Demand side response monitoring method based on multi-dimensional index monitoring

Abstract

The invention discloses a demand side response monitoring method based on multi-dimensional index monitoring, and relates to the technical field of power system operation and control, and the method comprises the steps: collecting and preprocessing multi-source heterogeneous demand side response data including a user historical load, a demand side response event and a user file; dynamically deducing a user incentive threshold, quantifying user behavior inertia, calculating a user demand side response participation willingness stability index, and constructing a user dynamic behavior portrait; based on the behavior kinetic parameters and the target demand side response event information, predicting the response behavior of the user by using a prediction model; and according to the prediction result and the demand side response demand, target users are optimized and screened, and a personalized demand side response scheduling scheme and an incentive strategy are generated. The invention aims to solve the problems of inaccurate response prediction, extensive scheduling, low user participation degree and insufficient response reliability of the traditional demand side response, and the overall operation efficiency, reliability and economical efficiency of the demand side response are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of power system operation and control, and in particular to a demand-side response monitoring method based on multi-dimensional indicator monitoring. Background Art

[0002] Demand-side response (DSR) is gaining increasing attention worldwide as a key means of enhancing power system flexibility, promoting renewable energy integration, and ensuring safe and stable grid operation. Traditional DSR management approaches typically rely on simple statistics of historical user response behavior, crude user categorization, and universal incentive mechanisms. However, user behavior in DSR is highly complex, dynamic, and individual.

[0003] Existing technologies have the following deficiencies in DSR management: traditional prediction models often lack a deep understanding of users' intrinsic behavioral motivations and obstacles, resulting in inaccurate predictions of users' willingness and response volume in specific DSR events, affecting the dispatchability of DSR resources; "one-size-fits-all" or incentive strategies based on simple classification are difficult to adapt to the behavioral characteristics of different users, often resulting in waste of incentive resources or poor incentive effects, and unable to effectively mobilize users' response potential; there is a lack of effective identification and targeted guidance of user behavioral inertia, making it difficult to fundamentally improve users' long-term participation enthusiasm and the stability and reliability of their response behavior; it is difficult to accurately assess the response reliability of individual DSR resources in different scenarios, which makes power grid dispatchers face greater uncertainty when relying on DSR resources. Summary of the Invention

[0004] The present invention provides a demand-side response monitoring method based on multi-dimensional indicator monitoring to solve the problems in the existing demand-side response management technology, such as inaccurate response prediction, extensive resource scheduling and incentives, failure to fully activate user participation potential, and difficulty in ensuring response reliability, caused by the lack of effective quantification and dynamic tracking of the deep dynamics of individual user behaviors, especially the critical points of their incentive responses, the inherent resistance to behavior changes, and the stability of long-term behavior patterns.

[0005] In view of the above problems, the present invention provides a demand-side response monitoring method based on multi-dimensional indicator monitoring, comprising: Collecting and preprocessing multi-source heterogeneous demand-side response data, wherein the multi-source heterogeneous demand-side response data includes user historical load data, demand-side response event data, and user profile data; Based on the pre-processed multi-source heterogeneous demand-side response data, dynamically inferring user incentive thresholds, quantifying user behavioral inertia, and calculating a user demand-side response participation willingness stability index; Predicting the user's response behavior in the target demand-side response event based on the user incentive threshold, user behavior inertia, user demand-side response participation willingness stability index, and target demand-side response event information; According to the predicted user response behavior and demand-side response requirements, target users are screened and a demand-side response scheduling plan and incentive strategy for the target users are generated.

[0006] Preferably, the method also includes: pushing personalized demand-side response event notifications and incentive information to target users based on their behavioral inertia index, incentive threshold estimation value, and demand-side response participation willingness stability index characteristics, and planning long-term behavioral guidance plans for specific users aimed at reducing their behavioral inertia and improving the stability of their participation willingness.

[0007] Preferably, the method also includes: continuously evaluating the comprehensive effects of demand-side response events, verifying the explanatory power of the behavioral inertia and incentive threshold parameters on user response behavior, and verifying the correlation between the demand-side response participation willingness stability index and the user's actual response stability and reliability; and based on the evaluation results, adaptively optimizing the weight coefficients of each sub-indicator in the behavioral inertia quantitative index system and the weight coefficients of each component in the calculation of the demand-side response participation willingness stability index.

[0008] The technical solution provided by this invention introduces and dynamically quantifies three behavioral dynamics parameters: user incentive threshold, behavioral inertia, and the stability index of demand-side response participation willingness. Applying these parameters to the entire process of optimizing demand-side response management can produce the following significant beneficial effects: By taking the intrinsic behavioral characteristics of user incentive thresholds, behavioral inertia, and the stability index of demand-side response willingness as key inputs, the prediction model of the present invention can more accurately predict the user's response probability and expected response volume in a specific DSR event, surpassing the traditional method that relies solely on historical load or simple user classification, and providing a more reliable basis for the effective scheduling of demand-side response resources. Based on the dynamic portrait of user incentive thresholds, behavioral inertia, and the stability index of demand-side response willingness, the present invention can intelligently screen the optimal target user combination and generate or recommend differentiated, more cost-effective incentive parameters for user groups with different behavioral characteristics. This not only avoids ineffective "pepper-sprinkling" incentives and reduces the implementation cost of demand-side response, but also improves the satisfaction of demand-side response needs and the reliability of responses.

[0009] Through personalized information push and a long-term behavioral guidance program designed to reduce user behavioral inertia and enhance the stability of participation willingness, this invention can fundamentally improve users' demand-side response participation experience and capabilities, cultivating more users into high-quality, highly reliable demand-side response resources. The introduced demand-side response participation willingness stability index provides a quantitative tool for assessing the stability and predictability of user demand-side response participation behavior, enabling users with more reliable behavior patterns to be prioritized when screening users and formulating scheduling plans, thereby significantly improving the overall reliability and planning of demand-side response scheduling and reducing grid scheduling risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flow chart of a demand-side response monitoring method based on multi-dimensional indicator monitoring according to the present invention. DETAILED DESCRIPTION

[0011] The present invention relates to a demand-side response monitoring method based on multi-dimensional indicator monitoring to solve the technical problems in the existing demand-side response management technology, such as inaccurate response prediction, extensive resource scheduling and incentives, failure to fully activate user participation potential, and difficulty in ensuring response reliability, which are caused by the lack of effective quantification and dynamic tracking of the deep dynamics of individual user behaviors, especially the critical points of their incentive responses, the inherent resistance to behavior changes, and the stability of long-term behavior patterns.

[0012] 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.

[0013] like Figure 1 A demand-side response monitoring method based on multi-dimensional indicator monitoring is shown in the flowchart. The method includes the following steps: First, data collection and preprocessing steps are performed. This step collects raw data related to demand-side response from multiple data sources and performs preprocessing operations such as cleaning, conversion, and alignment to provide a high-quality data foundation for subsequent analysis and modeling.

[0014] Next, we perform the parameterization and stability assessment of user behavior dynamics. This step, the core innovation of this invention, involves dynamically inferring the user's motivation threshold, quantifying their behavioral inertia, and calculating the stability index of their demand-side response willingness. Together, these parameters form a dynamic behavioral profile of the user.

[0015] Next, the demand-side response behavior prediction step is carried out. This step uses the prediction model to predict the user's response probability and expected response amount in a specific demand-side response event based on the user's latest behavioral dynamics parameters (including incentive threshold, behavioral inertia, demand-side response participation willingness stability index) and detailed information of the target demand-side response event.

[0016] Finally, the precise demand-side response scheduling and incentive strategy generation step is carried out. This step determines the participating users through intelligent screening based on the predicted user response behavior, demand-side response requirements, and the set optimization goals (such as cost minimization or reliability maximization), and generates differentiated and optimal demand-side response scheduling instructions and incentive strategies for different users or user groups.

[0017] A system for implementing the above method may include: Multi-source heterogeneous demand-side response data collection and preprocessing module, which is responsible for performing the aforementioned data collection and preprocessing steps.

[0018] User behavior dynamics parameterization and stability evaluation module, which is responsible for performing dynamic inference of user incentive thresholds, quantification of user behavior inertia, and calculation of the stability index of user demand-side response participation willingness.

[0019] A demand-side response behavior prediction module, which is responsible for performing the prediction of user response behavior.

[0020] A precise demand-side response scheduling and incentive strategy generation module, which is responsible for screening target users, formulating scheduling plans and generating incentive strategies.

[0021] The system can also include a module for guiding and activating user demand-side response behaviors, a module for system performance evaluation and iterative optimization, a module for data storage and management, and a visualization module for human-computer interaction and result display. Each module interacts with data and collaborates through predefined data interfaces.

[0022] Further, data sources: First, collect user historical load data.

[0023] Specifically, user historical load data is mainly collected through smart meters deployed on the user side, and obtained in batches through the system interface of the advanced metering architecture.

[0024] The data granularity of user historical load data is (for example, one data point is recorded every 1 minute) or 15-minute level (for example, one data point is recorded every 15 minutes). The data content of user historical load data is the time series data of the effective value of the electricity power consumption of each user.

[0025] Secondly, collect demand-side response event data.

[0026] Specifically, the demand-side response event data is obtained from the demand-side response management platform or the power market operation system through an application program interface or database docking.

[0027] The data content of the demand-side response event data is listed in detail as follows: including an event ID used to uniquely identify each demand-side response event; the time when the demand-side response event instruction is issued, that is, the initiation time; the effective period of the demand-side response event; the geographical area or specific user list affected by the demand-side response event, that is, the target area or user list; the incentive type used; and the specific incentive parameters corresponding to different incentive types; it also includes the response electricity or power promised by the user before the event occurs, that is, the response amount declared or contracted by the user.

[0028] Again, collect user profile data.

[0029] Specifically, the user profile data is obtained from a power marketing system, a user registration database or a customer relationship management system.

[0030] Further, a detailed description of the data preprocessing method: First, perform data cleaning operations.

[0031] Second, a timestamp alignment operation is performed.

[0032] Again, perform the baseline load calculation operation.

[0033] Specifically, baseline load is a benchmark reference for measuring the actual response amount of users in demand-side response events.

[0034] A commonly used method is the X-of-Y method (eg, the HighXofY method, or the MidXofY method).

[0035] Selection criteria for X and Y: The selection of X (the number of days selected) and Y (the historical range of days examined) should take into account the periodicity of user load, recent load trends, and the impact of special factors such as holidays. For example, for weekdays, Y can be the last 10-20 weekdays, and X can be 3-5 days within that range. For non-workdays, historical data for similar non-workdays should be selected. DSR event days, holidays, and days with abnormal load should be excluded.

[0036] Calculation steps: Determine the DSR event day type (such as weekdays, weekends, and specific holidays).

[0037] Select Y historical days of the same type.

[0038] Eliminate dates where load anomalies or DSR events also occurred.

[0039] For the remaining valid historical days, calculate the average load every 15 minutes (or other time granularity) during the period when the DSR event is expected to occur (for example, from 2:00 p.m. to 4:00 p.m.).

[0040] Filter out X days from Y historical days according to the High / Mid / LowXofY rule (for example, sort by total load or average load over a specific period and select the highest / middle / lowest X days).

[0041] The load curves of the X selected days during the corresponding period of the DSR event are averaged point by point to obtain the final baseline load curve.

[0042] Another alternative approach is based on time series regression analysis models.

[0043] Model form: For example, a linear regression model or a more complex machine learning model (such as SARIMA, Prophet) is built to predict the "normal" load of users in the absence of DSR events.

[0044] Features: The input features of the model can include historical load values ​​for the same period, day of the week, specific hour, holiday marks, recent average load, temperature, humidity, light intensity and other external environmental data.

[0045] Training and prediction: The model is trained using the user's historical data during non-DSR event periods, and then used to predict the baseline load during DSR events.

[0046] Then, the actual response amount calculation operation is performed.

[0047] Specifically, the formula for calculating the actual response volume is as follows: for each preset time granularity t (for example, every 15 minutes) during a demand-side response event, the user's actual response power value is calculated by subtracting the user's actual measured load power value at that time from the user's baseline load power value calculated at that time. A positive response volume generally indicates that the user successfully reduced their electricity load; a negative response volume indicates that the user increased their electricity load during that period (some special types of demand-side response projects may require users to increase their load).

[0048] Finally, perform preliminary feature engineering operations.

[0049] Specifically, during the data preprocessing stage, the following basic features related to user response behavior are extracted for subsequent analysis and modeling: Response delay: the time interval from the effectiveness of the instruction to the first significant deviation of the load from the baseline (such as a deviation of 5% and lasting for a time granularity).

[0050] Response persistence: The total length of time that the user load maintains an effective response state (e.g., below a preset percentage of the baseline).

[0051] Response completeness: the ratio of the actual total response power to the declared / contracted response power (or potential value).

[0052] Response stability: The standard deviation or coefficient of variation of the actual response power series during the response period, which measures the smoothness of the response.

[0053] Furthermore, the user incentive threshold is dynamically inferred. The incentive threshold refers to the ability to prompt a specific user to change from a "non-response" state to a "response" state (or to make the response amount reach a certain pre-set minimum response amount that is considered effective) under a specific situation (for example, the specific time, the user's load level at the time, external weather conditions, and other factors may all constitute part of the situation). The incentive threshold is a dynamically changing personalized parameter.

[0054] The present invention adopts a Bayesian update mechanism to integrate historical prior knowledge and newly observed data on each user's incentive threshold, thereby continuously optimizing the system's estimation of the user's incentive threshold.

[0055] The first step is to set the prior probability distribution of the excitation threshold.

[0056] Specifically, for users newly added to the system or with very sparse historical data on their participation in demand-side response events, due to the lack of direct, individual observational data, the prior probability distribution of their incentive threshold is primarily determined based on the historical statistical characteristics of incentive thresholds for the user's type (e.g., residential, small commercial, or large industrial). When selecting the form of the prior distribution, considering that the incentive threshold is typically a positive value and its distribution may exhibit a certain degree of skewness (i.e., asymmetry), a lognormal or gamma distribution can be used as the prior distribution for the incentive threshold. Of course, if historical data indicates that the distribution of incentive thresholds for a certain type of user is approximately symmetric, a normal distribution can also be used. For example, if the system chooses to use a lognormal distribution, its parameters (such as the mean and variance) can be derived from statistical analysis of historical incentive threshold data for groups of similar users (e.g., calculating the mean and variance of a sample, or inferring by constructing more complex group behavior models). If there is no direct historical incentive threshold data available in the system, the approximate range and distribution of the incentive threshold for that group of users can be indirectly inferred based on the average response rate data of similar users at different incentive levels in historical demand-side response events. For users who have accumulated a large amount of historical participation data in the system, the posterior probability distribution of the incentive threshold calculated by the system through the Bayesian update mechanism after their last participation in a demand-side response event will be directly used as the prior probability distribution for the new inference.

[0057] The second step is to construct the likelihood function.

[0058] Specifically, when a user experiences a new demand-side response event, the system obtains the specific incentive intensity provided by the event and the user's actual response outcome. The likelihood function describes the probability of observing the current actual response outcome, assuming the user's true incentive threshold is a specific value. Response outcomes can be quantified as binary (response / no response) or continuous (actual response power / electricity consumption). Likelihood function models typically assume that the probability of a user's response varies with the intensity of the provided incentive relative to their true IT. For example, a logistic function or probit function can be used.

[0059] The third step is to calculate the posterior probability distribution of the activation threshold. Specifically, according to Bayes' theorem, the posterior probability distribution is proportional to the product of the prior probability distribution and the likelihood function. Numerical calculation methods (such as discretization) or approximate inference methods (such as particle filtering, which maintains weighted particles and resampling) are used to obtain the posterior probability distribution of the activation threshold.

[0060] The fourth step outputs an estimate of the excitation threshold.

[0061] The fifth step reflects the dynamic update mechanism.

[0062] Specifically, after each new demand-side response event is processed, the updated posterior probability distribution of the incentive threshold will be used as the prior for the next inference to achieve dynamic approximation of the incentive threshold.

[0063] Furthermore, we quantify user behavioral inertia. Behavioral inertia refers to the inherent resistance or difficulty users experience in changing their baseline electricity usage patterns when faced with demand-side response incentives. Users with high behavioral inertia may exhibit delayed responses (delayed response start time), incomplete responses (actual response volume falls short of expectations), unstable responses (large power output fluctuations during the response period), or difficulty in sustaining responses, even when the incentive intensity exceeds their threshold. Behavioral inertia is a comprehensive metric; accurately quantifying it helps more accurately predict user responses and provides a basis for developing targeted user behavior guidance strategies.

[0064] User behavioral inertia is represented by a comprehensive index, the behavioral inertia index. This index is composed of the following sub-indicators through a weighted linear combination or nonlinear model: Response consistency factor: This factor measures the stability and consistency of a user's historical response behavior under conditions of sufficient incentives (a response is expected based on their incentive threshold). This factor is calculated by screening for events where the incentive intensity significantly exceeds the user's estimated incentive threshold at the time (for example, a ratio greater than 1.2). The factor then calculates the frequency of failure to generate an effective response (or response quality significantly below expectations) in these "should-respond" events. It also analyzes the variability or dispersion of actual responses or response patterns (e.g., response curve shape) under similar incentive margins. Greater variability or a higher frequency of response failures indicates poorer response consistency, resulting in a higher score for this factor. Recent historical events are given a higher weight.

[0065] Average Response Delay Factor: This factor reflects the average delay between a user receiving a demand-side response instruction and actually responding effectively. This factor is calculated as the weighted average of the response delays (defined in the Data Preprocessing section above) recorded across all demand-side response events in which the user has participated. Longer response delays are associated with higher scores, indicating greater behavioral inertia. Weighting can be applied exponentially, with more recent events receiving higher weights.

[0066] Response Incompleteness Factor: This factor reflects the extent to which a user's actual contribution to a demand-side response falls short of the expected or committed level. It is calculated based on the weighted average of the response completeness (defined in the data preprocessing section above) recorded across all of the user's historical demand-side response events. The lower the response completeness (i.e., the less thorough the response), the higher the score on this factor, indicating greater behavioral inertia. The weighting also favors recent events.

[0067] Response Instability Factor: This factor reflects the difficulty a user faces in maintaining the stability of their response output during the response period. It is calculated based on the weighted average of the response stability indicators (defined in the data preprocessing section above, such as the coefficient of variation or standard deviation of the actual response power series during the response period) recorded across all historical demand-side response events in which the user participated. The larger the response stability indicator value (i.e., greater power output fluctuations and more unstable response behavior), the higher the factor score, indicating greater behavioral inertia. The weighting also favors recent events.

[0068] Furthermore, the behavioral inertia index is calculated and updated: each of the above sub-indicators is normalized or standardized. The scores of all normalized sub-indicators are combined in a weighted linear manner to obtain the final behavioral inertia index. The initial value of the weight coefficient can be set by a domain expert, and then adaptive learning and optimization are performed based on the accumulated historical data through the system performance evaluation and iterative optimization module described in the present invention. Whenever a new demand-side response event data is processed, the system will recalculate all behavioral inertia sub-indicators related to the user, and update the user's behavioral inertia index accordingly, so as to timely reflect the latest status of the user's behavioral inertia.

[0069] Furthermore, the user's willingness to participate in demand-side response stability index is calculated. The demand-side response participation stability index is a comprehensive evaluation indicator designed to quantify the stability and predictability of a user's demand-side response participation behavior pattern within a specific evaluation period. The higher the index value, the more stable the user's willingness and behavior pattern in participating in demand-side response (i.e., the two core parameters, motivation threshold and behavioral inertia, themselves fluctuate less or show regularity). Therefore, their response behavior is more predictable and reliable for the system.

[0070] The Demand-Side Response Participation Intention Stability Index is calculated by comprehensively evaluating behavioral characteristics across the following dimensions, ultimately generating a standardized index value. This calculation is based on a configurable sliding time window (evaluation period), such as the user's most recent N demand-side response events or all behavioral data from the past M months.

[0071] Behavioral Inertia Stability Score: This measures the degree of fluctuation in a user's Behavioral Inertia Index over a set evaluation period. Obtain time series data on the user's Behavioral Inertia Index during the evaluation period. Calculate the coefficient of variation (standard deviation divided by the mean) or standard deviation of this time series as a volatility indicator. Normalize the calculated volatility indicator (for example, map it to a range of 0 to 1, where larger values ​​indicate greater instability). The Behavioral Inertia Stability Score is calculated by multiplying the full score for that sub-item by "1 minus the normalized volatility indicator."

[0072] Incentive Threshold Stability Score: This measures the degree of fluctuation in the estimated user incentive threshold over the specified evaluation period. This is similar to the calculation of the Behavioral Inertia Stability Score.

[0073] Incentive Threshold Trend Predictability Scoring: This assesses the regularity and predictability of the changing trends of user incentive thresholds within the set evaluation period. Time series trend analysis is performed on the estimated time series data of user incentive thresholds. For example, the least squares method can be used to perform linear regression fitting. The certainty of the trend fit is assessed, for example, by calculating the coefficient of determination (R-squared). A higher R-squared value indicates that the fitted linear trend explains more of the changes in incentive thresholds. The statistical significance of the fitted trend coefficient is also assessed (for example, by calculating the t-statistic and p-value). If the p-value is less than the preset significance level (e.g., 0.05), the trend coefficient is considered significant (whether significantly increasing, decreasing, or stable). Finally, a score is assigned based on the R-squared value and the significance of the trend coefficient (e.g., an R-squared greater than 0.7 and a p-value less than 0.05 is considered full marks, indicating high predictability; otherwise, scores decrease).

[0074] Furthermore, the comprehensive calculation and evaluation cycle selection of the demand-side response participation willingness stability index are as follows: the individual sub-item scores calculated through the above steps are weighted and summed to obtain the final demand-side response participation willingness stability index value. The initial value of the weight coefficient can be preset, and can be adaptively optimized and adjusted based on accumulated historical data during system operation. The higher the calculated demand-side response participation willingness stability index value, the more stable and predictable the user's demand-side response participation behavior pattern. The demand-side response participation willingness stability index is calculated based on a configurable sliding time window. The length of this time window (for example, the user's behavioral data from the past six months, or data from the 10 demand-side response events in which the user has recently participated) should be selected to fully reflect the user's recent behavioral characteristics and stability.

[0075] Furthermore, prediction of user demand-side response behavior.

[0076] Specifically, the present invention uses supervised learning methods to construct a prediction model, which can be combined and used according to specific circumstances.

[0077] For example, you can choose a gradient boosted decision tree model, random forest model, deep neural network model, logistic regression model, etc. You can make a comprehensive trade-off based on the amount of available training data, the complexity of the input features, the specific form of the prediction target (for example, predicting a probability value, a specific response value, or a response power curve), the system's available computing resources, and the requirements for the interpretability of the model's prediction results. You can use methods such as cross-validation to compare the actual prediction performance of different candidate models on a validation dataset to select the optimal model.

[0078] Construction of the prediction model input feature set: Use the dynamic user behavior parameters calculated in the previous steps as the core input features.

[0079] Specifically, the step of predicting the user's response behavior in the target demand-side response event based on the user incentive threshold, user behavioral inertia, user demand-side response participation willingness stability index and target demand-side response event information further includes: constructing a supervised learning prediction model, the input feature set of the prediction model includes at least the incentive intensity of the target demand-side response event, the user's incentive threshold estimate, the behavioral inertia index, the demand-side response participation willingness stability index, and the incentive margin composed of the ratio of the incentive intensity to the user's incentive threshold estimate.

[0080] The specific meanings and acquisition methods of these input features are as follows: Incentive intensity of target demand-side response event: The intensity of the incentive signal provided by the upcoming target demand-side response event.

[0081] User incentive threshold estimation value: the latest incentive threshold estimation value output by the aforementioned user behavior dynamics parameterization and stability evaluation module for the current target user when a target demand-side response event occurs.

[0082] User behavioral inertia index: the latest behavioral inertia index of the current target user when a target demand-side response event occurs, output by the aforementioned user behavior dynamics parameterization and stability evaluation module.

[0083] User's demand-side response participation willingness stability index: the latest demand-side response participation willingness stability index for the current target user when the target demand-side response event occurs, output by the aforementioned user behavior dynamics parameterization and stability evaluation module.

[0084] Incentive Margin: This directly reflects the extent to which the incentive intensity provided by the target demand-side response event exceeds the user's individualized incentive threshold. It can be calculated as the ratio of the "target demand-side response event's incentive intensity" divided by the "user's estimated incentive threshold," or as the difference between the "target demand-side response event's incentive intensity" and the "user's estimated incentive threshold."

[0085] The training process of the prediction model: Construct a training dataset: Each sample in the training dataset corresponds to the participation of an invited user in a historical demand-side response event.

[0086] Specifically, for each historical sample, the corresponding input feature set defined above needs to be extracted. When constructing features for historical samples, the dynamic parameters included, such as the user's incentive threshold estimate, behavioral inertia index, and demand-side response participation willingness stability index, should be the user's parameter values ​​at the time the historical demand-side response event occurred (these values ​​are obtained by retrospectively calculating earlier historical data, rather than using the current latest parameter values). At the same time, the target variable (i.e., label) needs to be defined for each historical sample. The specific form of the target variable depends on the goal of the prediction task.

[0087] Model training: On the training set, select a suitable loss function and optimization algorithm, and iteratively learn and adjust the internal parameters of the selected prediction model to minimize the difference between the model's prediction results on the training data and the true target variable (i.e., the loss function value).

[0088] The output form of the prediction result is the probability value of the user generating an effective response in this specific demand-side response event (this value is between 0 and 1, and the higher the value, the more likely the user is to respond). Alternatively, the model can also output the user's expected actual response power value at each preset time granularity (for example, every 15 minutes) of this demand-side response event, thereby forming an expected response power curve. This curve can be used for more refined grid scheduling arrangements and subsequent effect evaluation. By integrating the expected response power curve over the entire event duration, the user's expected total response power can also be obtained. In addition, the confidence level of the prediction result output by the model can be evaluated or adjusted based on the user's demand-side response participation willingness stability index value.

[0089] Furthermore, precise demand-side response scheduling and incentive strategies are generated.

[0090] Specifically, based on the specific demand-side response requirements proposed by the grid operator or load aggregator (for example, how many megawatts of total load needs to be reduced within a specific period of time in a specific area), combined with the operator's overall optimization goals (for example, pursuing the lowest incentive cost while meeting demand, or pursuing the highest response reliability, or a balance between the two), and making full use of known user behavior characteristics (especially their incentive threshold, behavioral inertia, demand-side response participation willingness stability index, and predicted response performance), through intelligent screening and optimization algorithms, an optimal user combination is selected to participate in the current demand-side response event, and differentiated and more effective incentive strategies are tailored for these selected users (or different user groups divided according to their behavioral characteristics).

[0091] Target user screening and combination optimization: can be seen as solving an optimization problem with constraints.

[0092] Optimization goals can be flexibly set based on actual operational needs and priorities. The following are some common optimization goals: The goal is to minimize the total incentive costs that need to be paid to users: the prerequisite is that the selected user group must be able to jointly meet the pre-defined total demand-side response requirements (for example, the total load reduction in a specific area reaches X megawatts), and this response must be achieved with a pre-defined reliability level (for example, there is a 90% probability that the selected user group can jointly achieve the target peak reduction amount).

[0093] The goal is to maximize the certainty or reliability of the overall expected response of the selected user portfolio: this presupposes that the selected user portfolio meets the total demand-side response requirements and that the total incentive cost expenditure does not exceed the preset budget cap. Under this objective, the system will favor users whose response behavior is more predictable and reliable. This can be achieved by comprehensively considering the user's predicted response probability and the stability index of their willingness to participate in demand-side response.

[0094] Alternatively, a multi-objective optimization approach can be employed: In practice, operators may be concerned with multiple objectives simultaneously (e.g., reducing costs while improving reliability). In this case, these objectives can be weighted and combined to form a single, comprehensive optimization objective. Alternatively, a specialized multi-objective optimization algorithm (e.g., one based on the concept of Pareto optimality) can be employed to obtain a set of Pareto optimal solutions that achieve different trade-offs between the various objectives. Decision-makers can then select a final solution based on their actual preferences.

[0095] When optimizing, a series of constraints need to be met, such as: The total demand-side response requirement must be satisfied: that is, the sum of the expected response amounts of all selected participating users (taking into account their response probabilities) must be greater than or equal to the demand-side response target value proposed by the power grid.

[0096] The total incentive cost must not exceed the budget cap constraint (if the operator sets a cost budget).

[0097] Constraints on the maximum response capacity of a single user: that is, the expected response volume of each selected user cannot exceed its own physical maximum response potential or the upper limit of its contracted response.

[0098] Screening logic and algorithm: The system preliminarily screens the candidate user pool based on the specific requirements of the current demand-side response event (such as region and user type).

[0099] For each user in the candidate user pool, the system uses its latest incentive threshold, behavioral inertia index, demand-side response participation willingness stability index, and the predicted response results of this event (response probability and expected response volume / curve) to comprehensively evaluate its "cost-effectiveness" or "comprehensive score".

[0100] Priority is given to users with a higher stability index of demand-side response participation willingness to improve the reliability and planning of the overall scheduling.

[0101] To find the optimal user combination, the following optimization algorithms can be selected: Heuristic algorithms: such as greedy algorithms, genetic algorithms, particle swarm optimization algorithms, etc., are suitable for large-scale or complex problems.

[0102] Exact optimization algorithms: Suitable for situations where the user pool is relatively small or the problem can be transformed into a standard mathematical programming problem.

[0103] Furthermore, layered or personalized incentive strategies are generated.

[0104] After determining the user mix that will ultimately participate in the demand-side response event, it is necessary to generate specific, effective, and potentially differentiated incentive strategies for these users (or different user groups divided according to their behavioral characteristics).

[0105] Dynamic division criteria for user behavior characteristic groups: Based on the combined characteristics of the three core behavioral parameters of each user, namely, the latest behavioral inertia index, the estimated incentive threshold, and the stability index of the demand-side response willingness to participate, users can be dynamically divided into different user groups with similar behavioral characteristics. The division methods include: Threshold division based on preset rules: Potential to be Unleashed: The behavioral inertia index is medium-high, the estimated incentive threshold is medium, and the demand-side response willingness stability index is medium. This type of user has responsiveness potential, but inertia needs to be overcome, and stability needs to be improved.

[0106] Cost-sensitive: Low behavioral inertia index, high estimated incentive threshold, and medium-high stability index of demand-side response willingness. Response willingness is stable, but higher incentives are required.

[0107] Risk volatility: High behavioral inertia index, medium-high incentive threshold estimate, and low demand-side response willingness stability index. High response uncertainty and high call risk.

[0108] Inertia Economy: High behavioral inertia index, low estimated incentive threshold, and medium-high stability index of demand-side response willingness. Insensitive to incentive price, but behavioral inertia is a major obstacle.

[0109] Automatic segmentation based on clustering algorithms: Use unsupervised learning algorithms such as K-Means clustering and hierarchical clustering to automatically aggregate users with similar feature combinations.

[0110] Differentiated incentive parameter generation: Based on the characteristics of each group, differentiated incentive parameters (including incentive level, method, information push focus, etc.) are automatically generated or recommended.

[0111] Rule-based approach: pre-set a set of incentive strategy rules for different groups of user behavior characteristics.

[0112] Potential to be discovered: It is recommended to provide incentives slightly higher than the estimated incentive threshold, and to combine with subsequent user behavior guidance measures to help overcome behavioral inertia.

[0113] Cost-sensitive: Recommendations provide incentive levels significantly higher than their estimated incentive threshold to induce a response, weighing costs and benefits.

[0114] Risk Volatility: Unless resources are extremely tight, avoid using this feature. If it is necessary, provide very high incentives, fully anticipate high uncertainty and potential failure risks, and strengthen real-time monitoring and behavioral guidance.

[0115] Inertia economy: The incentive level does not need to be set very high, but it must be accompanied by strong, targeted behavior change strategies and guidance measures to help overcome behavioral inertia and transform potential intentions into actual actions.

[0116] Model optimization-based methods: Methods such as reinforcement learning can also be used to dynamically optimize incentive strategies by learning historical data and system feedback.

[0117] After user screening, portfolio optimization, and the generation of differentiated incentive strategies, the system ultimately outputs a clear, specific, and executable demand-side response scheduling plan. This plan includes a list of selected participating users, specific incentive instructions for each user (including incentive level, start and end times, etc.), and an assessment of the expected overall scheduling results (e.g., estimated total response volume, total incentive costs, overall reliability level, etc.).

[0118] Further, the guidance and activation of user demand-side response behavior.

[0119] Specifically, based on the system's in-depth understanding of user behavior profiles (motivation threshold, behavioral inertia, and demand-side response participation willingness stability index), combined with current demand-side response event information, through personalized communication and long-term behavior shaping plans, we can increase user participation willingness, reduce behavioral inertia, and enhance response stability, thereby cultivating higher-quality and reliable demand-side response resources. The process is as follows: Precise push of personalized information and incentives: Push channels: Supports pushing information through multiple channels preferred by users or pre-agreed upon, such as mobile app messages, SMS, email, or voice prompts from smart home devices.

[0120] Content customization: Push information needs to be personalized based on the user's latest behavioral inertia index, incentive threshold estimate, and demand-side response participation willingness stability index to improve information effectiveness.

[0121] For users with high behavioral inertia: Provide specific action steps, simplified operation shortcuts, or encouraging words.

[0122] High incentive threshold users: Emphasize the incentive attractiveness or actual economic / non-economic benefits that can be obtained from this participation.

[0123] For users with low willingness to participate in demand-side response and stability: appropriately increase the push frequency), strengthen guidance and education, provide demand-side response knowledge and success cases, and help establish a stable response model.

[0124] High-demand side responds to users with stable participation intention: The information is more concise, focusing on gratitude and recognition, and can provide exclusive high-value participation opportunities or priority.

[0125] Push timing: Pay attention to timing, such as pre-notification before the event, reminder before the official start, immediate reminder at the beginning, and timely feedback on participation effects after the end.

[0126] Long-term behavioral guidance plan: For users who are assessed to have improvement potential but have a high behavioral inertia index or a low demand-side response participation willingness stability index, the system can plan a long-term, step-by-step "behavioral inertia reduction and willingness stability improvement plan" (i.e., "demand-side response behavior development plan").

[0127] The plan consists of different phases and objectives: Initial stage: The goal is to lower the participation threshold, help users gain initial positive success experience, and establish positive associations and initial participation habits.

[0128] Mid-term stage: After users establish initial habits, they are guided to try more types of more challenging demand-side response events, improve response capabilities, and explore sustainable and stable response models.

[0129] Later stage: When users form a stable habit of participating in demand-side response and can reliably contribute a large response volume, continuous high-quality participation is encouraged to help maximize their own benefits and the overall benefits of the power grid.

[0130] Implementation and dynamic adjustment of the plan: Based on the user behavior profile (motivation threshold, behavioral inertia, demand-side response participation willingness stability index and other characteristics), match the appropriate development path starting point and initial intervention strategy.

[0131] During the execution of the plan, user-related behavior data will be continuously monitored and the effectiveness of the plan will be dynamically evaluated.

[0132] Based on monitoring and evaluation results, the system dynamically adjusts subsequent intervention strategies. If users show significant progress, progress is accelerated; if they encounter difficulties, strategies are adjusted or additional support is provided. Gamification can be used to enhance engagement and retention.

[0133] Furthermore, the system's performance evaluation and iterative optimization comprehensively assess the operational performance of the entire demand-side response optimization system. Based on the evaluation results, the system's key models, core parameters, and strategies are continuously iteratively optimized. The goal is to achieve continuous improvement in the system's adaptive learning capabilities and overall performance. The process is as follows: Continuous evaluation of the comprehensive effects of demand-side response events: A multi-dimensional, standardized analysis and evaluation of the actual situation of each demand-side response event is required.

[0134] Response effect indicators: total response achievement rate, average response achievement rate of individual users, user response speed or delay distribution, user response continuity achievement rate, prediction model accuracy assessment, response probability prediction, response volume prediction, demand-side response target achievement degree assessment, incentive cost-benefit analysis, etc.

[0135] Correlation analysis between user behavior parameters and actual responses: Analyze the distribution of behavioral parameters such as the average demand-side response participation willingness stability index, behavioral inertia index, and incentive threshold estimation value of the actual participating users in this event, and compare them with the candidate user pool to verify the effectiveness of the screening strategy.

[0136] Analyze the actual response performance of users with different behavioral inertia index and incentive threshold estimation ranges under different incentive levels to verify the effectiveness of these behavioral parameters in explaining and predicting user behavior.

[0137] Core quantitative model performance verification and calibration: Continuously evaluate the effectiveness and accuracy of each quantitative model built in the User Behavior Dynamics Parameterization and Stability Assessment module. Regularly monitor whether there are significant changes in overall user behavior patterns (model drift). If so, retrain, adjust, or calibrate the model in a timely manner.

[0138] Effectiveness Analysis of Incentive Strategies: Compare and analyze the actual impact of different incentive strategies (tiered, differentiated, and personalized) on user groups with varying behavioral inertia, incentive thresholds, and demand-side response and participation stability indices. Conduct A / B testing on user groups with similar behavioral characteristics, comparing different strategies in terms of response rate, unit response cost, and user satisfaction. This will improve the precision and efficiency of incentives.

[0139] Adaptive optimization of model internal parameters and weight coefficients: Based on the accumulation of historical data in various dimensions and the results of continuous performance evaluation, the system needs to regularly and adaptively adjust and optimize internal key model parameters and important weight coefficients.

[0140] The system regularly uses the latest accumulated demand-side response events and user behavior data to retrain and optimize hyperparameters of the prediction model in the demand-side response behavior prediction module to adapt to the changing data distribution and user behavior patterns.

[0141] Based on the evaluation results, the weight coefficients of each sub-indicator in the behavioral inertia quantitative index system and the weight coefficients of each component in the calculation of the demand-side response participation willingness stability index are adaptively optimized.

[0142] The goal of optimizing the weights of the sub-indicators of the behavioral inertia index is to maximize the accuracy of the optimized behavioral inertia index in predicting future user response failures, or to minimize the error in the predicted response volume.

[0143] The goal of weight optimization of the components of the demand-side response participation willingness stability index is to maximize the prediction accuracy of the optimized demand-side response participation willingness stability index on the reliability and stability of users' future long-term response behavior.

[0144] Based on the evaluation results, the policy rule library (such as user behavior feature clustering thresholds, incentive adjustment coefficients) or internal parameters of related models are updated and adjusted in a timely manner.

[0145] Regularly generate system performance analysis reports, model parameter optimization and adjustment recommendations, demand-side response strategy improvement plans, and in-depth insights into user behavior characteristics. These outputs are fed back to demand-side response project operations managers to provide decision support and directly drive continuous improvement of the system's own parameters and strategies, forming a closed-loop, data-driven, and continuously learning and improving intelligent optimization system.

[0146] In a specific embodiment of the present invention, it is assumed that a power grid operator in a certain region plans to execute a demand-side response (DSR) event during the peak electricity consumption period on a summer weekday afternoon (specifically, from 2:00 p.m. to 4:00 p.m. on the same day, a total of two hours). The main goal of the DSR event is to effectively reduce the total electricity load in the target area by 5 megawatts (MW) during this period.

[0147] Step one: data collection and preprocessing.

[0148] Determination and collection of data sources: First, the system connects to the interfaces of the Advanced Metering Infrastructure (AMI) systems deployed in the target area for all users with installed smart meters. It obtains historical electricity load data recorded every 15 minutes over the past year for a total of, for example, 10,000 users in the area.

[0149] At the same time, the system connects to the demand-side response management platform to obtain records of all historical demand-side response events in which these users participated over the past year. These records include key information such as the time of occurrence and duration of each event, the type of incentive used (such as price subsidies or time-of-use electricity price adjustments), the specific incentive level or parameters, the user's declared response volume before the event, and the user's actual response volume during the event.

[0150] In addition, the system also obtained detailed profile information of these 10,000 users from the power company's marketing information system or user registration database, such as each user's unique ID, user type (for example, residential user, commercial office user, or small industrial user, etc.), and the main type of electrical equipment registered by the user (for example, whether there is high-power air conditioning, electric heating equipment, etc.).

[0151] In addition, the system also obtains hourly meteorological data of the area in the past year and the next few days by calling the API interface of an external public meteorological service provider, mainly including temperature and humidity information.

[0152] Data preprocessing operations: The system first cleans the collected historical load data for all users. This includes filling missing values ​​using linear interpolation based on adjacent valid data points. It also identifies any abnormal peaks or valleys (i.e., outliers) using the statistically-based "Three Sigma" criterion, and corrects these identified outliers using linear interpolation.

[0153] The timestamps of all time-related data (including load data, event data, meteorological data, etc.) are strictly aligned and unified to Beijing time (UTC+8 time zone), and the consistency of time granularity is ensured (for example, all are unified to 15-minute intervals).

[0154] For each customer, the system uses a preset baseline load calculation method. For example, for a demand-side response event scheduled to occur on a weekday afternoon, the baseline load calculation method is set as follows: select the customer's most recent 10 working days before the event (excluding any statutory holidays or dates on which the customer participated in other demand-side response events). From these 10 valid working days, select the five working days with the highest total power load (or the highest average load during the target period of 2:00 PM to 4:00 PM). Then, the load data for each 15-minute period between 2:00 PM and 4:00 PM for these five selected working days is averaged point by point, and the resulting average load curve is used as the customer's baseline load for this demand-side response event.

[0155] Based on the calculated baseline load of historical events and the actual load data of users in these events, the system further calculates a series of preliminary response behavior characteristic data, including the actual response amount of users in each historical demand-side response event, the response delay time from the start of the event to the user's effective response, the ratio of the user's actual response amount to its reported amount, i.e., the response completeness, and the standard deviation of the user's actual response power sequence during the response period, i.e., the response stability.

[0156] Step 2: Parameterization of user behavior dynamics and stability evaluation.

[0157] Next, for each user, the system infers incentive thresholds, quantifies behavioral inertia, and calculates a stability index for demand-side response willingness based on their preprocessed historical data. For simplicity, we use two typical users in the area as examples: User A (a medium-sized commercial office user) and User B (an ordinary residential user).

[0158] Dynamic inference of user incentive thresholds: For user A: Assume that user A participated in five documented demand-side response events in the past year. The system will analyze their specific performance in these five events. For example, in one event, when the incentive subsidy was 0.8 yuan per kilowatt-hour, user A actively responded, but in another event, when the incentive subsidy was only 0.5 yuan per kilowatt-hour, user A did not respond effectively. Combined with the prior assumptions about the incentive thresholds for the commercial office user group in the area (for example, assuming the system initially believes that the incentive thresholds for this type of user roughly obey a log-normal prior probability distribution with a mean of 0.7 yuan / kWh and a standard deviation of 0.2 yuan / kWh), the system uses the aforementioned Bayesian update mechanism (for example, using a particle filter algorithm to implement numerical calculations), and combines the five historical observation data to ultimately infer that the estimated incentive threshold for user A at the current moment is approximately 0.75 yuan / kWh, and provides a confidence interval. For example, the system is 95% confident that its true incentive threshold is between 0.65 yuan / kWh and 0.88 yuan / kWh.

[0159] For User B: Assume that User B has recently installed a smart meter and has no historical data on participating in any demand-side response events over the past year. In this case, the system will primarily determine the initial prior probability distribution of the incentive threshold based on the statistical characteristics of the average incentive threshold for the residential user group to which it belongs (for example, assuming that the system, through analysis of historical data from a large number of residential users, concludes that the incentive threshold for this group of users roughly follows a log-normal distribution with a mean of 0.4 yuan / kWh and a standard deviation of 0.15 yuan / kWh). Due to the lack of individual data, the system will assign a relatively wide confidence interval to the incentive threshold estimate, indicating a high degree of uncertainty.

[0160] Quantification of user behavior inertia: For user A: The system analyzed data from their five previous events and found that their average response delay was relatively short, their average response completeness was high, and their response stability (i.e., the standard deviation of the response power) was low. However, it also found that in one event where the incentive intensity was judged to be sufficiently high, user A did not respond effectively as expected. The system then calculated scores for multiple dimensions, including the "response consistency factor," "average response delay factor," "response incompleteness factor," and "response instability factor" (assuming these factors have the default weights of 30%, 20%, 30%, and 20% respectively when calculating the behavioral inertia index). After normalizing these scores, the final calculated behavioral inertia index for user A is 35 (assuming the behavioral inertia index ranges from 0 to 100, with lower values ​​indicating less behavioral inertia and a more proactive user).

[0161] For User B: Due to a lack of historical participation data, the system cannot calculate the various sub-factors of the Behavioral Inertia Index based on their individual behavior. In this case, the system can set their initial Behavioral Inertia Index to the average Behavioral Inertia Index value for that type of resident user, or to a neutral default value (e.g., 50), and simultaneously mark the Behavioral Inertia Index as having low confidence.

[0162] Calculation of the user demand-side response participation willingness stability index (assuming the stability assessment period is set to the user's most recent three demand-side response events, or behavioral data within the last three months): For user A, the system examined the series of estimated incentive thresholds for the three most recent evaluations (or within the last three months) (for example, the estimated incentive thresholds were 0.70, 0.72, and 0.75 RMB / kWh, respectively) and found relatively small fluctuations, indicating a degree of stability. Furthermore, the series of behavioral inertia indexes (for example, 30, 32, and 35) also showed relative stability. Further trend analysis of the time series of estimated incentive thresholds revealed predictable patterns (e.g., high R-squared values). Combining these three scores (behavioral inertia stability score, incentive threshold stability score, and incentive threshold trend predictability score), the system ultimately calculated user A's current demand-side response participation willingness stability index as 78 (assuming this index also ranges from 0 to 100, with higher values ​​indicating more stable and predictable user participation willingness and behavior patterns).

[0163] For user B: Due to insufficient historical data, it is impossible to form an effective time series for stability analysis. Therefore, the system is temporarily unable to accurately calculate its demand-side response participation willingness stability index, or it will mark it as a lowest-level default value and indicate that the confidence level of its evaluation result is extremely low.

[0164] Step 3: Prediction of user demand-side response behavior.

[0165] Obtaining information about the target demand-side response event: The system learns that this demand-side response event is scheduled for summer between 2:00 PM and 4:00 PM, with a fixed subsidy of 0.9 yuan per kilowatt-hour of electricity provided. The event will last for two hours. The system also obtains the average temperature forecast for the area between 2:00 PM and 4:00 PM that day, which is 32 degrees Celsius.

[0166] Calling a trained prediction model: Assume that the system has pre-trained a gradient boosting decision tree (GBDT) model using a large amount of historical demand-side response data to predict user response behavior.

[0167] Prepare input features for the user (taking user A as an example): The system will prepare a feature vector for user A to be input into the GBDT model. The feature vector will include: the incentive intensity of the target event (0.9 yuan / kWh), the estimated value of the current incentive threshold of user A (0.75 yuan / kWh), the current behavioral inertia index of user A (35), the current demand-side response participation willingness stability index of user A (78), the incentive margin calculated based on the incentive intensity and incentive threshold (for example, the incentive intensity is divided by the estimated value of the incentive threshold to get 1.2), and other related auxiliary features, such as the historical load characteristics of user A (for example, its average power consumption from 2 to 4 pm on recent weekdays), the current predicted temperature (32 degrees Celsius), the mark that the day is a weekday afternoon, etc.

[0168] Obtaining the prediction output (using user A as an example): User A's feature vector is input into the trained GBDT model. The model then calculates and outputs a prediction of user A's response behavior during this event. For example, the model predicts that user A has an 85% probability of making an effective response. Furthermore, the model predicts that user A will contribute an average of 20 kilowatts of power over the two hours of the event, for a total of 40 kilowatt-hours of power during the entire event. (The system performs similar response predictions for all other potential participants in the area, including user B.) Step 4: Generation of precise demand-side response scheduling and incentive strategies.

[0169] Screening and combination optimization of target users: The optimization goal set by the system is: to minimize the total incentive cost to be paid to these users while ensuring that the user combination finally selected can jointly meet the demand for reducing the total load by 5 megawatts, and when selecting users, priority will be given to those users with a higher demand-side response participation willingness stability index.

[0170] The system will use a greedy algorithm, such as one weighted by the demand-side response willingness stability index, to comprehensively sort all users and select users from them based on the expected response volume of all potential participating users, the predicted response probability, the demand-side response participation willingness stability index, and the incentive cost required for unit response volume (in this case, the incentive cost is fixed at 0.9 yuan per kilowatt-hour, so the unit response cost mainly depends on the user's expected unit response power).

[0171] For example, User A, due to a high stability index for demand-side response willingness and an ideal predicted response probability and unit response cost (relative to the expected response volume), is ranked higher and is selected by the system to participate in this event. User B, on the other hand, may not be selected or be given a lower priority due to insufficient historical data leading to high uncertainty in the prediction results or a high predicted unit response cost.

[0172] After the optimization algorithm calculates, the system ultimately selects, for example, 300 users to participate in this demand-side response event. The system estimates that these 300 users can collectively contribute approximately 5.1 MW of response capacity (slightly exceeding the target of 5 MW, with some margin), and calculates the total incentive cost to be approximately XX yuan.

[0173] Generate a personalized incentive strategy (in this embodiment, since the incentive standard is fixed at 0.9 yuan per kilowatt-hour, "personalization" is mainly reflected in the subsequent information push level. However, if the incentive standard itself is also optimizable, different incentive prices can be set for different users or groups): For example, for selected "high-quality and stable" users like User A, when the system sends incentive notifications later, in addition to informing the standard incentive information, it can also add an additional sentence of thanks for their consistent and stable support.

[0174] If the system selects some users who are assessed as "potential to be discovered" to participate while meeting the total demand, then the information pushed to them subsequently, in addition to incentive information, can also include some more specific energy-saving suggestions or response operation guidelines tailored to their energy usage characteristics.

[0175] Step 5: Guiding and activating user demand-side response behavior.

[0176] For example, if user B, despite having potential but not being prioritized due to insufficient data, is included in a "demand-side response behavior development plan" after this event, the system can proactively push energy-saving tips to user B or invite them to participate in smaller, more accessible demand-side response experiences (for example, "Participate in the 'Turn Up the AC One Degree' experience this weekend and receive small points.") The system also closely monitors changes in user behavior and parameter updates.

[0177] Step 6: Evaluation and iterative optimization of system performance.

[0178] After the demand-side response event actually ends, the system collects the actual power load data of all participating users from the AMI system. For example, it is found that user A actually responded to an average power of 18 kilowatts during the event.

[0179] The overall effectiveness of this event will be evaluated (for example, whether the actual total response volume met the target, the accuracy of the prediction, etc.), and the errors in the prediction model will be analyzed. Furthermore, the system will use User A's actual response data to update behavioral parameters such as their incentive threshold, behavioral inertia, and the demand-side response participation willingness stability index. (For example, if the response result was slightly lower than expected, the system may slightly increase their incentive threshold. However, if their response process was stable, their behavioral inertia index may be slightly reduced due to their good participation. Their demand-side response participation willingness stability index will also be recalculated based on the new incentive threshold and the stability of the behavioral inertia parameters.)

[0180] If long-term data analysis of a large number of users reveals that, among the various sub-factors that make up the behavioral inertia index, the correlation between a certain sub-factor (such as the "response instability factor") and the user's actual response failure has undergone a systematic change (for example, becoming stronger or weaker), then the system may adjust the weight coefficient of this sub-factor in calculating the comprehensive behavioral inertia index accordingly during the next iterative optimization.

[0181] Those skilled in the art will appreciate that the order of the above steps is not absolutely fixed. The order of certain steps may be adjusted, or certain steps may be combined or further subdivided, without affecting the core technical effect. Furthermore, the methods and systems of the present invention may also incorporate other auxiliary processing steps, such as confidence assessment of the final positioning result, historical data tracking and analysis, etc., depending on the needs of actual application scenarios.

[0182] The foregoing description is merely a specific embodiment of the present invention, and the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be readily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A demand-side response monitoring method based on multi-dimensional indicator monitoring, characterized in that: The method comprises: Collecting and preprocessing multi-source heterogeneous demand-side response data, wherein the multi-source heterogeneous demand-side response data includes user historical load data, demand-side response event data, and user profile data; Based on the pre-processed multi-source heterogeneous demand-side response data, dynamically inferring user incentive thresholds, quantifying user behavioral inertia, and calculating a user demand-side response participation willingness stability index; Predicting the user's response behavior in the target demand-side response event based on the user incentive threshold, user behavior inertia, user demand-side response participation willingness stability index, and target demand-side response event information; According to the predicted user response behavior and demand-side response requirements, target users are screened and a demand-side response scheduling plan and incentive strategy for the target users are generated.

2. A demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The dynamic inference of user incentive threshold includes: for the user, obtaining the incentive intensity of its historical demand-side response event from the pre-processed demand-side response event data, and determining its actual response result in the historical demand-side response event from the pre-processed user historical load data, using the Bayesian update mechanism to update the posterior probability distribution of the user incentive threshold, and outputting the user's current incentive threshold estimate.

3. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The quantification of user behavioral inertia includes: constructing a behavioral inertia quantification index system, the index system including at least one of a response consistency factor based on the user's historical response frequency and response pattern stability, an average response delay factor of the user's historical response delay, a response incompleteness factor inversely correlated with the user's historical response completeness, and a response instability factor based on the variability of the user's historical response power sequence; and calculating and outputting the user's current behavioral inertia index based on the index system.

4. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The calculation of the user demand-side response participation willingness stability index includes: calculating a behavioral inertia stability score based on the user's behavioral inertia index time series within a preset evaluation period; calculating an incentive threshold stability score based on the user's incentive threshold estimation value time series within the preset evaluation period; performing a time series trend analysis based on the user's incentive threshold estimation value time series within the preset evaluation period to evaluate the predictability of the incentive threshold trend, and calculating the incentive threshold trend predictability score accordingly; and calculating and outputting the user's current demand-side response participation willingness stability index by combining the behavioral inertia stability score, the incentive threshold stability score, and the incentive threshold trend predictability score.

5. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The collection and preprocessing of multi-source heterogeneous demand-side response data also includes: obtaining user electricity power series data through the smart metering infrastructure interface; obtaining detailed information of demand-side response events through the demand-side response management platform or the power market operation system interface, the detailed information including incentive type and incentive parameters; obtaining user basic information from the power marketing system or the user registration database interface; and performing data cleaning, timestamp alignment, user baseline load calculation before the demand-side response event, and actual response amount calculation of the user during the demand-side response event on various types of collected data.

6. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The step of predicting the user's response behavior in the target demand-side response event based on the user incentive threshold, user behavioral inertia, user demand-side response participation willingness stability index and target demand-side response event information further includes: constructing a supervised learning prediction model, the input feature set of the prediction model includes the incentive intensity of the target demand-side response event, the user's incentive threshold estimate, the behavioral inertia index, the demand-side response participation willingness stability index, and the incentive margin composed of the ratio of the incentive intensity to the user's incentive threshold estimate.

7. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The step of screening target users and generating a demand-side response scheduling plan and incentive strategy for the target users based on the predicted user response behavior and demand-side response requirements further includes: setting the optimization goal to minimize the total incentive cost expenditure or maximize the certainty of the expected response under the premise of satisfying the demand-side response demand and the preset response reliability level constraints; and, when screening target users, giving priority to users with a higher demand-side response participation willingness stability index to improve the overall reliability and planning of demand-side response scheduling.

8. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The generation of a demand-side response scheduling plan and incentive strategy for the target user further includes: dynamically dividing the user into different behavioral characteristic groups based on the user's behavioral inertia index, incentive threshold estimation value and demand-side response participation willingness stability index characteristics; and automatically generating or recommending differentiated incentive parameters for the different behavioral characteristic groups.

9. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The method also includes: pushing personalized demand-side response event notifications and incentive information to target users based on their behavioral inertia index, incentive threshold estimation value, and demand-side response participation willingness stability index characteristics, and planning a long-term behavior guidance plan for specific users aimed at reducing their behavioral inertia and improving the stability of their participation willingness.

10. The demand-side response monitoring method based on multi-dimensional indicator monitoring according to claim 1, characterized in that: The method also includes: continuously evaluating the comprehensive effects of demand-side response events, verifying the explanatory power of the behavioral inertia and incentive threshold parameters on user response behavior, and verifying the correlation between the demand-side response participation willingness stability index and the user's actual response stability and reliability; and based on the evaluation results, adaptively optimizing the weight coefficients of each sub-indicator in the behavioral inertia quantitative index system and the weight coefficients of each component in the calculation of the demand-side response participation willingness stability index.

Images