Self-adaptive load regulation and control method and system for smart power grid

Abstract

The invention discloses a self-adaptive load regulation and control method and system for a smart power grid, and relates to the technical field of smart power grids, and the method comprises the steps: collecting aggregated load data and auxiliary analysis data; a load response emergence critical point detection mechanism is adopted, the collective response sensitivity of the aggregated load is evaluated online, and whether the collective response sensitivity is in or close to a load response emergence critical point state or not is recognized; when the aggregation load is identified to be close to a load response emergence critical point state and a power grid regulation and control demand exists; a perturbation excitation signal with extremely small influence on a user is applied to the aggregation load or an internal probe load subset, and efficient regulation and control are realized by using a nonlinear response amplification effect in a load response emergence critical point state; and monitoring and evaluating a response effect in real time and feeding back the response effect for learning and iterative optimization. According to the method, efficient and self-adaptive load regulation and control are realized with extremely low cost and user disturbance by accurately identifying and utilizing the emergence critical behavior of the load, and valuable group response sensitivity dynamic information is generated.

Description

Technical Field

[0001] The present invention relates to the field of smart grids, and in particular to a method and system for adaptive load control of smart grids. Background Art

[0002] Load regulation in existing power systems plays a vital role in enhancing grid flexibility, promoting renewable energy integration, and ensuring stable system operation. However, traditional load regulation methods, such as those based on peak and valley electricity prices, direct load control, or high economic incentives, generally face bottlenecks.

[0003] First, regulation is costly and user participation is difficult to sustain. High subsidies increase grid operating costs, while mandatory direct controls or frequent price signals can easily lead to user resentment, causing "incentive fatigue" and limiting the enthusiasm and sustainability of user participation.

[0004] Secondly, there's a cognitive mismatch between the timing of regulation and the actual response state of the load group. Existing technologies generally lack real-time, accurate perception of the spontaneously evolving and dynamically changing "collective behavioral rhythms" or "collective response sensitivities" within the load group. This often results in regulatory interventions not being implemented during the "optimal window" when the group is most receptive and most likely to generate synergistic effects, thus impacting the efficiency and effectiveness of regulation.

[0005] Third, the deep value of aggregated load data has been insufficiently explored. Massive aggregated load data is primarily used for total volume forecasting or post-analysis, failing to fully tap into the deep information it contains about group dynamics, internal state evolution, and potential responses to minor disturbances. Summary of the Invention

[0006] The present invention provides an adaptive load control method and system for a smart grid to solve key technical problems commonly found in existing smart grid load control methods, such as high control costs, low user cooperation, mismatch between control timing and actual response status of load groups, and difficulty in accurately identifying and utilizing the dynamic response potential of load groups.

[0007] In view of the above problems, in a first aspect, the present invention provides a method for adaptive load control of a smart grid, comprising: collecting aggregated load data and auxiliary analysis data, wherein the auxiliary analysis data includes at least one of environmental data or time information; Based on the aggregated load data and auxiliary analysis data, a load response emergent critical point detection mechanism is adopted to online evaluate the collective response sensitivity of the aggregated load and identify whether the aggregated load is at or close to a load response emergent critical point state; Receiving a control demand instruction from the power grid, wherein the control demand instruction includes a target load adjustment amount and a response time requirement; When it is identified that the aggregated load is at or near the load response emergent critical point state, the currently optimal adaptive perturbation excitation strategy is selected or generated from a preset adaptive perturbation excitation strategy library in combination with the grid control demand instruction; applying a perturbation excitation signal to the aggregate load or a preselected subset of probe loads within the aggregate load according to the currently optimal adaptive perturbation excitation strategy, so as to actively utilize the inherent nonlinear response amplification effect of the aggregate load under the emergent critical point state of the load response to achieve effective regulation of the aggregate load; The actual response effect of the aggregate load after the perturbation excitation signal is applied is monitored and evaluated in real time, and the evaluation result of the actual response effect is fed back to learn and iteratively optimize the parameters of the load response emergent critical point detection mechanism and the adaptive perturbation excitation strategy library.

[0008] In a second aspect, the present invention further provides an adaptive load control system for a smart grid, comprising: A data acquisition and preprocessing module is used to acquire aggregated load data and auxiliary analysis data, and preprocess the acquired data to generate preprocessed aggregated load data and preprocessed auxiliary analysis data; a load response emergent critical point detector, connected to the data acquisition and preprocessing module, for online evaluating the collective response sensitivity of the aggregated loads based on the preprocessed aggregated load data and the preprocessed auxiliary analysis data, and identifying whether the aggregated loads are at or approaching a load response emergent critical point state, and outputting information on the load response emergent critical point state; a grid demand and strategy generation module, configured to receive a control demand instruction from the grid and, based on the load response emergent critical point state information output by the load response emergent critical point detector and the control demand instruction, select or generate a currently optimal adaptive perturbation excitation strategy from a preset adaptive perturbation excitation strategy library; an adaptive perturbation exciter, connected to the grid demand and strategy generation module, configured to apply a perturbation excitation signal to the aggregate load or a preselected subset of probe loads within the aggregate load according to the currently optimal adaptive perturbation excitation strategy, so as to actively utilize the inherent nonlinear response amplification effect of the aggregate load under the emergent critical point state of the load response; a load response monitoring and evaluation module, configured to monitor and evaluate in real time the actual response effect of the aggregate load after the adaptive perturbation exciter applies a perturbation excitation signal, and generate an evaluation result of the actual response effect; A learning and optimization module is connected to the load response monitoring and evaluation module, the load response emergent critical point detector and the grid demand and strategy generation module, and is used to receive the evaluation results and historical data of the actual response effect, and to learn and iteratively optimize the parameters of the load response emergent critical point detection mechanism and the adaptive perturbation excitation strategy library.

[0009] The technical solution provided by this application has at least the following technical effects or advantages: By accurately identifying and utilizing the emergent critical points of load response and applying extremely low-cost perturbation incentives at the moment when the system is most sensitive to perturbations, it is possible to leverage significant and predictable aggregate load responses with minimal effort, thereby achieving efficient load regulation at an incentive cost far lower than traditional demand-side response (even close to zero direct economic incentives) or with a smaller sacrifice in user experience.

[0010] By only implementing extremely minor, user-imperceptible or easily accepted perturbations during the window of opportunity when the system is at or near the critical point of emergent load response, this approach minimizes user aversion and incentive fatigue caused by strong, inopportune interference, fostering a long-term trusting relationship and a positive interactive ecosystem of proactive cooperation between users and the power grid or load aggregators. Regular or on-demand group response sensitivity scans are conducted on target load groups, dynamically and meticulously assessing their potential response willingness and ability under different conditions. This provides a more robust, near-real-time decision-making basis for grid planning, the construction of demand response resource pools, and bidding, and promotes a shift from static to dynamic understanding of group load plasticity.

[0011] The system uses a load response emergent critical point detection mechanism to perceive the actual collective response characteristics of the aggregate in real time, supplemented by a continuous closed-loop learning and optimization mechanism. This allows for greater adaptability and stability to dynamic changes in load composition, unpredictable user behavior, and external environmental disturbances. A continuously operating load response emergent critical point detector can, as a low-cost byproduct, produce a continuous data stream on the dynamic changes in the collective response sensitivity of aggregated loads over time and in the environment. This can serve as a new dimension for system status diagnosis, enabling earlier warnings of abnormal energy consumption behavior in groups, and enabling more refined user profiling and energy service innovation. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of a method for adaptive load control in a smart grid according to the present invention; Figure 2 This is an architecture diagram of an adaptive load control system for a smart grid. DETAILED DESCRIPTION

[0013] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention, rather than all, are shown in the drawings.

[0014] Example 1 like Figure 1 A flow chart of a method for adaptive load control in a smart grid is shown, and the method includes the following steps: Step 1: Data collection and preprocessing.

[0015] Specifically, data related to the aggregated load is continuously collected from multiple data sources in real time. The data mainly includes real-time operation data of the aggregated load and at least one auxiliary data for assisting in analyzing the current load status and environmental background.

[0016] More specifically, the real-time operational data of aggregated loads refers to high-frequency time-series data reflecting the total power consumption characteristics of the entire load aggregate, primarily the aggregate's total active power data, but may also include total reactive power, total current, and regional grid voltage. The high-frequency sampling frequency is at least in the second range, preferably in the sub-second range, to capture rapid dynamic changes in the load. The auxiliary analysis data may include environmental meteorological data, such as regional outdoor temperature, humidity, and light intensity; precise timestamp information for synchronization and timing analysis of all data; and grid-side status information, such as grid frequency and voltage stability indicators.

[0017] Furthermore, after data collection, a series of preprocessing operations need to be performed on the raw data to ensure data quality and extract valid information. The preprocessing operations include: Data cleaning. Specifically, for outliers that may appear during the data collection process (such as wild-point data outside the reasonable range due to communication failures or sensor errors), outlier identification and elimination or correction algorithms based on statistical distribution (such as the standard deviation multiple method, setting a reasonable multiple, such as 3 to 5 times the standard deviation) are used. For missing values ​​in the data, methods such as linear interpolation, neighboring value filling, or prediction filling based on time series models (such as the autoregressive moving average model) are used to fill them in, depending on the length and pattern of the missing values.

[0018] Time synchronization. Specifically, ensure that the timestamps of all data collected from different sources and devices are strictly aligned, with time errors controlled to the millisecond level. This can be achieved using the Network Time Protocol or a more precise synchronization mechanism.

[0019] Initial feature extraction. Specifically, some basic dynamic features are extracted from the preprocessed aggregated load data. These features may include, but are not limited to: the rate of change of load power per unit time, obtained by calculating the ratio of the power difference to the time difference between consecutive sampling points; the fluctuation amplitude of load power within a specific time window, characterized by calculating the difference between the maximum and minimum power values ​​within the window or the standard deviation of the power values; and the short-term autocorrelation of the load power series.

[0020] Step 2: Detect the emergent critical point of load response. Evaluate the collective response sensitivity of the current aggregated load in real time and online, and accurately identify whether it is in or about to enter a "load response emergent critical point" state.

[0021] In some embodiments, the load response emergent critical point detection mechanism includes an active perturbation detection process. The active perturbation detection process includes: Select and manage probe load subsets. Specifically, a small proportion of representative load units that can be precisely controlled and monitored by themselves are pre-selected from the entire aggregate load or pre-buried during system deployment to form a probe load subset. The principles for selecting the probe load subset are: the total capacity is extremely small relative to the entire aggregate load, usually not exceeding one to two percent of the total capacity of the aggregate; the power disturbance operation must be below the user's normal usage perception threshold, that is, it must not cause any perception or discomfort to the user in terms of vision, hearing, temperature and humidity perception, etc.; the probe load unit must have independent and precise power monitoring capabilities and communication and control capabilities to receive and execute external perturbation instructions. The optional probe load subset may include several dimmable public lighting circuits in large buildings, several terminal fan coil units of variable frequency air conditioners, and adjustable speed motors or electric heaters in certain non-critical auxiliary links in industrial production processes.

[0022] Design and apply standardized perturbation signals. Specifically, one or more standardized small-power perturbation signals are applied to the selected subset of probe loads through an adaptive perturbation exciter. The standardization refers to the fact that the waveform, amplitude, duration, etc. of the perturbation signal have preset specifications. For example, a typical perturbation signal may be: a small power drop step with an amplitude of 0.5 to 2 percent relative to the rated power of the probe load itself; the duration is from several seconds to tens of seconds, such as five to thirty seconds; the frequency of application may be a lower periodicity (such as a test every several minutes to more than ten minutes), or adaptively adjusted according to the rate of state change, or triggered on demand when a rapid assessment of the collective response sensitivity is required. The core is to ensure that this perturbation has extremely small energy for the overall aggregate load and is completely imperceptible to the user.

[0023] Observe and analyze the collective response. Specifically, while applying the standardized perturbation signal to the subset of probe loads, a high-frequency data acquisition system is used to synchronously and accurately monitor the changes in the total power of the entire aggregate load. The focus of the analysis is to identify and quantify whether the aggregate total load has a response that is significantly disproportionate to the input perturbation after the probe perturbation occurs. The response features that need to be extracted include: the aggregate response amplitude, which refers to the maximum power change of the aggregate total load due to the probe perturbation after deducting the natural fluctuation baseline of the background load (which can be obtained by extrapolating the load mean or trend for a period of time before the perturbation), or the average power change within a specific time window; the response delay, which refers to the time interval from the start of the probe perturbation to the aggregate response reaching a significant characteristic point (such as the maximum change rate point or half of the peak value); and the waveform pattern of the response.

[0024] Calculate the amplification coefficient and determine the critical state. Specifically, a key indicator, the amplification coefficient, is introduced to quantify the degree of nonlinear amplification of the collective response. The amplification coefficient is calculated by dividing the absolute value of the observed aggregate response amplitude by the absolute value of the power disturbance amplitude actually generated by the probe load subset. When this amplification coefficient is significantly greater than a certain benchmark value (for example, much greater than "one", the specific threshold can be dynamically set through historical data learning or statistical analysis, such as greater than five or ten), it is considered that the aggregate load exhibits a significant amplification effect on the perturbation. In addition, a synergy index can also be calculated, such as by calculating the correlation coefficient between the response of other non-probe loads in the aggregate and the probe disturbance (such as the Pearson correlation coefficient), or by measuring the improvement in waveform similarity through a dynamic time warping algorithm, or by increasing the amount of mutual information, etc., to characterize the enhancement of cooperative behavior in the aggregate. Finally, the amplification factor, coordination index and possible response pattern characteristics are comprehensively compared with the dynamically set threshold, or input into a pre-trained lightweight pattern recognition classifier (such as a support vector machine, decision tree or small neural network) for comprehensive judgment, so as to determine whether the aggregated load is currently at or close to the critical point of load response emergence.

[0025] In some embodiments, the load response emergent critical point detection mechanism may also include or be supplemented by an intrinsic fluctuation characteristic analysis process. The intrinsic fluctuation characteristic analysis process includes: Preprocess high-frequency load data. Specifically, use high-frequency (second-level or sub-second) power time series data of aggregated loads. Detrend the raw data (to remove slowly changing deterministic components, such as daily load curves) and de-periodicize (to remove obvious cyclical components, such as those caused by air conditioning cycled start and stop). Common processing methods include polynomial fitting detrending, differencing, signal separation based on seasonality and trend decomposition (such as the STL decomposition algorithm), or wavelet transform decomposition.

[0026] Extract statistical features related to critical phenomena. Specifically, calculate one or more statistical features from the preprocessed load intrinsic fluctuation sequence. These features may include: Long-range temporal correlation indices measure the "memory" or persistence of a time series over long time spans. For example, the Detrended Fluctuation Analysis (DFA) scaling index can be calculated using the DFA method, or the Hurst exponent can be calculated using the Rescaled Range Analysis (R / S analysis). When the values ​​of these indices deviate significantly from 0.5 (corresponding to a memoryless random walk) and approach 1 or higher, it indicates a strong long-range positive correlation between the series.

[0027] Power Spectral Density Analysis Features: Calculates the power spectral density of a load fluctuation sequence. If the power spectral density exhibits a unique power-law decay pattern in the low-frequency region, where the spectral density is inversely proportional to a power of the frequency (often referred to as "one-by-one alpha" noise or pink noise, where the alpha exponent is close to one), this is a common indicator of criticality.

[0028] Information entropy and complexity metrics: These are used to measure the complexity, uncertainty, or predictability of a time series. For example, approximate entropy or sample entropy can be calculated. These entropy values ​​may exhibit characteristic monotonic or non-monotonic changes (e.g., initially decreasing and then increasing) when transitioning from order to chaos or criticality. Indicators such as Lempel-Ziv complexity can also be calculated to reflect the rate at which new patterns emerge in a sequence.

[0029] Critical states are determined based on multi-indicator fusion. Specifically, because a single statistical feature may be affected by specific noise or load behavior, leading to misjudgment, a multi-indicator fusion strategy is employed to improve the robustness and accuracy of the judgment. A comprehensive "critical index" is defined. This index is a weighted combination of the aforementioned characteristic values ​​(such as long-range temporal correlation indicators, power spectral density characteristic parameters, and entropy values) after appropriate normalization, with specific weights (which can be set through expert experience or learned from historical data through machine learning methods) to form a comprehensive score. When this comprehensive critical index exceeds a dynamically set threshold, or when input into a pre-trained time series pattern recognition model (such as a hidden Markov model (HMM) or a long short-term memory (LSTM) network, a deep learning model specialized for processing time series data), and the output indicates a significant deviation from the normal fluctuation range and an approach to the critical region, the aggregated load is judged to be at or near the critical point of the load response emergence. Model parameters (such as the transition probability and emission probability of HMM, or the network weights of LSTM) should have a self-calibration mechanism, such as regular retraining or incremental learning using the latest data to adapt to the slow changes in load characteristics due to factors such as seasons and policies.

[0030] Step 3: Receive grid control requirements and generate an adaptive perturbation excitation strategy.

[0031] Specifically, it receives and interprets standardized control demand instructions from the grid dispatch center, virtual power plant management platform, or local energy management system. These instructions clearly define the control objectives, such as the desired absolute amount or relative percentage of load adjustment; the direction of adjustment (whether to reduce or increase load); the desired response start time, duration, and allowed response time window; the control urgency level (such as regular, priority, and emergency); and optional economic incentives or price signals associated with successful responses.

[0032] Furthermore, when the detection mechanism in step 2 determines that the current load aggregate is at or very close to the critical point of load response emergence, and at the same time receives a valid grid control demand instruction, this step is initiated to generate the optimal adaptive perturbation excitation strategy. The strategy generation process includes: Access and maintain an adaptive perturbation incentive strategy library. Specifically, the strategy library is a structured database that stores a variety of predefined perturbation incentive strategy templates or those optimized through historical experience learning. Each strategy template specifies in detail: the unique identifier of the incentive; the type of incentive (such as a small power reduction for a specific subset of probe loads, pushing designed energy-saving behavior prompts to user groups, and fine-tuning suggestions for the operating mode of smart appliances); the specific parameters of the incentive (such as the precise percentage of power adjustment, the content and form of information push, and the amplitude and range of mode adjustment); the expected response effect model of the strategy under different load backgrounds and different critical point intensity levels (such as the estimated achievable load adjustment amount, response speed, user acceptance probability, etc.); and the scenario label for which the strategy is most suitable (such as "high sensitivity state during the summer air conditioning peak period" and "moderate sensitivity state of nighttime lighting load").

[0033] Perform scenario matching and strategy selection. Specifically, based on the evaluation results output by the current load response emergent critical point detector (such as the specific quantitative level of collective response sensitivity and the proximity or intensity of the critical point), the specific parameters of the received grid control demand instruction (regulation target amount, direction, urgency, etc.), as well as current time information (such as peak, valley, and flat electricity price periods) and environmental conditions (such as auxiliary data such as current temperature and humidity), a set of candidate incentive strategies that preliminarily match the current scenario are screened from the adaptive perturbation incentive strategy library.

[0034] Optimization ranking and final decision making are performed. Specifically, the selected candidate incentive strategies are comprehensively evaluated and ranked according to one or more preset optimization objectives to select the optimal strategy under the current conditions. These optimization objectives may include: maximizing the "energy efficiency ratio" of regulation, that is, achieving the maximum effective load response with the minimum incentive cost (including direct energy or information incentive costs, as well as a quantitative assessment of the potential indirect impact on users); minimizing negative impacts on users, prioritizing strategies that minimize disruption to user comfort, work efficiency, or convenience; and meeting emergency grid regulation needs as quickly as possible. In some advanced implementations, this decision-making process can be assisted by an intelligent decision-making model, such as a reinforcement learning agent, whose state space consists of the current system state and grid demand, and whose action space consists of selectable incentive strategies or parameters. The reward function is designed based on the actual regulation effect and cost. Through continuous trial and error and learning, the agent can dynamically optimize its strategy selection ability. Alternatively, supervised learning models trained on historical data (such as decision trees, random forests, or Bayesian networks) can be used for intelligent recommendation.

[0035] Step 4: Apply adaptive perturbation excitation.

[0036] Specifically, this step is performed by the adaptive perturbation exciter. Based on the currently optimal adaptive perturbation excitation strategy generated in step 3, the adaptive perturbation exciter applies perturbation excitation signals to the aggregated load or specific controllable resources selected within it (such as the aforementioned subset of probe loads, or other flexible load units specified by the strategy). The energy intensity, range of action, and impact on the user's direct perception of these excitation signals are strictly controlled to extremely low levels.

[0037] Furthermore, the specific implementation of the process of applying the adaptive perturbation excitation depends on: Excitation execution interface. Specifically, the adaptive perturbation actuator has interfaces capable of communicating with and controlling different types of target loads. These interfaces may include: application programming interfaces (APIs) for smart end users; direct control interfaces for controllable devices within smart buildings or campuses; and flexible load control interfaces for industrial enterprises.

[0038] Excitation execution control. Specifically, the adaptive perturbation actuator ensures that the excitation signal is applied strictly according to the parameters defined in the selected strategy, such as timing (such as the exact start time and duration), amplitude (such as the percentage or absolute value of power adjustment), and frequency (such as the interval between repeated excitations). For coordinated excitation strategies involving multiple different load subsets, the synchronization or pre-set sequence of the excitation actions must also be precisely controlled.

[0039] State feedback mechanism. Specifically, after issuing an excitation instruction, the adaptive perturbation actuator obtains real-time feedback from the controlled object or its control node regarding the instruction execution status. For example, it confirms whether the instruction has been successfully received, is being executed, has been completed as required, and whether any errors or exceptions occurred during execution (and, if possible, obtains the error code or cause).

[0040] Step 5: Conduct load response monitoring and evaluation.

[0041] Specifically, after the adaptive perturbation actuator applies the excitation signal, comprehensive real-time monitoring of the actual response of the aggregated load is immediately initiated. The monitored data primarily includes high-frequency sequences of the aggregated total load power, as well as power data of specific sub-loads (such as probe loads) selected as excitation targets. It may also include user feedback related to the excitation event collected through specific channels (such as reviews from apps or customer service records).

[0042] Furthermore, the monitored response data needs to be quantitatively evaluated from multiple dimensions. The multi-dimensional effect evaluation includes: Calculation of actual load regulation. Specifically, this is achieved through the baseline subtraction method: First, based on the historical load data for a period of time before the application of the incentive, a suitable baseline load forecasting model (such as the sliding average method, the similar day load curve matching method, or a short-term load forecasting model based on regression analysis) is used to estimate the "what-if" load curve (i.e., baseline load) in the absence of incentives. Then, the actual monitored load curve during the incentive period is compared with this baseline load curve, and the difference between the two is the load response caused by the incentive.

[0043] Evaluation of response speed and sustainability. Specifically, quantitative indicators include: response delay, which is the time from the application of the excitation signal to the actual observation of a significant load change (such as reaching a predetermined change threshold or maximum rate of change); peak response time, which is the time from the start of the excitation to the load response reaching its maximum value; and effective response duration, which is the length of time the load response remains at a certain effective level (such as reaching more than 80% of the target regulation amount).

[0044] Evaluate the energy efficiency of regulation. Specifically, calculate the aggregate load response per unit stimulus input during this regulation event. For example, if active perturbation detection is used for stimulus, calculate the ratio of the actual aggregate response power to the probe perturbation power (i.e., the aforementioned amplification factor during the stimulus phase). If information-based stimulus is used, evaluate the average load reduction resulting from each information push.

[0045] Assess the impact on user experience. Specifically, if the incentive strategy may have a perceptible impact on users, collect and analyze relevant user experience metrics. For example, the number or percentage of users who received negative feedback (such as complaints, reports of decreased comfort, or operational inconvenience) within a period of time after the incentive event. For incentive programs that require users to actively choose to participate, also count the percentage of users who opt out after the incentive event. This ensures that negative impacts are minimized.

[0046] Evaluation of the state change of the emergent critical point of the load response. Specifically, analyze whether the stimulus event itself and the resulting collective response have an impact on the collective response sensitivity of the aggregate.

[0047] Evaluation of prediction accuracy. Specifically, the prediction results (e.g., estimated response volume, response speed, etc.) from the "expected effect model" upon which this incentive strategy was generated are compared with the actual observed effects to quantitatively evaluate the accuracy of the prediction model.

[0048] Finally, a structured evaluation report is automatically generated for each regulatory event, clearly presenting the quantitative results of the above evaluation indicators, relevant response curve charts, and descriptions of any abnormal situations that may occur.

[0049] Step 6: Learn and optimize.

[0050] Specifically, this step aims to utilize historically accumulated data and experience to continuously optimize the detection algorithm for emergent critical points of load response, the generation logic of the adaptive perturbation excitation strategy, and the content of the strategy library.

[0051] Furthermore, the learning and optimization process mainly includes: Historical data accumulation, management, and pattern recognition. Specifically, all relevant historical operating data requires long-term, structured storage. This data forms a learning database, which includes: raw aggregated load data and various auxiliary data; preprocessed characteristic data; detailed records of each load response emergent critical point detection process (such as probe perturbation parameters, collective response characteristic value sequence, and amplification coefficient value sequence during active detection; or various statistical characteristic indicator sequences during intrinsic fluctuation analysis); records of grid control demand instructions; details of each adaptive perturbation exciter's excitation strategy (type, parameters, target object, etc.) and execution feedback; and detailed effect evaluation reports generated after each control event. Based on this database, various data mining and machine learning techniques can be applied for pattern recognition and association analysis. For example, cluster analysis can be used to categorize load behavior patterns under different critical states and typical response patterns generated by different excitation strategies. Correlation analysis and regression modeling can be used to further explore the key factors affecting the generation and intensity of critical points, as well as the quantitative relationship between different excitation parameters and the final response effect.

[0052] Self-calibration of parameters for detectors of emergent critical points in load responses. Specifically, because factors such as the composition of load aggregates, user behavior, and the external environment can slowly change over time (so-called "concept drift" or "model drift"), the parameters of the detection models used to determine critical states (whether dynamic thresholds or classifiers in active detection, or statistical or time series models in intrinsic fluctuation analysis) must also be adaptive. Continuously monitor the performance of these models. For example, if the agreement between the predicted critical point and the subsequent "amplification effect" of actual stimuli decreases over time, this may indicate that the model parameters are no longer optimal. In this case, a parameter self-calibration mechanism should be triggered. For example, regular retraining or incremental learning of the relevant models using newly accumulated data that has been annotated with actual response effects (e.g., with the actual critical intensity level) is performed to update the model parameters. Dynamic thresholds can also be adaptively adjusted based on the statistical distribution (e.g., moving average and standard deviation) of recent relevant indicators (e.g., amplification coefficient).

[0053] Optimization of the adaptive perturbation excitation strategy library. Specifically including: Continuous retrospective evaluation of strategy effectiveness: Based on the large amount of control event data accumulated in the learning database, re-evaluate the actual average control effect (such as average response volume, average cost, average user impact, etc.) of each existing strategy in the strategy library in different scenarios and the stability of its effect (such as variance).

[0054] Optimization of strategy selection logic and recommendation models: If strategy generation relies on intelligent decision-making models (such as the aforementioned reinforcement learning agent or supervised learning recommendation model), the internal parameters of these models (such as the policy network weights or value network weights of the reinforcement learning agent, or the parameters of the supervised learning model) are continuously updated based on new environmental feedback (such as the reward signal of reinforcement learning, which integrates the control effect, cost, user feedback, etc.).

[0055] New strategy exploration and generation mechanisms: To avoid falling into local optima and discover new, potentially superior incentive strategies, certain exploration mechanisms are included. For example, when using reinforcement learning for strategy selection, a certain degree of randomness can be introduced into action selection. Alternatively, the parameters of certain high-quality strategies in the existing strategy library can be regularly subjected to small random perturbations or targeted mutations, and the effects observed to identify derivative strategies with better performance. Furthermore, under the premise that user influence is fully controllable, small-scale, controlled experiments can be designed to proactively test new perturbation combinations or incentive patterns that are theoretically feasible but not yet verified in the library.

[0056] Through the cyclic execution and closed-loop feedback of steps 1 to 6 above, the method of the present invention can achieve adaptive, efficient, and low-disturbance control of the load aggregate, and its overall performance will continue to improve with the increase of operating time and the accumulation of empirical data.

[0057] Example 2 This embodiment is based on the same inventive concept as the method for adaptive load regulation of a smart grid described in Example 1, and provides an adaptive load regulation system for a smart grid. The system can execute all or part of the steps of the method described in Example 1.

[0058] like Figure 2 As shown in the architecture diagram of an adaptive load control system for a smart grid, the system mainly includes: a data acquisition and preprocessing module, a load response emergent critical point detector, a grid demand and strategy generation module, an adaptive perturbation exciter, a load response monitoring and evaluation module, and a learning and optimization module.

[0059] Among them, the data acquisition and preprocessing module is responsible for collecting the operating data of the aggregated load and related auxiliary analysis data from multiple data sources in real time, and performing preprocessing operations such as data cleaning, time synchronization, and preliminary feature extraction on these raw data to provide high-quality data input for subsequent modules.

[0060] Furthermore, the data acquisition and preprocessing module may include: several data source interface units for connecting to different data acquisition channels; a data quality verification and cleaning unit with built-in multiple data anomaly detection and repair algorithms; a high-precision time synchronization unit to ensure the timing consistency of multi-source data; and a basic feature calculation unit for extracting preliminary features such as power change rate.

[0061] The load response emergent critical point detector is connected to the data acquisition and preprocessing module to obtain data, and adopts one or more of the aforementioned load response emergent critical point detection mechanisms (such as active perturbation detection mechanism, intrinsic fluctuation characteristic analysis mechanism, or a combination of the two) to evaluate the current collective response sensitivity of the aggregated load in real time online, and accurately determine whether it is in or approaching the load response emergent critical point state, and finally output an evaluation result or quantitative indicator about this state.

[0062] In some implementations, the load response emergent critical point detector may include: an active perturbation detection unit, which integrates probe load selection logic, standardized perturbation signal generation and control logic (coordinated with an adaptive perturbation actuator), collective response data capture and processing logic, and amplification coefficient and coordination calculation logic modules; an intrinsic fluctuation feature analysis unit, which integrates high-frequency data deep preprocessing logic (e.g., detrending and de-periodication) and parallel calculation logic modules for multiple statistical features related to critical phenomena (e.g., long-range temporal correlation, power spectral density characteristics, information entropy, etc.); and a critical state comprehensive judgment unit, which is responsible for integrating multivariate information from the active detection unit and / or the intrinsic fluctuation feature analysis unit, and ultimately outputting a judgment conclusion on the emergent critical point state of the load response through dynamic threshold comparison, pattern recognition classifier, or time series analysis model.

[0063] The grid demand and strategy generation module is responsible for receiving and parsing control demand instructions from the outside (such as the grid dispatching system). When the load response emergent critical point detector indicates that the system is in a favorable control opportunity (i.e., approaching the critical point), it combines specific control demand parameters and other current system context information (such as time and environment) from the adaptive perturbation excitation strategy library maintained within it, through intelligent matching, optimized sorting and decision logic, to select or dynamically generate the current optimal adaptive perturbation excitation strategy, and then issue this strategy instruction to the adaptive perturbation exciter.

[0064] Furthermore, the grid demand and strategy generation module may include: an external instruction receiving and parsing unit; a structured adaptive perturbation excitation strategy library and its management unit (responsible for the storage, update, and version control of the strategy); a situational awareness and feature extraction unit; and a strategy decision and optimization unit, which may be embedded with rule-based logic, heuristic algorithms, or more advanced machine learning decision models.

[0065] The adaptive perturbation exciter, in accordance with the current optimal adaptive perturbation excitation strategy instructions issued by the grid demand and strategy generation module, accurately applies perturbation excitation signals to the aggregated load or a specific subset selected therein through its diverse excitation execution interfaces, and can obtain real-time status feedback of the excitation execution.

[0066] Furthermore, the adaptive perturbation exciter may include: an excitation strategy instruction receiving and parsing unit; an excitation signal generation and precision control unit for different excitation types (such as power regulation, information push, mode switching) (ensuring the accuracy of signal parameters such as timing, amplitude, waveform, etc.); a communication and control interface adaptation unit for multiple types of targets (such as smart devices, user apps, industrial controllers); and a real-time monitoring and feedback unit for the excitation execution status.

[0067] The load response monitoring and evaluation module is responsible for immediately initiating comprehensive, real-time monitoring of the actual response of the aggregated load after the adaptive perturbation exciter applies excitation, and based on the monitoring data, quantitatively evaluating the actual effect of this regulation event from multiple dimensions (such as actual regulation amount, response speed, energy efficiency ratio, user impact, etc.), and finally generating a structured evaluation report.

[0068] Furthermore, the load response monitoring and evaluation module may include: a high-speed response data capture and synchronization unit; a baseline load dynamic estimation and correction unit; a multi-dimensional response indicator (covering effect, efficiency, impact, etc.) calculation and quantification unit; and an evaluation result visualization and report automatic generation unit.

[0069] The learning and optimization module is responsible for connecting all the other modules mentioned above, accumulating various types of data during the operation process (detection results, excitation records, evaluation reports, etc.) over a long period of time, and based on these data, using data mining, machine learning, operations optimization and other technologies to conduct continuous, closed-loop learning, calibration and optimization of core components such as the detection algorithm parameters of the emergent critical points of load response, the content and generation logic of the adaptive perturbation excitation strategy library, so as to continuously improve the overall control performance, adaptability and intelligence level of the system.

[0070] Furthermore, the learning and optimization module may include: a distributed historical database and its management unit; a multi-source heterogeneous data fusion and feature engineering unit; a parameter self-calibration and model update unit for the critical point detection model; a strategy effect backtracking evaluation, strategy parameter optimization and new strategy exploration and discovery unit for the incentive strategy library (advanced algorithms such as reinforcement learning can be integrated); and a system overall performance monitoring and diagnosis unit.

[0071] The embodiments disclosed in the present invention are intended to enable those skilled in the art to understand and implement the present invention. Based on the teachings of this specification, those skilled in the art may make various modifications to the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein with equivalents, without departing from the core ideas and spirit of the present invention. Such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention and should be included within the scope of protection of the present invention.

Claims

1. A method for adaptive load control in a smart grid, characterized in that: include: collecting aggregated load data and auxiliary analysis data, wherein the auxiliary analysis data includes at least one of environmental data or time information; Based on the aggregated load data and auxiliary analysis data, a load response emergent critical point detection mechanism is adopted to online evaluate the collective response sensitivity of the aggregated load and identify whether the aggregated load is at or close to a load response emergent critical point state; Receiving a control demand instruction from the power grid, wherein the control demand instruction includes a target load adjustment amount and a response time requirement; When it is identified that the aggregated load is at or near the load response emergent critical point state, the currently optimal adaptive perturbation excitation strategy is selected or generated from a preset adaptive perturbation excitation strategy library in combination with the grid control demand instruction; applying a perturbation excitation signal to the aggregate load or a preselected subset of probe loads within the aggregate load according to the currently optimal adaptive perturbation excitation strategy, so as to actively utilize the inherent nonlinear response amplification effect of the aggregate load under the emergent critical point state of the load response to achieve effective regulation of the aggregate load; The actual response effect of the aggregate load after the perturbation excitation signal is applied is monitored and evaluated in real time, and the evaluation result of the actual response effect is fed back to learn and iteratively optimize the parameters of the load response emergent critical point detection mechanism and the adaptive perturbation excitation strategy library.

2. The method for adaptive load control of a smart grid according to claim 1, wherein: The load response emergent critical point detection mechanism includes an active perturbation detection process, and the active perturbation detection process includes: selecting the probe load subset in the aggregate load; applying a preset standardized perturbation signal to the selected subset of probe loads; monitoring the magnitude of the overall response of the aggregated load to the normalized perturbation signal; Calculating an amplification factor between the overall response amplitude and the normalized perturbation signal input; Based on the calculated amplification factor and a preset threshold, it is determined whether the aggregated load is at or close to the load response emergent critical point state.

3. The method for adaptive load control of a smart grid according to claim 1, wherein: The load response emergent critical point detection mechanism further includes an intrinsic fluctuation characteristic analysis process, which includes: Preprocessing the aggregated load data to extract an intrinsic fluctuation series; Analyzing at least one statistical feature of the intrinsic fluctuation sequence, wherein the statistical feature is selected from a long-range time correlation index, a power spectral density feature, an information entropy index, or a complexity index; Based on the change of the statistical characteristics and a preset critical phenomenon criterion model, it is determined whether the aggregated load is in or close to the load response emergent critical point state.

4. The method for adaptive load control of a smart grid according to claim 1, wherein: The collection and aggregation of load data and auxiliary analysis data include: Collecting real-time power data of the aggregated load at a sampling frequency of seconds or sub-seconds; synchronously collecting the auxiliary analysis data, wherein when the auxiliary analysis data includes environmental data, the environmental data includes at least one of temperature, humidity, and light intensity; Data cleaning, time synchronization and feature extraction preprocessing are performed on the collected aggregated load data and the auxiliary analysis data.

5. The method for adaptive load control of a smart grid according to claim 1, wherein: The selecting or generating the currently optimal adaptive perturbation excitation strategy from a preset adaptive perturbation excitation strategy library includes: Maintaining the adaptive perturbation excitation strategy library, which stores a plurality of perturbation excitation strategies, each of which defines an excitation type, a target object, a signal parameter, an expected response effect model, and an applicable load response emergent critical point state intensity level and a scenario label; Based on the intensity of the currently identified load response emergent critical point state, the specific parameters of the power grid regulation demand instruction, and the time and environmental conditions reflected by the auxiliary analysis data; Screening a subset of candidate strategies from the adaptive perturbation excitation strategy library through a context matching algorithm, and sorting and selecting the subset of candidate strategies according to a preset optimization objective function to determine the current optimal adaptive perturbation excitation strategy; The optimization objective function uses at least one of maximizing the expected regulation energy efficiency ratio, minimizing the user perception impact, or satisfying the regulation demand as the fastest as an optimization criterion.

6. The method for adaptive load control of a smart grid according to claim 1, wherein: The applying of the perturbation excitation signal ensures that the perturbation excitation signal has a preset waveform, a preset amplitude, a preset duration, and a preset application frequency, wherein the amplitude and the duration are controlled within a range such that a degree of direct perception by the user is below a preset low threshold; The perturbation excitation signal is applied to the aggregate load or the subset of probe loads via a standardized digital communication interface.

7. The method for adaptive load control of a smart grid according to claim 1, wherein: The feedback of the evaluation results of the actual response effect is used to learn and iteratively optimize the parameters of the load response emergent critical point detection mechanism and the adaptive perturbation excitation strategy library, including: The learning database is constructed by using long-term accumulated historical aggregated load data, auxiliary analysis data, indicator sequences output by the load response emergent critical point detection mechanism, applied stimulus records, and evaluation data of the actual load response effect; Based on the data in the learning database, periodically or event-triggeredly self-calibrate or retrain the dynamic thresholds or model parameters in the load response emergent critical point detection mechanism to adapt to the drift of the characteristics and behavior patterns of the aggregated load; By analyzing the strategies, situations and effect data of historical incentive events in the learning database, the strategy parameters in the adaptive perturbation incentive strategy library are optimized, the expected effect model of the strategy is updated, inefficient strategies are eliminated, and new effective perturbation incentive strategies are explored and introduced through reinforcement learning mechanism or heuristic search mechanism.

8. An adaptive load control system for a smart grid, characterized in that: The system is used to perform the method according to any one of claims 1 to 7, and the system includes: A data acquisition and preprocessing module is used to acquire aggregated load data and auxiliary analysis data, and preprocess the acquired data to generate preprocessed aggregated load data and preprocessed auxiliary analysis data; a load response emergent critical point detector, connected to the data acquisition and preprocessing module, for online evaluating the collective response sensitivity of the aggregated loads based on the preprocessed aggregated load data and the preprocessed auxiliary analysis data, and identifying whether the aggregated loads are at or approaching a load response emergent critical point state, and outputting information on the load response emergent critical point state; a grid demand and strategy generation module, configured to receive a control demand instruction from the grid and, based on the load response emergent critical point state information output by the load response emergent critical point detector and the control demand instruction, select or generate a currently optimal adaptive perturbation excitation strategy from a preset adaptive perturbation excitation strategy library; an adaptive perturbation exciter, connected to the grid demand and strategy generation module, configured to apply a perturbation excitation signal to the aggregate load or a preselected subset of probe loads within the aggregate load according to the currently optimal adaptive perturbation excitation strategy, so as to actively utilize the inherent nonlinear response amplification effect of the aggregate load under the emergent critical point state of the load response; a load response monitoring and evaluation module, configured to monitor and evaluate in real time the actual response effect of the aggregate load after the adaptive perturbation exciter applies a perturbation excitation signal, and generate an evaluation result of the actual response effect; A learning and optimization module is connected to the load response monitoring and evaluation module, the load response emergent critical point detector and the grid demand and strategy generation module, and is used to receive the evaluation results and historical data of the actual response effect, and to learn and iteratively optimize the parameters of the load response emergent critical point detection mechanism and the adaptive perturbation excitation strategy library.

Images