Virtual power plant resource aggregation electric appliance power consumption feature extraction method, system and device

Abstract

The application discloses a virtual power plant resource aggregation electric appliance power consumption feature extraction method, system and device, relates to the field of artificial intelligence, solves the technical problem that the existing technology lacks the integration of field knowledge, leading to the disconnection between the analysis results and the real needs of users, and restricting the resource aggregation efficiency of the virtual power plant; two-dimensional data slices are obtained by data processing on multi-source data, and mode labels are obtained through a pattern recognition model; a Gaussian mixture model is constructed with a time parameter, and the Gaussian mixture model is solved and updated to obtain a power consumption feature parameter set, achieving semantic improvement from original data to user understandable behavior, depicting the complex statistical rules and multi-peak characteristics of user behavior in the time dimension, and the obtained power consumption feature parameter set can truly reflect the personalized needs and behavior habits of users, has high precision and dynamic adaptability, and provides accurate data basis for efficient scheduling of aggregated resources of the virtual power plant.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically a method, system, and device for extracting electrical consumption characteristics of appliances from virtual power plant resource aggregation. Background Technology

[0002] With the construction of new power systems, virtual power plants, as an important form of aggregating distributed resources, place higher demands on the ability to accurately perceive and model massive terminal loads. The widespread adoption of advanced metering systems has provided a technological foundation for acquiring terminal appliance-level electricity consumption data. However, how to effectively extract information containing user behavior and appliance characteristics from this data to support the intelligent operation of virtual power plants remains a key issue that urgently needs to be addressed.

[0003] Existing methods generally adopt a data intelligence paradigm, but lack the integration of domain knowledge, resulting in a disconnect between the analysis results and users' real needs and behavioral habits. This creates a cognitive gap and restricts the virtual power plant's in-depth understanding and efficient utilization of aggregated resources. Therefore, the method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation still needs further improvement. Summary of the Invention

[0004] This application aims to solve at least one of the technical problems existing in the prior art; to this end, this application proposes a method, system and device for extracting electrical electricity consumption characteristics of virtual power plant resource aggregation, in order to solve the technical problem that the lack of integration of domain knowledge in the prior art leads to the analysis results being out of touch with the actual needs of users, thereby restricting the resource aggregation efficiency of virtual power plants.

[0005] To achieve the above objectives, the first aspect of this application provides a method for extracting electrical appliance power consumption characteristics from virtual power plant resource aggregation, including: Acquire multi-source data of the target electrical appliance; the multi-source data includes active power sequence and timestamp sequence; Two-dimensional data slices are obtained by preprocessing multi-source data and constructing two-dimensional tensors; Two-dimensional data slices are used to obtain pattern labels through a pattern recognition model; Time parameters are determined based on pattern labels and multi-source data; Construct a Gaussian mixture model based on time parameters; The Gaussian mixture model is solved and updated using the real-time incremental expectation-maximization algorithm to obtain the set of electricity consumption characteristic parameters corresponding to the time parameters.

[0006] Through the steps described above, this application utilizes a pattern recognition model to achieve a semantic leap from raw data to user-understandable behavior, effectively bridging the gap between data and cognition. By constructing a Gaussian mixture model, it accurately depicts the complex statistical patterns and multi-peak distribution characteristics of user electricity consumption behavior over time. Simultaneously, the Gaussian mixture model is continuously updated online, enabling its feature parameter set to dynamically adapt to the evolution of user habits. This allows it to realistically and accurately reflect users' personalized needs and behavioral patterns, combining high precision with strong dynamic adaptability. This provides solid and reliable data support for virtual power plants to achieve deep perception and efficient scheduling of aggregated resources.

[0007] Furthermore, the process of obtaining two-dimensional data slices by preprocessing multi-source data and constructing two-dimensional tensors includes: Extract active power sequence and timestamp sequence from multi-source data; The normalized active power sequence is obtained by performing global normalization on several power values ​​in the active power sequence. Define a sliding window and use the sliding window to prune the timestamp sequence and the normalized active power sequence to obtain several timestamp slices and power slices; the sliding window includes a window length and a sliding step size; A normalized timestamp slice is obtained by performing local normalization on several timestamps in the timestamp slice. The power slice and the normalized timestamp slice are concatenated along the channel dimension to obtain a two-dimensional data slice.

[0008] This application first performs global normalization on the original active power sequence to eliminate dimensional differences between different electrical appliances. Then, it uses a sliding window to prune the time and power sequences, extracting data segments with temporal locality. Next, it performs local normalization on the timestamp slices to weaken the influence of absolute time and strengthen the relative time relationship. Finally, it concatenates the processed power and timestamp slices in the channel dimension to form a two-dimensional tensor that combines power change information and temporal structure. Through multi-level normalization and structured data construction, it not only improves the consistency and comparability of the data, but also retains key local temporal features, providing high-quality and interpretable input for subsequent pattern recognition models. This enhances the accuracy and robustness of electricity consumption feature extraction from the data source.

[0009] Furthermore, the pattern recognition model is constructed using an improved convolutional neural network model; The improved convolutional neural network consists of several convolutional-pooling blocks and a fully connected classifier; The kernel size in the convolution-pooling block is adjusted to (1, 5); the adjusted kernel allows the convolution operation to slide only in the time dimension, processing the power channel and the time channel independently, thus avoiding feature confusion. The activation function in the convolution-pooling block is adjusted to the ReLU activation function; replacing the Sigmoid activation function with the ReLU activation function effectively alleviates the gradient vanishing problem in deep network training; A Dropout layer is added after the fully connected layer of the network in the fully connected classifier. The hyperparameter of the Dropout layer is the dropout rate. The loss function of the pattern recognition model is the cross-entropy loss function.

[0010] Furthermore, obtaining pattern labels from two-dimensional data slices using a pattern recognition model includes: Extract two-dimensional data slices; Two-dimensional data slices are input into a pattern recognition model, and then pass through several convolutional-pooling blocks and a fully connected classifier. The Softmax layer in the fully connected classifier outputs the probability of the two-dimensional data slices corresponding to several pattern labels. The pattern label corresponding to the maximum value among the probabilities is used as the pattern label of the two-dimensional data slice.

[0011] Furthermore, the determination of time parameters based on pattern labels and multi-source data includes: Extract several two-dimensional data slices under the same pattern label; When there are continuous two-dimensional data slices, extract the timestamp sequence of several consecutive two-dimensional data slices in multi-source data; The time parameters are obtained by performing parameter statistics on the timestamp sequence; the time parameters include the start time, end time, duration and interval corresponding to several identical pattern labels; the interval refers to the time interval between the end time of the previous pattern label and the start time of the current pattern label.

[0012] Furthermore, the construction of the Gaussian mixture model based on time parameters includes: Extract time parameters; the time parameters include the start time, end time, duration, and interval corresponding to several identical pattern labels; A Gaussian mixture model is constructed for several time parameters under the pattern label; the Gaussian mixture model satisfies: ;in, Let d represent the d-th time parameter; d represents the number corresponding to the time parameter; u represents the number of the Gaussian component; and U represents the total number of Gaussian components. This is represented as the model parameter corresponding to the d-th time parameter; The mixing weights are represented as the Gaussian components of the d-th time parameter. and These are expressed as mean and variance, respectively. and ; Represented as a Gaussian probability density function; The Gaussian probability density function satisfies: .

[0013] This application first extracts time parameters representing key temporal information of user behavior from accurately identified identical pattern labels, transforming abstract electricity consumption behavior into structured data that can be quantified and analyzed. Based on this, a Gaussian mixture model is constructed for each type of time parameter, utilizing its multimodal characteristics to accurately characterize the complex probability distribution of user behavior in the time dimension. This not only achieves a precise description of user habits from qualitative patterns to quantitative parameters, but also reveals the inherent statistical regularities and uncertainties of behavior through probabilistic models. As a result, a set of electricity consumption characteristics that can deeply reflect the real needs of users and possess both interpretability and statistical robustness is output, providing a key basis for virtual power plants to conduct refined resource prediction and adaptive scheduling.

[0014] Furthermore, the step of obtaining the electricity consumption characteristic parameter set corresponding to the time parameters by solving and updating the Gaussian mixture model through the real-time incremental expectation-maximization algorithm includes: The real-time incremental expectation-maximization algorithm is used to calculate the set of electricity consumption characteristic parameters corresponding to the Gaussian mixture model with time parameters in a fixed data pool; the size of the fixed data pool is... It is a fixed value; the specific value is set based on experience. The number of time parameters in the fixed data pool reaches At that time, the time parameters in the fixed data pool are removed according to the first-in-first-out principle, and the new time parameters are added to the fixed data pool; The update of the electricity consumption characteristic parameter set includes mixed weight update, mean update and variance update; The hybrid weight update satisfies: ;in, This represents the mixed weight updated on the u-th Gaussian component for the d-th time parameter; m is the index of the time parameter in the fixed data pool, and M is the total number of time parameters in the fixed data pool. Quantization represents the influence of time parameters in a fixed data pool on the calculation of new mixed weights, satisfying... ; This represents the posterior probability that the m-th time parameter sample in a fixed data pool is generated by the u-th Gaussian component. The mean update satisfies: ;in, This is represented as the d-th time parameter with number m in a fixed data pool; It is represented as the mean of the d-th time parameter after updating on the u-th Gaussian component; It represents the posterior probability of the first time parameter in a fixed data pool being generated by the u-th Gaussian component; The variance update satisfies: ;in, It is expressed as the variance of the d-th time parameter after updating on the u-th Gaussian component; The set of electricity consumption characteristic parameters corresponding to the time parameters is determined based on the updated mixed weights, mean, and variance.

[0015] Furthermore, the posterior probability satisfies: Where h and u are both numbers representing Gaussian components; It is represented as the Gaussian probability density function corresponding to the d-th time parameter in the m-th data pool.

[0016] This application introduces a fixed-capacity data pool, retaining only the most recent time parameters, and employs a first-in-first-out (FIFO) mechanism to ensure the timeliness of the data used. During model updates, the algorithm assigns time-decay-based weights to each data point, making the model more focused on recent user behavior. Based on this, the algorithm dynamically re-estimates the mixture weights, mean, and variance of the Gaussian mixture model using posterior probability, thereby continuously and efficiently capturing and learning the evolution trend of user electricity consumption habits. This ensures that the output set of electricity consumption characteristic parameters is always synchronized with the user's current real needs, significantly improving the virtual power plant's real-time perception capability and modeling accuracy of the dynamic characteristics of aggregated resources. This provides a reliable and adaptive data model foundation for achieving accurate and flexible load forecasting and scheduling decisions.

[0017] A second aspect of the present invention provides an electrical appliance power consumption feature extraction system for virtual power plant resource aggregation, comprising: a data acquisition module and a data analysis module; the data acquisition module and the data analysis module are connected together; The data acquisition module acquires multi-source data of the target electrical appliance through a data acquisition device; the multi-source data includes an active power sequence and a timestamp sequence. The data analysis module includes a data processing unit, a parameter determination unit, and a result generation unit; The data processing unit obtains two-dimensional data slices by performing data preprocessing on multi-source data and constructing two-dimensional tensors; The parameter determination unit: obtains pattern labels from two-dimensional data slices through a pattern recognition model; determines time parameters based on the pattern labels and multi-source data; The result generation unit: constructs a Gaussian mixture model based on time parameters; and obtains the set of electricity consumption characteristic parameters corresponding to the time parameters by solving and updating the parameters of the Gaussian mixture model through a real-time incremental expectation-maximization algorithm.

[0018] Another aspect of the present invention provides an appliance power consumption feature extraction device for virtual power plant resource aggregation, comprising: a processor and a storage medium; the storage medium includes instructions, and the processor is configured to execute the instructions to implement the method described in the first aspect and any possible implementation thereof; the appliance power consumption feature extraction device for virtual power plant resource aggregation may be an electronic device or a chip in an electronic device.

[0019] Compared with the prior art, the beneficial effects of this application are: 1. This application utilizes a pattern recognition model to extract predefined pattern labels with clear behavioral semantics based on domain knowledge from multi-source electrical appliance data, achieving semantic enhancement from raw data to user-understandable behavior, directly bridging the cognitive gap. Furthermore, time parameters are extracted based on the pattern labels, and a Gaussian mixture model is constructed to accurately characterize the complex statistical patterns and multi-peak characteristics of user behavior over time. Finally, a real-time incremental expectation-maximization algorithm is used to continuously update the model, enabling the feature parameter set to dynamically adapt to changes in user habits. This results in a final output set of electricity consumption feature parameters that is not only highly interpretable and accurately reflects users' personalized needs and behavioral habits, but also possesses high precision and dynamic adaptability. This provides an accurate and reliable data foundation for virtual power plants to deeply understand and efficiently schedule aggregated resources.

[0020] 2. This application improves the accuracy and generalization ability of pattern recognition by modifying the convolutional neural network model: it uses convolutional kernels of a specific size to specifically capture fine-grained local time dependencies in active power sequences, and combines them with the ReLU activation function to enhance nonlinear expression capabilities; at the same time, it introduces a Dropout layer after the fully connected layer to effectively alleviate the overfitting problem, ensuring the model's adaptability to diverse user behaviors. The optimized model can map preprocessed two-dimensional data slices into semantic pattern labels based on domain knowledge definitions with high accuracy and robustness. This not only achieves reliable parsing from low-level data to high-level electricity consumption behavior, providing an accurate semantic foundation for subsequent steps, but also significantly enhances the accuracy and practicality of the entire feature extraction method at the core algorithm level.

[0021] 3. This application utilizes a fixed-size data pool to retain only the most recent time parameters and ensures data timeliness through a first-in, first-out (FIFO) mechanism. During parameter updates, the algorithm assigns time decay weights to each data point, making the model more focused on recent behavior. Based on posterior probabilities, it re-estimates the mixture weights, mean, and variance of the Gaussian mixture model, enabling it to continuously and efficiently capture and learn the evolution of users' electricity consumption habits. This ensures that the output set of electricity consumption characteristic parameters remains synchronized with the current real needs of users, thereby significantly improving the real-time performance and accuracy of the virtual power plant's understanding of the dynamic characteristics of aggregated resources. This provides a reliable and adaptive data model foundation for implementing precise and flexible load forecasting and scheduling decisions. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the method for extracting electrical appliance power consumption characteristics from virtual power plant resource aggregation in this application; Figure 2 This is a schematic diagram illustrating the principle of the electrical appliance power consumption feature extraction system for virtual power plant resource aggregation in this application. Detailed Implementation

[0024] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0025] Please see Figure 1 The first aspect of this application provides a method for extracting electrical appliance power consumption features from virtual power plant resource aggregation, including: Acquire multi-source data for the target electrical appliance; the multi-source data includes active power sequences and timestamp sequences; Two-dimensional data slices are obtained by preprocessing multi-source data and constructing two-dimensional tensors; Two-dimensional data slices are used to obtain pattern labels through a pattern recognition model; Time parameters are determined based on pattern labels and multi-source data; Construct a Gaussian mixture model based on time parameters; The Gaussian mixture model is solved and updated using the real-time incremental expectation-maximization algorithm to obtain the set of electricity consumption characteristic parameters corresponding to the time parameters.

[0026] In this embodiment, two-dimensional data slices are obtained by preprocessing multi-source data and constructing two-dimensional tensors, including: Extract the active power sequence and timestamp sequence from the multi-source data; in this embodiment, the active power sequence is represented as follows: Its corresponding timestamp sequence is represented as Where i represents the index of the data point in the sequence; N represents the total number of data points in the sequence; The normalized active power sequence is obtained by performing global normalization on several power values ​​in the active power sequence. In this embodiment, the global normalization process satisfies: ;in, and Represented as the minimum and maximum values ​​in the historical active power; Represented as Power value after global normalization; A sliding window is defined, and it is used to prune the timestamp sequence and the normalized active power sequence to obtain several timestamp slices and power slices. The sliding window includes a window length and a sliding step size. In this embodiment, the window length L is set to 28 sampling points, and the sliding step size is set to 1. In this embodiment, the timestamp slice is represented as follows: Timestamp slices are represented as Where k represents the slice number; A normalized timestamp slice is obtained by performing local normalization on several timestamps in the timestamp slice. In this embodiment, local standardization satisfies: ;in, This is represented as the e-th timestamp in the k-th slice. This represents the first timestamp in the k-th slice; T represents the time constant, the specific value of which is set according to experience. In this embodiment, T is set to 600 seconds; it is used to normalize the time scale to the vicinity of the [0, 1] interval, which is convenient for the subsequent learning of the pattern recognition model; Represented as The timestamp value after partial standardization; The power slice and the normalized timestamp slice are concatenated along the channel dimension to obtain a two-dimensional data slice; in this embodiment, the timestamp slice is concatenated after the power slice.

[0027] This embodiment first performs global normalization on the original active power sequence to eliminate dimensional differences between different electrical appliances. Then, a sliding window is used to prune the power and time series, extracting data segments with temporal locality. Next, local standardization is applied to the timestamp slices, weakening the influence of absolute time while strengthening the expression of relative time relationships. Finally, the processed power sequence and timestamp slices are concatenated along the channel dimension to construct a two-dimensional tensor that integrates power change information and temporal structure. Through multi-level normalization and structured data processing, not only is the consistency and comparability of the data significantly improved, but key local temporal features are also effectively preserved, providing high-quality, highly interpretable input for subsequent pattern recognition models. This enhances the accuracy and robustness of electricity consumption feature extraction from the data source.

[0028] The pattern recognition model in this embodiment is constructed using an improved convolutional neural network model; The improved convolutional neural network consists of several convolutional-pooling blocks and a fully connected classifier; the convolutional neural network in this embodiment uses the LeNet-5 convolutional neural network; the convolutional neural network in this embodiment uses 2 convolutional-pooling blocks and 1 fully connected classifier; The kernel size in the convolution-pooling block is adjusted to (1, 5); the adjusted kernel allows the convolution operation to slide only in the time dimension, processing the power channel and the time channel independently, thus avoiding feature confusion. The activation function in the convolution-pooling block is adjusted to the ReLU activation function; replacing the Sigmoid activation function with the ReLU activation function effectively alleviates the gradient vanishing problem in deep network training; A Dropout layer is added after the fully connected layer in the fully connected classifier network. The hyperparameter of the Dropout layer is the dropout rate; the specific value is set according to experience, and in this embodiment it is set to 0.2. Adding the Dropout layer enhances the model's generalization ability and prevents overfitting. The loss function of the pattern recognition model is the cross-entropy loss function; in this embodiment, the loss function L satisfies: Where B represents the batch size, j represents the pattern label number, MB represents the total number of pattern labels, and b represents the sample number. Let the one-hot vector of the b-th sample be denoted as the first... Each component, that is, for a sample If it belongs to the first Each pattern label, then ,otherwise ; This represents the predicted probability that the b-th sample belongs to the j-th pattern label.

[0029] In this embodiment, obtaining pattern labels from two-dimensional data slices using a pattern recognition model includes: Extract two-dimensional data slices; Two-dimensional data slices are input into a pattern recognition model, and then pass through several convolutional-pooling blocks and a fully connected classifier. The Softmax layer in the fully connected classifier outputs the probability of the two-dimensional data slices corresponding to several pattern labels. Use the pattern label corresponding to the maximum value in the probability as the pattern label for the two-dimensional data slice.

[0030] This embodiment employs convolutional kernels of a specific size to accurately capture fine-grained local time dependencies in the active power sequence, and combines this with the ReLU activation function to enhance the model's nonlinear expressive power. Simultaneously, a Dropout mechanism is introduced after the fully connected layer to effectively suppress overfitting and improve the model's adaptability to diverse user electricity consumption behaviors. The optimized model can map preprocessed two-dimensional data slices into semantic pattern labels based on domain knowledge definitions with high precision and robustness. This not only achieves reliable parsing from the underlying raw data to the higher-level electricity consumption behavior semantics, providing a solid and accurate semantic foundation for subsequent analysis, but also significantly enhances the accuracy, robustness, and practical value of the entire feature extraction method at the core algorithm level.

[0031] The determination of time parameters based on pattern labels and multi-source data in this embodiment includes: Extract several two-dimensional data slices under the same pattern label; When there are continuous two-dimensional data slices, extract the timestamp sequence of several consecutive two-dimensional data slices in multi-source data; The time parameters are obtained by performing parameter statistics on the timestamp sequence. The time parameters include the start time, end time, duration, and interval corresponding to several identical pattern labels. The interval is the time interval between the end time of the previous pattern label and the start time of the current pattern label.

[0032] The construction of a Gaussian mixture model based on time parameters in this embodiment includes: Extract time parameters; time parameters include start time, end time, duration, and interval corresponding to several identical pattern labels; Construct a Gaussian mixture model for several time parameters under the pattern label; the Gaussian mixture model satisfies: ;in, Let d represent the d-th time parameter; d represents the number corresponding to the time parameter; u represents the number of the Gaussian component; and U represents the total number of Gaussian components. This is represented as the model parameter corresponding to the d-th time parameter; in this embodiment, ; The mixing weights are represented as the Gaussian components of the d-th time parameter. and These are expressed as mean and variance, respectively. and ; Represented as a Gaussian probability density function; The Gaussian probability density function satisfies: .

[0033] This embodiment first extracts time parameters representing key temporal information of user behavior from accurately identified similar pattern labels, transforming abstract electricity consumption behavior into quantifiable and analyzable structured data. Based on this, a Gaussian mixture model is constructed for each type of time parameter, making full use of its multi-peak characteristics to accurately characterize the complex probability distribution of user behavior in the time dimension. This not only achieves the accurate transformation from qualitative electricity consumption patterns to quantitative time parameters, but also reveals the statistical regularities and inherent uncertainties behind user behavior through probabilistic modeling. The resulting electricity consumption characteristics can deeply reflect the real needs of users and also have good interpretability and statistical robustness, providing solid and crucial data support for virtual power plants to carry out refined resource prediction and adaptive scheduling.

[0034] In this embodiment, the electricity consumption characteristic parameter set corresponding to the time parameters is obtained by solving and updating the Gaussian mixture model using the real-time incremental expectation-maximization algorithm, including: The real-time incremental expectation-maximization algorithm is used to calculate the set of electricity consumption characteristic parameters corresponding to a Gaussian mixture model with fixed time parameters in a fixed data pool; the size of the fixed data pool is... This is a fixed value, and the specific value is set based on experience. In this embodiment, it is set to 200. The number of time parameters in the fixed data pool has reached At that time, the time parameters in the fixed data pool are removed according to the first-in-first-out principle, and the new time parameters are added to the fixed data pool. The update of the electricity consumption characteristic parameter set includes mixed weight update, mean update and variance update; Mixed weight updates satisfy: ;in, This represents the mixed weight updated on the u-th Gaussian component for the d-th time parameter; m is the index of the time parameter in the fixed data pool, and M is the total number of time parameters in the fixed data pool. Quantization represents the influence of time parameters in a fixed data pool on the calculation of new mixed weights, satisfying... ,and ; This represents the posterior probability that the m-th time parameter sample in the data pool is generated by the u-th Gaussian component; The mean update satisfies: ;in, This is represented as the d-th time parameter with number m in a fixed data pool; It is represented as the mean of the d-th time parameter after updating on the u-th Gaussian component; It represents the posterior probability of the first time parameter in a fixed data pool being generated by the u-th Gaussian component; Variance update satisfies: ;in, It is expressed as the variance of the d-th time parameter after updating on the u-th Gaussian component; The set of electricity consumption characteristic parameters corresponding to the time parameters is determined based on the updated mixed weights, mean, and variance.

[0035] The posterior probability in this embodiment satisfies: Where h and u are both numbers representing Gaussian components; It is represented as the Gaussian probability density function corresponding to the d-th time parameter in the m-th data pool.

[0036] Please see Figure 2 The second aspect of this application provides an electrical appliance power consumption feature extraction system for virtual power plant resource aggregation, including: a data acquisition module and a data analysis module; the data acquisition module and the data analysis module are electrically and / or communicatively connected. Data acquisition module: Acquires multi-source data of the target electrical appliance through data acquisition equipment; the multi-source data includes active power sequence and timestamp sequence; the data acquisition equipment in this embodiment includes several sensors, etc. The data analysis module includes a data processing unit, a parameter determination unit, and a result generation unit; Data processing unit: Obtains two-dimensional data slices by performing data preprocessing on multi-source data and constructing two-dimensional tensors; Parameter determination unit: Obtains pattern labels from two-dimensional data slices using a pattern recognition model; determines time parameters based on the pattern labels and multi-source data; Result generation unit: Constructs a Gaussian mixture model based on time parameters; The Gaussian mixture model is solved and updated using the real-time incremental expectation-maximization algorithm to obtain the set of electricity consumption characteristic parameters corresponding to the time parameters.

[0037] Another embodiment of this application provides an appliance power consumption feature extraction device for virtual power plant resource aggregation, comprising: a processor and a storage medium; the storage medium includes instructions, and the processor is used to execute the instructions to implement the method described in the first aspect embodiment and any possible implementation of the first aspect embodiment; the appliance power consumption feature extraction device for virtual power plant resource aggregation may be an electronic device or a chip in an electronic device.

[0038] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0039] The working principle of this application is as follows: Multi-source data of the target electrical appliance is acquired; two-dimensional data slices are obtained by preprocessing the multi-source data and constructing two-dimensional tensors; pattern labels are obtained from the two-dimensional data slices using a pattern recognition model; time parameters are determined based on the pattern labels and multi-source data; a Gaussian mixture model is constructed based on the time parameters; the Gaussian mixture model is solved and updated using a real-time incremental expectation-maximization algorithm to obtain the electricity consumption characteristic parameter set corresponding to the time parameters. The pattern recognition model achieves semantic enhancement from raw data to user-understandable behavior, directly bridging the cognitive gap; the Gaussian mixture model is constructed to accurately characterize the complex statistical patterns and multi-peak characteristics of user behavior in the time dimension; the Gaussian mixture model is continuously updated, enabling the characteristic parameter set to dynamically adapt to changes in user habits. This ensures that the electricity consumption characteristic parameter set truly reflects the user's personalized needs and behavioral habits, possessing high accuracy and dynamic adaptability. This provides an accurate and reliable data foundation for the virtual power plant's deep understanding and efficient scheduling of aggregated resources, avoiding the problem of existing technologies lacking domain knowledge integration, leading to a disconnect between analysis results and real user needs, thus restricting the resource aggregation efficiency of the virtual power plant.

[0040] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application 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 methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. A method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation, characterized in that, include: Acquire multi-source data of the target electrical appliance; the multi-source data includes active power sequence and timestamp sequence; Two-dimensional data slices are obtained by preprocessing multi-source data and constructing two-dimensional tensors; Two-dimensional data slices are used to obtain pattern labels through a pattern recognition model; Time parameters are determined based on pattern labels and multi-source data; Construct a Gaussian mixture model based on time parameters; The Gaussian mixture model is solved and updated using the real-time incremental expectation-maximization algorithm to obtain the set of electricity consumption characteristic parameters corresponding to the time parameters.

2. The method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation according to claim 1, characterized in that, The process of obtaining two-dimensional data slices by preprocessing multi-source data and constructing two-dimensional tensors includes: Extract active power sequence and timestamp sequence from multi-source data; The normalized active power sequence is obtained by performing global normalization on several power values ​​in the active power sequence. Define a sliding window and use the sliding window to prune the timestamp sequence and the normalized active power sequence to obtain several timestamp slices and power slices; the sliding window includes a window length and a sliding step size; A normalized timestamp slice is obtained by performing local normalization on several timestamps in the timestamp slice. The power slice and the normalized timestamp slice are concatenated along the channel dimension to obtain a two-dimensional data slice.

3. The method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation according to claim 1, characterized in that, The pattern recognition model is constructed using an improved convolutional neural network model; The improved convolutional neural network consists of several convolutional-pooling blocks and a fully connected classifier; The kernel size in the convolution-pooling block is adjusted to (1, 5); The activation function in the convolution-pooling block is adjusted to the ReLU activation function; A Dropout layer is added after the fully connected layer of the network in the fully connected classifier. The hyperparameter of the Dropout layer is the dropout rate. The loss function of the pattern recognition model is the cross-entropy loss function.

4. The method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation according to claim 1, characterized in that, The step of obtaining pattern labels from two-dimensional data slices through a pattern recognition model includes: Extract two-dimensional data slices; Two-dimensional data slices are input into a pattern recognition model, and then pass through several convolutional-pooling blocks and a fully connected classifier. The Softmax layer in the fully connected classifier outputs the probability of the two-dimensional data slices corresponding to several pattern labels. The pattern label corresponding to the maximum value among the probabilities is used as the pattern label of the two-dimensional data slice.

5. The method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation according to claim 1, characterized in that, The determination of time parameters based on pattern labels and multi-source data includes: Extract several two-dimensional data slices under the same pattern label; When there are continuous two-dimensional data slices, extract the timestamp sequence of several consecutive two-dimensional data slices in multi-source data; The time parameters are obtained by performing parameter statistics on the timestamp sequence; the time parameters include the start time, end time, duration and interval corresponding to several identical pattern labels.

6. The method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation according to claim 1, characterized in that, The construction of the Gaussian mixture model based on time parameters includes: Extract time parameters; the time parameters include the start time, end time, duration, and interval corresponding to several identical pattern labels; A Gaussian mixture model is constructed for several time parameters under the pattern label; the Gaussian mixture model satisfies: ;in, Let d represent the d-th time parameter; d represents the number corresponding to the time parameter; u represents the number of the Gaussian component; and U represents the total number of Gaussian components. This is represented as the model parameter corresponding to the d-th time parameter; The mixing weights are represented as the Gaussian components of the d-th time parameter. and These are expressed as mean and variance, respectively. and ; Represented as a Gaussian probability density function; The Gaussian probability density function satisfies: 。 7. The method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation according to claim 1, characterized in that, The method of obtaining the set of electricity consumption characteristic parameters corresponding to the time parameters by solving and updating the parameters of the Gaussian mixture model through the real-time incremental expectation-maximization algorithm includes: The real-time incremental expectation-maximization algorithm is used to calculate the set of electricity consumption characteristic parameters corresponding to the Gaussian mixture model with time parameters in a fixed data pool; the size of the fixed data pool is... It is a fixed value; The number of time parameters in the fixed data pool reaches At that time, the time parameters in the fixed data pool are removed according to the first-in-first-out principle, and the new time parameters are added to the fixed data pool; The update of the electricity consumption characteristic parameter set includes mixed weight update, mean update and variance update; The hybrid weight update satisfies: ;in, This represents the mixed weight updated on the u-th Gaussian component for the d-th time parameter; m is the index of the time parameter in the fixed data pool, and M is the total number of time parameters in the fixed data pool. Quantization represents the influence of time parameters in a fixed data pool on the calculation of new mixed weights, satisfying... ; This represents the posterior probability that the m-th time parameter in a fixed data pool is generated by the u-th Gaussian component. The mean update satisfies: ;in, This is represented as the d-th time parameter with number m in a fixed data pool; It is represented as the mean of the d-th time parameter after updating on the u-th Gaussian component; It represents the posterior probability of the first time parameter in a fixed data pool being generated by the u-th Gaussian component; The variance update satisfies: ;in, It is expressed as the variance of the d-th time parameter after updating on the u-th Gaussian component; The set of electricity consumption characteristic parameters corresponding to the time parameters is determined based on the updated mixed weights, mean, and variance.

8. The method for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation according to claim 7, characterized in that, The posterior probability satisfies: Where h and u are both numbers representing Gaussian components; It is represented as the Gaussian probability density function corresponding to the d-th time parameter in the m-th data pool.

9. A system for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation, characterized in that: include: Interconnected data acquisition and data analysis modules; The data acquisition module acquires multi-source data of the target electrical appliance through a data acquisition device; the multi-source data includes an active power sequence and a timestamp sequence. The data analysis module includes a data processing unit, a parameter determination unit, and a result generation unit; The data processing unit obtains two-dimensional data slices by performing data preprocessing on multi-source data and constructing two-dimensional tensors; The parameter determination unit: obtains pattern labels from two-dimensional data slices through a pattern recognition model; determines time parameters based on the pattern labels and multi-source data; The result generation unit: constructs a Gaussian mixture model based on time parameters; and obtains the set of electricity consumption characteristic parameters corresponding to the time parameters by solving and updating the parameters of the Gaussian mixture model through a real-time incremental expectation-maximization algorithm.

10. A device for extracting electrical appliance consumption characteristics from virtual power plant resource aggregation, characterized in that, include: A processor and a storage medium; the storage medium includes instructions, and the processor is used to execute the instructions to implement the method for extracting electrical appliance power consumption features of virtual power plant resource aggregation as described in any one of claims 1-8.

Images