A household electricity consumption behavior recognition method, system, device and medium based on user global behavior mining
Patent Information
- Application Number
- CN202610602790.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-06
- Publication Date
- 2026-09-15
Smart Images
Figure CN122758084A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electricity consumption behavior identification technology, and in particular to a method, system, device and medium for household electricity consumption behavior identification based on the mining of user behavior across the entire domain. Background Technology
[0002] With rapid economic development and the continuous improvement of people's living standards, household electricity demand has increased significantly, and electricity consumption behavior has become increasingly complex, thus placing higher demands on intelligent electricity management and energy efficiency improvement. Current research indicates that by providing refined electricity consumption information feedback to help users optimize their electricity consumption behavior, household energy conservation and emission reduction can be achieved. Therefore, in this context, load identification technology, as a key means of intelligent electricity management, automatically identifies the operating status and electricity consumption composition of household appliances by analyzing power load curves, providing crucial data support for demand response, load forecasting, and electricity safety monitoring.
[0003] Traditional load identification methods primarily rely on electrical characteristic parameters, classifying and identifying appliance types through cluster analysis, neural networks, or hidden Markov models. However, these existing identification methods have significant limitations: firstly, the brands, models, years of use, and installation environments of appliances vary among different households, leading to substantial heterogeneity in the electrical characteristics of the same type of appliance. This makes it difficult for models relying solely on electrical characteristics to maintain good generalization ability across different users. Secondly, different users have different lifestyles and electricity consumption patterns, making it difficult to fully capture the diversity and individual characteristics of user behavior using only limited electrical signals, thus limiting the accuracy of behavior identification.
[0004] To overcome these limitations, some research has begun to incorporate user behavior knowledge to assist load identification. For example, existing inventions enhance load decomposition through spatiotemporal behavioral features, utilizing the temporal regularity and spatial distribution characteristics of appliance usage to improve identification accuracy. However, existing methods often integrate the extraction processes of behavioral and electrical features into a single deep neural network model. This means that while the model learns the general electrical characteristics of appliances, it must also consider the specific behavioral patterns of users. This tightly coupled design leads to decreased generalization ability of the model in cross-user applications and makes it difficult to flexibly adjust to changes in individual user behavior, failing to effectively balance the contradiction between the model's universality and the uniqueness of user electricity consumption behavior. Summary of the Invention
[0005] In view of the aforementioned existing problems, the present invention is proposed.
[0006] Therefore, this invention provides a method, system, device, and medium for identifying household electricity consumption behavior based on user full-domain behavior mining, which solves the problem that existing load identification methods cannot balance the contradiction between the generalization ability of the model and the uniqueness of user electricity consumption behavior because they integrate behavioral features and electrical features into a single model.
[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for identifying household electricity consumption behavior based on user-wide behavior mining, including: Collect user appliance-level power consumption data and home network device connection and usage data, and perform preprocessing; Based on the preprocessed data, a load identification model is constructed and trained to obtain preliminary identification results; Based on the preliminary identification results, a behavior enhancement model is constructed by combining the user's full-domain behavior characteristics, and the electricity consumption behavior identification results after behavior information enhancement are output through the behavior enhancement model. The load identification model and the behavior enhancement model are deployed in a real home environment. User behavior data and load identification results are collected regularly, and the behavior enhancement model is incrementally learned to achieve adaptive updates.
[0008] As a preferred embodiment of the household electricity consumption behavior identification method based on user full-domain behavior mining described in this invention, the step of constructing a behavior enhancement model by combining user full-domain behavior features and outputting the electricity consumption behavior identification result after behavior information enhancement through the behavior enhancement model includes: The probability vectors of various electrical appliances output by the load identification model are concatenated with the encoded user global behavior features to generate the input feature vector of the behavior enhancement model. The corresponding real appliance type labels are used as supervision labels for the behavior enhancement model. The first ensemble learning algorithm is used to train the behavior enhancement model to obtain the trained behavior enhancement model. The input feature vector corresponding to the current electricity consumption event is input into the behavior enhancement model, and the probability distribution of various electrical appliances is output as the final behavior identification result.
[0009] As a preferred embodiment of the household electricity consumption behavior identification method based on user-wide behavior mining described in this invention, the encoding operation of the user-wide behavior features includes: The network connection device identifier is converted into a binary feature vector using the first encoding method; The second encoding method is used to convert the electricity consumption area information and electricity consumption time period information into corresponding area feature vectors and time period feature vectors; The real-time network speed of each device and the original values of the appliance usage status vectors identified in the previous moment are directly used as numerical feature vectors.
[0010] As a preferred embodiment of the household electricity consumption behavior identification method based on user full-domain behavior mining described in this invention, the step of deploying the load identification model and the behavior enhancement model in a real household environment, periodically collecting user behavior data and load identification results, and incrementally learning the behavior enhancement model to achieve adaptive model updates includes: In a real-world home environment, all network parameters of the trained load identification model are fixed so that the load identification model remains unchanged during online applications. According to the preset update cycle, collect user behavior data and corresponding load identification results generated in the actual home environment since the last update, and construct the user behavior data and load identification results into an incremental training sample set; A regularization-based incremental learning strategy is adopted, and the parameters of the current behavior enhancement model are adjusted using the incremental training sample set. This includes: calculating the importance weights of the original parameters of the behavior enhancement model to the historical task; and when updating the model parameters using the incremental training sample set, applying additional constraint terms to parameters whose importance weights exceed a preset threshold to limit the magnitude of change. The behavior enhancement model with adjusted parameters will be used as the current model for online identification in the next update cycle.
[0011] As a preferred embodiment of the household electricity consumption behavior identification method based on user full-domain behavior mining described in this invention, the step of collecting user household appliance-level electricity consumption data and household network device connection and usage data, and performing preprocessing, includes: The system collects user home appliance-level electricity consumption data through invasive devices, including smart wall sockets and power strips; it also collects the connection status of network devices in the home LAN, network speed data of each connected device, current electricity consumption area information, and current electricity consumption time information. All collected data is cleaned, including removing outliers and filling in missing values; Data from different sources are aligned using the same timestamp; continuous electrical parameters in the aligned data are normalized; and discrete tag information is numerically mapped and encoded. The processed data will be used as a multi-source dataset for subsequent model building and training.
[0012] As a preferred embodiment of the household electricity consumption behavior identification method based on user full-domain behavior mining described in this invention, the construction of the load identification model includes: A four-layer fully connected deep neural network is used as the basic architecture of the load identification model; The number of nodes in the input layer is consistent with the dimension of the electrical feature vector, which includes active power, reactive power, effective voltage value, effective current value, power factor, amplitude and phase of the 1st to 7th harmonics of voltage, and amplitude and phase of the 1st to 7th harmonics of current. After the input layer, a first hidden layer and a second hidden layer are set sequentially. The number of neurons in each hidden layer is set, and a Dropout operation is connected after each hidden layer. The number of neurons in the output layer is the same as the total number of appliance types to be identified, and an activation function is used in the output layer to convert the output value into a probability distribution for each type of appliance.
[0013] As a preferred embodiment of the household electricity consumption behavior identification method based on user full-domain behavior mining described in this invention, the training of the load identification model includes: The preprocessed multi-source dataset is divided into training set, validation set and test set according to user dimension and time order. The training set is used for model parameter learning, the validation set is used for early stopping monitoring during training, and the test set is used for performance verification after training is completed. Initialize the model parameters, train the deep neural network using the training set, and use the classification cross-entropy as the loss function and set the initial learning rate to 0.001 during the training process; During training, the current learning rate is multiplied by a decay coefficient and gradually reduced after a preset number of training rounds. At the same time, an early stopping strategy for the validation set is adopted. When the loss value on the validation set no longer decreases for several consecutive training rounds, the training is automatically terminated, and the model parameters corresponding to the minimum loss on the validation set are saved as the optimal load recognition model. The trained optimal load identification model is applied to the test set, and the probability vectors of various electrical appliances are output as preliminary identification results.
[0014] Secondly, the present invention provides a household electricity consumption behavior identification system based on user full-domain behavior mining, comprising: The data acquisition module is used to collect user's home appliance power consumption data and home network device connection and usage data, and to perform preprocessing. The load identification model construction module is used to construct and train a load identification model based on the preprocessed data to obtain preliminary identification results. The behavior enhancement model construction module is used to construct a behavior enhancement model based on the preliminary identification results and combined with the user's full-domain behavior characteristics, and to output the electricity consumption behavior identification results after behavior information enhancement through the behavior enhancement model. The adaptive update module is used to deploy the load identification model and the behavior enhancement model in a real home environment, periodically collect user behavior data and load identification results, and perform incremental learning on the behavior enhancement model to achieve adaptive updates of the model.
[0015] Thirdly, the present invention provides an electronic device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor, when executing the computer-executable instructions, implements the steps of a method for identifying household electricity consumption behavior based on user global behavior mining.
[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of a method for identifying household electricity consumption behavior based on user-wide behavior mining.
[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: By introducing and mining comprehensive behavioral information, this invention greatly enriches and expands the knowledge dimensions of user behavior used for identification, solving the problem of poor identification performance caused by traditional methods relying only on limited electrical information. This invention ensures a relatively independent model architecture for the extraction processes of electrical features and behavioral features, allowing the load identification model to focus on learning the universal electrical characteristics of appliances, possessing cross-user and cross-household versatility; while the behavior enhancement model can evolve independently, focusing on capturing and adapting to the unique behavioral patterns of specific users. This separation design cleverly improves the overall generalization of the method while ensuring the flexibility of adaptive learning of individual user electricity consumption behavior. This invention can provide more accurate load identification results for users with different electricity consumption habits, not only directly improving the intelligence level of home energy management and user experience, but also providing a reliable data foundation and technical support for power grid companies to implement refined demand-side response and formulate effective energy-saving strategies, thus promoting smart electricity technology to a higher stage. Attached Figure Description
[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of the overall process logic of a household electricity consumption behavior identification method based on user full-domain behavior mining, provided in an embodiment of the present invention.
[0020] Figure 2This is a flowchart of the behavior enhancement model processing of a household electricity consumption behavior identification method based on user global behavior mining, provided as an embodiment of the present invention.
[0021] Figure 3 This is a schematic diagram of the adaptive learning model of a household electricity consumption behavior identification method based on user global behavior mining, provided in an embodiment of the present invention. Detailed Implementation
[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0023] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for identifying household electricity consumption behavior based on user-wide behavior mining is provided, such as... Figure 1 The specific steps shown are as follows: S100: Collects user's home appliance power consumption data and home network device connection and usage data, and performs preprocessing; S200: Based on the preprocessed data, a load identification model is constructed and trained to obtain preliminary identification results; S300: Based on the preliminary identification results, a behavior enhancement model is constructed by combining the user's full-domain behavior characteristics, and the behavior enhancement model outputs the electricity consumption behavior identification results after the behavior information is enhanced. S400: Deploys the load identification model and behavior enhancement model into a real home environment, regularly collects user behavior data and load identification results, and performs incremental learning on the behavior enhancement model to achieve adaptive updates of the model.
[0024] It should be noted that traditional load identification methods mainly rely on electrical features for clustering or classification analysis, which has significant limitations. Furthermore, relying solely on electrical signals makes it difficult to capture the diversity of user behavior, limiting the model's generalization ability and the accuracy of behavior identification. To overcome these limitations, some studies have begun to introduce user behavior knowledge to assist identification, enhancing load decomposition through spatiotemporal behavioral features. However, existing methods often integrate the behavioral feature extraction process with the electrical feature extraction process into a single model, resulting in the model's inability to effectively balance generalization and the uniqueness of user behavior.
[0025] To address the problems of the existing technologies, this embodiment provides steps S100-S400, which introduce and mine comprehensive behavioral information, greatly enriching and expanding the knowledge dimensions of user behavior used for identification. This solves the problem of poor identification performance caused by traditional methods relying only on limited electrical information. The present invention ensures a relatively independent model architecture for the extraction of electrical and behavioral features. This allows the load identification model to focus on learning the universal electrical characteristics of appliances, possessing cross-user and cross-household versatility; while the behavior enhancement model can evolve independently, focusing on capturing and adapting to the unique behavioral patterns of specific users. This separation design cleverly improves the overall generalization of the method while ensuring the flexibility of adaptive learning of individual user electricity consumption behavior. The present invention can provide more accurate load identification results for users with different electricity consumption habits, directly improving the intelligence level of home energy management and user experience. It also provides a reliable data foundation and technical support for power grid companies to implement refined demand-side response and formulate effective energy-saving strategies, propelling smart electricity technology to a higher stage.
[0026] Example 2, refer to Figure 2 and Figure 3 As another embodiment, this embodiment provides a specific implementation of a method for identifying household electricity consumption behavior based on user full-domain behavior mining, in order to illustrate the technical means of this invention.
[0027] In this embodiment of the invention, step S100 involves collecting user appliance-level power consumption data and home network device connection and usage data, and performing preprocessing; including the following sub-step A1: In A1: User appliance-level electricity consumption data is collected through intrusive devices, including smart wall sockets and power strips; at the same time, the connection status of network devices in the home LAN, network speed data of each connected device, current electricity consumption area information, and current electricity consumption time information are also collected; All collected data is cleaned, including removing outliers and filling in missing values; Data from different sources are aligned using the same timestamp; continuous electrical parameters in the aligned data are normalized; and discrete tag information is numerically mapped and encoded. The processed data will be used as a multi-source dataset for subsequent model building and training.
[0028] Specifically, the collection of user appliance-level electricity consumption data and home network device connection and usage data through invasive devices should meet the following requirements: The data labeling information, such as appliance type labeling and region labeling, is accurate and error-free. The data should include active power at time intervals of less than 2 seconds. P reactive power QRMS voltage U rms RMS value of current I rms Power factor PF Electrical information such as voltage, current, amplitude and phase of the 1st to 7th harmonics, network information such as the connection status of network devices in the home LAN and the network speed of each device, as well as other information such as the area and time of electricity use; The data time span should ensure that it reflects users' daily electricity consumption habits, and the data time span should be greater than one month; the data should come from multiple user households to ensure that it reflects users' general electricity consumption habits, and the number of user households should be greater than 10.
[0029] Specifically, the amplitudes of the 1st to 7th harmonics of the voltage and phase and the amplitude of the 1st to 7th harmonics of the current and phase It is the result obtained by performing a Fast Fourier Transform on the original voltage and current signals, where The calculation formulas for other electrical signal quantities are as follows: It should be noted that the data is collected autonomously through intrusive devices, including smart wall sockets and power strips. To ensure that the smart socket can calculate reliable harmonic information, the data acquisition frequency of the smart socket should be greater than 1.5kHz.
[0030] Specifically, data preprocessing includes cleaning, alignment, and normalization to prepare the data for model training. Data cleaning involves handling missing values and correcting erroneous values. Data alignment refers to standardizing the format by mapping discrete labels such as appliance type and region. Finally, the Z-score method is used to complete data normalization.
[0031] It should be noted that step S100 above synchronously collects multi-source heterogeneous data such as electrical parameters, home network device connection status, network speed, power consumption area, and power consumption time through invasive devices, constructing a rich dataset containing user behavior information, providing complete feature dimension support for subsequent model training, and solving the problem that traditional methods rely solely on electrical information, resulting in insufficient behavior recognition capabilities.
[0032] In this embodiment of the invention, step S200 above constructs and trains a load identification model based on the preprocessed data to obtain preliminary identification results; it includes the following sub-steps B1 and B2: In B1: The construction of the load identification model includes: A four-layer fully connected deep neural network is used as the basic architecture of the load identification model; The number of nodes in the input layer is consistent with the dimension of the electrical feature vector. The electrical feature vector includes active power, reactive power, effective voltage value, effective current value, power factor, amplitude and phase of the 1st to 7th harmonics of voltage, and amplitude and phase of the 1st to 7th harmonics of current. After the input layer, set the first hidden layer and the second hidden layer in sequence, set the number of neurons in each hidden layer, and connect the Dropout operation after each hidden layer; The number of neurons in the output layer is the same as the total number of appliance types to be identified, and an activation function is used in the output layer to convert the output value into a probability distribution for each type of appliance.
[0033] Specifically, the load identification model uses a four-layer fully connected deep neural network as its basic architecture. The number of nodes in the input layer of this deep neural network is consistent with the dimension of the electrical feature vector. The electrical feature vector specifically includes active power, reactive power, effective voltage value, effective current value, power factor, amplitude and phase of the 1st to 7th harmonics of voltage (14 dimensions in total), amplitude and phase of the 1st to 7th harmonics of current (14 dimensions in total), for a total of 33 input features.
[0034] Specifically, a first hidden layer and a second hidden layer are set sequentially after the input layer. The number of neurons in the first hidden layer is set to 64, and the number of neurons in the second hidden layer is set to 32. Both hidden layers use the ReLU function as the activation function, and a Dropout operation is added after each hidden layer with a dropout rate of 0.3 to prevent the model from overfitting.
[0035] Specifically, the number of neurons in the output layer is the same as the total number of appliance types to be identified. Appliance types include air conditioners, refrigerators, washing machines, microwave ovens, electric kettles, computers, televisions, and lighting fixtures. The output layer uses the Softmax function as the activation function to convert the output values into probability distributions for various appliance types.
[0036] In an optional embodiment, the load identification model can also be constructed using a random forest algorithm, which classifies electrical feature samples by integrating multiple decision trees. Each tree searches for the optimal split node in a randomly selected feature subset, and finally, the prediction results of all trees are voted on to determine the type of electrical appliance.
[0037] In an optional embodiment, the load identification model can also be constructed using the gradient boosting decision tree algorithm in the ensemble learning model. Multiple decision trees are trained iteratively, with each new tree used to fit the prediction residual of the previous tree. The prediction results of all trees are weighted and summed to output the final appliance classification result.
[0038] In B2: Training the load identification model includes: The preprocessed multi-source dataset is divided into training set, validation set and test set according to user dimension and time order. The training set is used for model parameter learning, the validation set is used for early stopping monitoring during training, and the test set is used for performance verification after training is completed. Initialize the model parameters, train the deep neural network using the training set, and use the classification cross-entropy as the loss function during training with an initial learning rate of 0.001. During training, the current learning rate is multiplied by a decay coefficient and gradually reduced after a preset number of training rounds. At the same time, an early stopping strategy for the validation set is adopted. When the loss value on the validation set no longer decreases for several consecutive training rounds, the training is automatically terminated, and the model parameters corresponding to the minimum loss on the validation set are saved as the optimal load recognition model. The trained optimal load identification model is applied to the test set, and the probability vectors of various electrical appliances are output as preliminary identification results.
[0039] Specifically, the training of the load identification model first divides the preprocessed multi-source dataset into training set, validation set and test set according to user dimension and time order, with a division ratio of 70%:15%:15%. The total number of training data samples is no less than 4 million time point samples, and the number of samples for each type of appliance is relatively balanced. For a few categories with a small number of samples, oversampling technology is used for processing.
[0040] Specifically, the model parameters are initialized using the Xavier method. The deep neural network is trained using a training set, with classification cross-entropy as the loss function. The optimizer is the Adam algorithm, with an initial learning rate of 0.001, a batch size of 128, and Adam's parameters β1 and β2 set to 0.9 and 0.999, respectively. During training, a step-wise learning rate decay strategy is employed, multiplying the current learning rate by 0.5 every 20 training epochs to gradually decrease the learning rate. Simultaneously, a validation set early stopping strategy is used, automatically terminating training when the loss value on the validation set no longer decreases for 10 consecutive training epochs, and saving the model parameters corresponding to the minimum validation set loss as the optimal load recognition model.
[0041] Specifically, after the model training is completed, the optimal load identification model is applied to an independent test set for performance evaluation. The evaluation metrics include accuracy, precision, recall, and F1 score. The test verifies that the overall accuracy of the model is not less than 85%, and the F1 score of each type of appliance is not less than 80%. The probability vectors of each type of appliance are output as the preliminary identification results.
[0042] It should be noted that step S200 above uses a large-scale multi-user dataset to train the load recognition model, enabling the model to learn the general electrical characteristics of electrical appliances and have good generalization ability across users and households. This provides a reliable initial probability output for the subsequent behavior enhancement model and avoids overfitting problems caused by insufficient data from a single user.
[0043] In this embodiment of the invention, step S300, based on the preliminary identification results, constructs a behavior enhancement model by combining the user's overall behavior characteristics, and outputs the electricity consumption behavior identification result enhanced by the behavior information through the behavior enhancement model; including the following sub-step C1: In C1: The probability vectors of various electrical appliances output by the load identification model are concatenated with the encoded user global behavior features to generate the input feature vector of the behavior enhancement model. The corresponding real appliance type labels are used as supervision labels for the behavior enhancement model. The first ensemble learning algorithm is used to train the behavior enhancement model to obtain the trained behavior enhancement model. The input feature vector corresponding to the current electricity consumption event is input into the behavior enhancement model, and the probability distribution of various electrical appliances is output as the final behavior identification result.
[0044] Specifically, the global behavior enhancement model takes the output of the load identification model as input. Based on the initial identification of electrical equipment by the load identification model, it combines the user's global behavior characteristics to complete the construction of the behavior enhancement model and outputs the behavior identification results after the behavior information is enhanced. In this embodiment, the behavior enhancement model is constructed through a random forest architecture.
[0045] Specifically, the input features of the behavior enhancement model are organized through vector concatenation. On one hand, they come from the probability vectors of various electrical appliances output by the load identification model, the dimension of which is the same as the number of appliance categories N. On the other hand, they come from the encoded global behavior features, including network information such as the connection status of network devices in the home LAN and the network speed of each device, as well as other information such as the electricity usage area and time period. In addition, it also includes information on the usage of other electrical appliances at the previous moment, where the usage of other electrical appliances at the previous moment is an N-dimensional 0 / 1 vector, used to consider the usage correlation between appliances.
[0046] In this embodiment of the invention, the encoding operation of user global behavior features includes: The network connection device identifier is converted into a binary feature vector using the first encoding method; The second encoding method is used to convert the electricity consumption area information and electricity consumption time period information into corresponding area feature vectors and time period feature vectors; The real-time network speed of each device and the original values of the appliance usage status vectors identified in the previous moment are directly used as numerical feature vectors.
[0047] Specifically, in this embodiment, the first encoding method is multi-hot encoding, which is used to convert the network connection device identifier into a binary feature vector. That is, each preset network device type corresponds to a bit in the binary feature vector. The bit corresponding to the currently connected device is 1, and the bit corresponding to the unconnected device is 0, thereby simultaneously representing the simultaneous connection status of multiple devices. The second encoding method is one-hot encoding, which is used to convert the power consumption area information and power consumption time period information into the corresponding area feature vector and time period feature vector. That is, each possible power consumption area or power consumption time period corresponds to an independent binary bit. Only the bit corresponding to the currently occurring area or time period is 1, and the other bits are 0.
[0048] It should be noted that multi-hot coding can effectively handle situations where multiple devices are connected simultaneously in a network, avoiding the problem that a single coding method cannot express the coexistence information of multiple devices; one-hot coding can transform discrete and unordered category information into numerical features, eliminating the risk of artificially assigning incorrect ordinal relationships to categories that do not have numerical magnitude relationships, while ensuring the equidistance of each category in the feature space, enabling the behavior enhancement model to learn the correlation between each category and the use of electrical appliances without bias.
[0049] Specifically, in this embodiment, the behavior enhancement model uses the random forest algorithm, with hyperparameters set to 100 trees, a maximum tree depth of 20, and a minimum number of samples per leaf node of 5. During model training, the probability vectors of various electrical appliances output by the load identification model are concatenated with the encoded user global behavior features to generate the input feature vector of the behavior enhancement model. The corresponding real electrical appliance type labels are used as the supervision labels for the behavior enhancement model. To address the imbalance between positive and negative samples in electrical appliance usage events, different weights are assigned to samples of different categories, resulting in the trained behavior enhancement model.
[0050] Specifically, such as Figure 2 As shown, the input feature vector corresponding to the current electricity consumption event is input into the trained behavior enhancement model. The behavior enhancement model takes the soft output probability of the load identification model as one of its inputs. By learning the correlation between behavioral features and appliance usage, the original probability distribution is recalibrated and corrected. Finally, the corrected probability distribution of various appliances is output as the electricity consumption behavior identification result after behavior information enhancement.
[0051] It should be noted that in step S300 above, the probability vector output by the load identification model is concatenated with the user's global behavioral features and then input into the behavior enhancement model. This achieves a decoupling design between electrical feature extraction and behavior pattern learning, enabling the behavior enhancement model to focus on capturing the unique behavior of specific users and to make targeted corrections and refinements to the basic identification results, thereby significantly improving the identification accuracy of personalized electricity behavior.
[0052] In this embodiment of the invention, step S400 deploys the load identification model and the behavior enhancement model in a real home environment, periodically collects user behavior data and load identification results, and performs incremental learning on the behavior enhancement model to achieve adaptive updates of the model; including the following sub-step D1: In D1: In a real-world home environment, all network parameters of the trained load identification model are fixed so that the load identification model remains unchanged during online applications; According to the preset update cycle, collect user behavior data and corresponding load identification results generated in the actual home environment since the last update, and construct an incremental training sample set from the user behavior data and load identification results; A regularization-based incremental learning strategy is adopted to adjust the parameters of the current behavior enhancement model using an incremental training sample set. This includes: calculating the importance weights of the original parameters of the behavior enhancement model to the historical tasks; and when updating the model parameters using the incremental training sample set, applying additional constraint terms to parameters whose importance weights exceed a preset threshold to limit the magnitude of change. The behavior enhancement model with adjusted parameters will be used as the current model for online identification in the next update cycle.
[0053] Specifically, in this embodiment, the preset update cycle is set to once a week. User behavior data and corresponding load identification results generated in the actual home environment since the last update are collected. The user behavior data and load identification results are used to construct an incremental training sample set for incremental learning of the behavior enhancement model. A regularized incremental learning strategy is adopted, and the parameters of the current behavior enhancement model are adjusted using the incremental training sample set. Specifically, the importance weights of the original parameters of the behavior enhancement model to historical tasks are calculated using elastic weight consolidation technology. When updating the model parameters using the incremental training sample set, additional constraint terms are applied to parameters whose importance weights exceed a preset threshold to limit their change. The preset threshold is set to 0.8 according to the parameter importance distribution, thereby protecting old knowledge from being forgotten.
[0054] Specifically, such as Figure 3The diagram illustrates the adaptive update process of the model. To avoid the enormous overhead of retraining and overcome catastrophic forgetting, a regularization-based incremental learning strategy is employed. Elastic weight consolidation technology is used to calculate the importance of network parameters to previously learned tasks. When updating the model using new data, constraints are imposed on important parameters, penalizing significant changes and thus protecting old knowledge from being forgotten. Before model deployment, necessary evaluations are required. In addition to assessing conventional metrics such as accuracy on new data, a key evaluation of the model's performance on older tasks is crucial to ensure that catastrophic forgetting has not occurred.
[0055] It should be noted that the parameters of the S400 fixed load identification model in the above steps maintain its general identification capability. At the same time, a regularized incremental learning strategy is used to periodically fine-tune the behavior enhancement model. This allows the system to track the dynamic changes in users' electricity consumption habits while avoiding catastrophic forgetting, enabling the system to maintain high-precision personalized identification performance over a long period of time.
[0056] Example 3: This example provides a household electricity consumption behavior identification system based on user full-domain behavior mining, including: The data acquisition module is used to collect user's home appliance power consumption data and home network device connection and usage data, and to perform preprocessing. The load identification model building module is used to build and train a load identification model based on the preprocessed data to obtain preliminary identification results. The behavior enhancement model construction module is used to build a behavior enhancement model based on the preliminary identification results and combined with the user's full-domain behavior characteristics, and output the electricity consumption behavior identification results after the behavior information is enhanced through the behavior enhancement model. The adaptive update module is used to deploy the load identification model and behavior enhancement model into a real home environment, regularly collect user behavior data and load identification results, and perform incremental learning on the behavior enhancement model to achieve adaptive updates of the model.
[0057] It should be noted that the technical solution of the household electricity consumption behavior identification system based on user full-domain behavior mining is based on the same concept as the technical solution of the household electricity consumption behavior identification method based on user full-domain behavior mining described above. For details not described in detail in the technical solution of the household electricity consumption behavior identification system based on user full-domain behavior mining in this embodiment, please refer to the description of the technical solution of the household electricity consumption behavior identification method based on user full-domain behavior mining described above.
[0058] The above-mentioned unit modules can be embedded in the processor of the electronic device in hardware form or independent of it, or they can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of the above modules.
[0059] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for identifying household electricity consumption behavior based on user behavior mining. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.
[0060] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method proposed in the above embodiments.
[0061] The storage medium proposed in this embodiment belongs to the same inventive concept as the method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0062] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory, random access memory, flash memory, hard disk, or optical disk, and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute the method of the embodiments of the present invention.
[0063] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the present invention.
Claims
1. A method for identifying household electricity consumption behavior based on user-wide behavior mining, characterized in that, include: Collect user appliance-level power consumption data and home network device connection and usage data, and perform preprocessing; Based on the preprocessed data, a load identification model is constructed and trained to obtain preliminary identification results; Based on the preliminary identification results, a behavior enhancement model is constructed by combining the user's full-domain behavior characteristics, and the electricity consumption behavior identification results after behavior information enhancement are output through the behavior enhancement model. The load identification model and the behavior enhancement model are deployed in a real home environment. User behavior data and load identification results are collected regularly, and the behavior enhancement model is incrementally learned to achieve adaptive updates.
2. The method for identifying household electricity consumption behavior based on user-wide behavior mining as described in claim 1, characterized in that, The step of constructing a behavior enhancement model by combining user's full-domain behavior features, and outputting the electricity consumption behavior identification result after behavior information enhancement through the behavior enhancement model, includes: The probability vectors of various electrical appliances output by the load identification model are concatenated with the encoded user global behavior features to generate the input feature vector of the behavior enhancement model. The corresponding real appliance type labels are used as supervision labels for the behavior enhancement model. The first ensemble learning algorithm is used to train the behavior enhancement model to obtain the trained behavior enhancement model. The input feature vector corresponding to the current electricity consumption event is input into the behavior enhancement model, and the probability distribution of various electrical appliances is output as the final behavior identification result.
3. The method for identifying household electricity consumption behavior based on user-wide behavior mining as described in claim 2, characterized in that, The encoding operations for the user's global behavior features include: The network connection device identifier is converted into a binary feature vector using the first encoding method; The second encoding method is used to convert the electricity consumption area information and electricity consumption time period information into corresponding area feature vectors and time period feature vectors; The real-time network speed of each device and the original values of the appliance usage status vectors identified in the previous moment are directly used as numerical feature vectors.
4. The method for identifying household electricity consumption behavior based on user-wide behavior mining as described in claim 3, characterized in that, The process of deploying the load identification model and the behavior enhancement model in a real home environment, periodically collecting user behavior data and load identification results, and incrementally learning the behavior enhancement model to achieve adaptive updates includes: In a real-world home environment, all network parameters of the trained load identification model are fixed so that the load identification model remains unchanged during online applications. According to the preset update cycle, collect user behavior data and corresponding load identification results generated in the actual home environment since the last update, and construct the user behavior data and load identification results into an incremental training sample set; A regularization-based incremental learning strategy is adopted, and the parameters of the current behavior enhancement model are adjusted using the incremental training sample set. This includes: calculating the importance weights of the original parameters of the behavior enhancement model to the historical task; and when updating the model parameters using the incremental training sample set, applying additional constraint terms to parameters whose importance weights exceed a preset threshold to limit the magnitude of change. The behavior enhancement model with adjusted parameters will be used as the current model for online identification in the next update cycle.
5. The method for identifying household electricity consumption behavior based on user-wide behavior mining as described in claim 1, characterized in that, The process of collecting and preprocessing user appliance-level electricity consumption data and home network device connection and usage data includes: The system collects user home appliance-level electricity consumption data through invasive devices, including smart wall sockets and power strips; it also collects the connection status of network devices in the home LAN, network speed data of each connected device, current electricity consumption area information, and current electricity consumption time information. All collected data is cleaned, including removing outliers and filling in missing values; Data from different sources are aligned using the same timestamp; continuous electrical parameters in the aligned data are normalized; and discrete tag information is numerically mapped and encoded. The processed data will be used as a multi-source dataset for subsequent model building and training.
6. The method for identifying household electricity consumption behavior based on user-wide behavior mining as described in claim 5, characterized in that, The construction of the load identification model includes: A four-layer fully connected deep neural network is used as the basic architecture of the load identification model; The number of nodes in the input layer is consistent with the dimension of the electrical feature vector, which includes active power, reactive power, effective voltage value, effective current value, power factor, amplitude and phase of the 1st to 7th harmonics of voltage, and amplitude and phase of the 1st to 7th harmonics of current. After the input layer, a first hidden layer and a second hidden layer are set sequentially. The number of neurons in each hidden layer is set, and a Dropout operation is connected after each hidden layer. The number of neurons in the output layer is the same as the total number of appliance types to be identified, and an activation function is used in the output layer to convert the output value into a probability distribution for each type of appliance.
7. The method for identifying household electricity consumption behavior based on user-wide behavior mining as described in claim 6, characterized in that, The training of the load identification model includes: The preprocessed multi-source dataset is divided into training set, validation set and test set according to user dimension and time order. The training set is used for model parameter learning, the validation set is used for early stopping monitoring during training, and the test set is used for performance verification after training is completed. Initialize the model parameters, train the deep neural network using the training set, and use the classification cross-entropy as the loss function and set the initial learning rate to 0.001 during the training process; During training, the current learning rate is multiplied by a decay coefficient and gradually reduced after a preset number of training rounds. At the same time, an early stopping strategy for the validation set is adopted. When the loss value on the validation set no longer decreases for several consecutive training rounds, the training is automatically terminated, and the model parameters corresponding to the minimum loss on the validation set are saved as the optimal load recognition model. The trained optimal load identification model is applied to the test set, and the probability vectors of various electrical appliances are output as preliminary identification results.
8. A household electricity consumption behavior identification system based on user full-domain behavior mining, employing the household electricity consumption behavior identification method based on user full-domain behavior mining as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect user's home appliance power consumption data and home network device connection and usage data, and to perform preprocessing. The load identification model construction module is used to construct and train a load identification model based on the preprocessed data to obtain preliminary identification results. The behavior enhancement model construction module is used to construct a behavior enhancement model based on the preliminary identification results and combined with the user's full-domain behavior characteristics, and to output the electricity consumption behavior identification results after behavior information enhancement through the behavior enhancement model. The adaptive update module is used to deploy the load identification model and the behavior enhancement model in a real home environment, periodically collect user behavior data and load identification results, and perform incremental learning on the behavior enhancement model to achieve adaptive updates of the model.
9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and when the processor executes the computer-executable instructions, it implements the steps of the household electricity consumption behavior identification method based on user global behavior mining as described in any one of claims 1 to 7.
10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the computer-executable instructions are executed by the processor, they implement the steps of the household electricity consumption behavior identification method based on user global behavior mining as described in any one of claims 1 to 7.