using clustering and deep learning fusion algorithms, and several historical regional user portraits are generated; According to the historical regional user portraits of different regions and the corresponding pre-processed historical multimodal monitoring data, a multimodal feature fusion algorithm is used to build a multimodal feature fusion model and generate several historical multimodal fusion features; Based on several historical multimodal fusion features, a deep learning algorithm is used to build a resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results; Based on the analysis results of several historical residents' energy consumption behaviors, a regional power dispatch model was constructed using an enhanced reinforcement learning algorithm.

4. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 2 is characterized by: The regional user portrait generation model is constructed based on the FCM-RF algorithm, and the regional user portrait generation model includes a group user information clustering module constructed based on the FCM clustering algorithm and a regional user portrait generation module constructed based on the RF algorithm, which are connected in sequence; The multimodal feature fusion model is constructed based on the GCN-LSTM-Attention algorithm, and the multimodal feature fusion model includes a graph feature extraction module constructed based on the GCN algorithm, a data feature extraction module constructed based on the LSTM algorithm, and a weighted fusion module constructed based on the Attention mechanism, and the graph feature extraction module and the data feature extraction module are both connected to the weighted fusion module; The resident energy consumption behavior analysis model is constructed based on the CFWA-MLP algorithm, and the resident energy consumption behavior analysis model includes a network parameter optimization module constructed based on the CFWA algorithm and a resident energy consumption behavior analysis module constructed based on the MLP algorithm; The regional power dispatching model is constructed based on the ML-DQN algorithm, and the regional power dispatching model includes a meta-learning module constructed based on the ML algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is respectively connected to the deep Q network, the experience replay pool and the meta-learning module, and the meta-learning module is connected to the experience replay pool.

5. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 4 is characterized by: Based on several historical multimodal fusion features, a deep learning algorithm is used to build a resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results, including the following steps: Using the CFWA-MLP algorithm, an initial resident energy consumption behavior analysis model is constructed, which includes a network parameter optimization module and an initial resident energy consumption behavior analysis module; Use the network parameter optimization module to iteratively optimize the initial network parameters of the initial resident energy consumption behavior analysis module to obtain the optimal initial network parameters; According to the optimal initial network parameters, the initial resident energy consumption behavior analysis module is optimized to obtain an optimized resident energy consumption behavior analysis module; According to several historical multimodal fusion features, the optimized resident energy consumption behavior analysis module is optimized and trained to obtain the final resident energy consumption behavior analysis module; The network parameter optimization module and the final resident energy consumption behavior analysis module are integrated to obtain the final resident energy consumption behavior analysis model and generate several historical resident energy consumption behavior analysis results.

6. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 4 is characterized by: Based on the analysis results of several historical residents' energy consumption behaviors, an enhanced reinforcement learning algorithm is used to build a regional power dispatch model, which includes the following steps: The regional power dispatch problem is used as the simulation environment of the DQN algorithm, and the regional power dispatch problem is divided into several subtasks; Define meta-feature parameters in the historical analysis results of residents' energy consumption behavior, and extract several historical meta-features of the historical analysis results of residents' energy consumption behavior based on the meta-feature parameters; According to several historical meta-features of the analysis results of different historical residents’ energy consumption behaviors, ML algorithms are used for training under several subtasks to obtain the initial meta-learning module and several historical meta-learning experiences; Use the initial meta-learning module to initialize the deep Q network of the DQN algorithm, and update the initial meta-learning module to obtain an updated meta-learning module; Set the environment state, action space, and reward function of the DQN algorithm, and build the agent and experience replay pool to obtain the initial reinforcement learning module; Based on the analysis results of several historical residents’ energy consumption behaviors, the initial reinforcement learning module is trained and optimized to obtain the final reinforcement learning module and several historical regional power dispatch experiences; Several historical meta-learning experiences and several historical regional power dispatching experiences are stored in the experience replay pool, and combined with the final reinforcement learning module and the updated meta-learning module to obtain the regional power dispatching model.

7. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 4 is characterized by: According to some pre-processed real-time resident information, a regional user portrait generation model is used to generate a regional user portrait to obtain a real-time regional user portrait of the target area, including the following steps: Input some pre-processed real-time resident information into the regional user portrait generation model; Using the group user information clustering module, clustering is performed on a number of pre-processed real-time resident information to obtain a number of real-time clustering centers; According to the pre-processed real-time resident information corresponding to several real-time clustering centers, a regional user portrait generation module is used to generate a regional user portrait to obtain a real-time regional user portrait of the target area.

8. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 4 is characterized by: According to the real-time regional user portrait and some pre-processed real-time multimodal monitoring data, a multimodal feature fusion model is used to perform multimodal feature fusion to obtain the corresponding real-time multimodal fusion features, including the following steps: Input the real-time regional user portrait and several pre-processed real-time multimodal monitoring data into the multimodal feature fusion model; Use the graph feature extraction module to extract the real-time graph features of the real-time regional user portrait, and use the data feature extraction module to extract the real-time data features of several pre-processed real-time multimodal monitoring data; According to the preset attention weights of the weighted fusion module, the real-time graph features and real-time data features are weightedly fused to obtain the corresponding real-time multimodal fusion features.

9. The method for analyzing residents' energy consumption behavior based on multimodal data according to claim 6, characterized in that: According to the real-time analysis results of residents' energy consumption behavior, regional power dispatching model is used to perform regional power dispatching and obtain the real-time regional power dispatching plan for the target area, including the following steps: The real-time analysis results of residents' energy consumption behavior are input into the regional electric energy dispatch model, and several real-time meta-features of the real-time analysis results of residents' energy consumption behavior are extracted according to the meta-feature parameters; According to a number of real-time meta-features, the corresponding real-time meta-learning experience is matched in the experience replay pool, and the regional power dispatch model is updated according to the real-time meta-learning experience to obtain an updated regional power dispatch model; According to the real-time analysis results of residents' energy consumption behavior, the corresponding real-time regional power dispatch experience is matched in the experience playback pool; According to the real-time analysis results of residents' energy consumption behavior and the real-time regional power dispatching experience, the updated regional power dispatching model is used to carry out regional power dispatching and obtain the real-time regional power dispatching plan for the target area.

10. A resident energy consumption behavior analysis system based on multimodal data, used to implement the resident energy consumption behavior analysis method according to any one of claims 1 to 9, characterized in that: The system includes a cloud data center and several data acquisition devices. The cloud data center is respectively communicated with the several data acquisition devices, and the cloud data center includes a model building unit, a regional user portrait generation unit, a multimodal fusion unit, a resident energy consumption behavior analysis unit and a regional power scheduling unit which are connected in sequence.

Images