A control method and system of a power selling device

Abstract

This invention relates to the field of electricity sales control technology, and discloses a control method and system for an electricity sales device. The system includes a dynamic electricity price calculation module, a load characteristic acquisition module, an optimization decision-making module, an execution control module, and a feedback adjustment module. The dynamic electricity price calculation module generates a benchmark electricity price curve; the load characteristic acquisition module acquires electricity load data and transmits it to the electricity consumption pattern analysis module, which outputs load classification labels; the optimization decision-making module combines the benchmark electricity price curve and load classification labels, and uses an artificial bee colony algorithm to optimize the electricity sales strategy and generate electricity price adjustment instructions; the execution control module determines the equipment control parameter sequence through a tabu search algorithm and transmits it to the electricity metering equipment; the feedback adjustment module collects user response data, processes it through wavelet noise reduction, and then transmits it to the dynamic electricity price calculation module and the optimization decision-making module. This system achieves dynamic optimization and precise control of the electricity sales process, adapting to the needs of the developing electricity market.

Description

Technical Field

[0001] This invention relates to the field of power sales control technology, specifically to a control method and system for a power sales device. Background Technology

[0002] With the deepening of electricity market reforms, the traditional single-price model is no longer adequate to meet diversified electricity demands and flexible market trading mechanisms. Currently, the electricity supply and demand relationship is characterized by dynamic changes, and user electricity consumption behavior is becoming increasingly complex, with significant differences among different types of users in terms of electricity consumption periods and load intensity. At the same time, the large-scale integration of renewable energy generation has increased the uncertainty of grid operation. Therefore, how to optimize the allocation of electricity resources and guide users to use electricity rationally has become a focus of industry attention.

[0003] The control systems of existing electricity sales devices often rely on fixed electricity pricing strategies, lacking real-time analysis of market transaction data and accurate understanding of user load characteristics. Electricity pricing is largely based on empirical adjustments, making it difficult to dynamically optimize according to market fluctuations and user electricity consumption patterns. This results in inefficient allocation of electricity resources, failing to adequately incentivize users to use electricity during off-peak hours and failing to meet the grid's peak-shaving and valley-filling operational requirements.

[0004] The methods for collecting and analyzing user electricity consumption data are relatively outdated, mostly relying on offline analysis methods. This makes it impossible to capture changes in user electricity consumption behavior in real time, resulting in a lag in adjustments to electricity sales strategies. In the control execution phase, the determination of equipment control parameters lacks scientific algorithm support, easily leading to insufficient control accuracy and affecting the stability of electricity metering and sales processes. Simultaneously, the processing of user response data is not refined enough; noise interference leads to inaccurate feedback information, making it difficult to effectively use for subsequent electricity price adjustments and strategy optimization, creating a vicious cycle. Summary of the Invention

[0005] The purpose of this invention is to provide a control method and system for an electricity sales device to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a control system for an electricity sales device, the system comprising:

[0007] The system includes a dynamic electricity price calculation module, a load characteristic acquisition module, an optimization decision-making module, an execution control module, and a feedback adjustment module.

[0008] The dynamic electricity price calculation module is used to generate a benchmark electricity price curve based on electricity market transaction data; the load feature acquisition module is connected to the electricity user terminal equipment to acquire electricity load data in real time and extract features, and transmit the load feature matrix to the electricity consumption pattern analysis module.

[0009] The electricity consumption pattern analysis module performs incremental learning on the load feature matrix based on the online sequence extreme learning machine algorithm and outputs load classification labels.

[0010] The optimization decision module optimizes the electricity sales strategy and generates electricity price adjustment instructions based on the benchmark electricity price curve and load classification labels using the artificial bee colony algorithm.

[0011] The execution control module receives the electricity price adjustment instruction, determines the sequence of equipment control parameters through a tabu search algorithm, and transmits the control signal to the power metering equipment.

[0012] The feedback adjustment module collects user response data, processes the feedback signal using wavelet denoising technology, and transmits the purified response feature vector to the dynamic electricity price calculation module and the optimization decision module.

[0013] Preferably, the load feature acquisition module includes a multi-source sensor access unit, a sliding window processing unit, and a feature dimensionality reduction unit; the multi-source sensor access unit acquires the effective value of current, voltage harmonic distortion rate, and power factor angle output by the smart meter; the sliding window processing unit sets the dynamic time window length and extracts continuous load segments according to a preset sampling period; the feature dimensionality reduction unit uses principal component analysis to compress high-dimensional load data and outputs the dimensionality-reduced load feature matrix.

[0014] Preferably, the electricity consumption pattern analysis module performs the following operation process: initialize the online sequence extreme learning machine network structure, set the number of input layer nodes to be equal to the load feature dimension; divide the real-time acquired load feature matrix into sequential data blocks; perform incremental training on each data block, update the hidden layer to the output layer weight matrix; calculate the load pattern confidence based on the output layer activation value, and output the load classification label when the confidence exceeds the threshold.

[0015] Preferably, the optimization decision module includes a strategy generation unit and a risk assessment unit; the strategy generation unit generates an initial set of electricity price strategies based on the benchmark electricity price curve; the risk assessment unit calculates the load response sensitivity coefficient of each strategy; in the artificial bee colony algorithm, the hired bee stage adjusts the strategy parameters according to the sensitivity coefficient, the observation bee stage uses a roulette wheel selection mechanism to screen high-quality strategies, and the scout bee stage resets low-fitness strategies until the optimal electricity price adjustment command is output.

[0016] Preferably, the execution control module executes the equipment parameter configuration process as follows: after receiving the electricity price adjustment instruction, it maps it to a set of equipment control constraints; the tabu search algorithm initializes the feasible solution neighborhood structure; the objective function value of the candidate solution is evaluated in the current solution neighborhood; the tabu list is updated to record the historical movement direction; and when the maximum number of iterations is reached, the equipment control parameter sequence is output to the power metering device.

[0017] Preferably, the feedback adjustment module includes a response acquisition unit, a signal processing unit, and a feature fusion unit; the response acquisition unit acquires user power consumption change rate data; the signal processing unit uses wavelet denoising technology to decompose the original signal and eliminates high-frequency noise through multi-scale threshold processing; the feature fusion unit extracts the time-domain statistical features and frequency-domain energy features of the denoised signal to generate a response feature vector.

[0018] Preferably, the feedback adjustment module further includes a closed-loop update unit; the closed-loop update unit inputs the response feature vector into the chaotic optimization algorithm to generate the correction coefficient of the benchmark electricity price curve; the correction coefficient is simultaneously transmitted to the curve generation unit of the dynamic electricity price calculation module and the strategy evaluation unit of the optimization decision module.

[0019] Preferably, the system further includes an anomaly detection module; the anomaly detection module receives the load classification label output by the power consumption pattern analysis module and the equipment control parameter sequence sent by the execution control module; it uses the isolated forest algorithm to construct a multi-dimensional feature space to detect abnormal data points that deviate from the preset operating range; when an anomaly is detected, it generates an interrupt command and transmits it to the execution control module.

[0020] Preferably, the anomaly detection module executes a real-time monitoring process: establishing a joint feature vector containing load characteristics, control parameters, and ambient temperature and humidity; constructing an isolation tree set through random projection; calculating the anomaly score for each data point and setting a dynamic alarm threshold; and triggering a device control parameter reset command when the anomaly score exceeds the threshold.

[0021] Preferably, the present invention further includes a control method for an electricity sales device, applied to a control system of an electricity sales device as described above, the method comprising:

[0022] The dynamic electricity price calculation module acquires electricity market transaction data and generates a benchmark electricity price curve.

[0023] The load feature acquisition module collects power user load data in real time and extracts the load feature matrix.

[0024] The load feature matrix is ​​processed by the online sequence extreme learning machine algorithm through the power consumption pattern analysis module, and load classification labels are output.

[0025] By optimizing the decision-making module and combining the benchmark electricity price curve and load classification labels, an artificial bee colony algorithm is used to generate electricity price adjustment instructions.

[0026] The execution control module uses a tabu search algorithm to parse the electricity price adjustment command and outputs a sequence of equipment control parameters.

[0027] User response data is collected through the feedback adjustment module, and the feedback signal is processed using wavelet noise reduction technology.

[0028] The parameters of the benchmark electricity price curve are adjusted using a chaotic optimization algorithm through a closed-loop update unit.

[0029] The anomaly detection module monitors abnormal operating data and triggers the device control parameter reset mechanism.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] Through the collaborative operation of multiple modules, intelligent and refined management of the electricity sales process has been achieved. The dynamic electricity price calculation module generates a benchmark electricity price curve based on electricity market transaction data, enabling electricity price setting to closely follow market dynamics, breaking away from the limitations of traditional experience-based pricing, and allowing electricity prices to better reflect real-time changes in market supply and demand.

[0032] The load feature acquisition module acquires electricity load data in real time and extracts features. Combined with the online sequence extreme learning machine algorithm used in the electricity consumption pattern analysis module, it can incrementally learn the load feature matrix and output load classification labels. This method can capture subtle changes in user electricity consumption behavior in real time, accurately classify user electricity consumption patterns, and provide a realistic basis for optimizing subsequent electricity sales strategies, making the strategies more targeted.

[0033] The optimization decision-making module uses the benchmark electricity price curve and load classification labels to optimize the electricity sales strategy and generate electricity price adjustment instructions using the artificial bee colony algorithm. This makes the formulation of the electricity sales strategy no longer dependent on subjective judgment, but achieves global optimization through intelligent algorithms. It can find a more effective electricity price adjustment scheme in complex market environments and diverse user load characteristics.

[0034] After receiving the electricity price adjustment command, the execution control module uses a tabu search algorithm to determine the sequence of equipment control parameters, which improves the scientific nature and accuracy of equipment control, ensures that the power metering equipment can operate stably according to the optimized strategy, reduces errors caused by manually setting parameters, and ensures the smooth execution of the electricity sales process.

[0035] After collecting user response data, the feedback adjustment module uses wavelet denoising technology to process the feedback signal. The purified response feature vector is transmitted to the dynamic electricity price calculation module and the optimization decision module, which effectively reduces the interference of noise on the feedback information, making the feedback data more realistic and reliable. This provides high-quality reference information for electricity price calculation and strategy optimization, forming a closed-loop optimization system that promotes the continuous evolution of electricity price setting and electricity sales strategies towards a more reasonable direction. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the working principle of the control system of the power sales device described in this invention;

[0037] Figure 2 This is a flowchart of the load characteristic acquisition module;

[0038] Figure 3 To optimize the flowchart of the decision-making module;

[0039] Figure 4 The flowchart for the feedback adjustment module;

[0040] Figure 5 This is a flowchart of a closed-loop update unit. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Please see Figure 1 The present invention provides a control system for an electricity sales device, the system comprising:

[0043] The system includes a dynamic electricity price calculation module, a load characteristic acquisition module, an optimization decision-making module, an execution control module, and a feedback adjustment module.

[0044] The dynamic electricity price calculation module generates a benchmark electricity price curve based on electricity market transaction data, which reflects the electricity supply and demand relationship at different times.

[0045] The load feature acquisition module connects to the power user terminal equipment, collects electricity load data in real time, extracts features, and forms a load feature matrix. The electricity consumption pattern analysis module uses an online sequential extreme learning machine algorithm to incrementally learn the load feature matrix and outputs load classification labels to distinguish the electricity consumption behavior patterns of different users.

[0046] The optimization decision-making module combines the benchmark electricity price curve and load classification labels to optimize the electricity sales strategy through artificial bee colony algorithm and generate electricity price adjustment instructions.

[0047] After receiving the electricity price adjustment command, the execution control module uses a tabu search algorithm to determine the sequence of equipment control parameters and transmits the control signal to the power metering equipment.

[0048] The feedback adjustment module collects user response data, processes the feedback signal using wavelet denoising technology, and feeds back the purified response feature vector to the dynamic electricity price calculation module and the optimization decision module to form a closed-loop control mechanism.

[0049] Example 1: See Figure 2The load characteristic acquisition module collects real-time operating data from power user terminal equipment through a multi-source sensor access unit, including RMS current, voltage harmonic distortion rate, and power factor angle. The RMS current reflects the real-time power consumption of the load, the voltage harmonic distortion rate characterizes the power quality of the grid, and the power factor angle reflects the phase characteristics of the load. This raw data is standardized to form a unified digital signal stream, which is then transmitted to the sliding window processing unit. The sliding window processing unit employs a dynamic adjustment mechanism; the window length is automatically adjusted according to the load fluctuation amplitude. A longer window is used when the load change is gradual to smooth noise, while a shorter window is used when the load fluctuates drastically to capture detailed features. The window sliding step size is synchronized with the power system's sampling period to ensure data continuity. Each load segment captured by each window is normalized to eliminate the impact of dimensional differences on subsequent analysis.

[0050] The feature dimensionality reduction unit employs principal component analysis (PCA) to compress the high-dimensional load data output by the sliding window. This method first calculates the covariance matrix of the dataset and determines the principal component directions through eigenvalue decomposition. Several principal components with cumulative contribution rates exceeding a preset threshold are selected to form a low-dimensional feature space. The dimensionality-reduced load feature matrix retains the main variation patterns of the original data while significantly reducing the data volume and improving subsequent processing efficiency. The PCA calculation process uses an incremental update mechanism to adapt to dynamic changes in load characteristics. The dimensionality-reduced feature matrix is ​​sorted by timestamp to form a time-series feature sequence for further processing by the power consumption pattern analysis module.

[0051] The power consumption pattern analysis module uses an online sequence extreme learning machine algorithm to incrementally identify load patterns. During network initialization, the number of input layer nodes is set to the dimension of the load feature matrix, the number of hidden layer nodes is determined according to empirical rules, and the number of output layer nodes is consistent with the number of load classification categories. The hidden layer uses randomly generated fixed weights to connect to the input layer, while the output layer weights are dynamically adjusted through online learning. The real-time input load feature matrix is ​​divided into continuous data blocks, and each data block is sequentially input into the network for training. During training, the hidden layer performs nonlinear transformations on the input features, and the output layer updates the weight matrix using recursive least squares to adapt to new data patterns.

[0052] During online learning, the network calculates the output layer activation value for each data block and converts it into a probability distribution using the Sigmoid function. Based on the probability distribution, it calculates the confidence level of the current load pattern. When the confidence level exceeds a set threshold, the module outputs the corresponding classification label. The label system covers typical electricity consumption patterns, such as residential, industrial, and commercial electricity consumption, and also includes mixed and abnormal patterns. For input data with insufficient confidence, the module marks it as a pending pattern and triggers a feature enhancement mechanism, requiring the load feature acquisition module to provide a more refined feature representation.

[0053] The network weight updates employ regularization to prevent overfitting, and retain some historical data as a validation set to monitor the model's generalization performance. When a performance degradation is detected, the number of hidden layer nodes is automatically adjusted or some connection weights are reinitialized. The incremental learning characteristics of the online sequential extreme learning machine enable it to adapt to the slow shifts in electricity user consumption patterns, such as seasonal changes or load characteristic alterations caused by equipment upgrades. The module periodically saves model snapshots, supporting historical state backtracking and anomaly diagnosis.

[0054] The load feature acquisition module and the electricity consumption pattern analysis module work together to form a closed-loop optimization mechanism. The classification labels output by the electricity consumption pattern analysis module are fed back to the load feature acquisition module, guiding the dynamic adjustment of the sliding window length and feature dimensionality reduction strategy. For load patterns that are more difficult to identify, the acquisition module automatically increases the sampling frequency or expands the feature extraction dimensions, providing richer data support. The data interfaces of the two modules use standardized protocols to ensure compatibility with equipment from different manufacturers. The system operation log records the complete feature processing and pattern recognition chain, supporting offline analysis and parameter optimization.

[0055] The online learning capability of the electricity consumption pattern analysis module allows it to be operational from the initial system deployment stage, without waiting for a large accumulation of historical data. As the running time increases, the model continuously learns from new data, gradually improving classification accuracy. The module's built-in anomaly detection mechanism can identify load data that does not conform to known patterns, triggering alarms and recording abnormal samples for subsequent analysis. For newly emerging stable electricity consumption patterns, the module expands the categories by extending the output layer nodes, maintaining the integrity of the classification system.

[0056] The temporal characteristics of the load feature matrix are fully considered in the network design, and the hidden layer includes a short-term memory structure to capture the dynamic patterns of load changes. The output layer weight update employs a recursive algorithm with a forgetting factor to balance the influence weights of new and old data. The module's operational status monitoring interface displays the classification result confidence, model update frequency, and historical accuracy curves in real time, facilitating maintenance personnel's understanding of the system's working status. The module supports remote parameter configuration and model import / export, meeting the requirements of distributed deployment.

[0057] The system achieves real-time perception and dynamic classification of power user load characteristics through the collaboration of a load feature acquisition module and an electricity consumption pattern analysis module. The multi-level feature processing of the acquisition module ensures the representativeness and simplicity of the input data, while the incremental learning algorithm of the analysis module adapts to the non-stationary characteristics of the power scenario. The entire mechanism enables online identification and tracking of user electricity consumption patterns without overly relying on historical data.

[0058] Example 2: See Figure 3The optimization decision-making module receives the benchmark electricity price curve from the dynamic electricity price calculation module and the load classification labels from the electricity consumption pattern analysis module through the strategy generation unit. The benchmark electricity price curve contains electricity price fluctuation information for different time periods, reflecting changes in the supply and demand relationship in the electricity market. The load classification labels identify the electricity consumption behavior characteristics of user groups, such as typical patterns of residential, industrial, or commercial electricity consumption. Based on these input data, the strategy generation unit constructs an initial set of electricity price strategies, including various forms such as time-of-use pricing, tiered pricing, and real-time pricing. Each strategy is associated with a combination of parameters, such as the time period division method, price gradient setting, and adjustment frequency.

[0059] The risk assessment unit analyzes the generated initial strategy set and calculates the load response sensitivity coefficient for each strategy. This coefficient is determined through a combination of historical data analysis and real-time monitoring, reflecting the intensity of the response of different user groups to price changes. Residential users are generally more sensitive to price changes, while industrial users' responses may be relatively lagging. The risk assessment unit establishes a multi-dimensional evaluation system, comprehensively considering factors such as the economic benefits, implementation feasibility, and user acceptance of the strategies. Evaluation indicators include expected returns, load shifting potential, and user satisfaction predictions.

[0060] The Artificial Bee Colony Algorithm (APA) performs the strategy search and optimization process in the optimization decision-making module. During the initialization phase, the initial strategy set is mapped to the locations of food sources in the bee colony, with each location representing a possible electricity pricing strategy. In the hired bee phase, a local search is performed on the current strategy, exploring the neighborhood solution space by fine-tuning the strategy parameters. The parameter adjustment magnitude is correlated with the load response sensitivity coefficient, employing more refined parameter tuning for highly sensitive user groups. In the observation bee phase, selection is made based on strategy fitness, with the fitness function comprehensively considering expected returns and risk indicators. A roulette wheel selection mechanism ensures that high-quality strategies receive more optimization opportunities while maintaining a certain level of diversity.

[0061] The reconnaissance bee phase is responsible for escaping local optima. When certain strategies fail to significantly improve fitness after multiple optimizations, these strategies are reset to new random parameter combinations. The reset process preserves the basic type characteristics of the strategies, only adjusting the specific parameter values. During algorithm iteration, the strategy set continuously evolves, gradually converging to the optimal solution region. The final output electricity price adjustment instruction contains detailed information such as time period divisions, price levels, and execution sequence, providing clear operational guidelines for the execution control module.

[0062] The execution control module receives the electricity price adjustment instructions generated by the optimization decision module and converts them into a set of equipment control constraints. These constraints include technical parameters such as voltage adjustment range, power limit threshold, and switching time interval. These conditions ensure that electricity price adjustments are implemented within the safe operating range of the power equipment. The module establishes a mapping relationship between equipment control parameters and electricity price strategies, clarifying the strategic objectives and expected effects corresponding to each control action.

[0063] The tabu search algorithm is responsible for finding the optimal sequence of equipment control parameters within the execution control module. During the algorithm initialization phase, the feasible solution neighborhood structure is defined, and the rules for generating candidate solutions are determined. The neighborhood of the current solution contains all feasible parameter combinations that meet the constraints, with each combination representing a possible equipment control scheme. The objective function evaluation comprehensively considers factors such as technical feasibility, implementation cost, and user impact, calculating a comprehensive score for each candidate solution.

[0064] During the search, the algorithm maintains a tabu list to record recent movement directions, avoiding repeated visits to the same solution space. The tabu timeout is dynamically adjusted according to the search progress, using a shorter timeout in the early stages to maintain flexibility, and extending the timeout as convergence approaches to enhance local search capabilities. The amnesty criterion allows breaking tabu restrictions to access certain high-quality solutions, preventing the missing of potential optimization opportunities. When the preset maximum number of iterations is reached or the convergence condition is met, the algorithm terminates and outputs the optimal device control parameter sequence.

[0065] The output control parameter sequence is transmitted to power metering equipment, including smart meters, remote terminal units, and other terminal devices, via standard communication protocols. The control signals contain precise timestamps and parameter values ​​to ensure synchronized adjustment operations across all devices. The module monitors the execution status of control commands in real time and collects feedback data from the devices to verify the adjustment effect. In case of abnormal execution, the module initiates an emergency handling procedure, automatically reverting to a safe state or triggering a re-optimization process.

[0066] The collaborative work of the optimization decision-making module and the execution control module forms a complete strategy optimization and execution chain. The decision-making module focuses on global optimization at the strategy level, while the control module is responsible for the precise implementation of technical details. Data interaction between the two modules adopts a standardized format to ensure the accuracy and timeliness of information transmission. The system operation log records the complete decision optimization process and control execution trajectory, supporting post-event analysis and parameter tuning.

[0067] Example 3: See Figure 4The response acquisition unit of the feedback adjustment module acquires user electricity consumption change data through the communication interface of the power metering equipment. This data is recorded at fixed time intervals, forming a raw response signal in time series form. During the acquisition process, the system automatically aligns with the electricity price adjustment time points and marks the changes in electricity consumption before and after the policy implementation. The raw signal contains various interference components, such as metering errors, instantaneous fluctuations, and abnormal electricity consumption behavior, which need to be purified by the signal processing unit. The signal processing unit uses wavelet denoising technology and selects the Daubechies series wavelet basis functions to decompose the signal at multiple scales. The number of decomposition levels is automatically determined according to the signal sampling rate and is usually set to cover the main load change cycle. At each decomposition scale, an adaptive threshold is determined by calculating the statistical characteristics of the wavelet coefficients.

[0068]

[0069] in Indicates the first Threshold for layer wavelet coefficients, It is the estimated standard deviation of the wavelet coefficients of this layer. The threshold function effectively suppresses noise while preserving valid signal characteristics. A soft thresholding method is used to shrink the wavelet coefficients and eliminate high-frequency noise components. The reconstruction process is performed layer by layer, ultimately yielding the denoised user response signal. The signal processing unit simultaneously detects and marks anomalies in the reconstructed signal, which may reflect specific power consumption events or equipment malfunctions.

[0070] The feature fusion unit extracts multi-dimensional features from the denoised signal. Time-domain analysis calculates the signal's statistical characteristics, including mean, variance, skewness, and kurtosis, describing the central tendency and distribution pattern of electricity consumption changes. Frequency-domain analysis uses wavelet packet transform to decompose the signal energy into different frequency bands, calculating the relative energy proportion and entropy value of each band. Time-frequency joint analysis calculates short-time feature sequences through a sliding window to capture the dynamic characteristics of load changes. The extracted features are standardized and organized into multi-dimensional feature vectors in chronological order. The dimensions of the feature vectors are dynamically adjusted according to the analysis requirements, minimizing redundancy while ensuring information integrity.

[0071] The feature fusion process employs a hierarchical structure. Low-level features reflect specific details of electricity consumption changes, while high-level features characterize the overall response pattern. Low-level features include directly observable indicators such as hourly electricity consumption change rate, daily load factor, and time-of-day shift volume. Mid-level features are obtained through time aggregation calculations, such as daily average change rate and weekly fluctuation index. High-level features are generated using pattern recognition methods, such as abstract indicators like response latency, policy sensitivity, and behavioral consistency. Relationships are established between the features at each level to form a complete feature system.

[0072] The generation process of the response feature vector is adaptive; the system dynamically adjusts the feature extraction strategy based on the user's historical response characteristics. For users exhibiting significant periodicity, the extraction weight of frequency domain features is increased; for users with higher randomness, the focus is on capturing time domain statistical features. The feature vector includes a quality evaluation index, reflecting the reliability and representativeness of each feature. The quality evaluation is calculated based on indicators such as the stability, discriminative power, and relevance of the feature values.

[0073] The feedback adjustment module employs a pipelined architecture for data processing, with each processing stage operating in parallel to improve real-time performance. The response acquisition unit continuously receives new electricity consumption data, the signal processing unit processes the data stream using a sliding window approach, and the feature fusion unit generates feature vectors at fixed intervals. Data buffers are set between each stage of the pipeline to balance synchronization issues caused by differences in processing speed. An internal status monitoring mechanism tracks data processing progress in real time, detecting and correcting data loss or anomalies.

[0074] The response feature vector output by the module is transmitted to the dynamic electricity price calculation module and the optimization decision module. The transmission protocol uses a lightweight data format that includes timestamps, feature values, and metadata. The dynamic electricity price calculation module uses the feature vector to evaluate the actual effect of the current electricity price strategy and identify the time periods and magnitudes that need to be adjusted. The optimization decision module analyzes the differences in response characteristics among different user groups and optimizes the target parameters and constraints of subsequent strategies. The closed-loop flow of feedback data enables the system to continuously adapt to changes in the electricity market and user behavior.

[0075] The configuration parameters of the feedback adjustment module are stored in an editable configuration file, supporting online updates and dynamic adjustments. Key parameters include wavelet type selection for signal processing, decomposition layer setting, feature extraction dimension, and fusion weights. The parameter adjustment interface provides visual aids to intuitively demonstrate the differences in processing effects under different parameter settings. The module maintains a versioned set of parameters, supporting quick rollback and historical configuration comparison analysis.

[0076] The module's anomaly handling mechanism is designed specifically for the characteristics of power data. For data loss caused by communication interruptions, interpolation based on historical patterns is used to supplement it; for outliers significantly exceeding reasonable ranges, data verification and re-acquisition processes are automatically triggered; for persistent systematic deviations, diagnostic reports are generated to indicate possible equipment failures or human interference. Detailed logs are recorded throughout the anomaly handling process, including anomaly type, occurrence time, handling method, and result status.

[0077] The storage management of response feature vectors employs a hierarchical strategy. Recently accessed high-frequency data is stored in an in-memory database, medium-term data is stored on a high-speed disk array, and long-term historical data is archived to a distributed file system. The data compression algorithm automatically selects a compression strategy based on the sparsity of the feature vectors, balancing storage space and access efficiency. The indexing mechanism is organized according to multiple dimensions such as time range, user type, and feature category, supporting efficient querying and analysis.

[0078] The performance optimization of the feedback adjustment module focuses on balancing processing latency and resource consumption. Computationally intensive tasks such as wavelet transforms are performed using multi-threaded parallel computation, memory usage is optimized through object pooling, and disk I / O is accelerated through pre-fetching and caching mechanisms. The module's running status is monitored in real time, including metrics such as CPU utilization, memory usage, and task queue depth, and resource allocation strategies are dynamically adjusted. When the system load is high, a degradation processing mode is automatically activated, temporarily reducing feature extraction dimensions or extending processing cycles to ensure the continuous operation of core functions.

[0079] The module's scalability design supports linear growth in processing power. New compute nodes can be seamlessly added to the processing cluster, with data processing tasks distributed via a consistent hashing algorithm. The horizontal scaling mechanism enables the system to handle increased load due to a larger user base or more frequent data collection. The module interface design follows a loosely coupled principle, with interactions with upstream and downstream modules conducted through standardized protocols, facilitating deployment and integration in a distributed environment.

[0080] The testing and verification of the feedback adjustment module employs a combination of real historical data and simulation data. Test cases cover various scenarios, including typical power consumption patterns, extreme load conditions, and abnormal events. Verification metrics include signal noise reduction effectiveness, feature extraction accuracy, and processing timeliness. Performance bottlenecks and functional defects discovered during testing are gradually resolved through iterative optimization, ultimately resulting in a stable and reliable processing flow.

[0081] Example 4: See Figure 5 The closed-loop update unit in the feedback adjustment module dynamically adjusts the system operating parameters using a chaotic optimization algorithm. Its core function is to generate correction coefficients for the benchmark electricity price curve based on user response feature vectors. This unit receives response feature vectors processed by the feature fusion unit, which contain user electricity consumption behavior data that has undergone noise reduction and feature extraction. In actual operation, the workflow of the closed-loop update unit can be illustrated with a specific example:

[0082] The system collected the following typical response data regarding changes in electricity consumption behavior among users in a commercial area after the adjustment of electricity pricing policies:

[0083] Time point Time period type Electricity consumption change rate Load transfer amount Response delay Price sensitivity Behavioral consistency D108:00 Peak hours -12.5% 15.2kWh 2 cycles 0.78 0.92 D112:00 Normal period +5.3% -8.7kWh 1 cycle 0.65 0.85 D118:00 Peak hours -9.8% 12.1kWh 3 cycles 0.72 0.88 D208:00 Peak hours -11.2% 14.5kWh 2 cycles 0.75 0.91 D212:00 Normal period +4.1% -7.9kWh 1 cycle 0.63 0.84

[0084] The time points in the table represent the monitoring points after the electricity price adjustment, and the time period types distinguish different time periods under the electricity pricing strategy. The electricity consumption change rate reflects the percentage change compared to the benchmark electricity consumption, and the load shift represents the absolute value of electricity consumption shifted between high and low electricity price periods. Response latency records the number of time periods required for users to exhibit significant behavioral changes after an electricity price adjustment. Price sensitivity measures the intensity of users' reactions to price changes, and behavioral consistency measures the stability of users' response patterns.

[0085] The closed-loop update unit first normalizes these response data to eliminate dimensional differences between different features. The processed data is then input into the chaotic optimization algorithm. During the algorithm's initialization phase, a set of candidate solutions for correction coefficients is randomly generated within the solution space. Each candidate solution contains adjustment parameters for each time period of the benchmark electricity price curve, such as the peak-period price fluctuation ratio, the base price during normal periods, and the duration of the flat periods. Chaotic optimization searches for the optimal solution through iterative search, utilizing the ergodic and random characteristics of chaotic motion during the search process to avoid getting trapped in local optima.

[0086] During the algorithm evaluation phase, each candidate solution is simulated to calculate its impact on user response characteristics. The evaluation process considers multiple aspects of matching, including the closeness between the expected and actual observed values ​​of electricity consumption change rate, the rationality of the load transfer distribution, and the consistency of user behavior. The evaluation results are converted into fitness values ​​to guide subsequent search direction adjustments. Information on historical high-quality solutions is retained during the search process, but the exploration of new areas is not completely excluded, maintaining appropriate search diversity.

[0087] After several iterations, the algorithm converges to a set of optimal correction coefficients. These coefficients include suggestions for adjusting the original benchmark electricity price curve, such as appropriately increasing the price fluctuation range for certain periods or extending the duration of specific price periods. The correction coefficients also include suggestions for adjusting strategy evaluation indicators, such as redefining the calculation method for price sensitivity or correcting the weighting of load transfer. These adjustment suggestions are synchronously transmitted to the dynamic electricity price calculation module and the optimization decision-making module.

[0088] After receiving the correction coefficient, the dynamic electricity price calculation module updates its internal curve generation logic. The new benchmark electricity price curve maintains the overall trend while making targeted adjustments based on actual user responses. For example, for user groups showing consistently high price sensitivity during peak hours (as shown in the table), the peak-valley price difference is appropriately increased; for cases of abnormal electricity consumption growth during off-peak hours, the rationality of off-peak pricing is reassessed. The curve update process retains historical version records, supporting rollback operations when necessary.

[0089] The strategy evaluation unit of the optimization decision-making module adjusts its evaluation criteria based on correction coefficients. The weights of user response-related parameters in the evaluation indicators are recalibrated to more accurately reflect actual operational results. The parameters of the artificial bee colony algorithm in the strategy generation process are also adjusted accordingly, such as changing the search step size of hired bees or the selection pressure of observation bees, to make strategy optimization more consistent with current user behavior characteristics. The adjusted evaluation system generates more realistic electricity price adjustment instructions in subsequent strategy optimizations.

[0090] The operating cycle of the closed-loop update unit is dynamically adjusted according to the system's operating status. During periods of significant fluctuation in the electricity market or marked changes in user behavior, the update cycle is shortened to quickly adapt to these changes; during stable operation, the cycle is appropriately extended to reduce computational overhead. Each update operation generates a detailed adjustment log, recording the trajectory of changes in the correction coefficients and their impact assessment. These logs are used to analyze the long-term effects of system parameter adjustments, accumulating experience for subsequent optimization.

[0091] The unit's built-in anomaly detection mechanism monitors the rationality of the correction coefficients. When a coefficient value is detected to exceed a preset reasonable range or deviate significantly from historical trends, a review process is triggered. The review process analyzes possible causes, such as abnormal data collection, sudden market events, or improper algorithm parameter settings, and takes corresponding measures based on the analysis results. For confirmed anomalies, coefficient application is suspended and an emergency handling process is initiated; for reasonable major adjustments, a special explanation is generated to record the basis for the decision.

[0092] The application of adjustment coefficients employs a gradual strategy to avoid drastic impacts on system operation. New coefficients are typically implemented in phases, first piloted with a small user group or during a specific time period to verify their effectiveness before gradually expanding their application. Detailed comparative data is collected during the pilot verification period to evaluate the differences in effectiveness before and after the adjustment. This gradual approach allows the system to continuously optimize and adjust its strategies while maintaining stable operation.

[0093] The closed-loop update unit interacts with other modules of the system using an asynchronous communication mechanism. Coefficient update requests are placed in a message queue, and the receiving module determines the processing time based on its own state. This design avoids system blocking caused by differences in module processing speeds, improving overall operational efficiency. Message passing uses a reliable transmission protocol to ensure accurate delivery and complete execution of update commands. The unit simultaneously monitors the status feedback from other modules to promptly understand the application effect of coefficients and identify potential problems.

[0094] The unit's performance monitoring tracks key operational metrics in real time, including algorithm convergence speed, computational resource usage, and communication latency. Monitoring data guides resource allocation and parameter tuning to maintain efficient unit operation. When performance degradation or anomalies are detected, diagnostic and recovery processes are automatically triggered. Long-term performance data accumulation forms an operational characteristic model of the unit, supporting predictive maintenance and capacity planning.

[0095] During system upgrades or maintenance, the closed-loop update unit supports pause and resume operations. The paused state saves the current algorithm state and intermediate results, and resumes execution from the point of interruption. This mechanism ensures that system maintenance does not affect the long-term optimization process and facilitates planned algorithm parameter adjustments and functional testing. The unit provides multiple data export formats, supports offline analysis and third-party tool processing, and expands the system's data analysis capabilities.

[0096] Example 5: The anomaly detection module constructs a multi-dimensional monitoring system by receiving load classification tags from the power consumption pattern analysis module and equipment control parameter sequences from the execution control module. This module uses the isolated forest algorithm to analyze operational data and establish a joint feature space including load characteristics, control parameters, and environmental variables. The dimensions of the feature space are dynamically adjusted according to the system configuration, including both real-time collected monitoring data and calculated derived indicators. During the module initialization phase, samples are randomly extracted from historical normal operation data to construct an isolated tree set. Each tree is generated by recursively and randomly selecting features and partition values ​​until all sample points are isolated or the maximum depth of the tree is reached.

[0097] During real-time monitoring, the module continuously receives data streams from various subsystems. Load characteristic data includes electrical measurements such as RMS current, voltage harmonic distortion rate, and power factor angle; this data has already been preprocessed and categorized by the power consumption pattern analysis module. Equipment control parameter sequences contain adjustment commands executed by the control module, such as voltage setpoints, power limit thresholds, and switching time parameters. Environmental variable data is collected through dedicated sensors, including physical quantities that may affect equipment operating status, such as temperature, humidity, and noise levels. All input data is time-aligned and formatted to form a standardized joint feature vector.

[0098] The Isolation Forest algorithm determines the degree of anomaly by calculating the path length of data points in the isolation trees. Normal data points are usually located in deeper levels of the tree and require more random partitioning to be isolated; anomaly data points are often isolated at shallower levels, exhibiting significantly different distribution characteristics. The module calculates anomaly scores for each data point, which comprehensively reflect the deviation of its average path depth across all isolation trees from the standard path. The score range is normalized for consistent interpretation and threshold setting.

[0099] The dynamic alarm threshold adaptively adjusts based on the system's operating status. Threshold calculation considers the statistical distribution of historical anomaly scores and employs a sliding window mechanism to track the latest data characteristics. The window size automatically adjusts based on the stability of the monitored object; a larger window smooths random fluctuations for slowly changing load types, while a smaller window responds quickly to changes for volatile industrial equipment. When an anomaly score exceeds the current threshold, the module generates an alarm event containing detailed diagnostic information. Alarm events are categorized into different severity levels, ranging from advisory notifications to emergency intervention requests, each corresponding to a different processing procedure.

[0100] The device control parameter reset command is the core output of the anomaly detection module. Before generating the command, the module performs a multi-level verification process to confirm the authenticity of the anomaly. The verification process includes checking the status of relevant sensors, analyzing the duration pattern of the anomaly event, and comparing with similar historical cases. After confirming the anomaly, the command contains specific reset parameter recommendations, which refer to the device's safe operating specifications and recent stable state records. The reset strategy may involve completely restoring the default parameters or may use a gradual adjustment to progressively correct the anomaly state; the specific choice depends on the anomaly type and the current system load.

[0101] During exception handling, the module maintains close interaction with the execution control module. Reset command transmission employs a reliable communication protocol, with priority flags to ensure timely processing. Confirmation information and status updates returned by the execution control module are used to verify the reset effect. If the abnormal state persists or worsens, the module may escalate handling measures, such as expanding the parameter reset range or triggering system-level protection mechanisms. The entire process is logged in a detailed operation log, including descriptions of the exception characteristics, measures taken, execution results, and subsequent observations.

[0102] The module's isolation tree set is updated periodically to maintain detection performance. The update strategy combines time decay and performance evaluation mechanisms. Newer data samples receive higher weights, while older data are gradually removed from the training set. Simultaneously, the detection accuracy of each isolation tree is continuously monitored, eliminating trees with declining performance and replacing them with newly generated trees. The update process is executed asynchronously in the background, without affecting real-time monitoring. The module maintains multiple versions of the tree set, supporting rapid rollback to a previous stable state to address potential issues introduced by algorithm updates.

[0103] The anomaly detection module's configuration interface offers a wealth of customization options. Users can adjust parameters such as the number and depth of isolation trees, the calculation method for anomaly scores, and the sensitivity of alarm thresholds. Configuration changes are only applied to the production environment after simulation verification to avoid improper settings affecting system stability. The module provides visualization tools to display the feature space distribution and anomaly detection boundaries, helping to understand algorithm behavior. These tools are invaluable for analyzing complex anomaly patterns and optimizing detection strategies.

[0104] Performance optimization for this module focuses on balancing processing latency and resource consumption. Computationally intensive tasks, such as isolation tree construction and anomaly score calculation, employ parallel processing techniques to fully utilize multi-core processor capabilities. Memory management is optimized through object pooling and caching mechanisms to reduce the overhead of frequent allocation and deallocation. The data preprocessing pipeline is designed for real-time requirements, using lock-free data structures to avoid thread blocking. Performance monitoring tracks key metrics in real time, such as throughput, latency distribution, and resource utilization, dynamically adjusting task scheduling strategies.

[0105] The anomaly detection module's scalable design supports distributed deployment. In a multi-node environment, monitoring tasks can be distributed across different nodes based on load type or region. A global coordinator is responsible for aggregating the detection results from all nodes and identifying systemic anomalies across nodes. The distributed architecture improves the system's ability to handle large-scale monitoring data and enhances system resilience in the event of a single node failure. Inter-node communication employs an efficient serialization protocol to minimize network transmission overhead.

[0106] The module's testing and verification employs a combination of real-world operational data and simulated anomalies. The test case library includes various typical anomaly scenarios, such as equipment failure, human interference, communication anomalies, and sudden environmental changes. Each test case provides a detailed description of the anomaly characteristics, expected detection results, and suggested handling. Regular regression testing ensures that algorithm updates do not reduce existing detection capabilities. During testing, special attention is paid to balancing false positives and false negatives, and algorithm parameters are adjusted to minimize both types of errors as much as possible.

[0107] After integration into the system, the module continuously collects operational data for subsequent optimization. The results of anomaly handling and subsequent analysis are fed back to the detection algorithm, forming a closed-loop learning mechanism. With accumulated operational experience, the algorithm gradually adapts to the data characteristics of specific scenarios, improving detection accuracy. The module maintains a knowledge base recording historical anomalies and handling experiences, supporting case-based reasoning and decision-making.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A control system of a power vending device, characterized by, It includes a dynamic electricity price calculation module, a load characteristic acquisition module, an optimization decision-making module, an execution control module, and a feedback adjustment module; The dynamic electricity price calculation module is used to generate a benchmark electricity price curve based on electricity market transaction data; the load feature acquisition module is connected to the electricity user terminal equipment to acquire electricity load data in real time and extract features, and transmit the load feature matrix to the electricity consumption pattern analysis module. The electricity consumption pattern analysis module performs incremental learning on the load feature matrix based on the online sequence extreme learning machine algorithm and outputs load classification labels. The optimization decision module optimizes the electricity sales strategy and generates electricity price adjustment instructions based on the benchmark electricity price curve and load classification labels using the artificial bee colony algorithm. The execution control module receives the electricity price adjustment instruction, determines the sequence of equipment control parameters through a tabu search algorithm, and transmits the control signal to the power metering equipment. The feedback adjustment module collects user response data, processes the feedback signal using wavelet denoising technology, and transmits the purified response feature vector to the dynamic electricity price calculation module and the optimization decision module. The optimization decision-making module includes a strategy generation unit and a risk assessment unit; the strategy generation unit generates an initial electricity price strategy set based on the benchmark electricity price curve; the risk assessment unit calculates the load response sensitivity coefficient of each strategy; In the artificial bee colony algorithm, the hired bee stage adjusts the strategy parameters based on the sensitivity coefficient, the observation bee stage uses a roulette wheel selection mechanism to select high-quality strategies, and the scout bee stage resets low-fitness strategies until the optimal electricity price adjustment command is output.

2. The control system of a power vending device according to claim 1, wherein The load feature acquisition module includes a multi-source sensor access unit, a sliding window processing unit, and a feature dimensionality reduction unit. The multi-source sensor access unit acquires the effective value of current, voltage harmonic distortion rate, and power factor angle output by the smart meter. The sliding window processing unit sets the dynamic time window length and extracts continuous load segments according to a preset sampling period. The feature dimensionality reduction unit uses principal component analysis to compress high-dimensional load data and outputs the dimensionality-reduced load feature matrix.

3. The control system of a power vending device according to claim 1, wherein The electricity consumption pattern analysis module performs the following operation: initializes the online sequence extreme learning machine network structure, sets the number of input layer nodes to be equal to the load feature dimension; divides the real-time acquired load feature matrix into sequential data blocks; performs incremental training on each data block, updates the hidden layer to the output layer weight matrix; calculates the load pattern confidence based on the output layer activation value, and outputs the load classification label when the confidence exceeds the threshold.

4. The control system of a power vending device according to claim 3, wherein The execution control module executes the equipment parameter configuration process as follows: after receiving the electricity price adjustment instruction, it maps it to a set of equipment control constraints; the tabu search algorithm initializes the feasible solution neighborhood structure; the objective function value of the candidate solution is evaluated in the current solution neighborhood; the tabu list is updated to record the historical movement direction; and when the maximum number of iterations is reached, the equipment control parameter sequence is output to the power metering equipment.

5. The control system of the power sales device according to claim 1, characterized in that, The feedback adjustment module includes a response acquisition unit, a signal processing unit, and a feature fusion unit; the response acquisition unit acquires user electricity consumption change rate data; the signal processing unit uses wavelet noise reduction technology to decompose the original signal and eliminates high-frequency noise through multi-scale threshold processing. The feature fusion unit extracts the time-domain statistical features and frequency-domain energy features of the denoised signal to generate a response feature vector.

6. The control system of the electricity sales device according to claim 5, characterized in that, The feedback adjustment module further includes a closed-loop update unit; the closed-loop update unit inputs the response feature vector into the chaotic optimization algorithm to generate the correction coefficient of the benchmark electricity price curve; the correction coefficient is simultaneously transmitted to the curve generation unit of the dynamic electricity price calculation module and the strategy evaluation unit of the optimization decision module.

7. The control system of the power sales device according to claim 1, characterized in that, The system also includes an anomaly detection module; the anomaly detection module receives the load classification labels output by the power consumption pattern analysis module and the equipment control parameter sequence sent by the execution control module; it uses the isolated forest algorithm to construct a multi-dimensional feature space to detect abnormal data points that deviate from the preset operating range; when an anomaly is detected, it generates an interrupt command and transmits it to the execution control module.

8. The control system of the electricity sales device according to claim 7, characterized in that, The anomaly detection module executes a real-time monitoring process: establishing a joint feature vector containing load characteristics, control parameters, and ambient temperature and humidity; constructing an isolation tree set through random projection; calculating the anomaly score for each data point and setting a dynamic alarm threshold; and triggering a device control parameter reset command when the anomaly score exceeds the threshold.

9. A control method for an electricity sales device, applied to the control system of an electricity sales device as described in any one of claims 1-8, characterized in that, The method includes: The dynamic electricity price calculation module acquires electricity market transaction data and generates a benchmark electricity price curve. The load feature acquisition module collects power user load data in real time and extracts the load feature matrix. The load feature matrix is ​​processed by the online sequence extreme learning machine algorithm through the power consumption pattern analysis module, and load classification labels are output. By optimizing the decision-making module and combining the benchmark electricity price curve and load classification labels, an artificial bee colony algorithm is used to generate electricity price adjustment instructions. The execution control module uses a tabu search algorithm to parse the electricity price adjustment command and outputs a sequence of equipment control parameters. User response data is collected through the feedback adjustment module, and the feedback signal is processed using wavelet noise reduction technology. The parameters of the benchmark electricity price curve are adjusted using a chaotic optimization algorithm through a closed-loop update unit. The anomaly detection module monitors abnormal operating data and triggers the device control parameter reset mechanism.

Images