Power supply management method, system and equipment based on household energy storage power supply system

Abstract

The invention relates to the field of energy storage power supply management, in particular to a power supply management method, system and equipment based on a household energy storage power supply system. The method comprises the following steps: collecting a household power utilization monitoring data stream, carrying out equipment use behavior preference mining and multi-scale state transition deep learning, and generating a user equipment behavior prediction model; mining power utilization behaviors of power utilization terminals one by one based on the household power utilization monitoring data flow, performing multi-time-point power utilization demand prediction based on a user equipment behavior prediction model, and constructing a multi-time-point power utilization demand prediction map of the user; and obtaining meteorological environment parameters of a current user family area, performing climate disturbance demand adjustment on the power demand prediction maps at the multiple time points based on the meteorological environment parameters, and constructing a climate disturbance power demand prediction map. According to the invention, through dynamic user demand changes, the energy efficiency management level and the energy allocation precision of long-term operation of the energy storage system are improved.

Description

Technical Field

[0001] The present invention relates to the field of energy storage power supply management, and in particular to a power supply management method, system and device based on a household energy storage power supply system. Background Art

[0002] With the continuous optimization of energy structures and the growing popularity of green and low-carbon concepts, home energy storage power systems are gaining widespread adoption as a key technology for achieving distributed energy utilization and improving household energy autonomy. Especially with the rapid development of distributed generation systems based on renewable energy sources such as solar and wind power, home energy storage systems not only effectively improve energy efficiency but also play a key role in peak load shifting, emergency power supply, and electricity cost optimization. However, with the increasing number of home energy storage devices and the increasing complexity of their systems, achieving efficient, intelligent, and secure power management has become a major challenge facing current home energy systems.

[0003] Traditional power management methods for home energy storage systems mostly employ static or rule-based control strategies, relying on preset thresholds and fixed parameters for energy allocation and switching. While these approaches may meet basic power management needs in the early stages, they often lack flexibility and intelligence when faced with variable loads, complex electricity pricing mechanisms, and dynamic renewable energy output. Furthermore, existing power management systems generally lack deep awareness of the energy storage system's operating status, making it difficult to accurately assess and respond in real time to key factors such as battery health, power fluctuations, and grid status changes. This can easily lead to problems such as battery overcharge and overdischarge, energy waste, and power outages, potentially jeopardizing the stability and safety of home electricity use. With the development of emerging technologies such as artificial intelligence, big data analytics, and the Internet of Things, power management for home energy storage systems is gradually evolving towards intelligent and automated approaches. Traditional methods, in particular, are unable to meet the demands of modern smart home systems in areas such as multi-energy coordination, dynamic scheduling optimization, load forecasting, and anomaly identification. Therefore, there is an urgent need for a power management approach with real-time, highly adaptable, and multi-objective optimization capabilities to improve the overall operational efficiency and safety of energy storage systems. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a power management method, system and device based on a home energy storage power system to solve at least one of the above technical problems.

[0005] To achieve the above objectives, the present invention provides a power management method based on a household energy storage power supply system, comprising the following steps: Step S1: Collect household electricity monitoring data streams, conduct device usage behavior preference mining and multi-scale state transition deep learning, and generate a user device behavior prediction model; Step S2: mining the electricity consumption behavior of each electricity terminal based on the household electricity consumption monitoring data stream, and performing multi-time point electricity demand prediction based on the user device behavior prediction model to construct a user's multi-time point electricity demand prediction map; Step S3: obtaining meteorological environment parameters of the current user's home area, and adjusting the electricity demand forecast graphs for the multiple time points based on the meteorological environment parameters to construct a climate disturbance electricity demand forecast graph; Step S4: Analyze the electricity demand forecast map for climate disturbances by electricity consumption period, perform adaptive energy storage capacity analysis, and construct an adaptive power supply energy storage strategy; Step S5: pre-allocate energy storage capacity based on the adaptive power storage strategy, perform multi-objective coordinated control, and build an intelligent energy storage power control engine; Step S6: Perform online optimization of strategy parameters and iterative decision optimization on the intelligent energy storage power supply control engine to build a power supply iterative optimization management model.

[0006] In the present invention, a power management system based on a home energy storage power supply system is provided, which is used to execute the power management method based on the home energy storage power supply system as described above, comprising: The behavior prediction module is used to collect household electricity monitoring data streams, conduct device usage behavior preference mining and multi-scale state transition deep learning, and generate user device behavior prediction models; An electricity demand forecasting module is used to mine the electricity consumption behavior of each electricity terminal based on the household electricity consumption monitoring data stream, and to forecast electricity demand at multiple time points based on the user device behavior prediction model, thereby constructing a user's electricity demand forecast graph at multiple time points; A climate disturbance analysis module is used to obtain meteorological environment parameters of the current user's home area, adjust the electricity demand forecast graphs for the multiple time points based on the meteorological environment parameters, and construct a climate disturbance electricity demand forecast graph; Adaptive energy storage module, used to analyze electricity demand forecasts for climate disturbances, analyze adaptive energy storage capacity, and build adaptive power storage strategies; The capacity pre-allocation module is used to pre-allocate energy storage capacity based on the adaptive power storage strategy, perform multi-objective coordinated control, and build an intelligent energy storage power control engine; The iterative optimization module is used to perform online tuning of strategy parameters and iterative decision optimization on the intelligent energy storage power supply control engine, and to build an iterative optimization management model for the power supply.

[0007] The present invention further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the power management method based on the home energy storage power system described above are implemented.

[0008] The present invention achieves the following beneficial effects: accurately modeling the long-term and short-term electricity usage habits of multiple household electrical devices; introducing multi-scale state transfer learning to capture electricity usage behavior switching patterns at different time granularities, improving the dynamic responsiveness of behavior prediction; constructing a user device behavior prediction model through deep learning mechanisms with adaptive learning capabilities, automatically adjusting the model structure as user habits change; laying a highly personalized, dynamically updated model foundation for electricity demand forecasting at subsequent time points, enhancing the behavioral consistency and actual accuracy of the predictions; independently modeling and behavioral decomposition of different household electrical devices (such as refrigerators, air conditioners, and water heaters) for more accurate predictions; utilizing the behavior prediction model to predict electricity usage at multiple time points, enabling early identification of peak load and conflicting periods; and constructing a highly visualized electricity demand forecast map, facilitating subsequent policy module strategy allocation for peak load and peak control. This enables the transformation of household energy storage systems from "responsive management" to "prediction-driven scheduling," significantly improving the forward-looking nature of system scheduling. The system effectively integrates the impact of meteorological parameters (such as temperature, humidity, and light intensity) on household loads (especially air conditioning and heating) to achieve environmentally driven forecast corrections. Adjustments to the original electricity demand map under climate disturbances enhance the environmental sensitivity of the forecast map and its consistency with reality. The climate disturbance mechanism improves the system's ability to regulate and control extreme weather changes (such as cold waves and heat waves). The system's ability to respond to future uncontrollable environmental factors is significantly enhanced, ensuring the stability and adaptability of the power management strategy. Based on the adjusted forecast map, periods of high electricity consumption and low electricity consumption can be accurately identified, providing the optimal time window for energy storage strategy formulation. Adaptive energy storage capacity analysis can dynamically adjust energy storage release plans based on predicted load trends to avoid overcharging or energy waste. Energy storage strategies are no longer based on fixed threshold settings, but have the ability to adapt to the load structure of the time period, improving system economy and stability. This strategy provides a decision-making basis for "on-demand power supply" and "load peak shaving and valley filling" in actual operation. Achieve coordinated control of the energy storage system under multiple constraints such as load demand, energy storage life, and electricity price fluctuations; the energy storage capacity pre-allocation mechanism ensures the reliability of power supply for high-priority behaviors and high-impact loads; the constructed control engine can dynamically adjust the energy allocation method under different load scenarios to improve the overall scheduling accuracy; strengthen the "intelligent response" capability of the energy storage system, so that it has strategic flexibility and task priority configuration mechanism in complex situations. Introduce an online tuning mechanism so that the system can continuously learn changes in user behavior and environmental disturbances, and achieve "optimization while running"; the iterative optimization mechanism realizes self-evolution of strategies based on actual operation data feedback, reducing scheduling deviation and failure probability; the constructed power supply iterative optimization management model can continuously enhance the system's adaptability in different load structures and user demand change scenarios; improve the energy efficiency management level of the energy storage system in long-term operation, extend battery life and optimize energy utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a schematic flow chart of the steps of a power management method based on a household energy storage power system according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Schematic diagram of the detailed implementation steps of step S2. DETAILED DESCRIPTION

[0010] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0011] This application provides a power management method, system, and device based on a home energy storage power system. The execution entities of the power management method, system, and device based on a home energy storage power system include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc. equipped with the system, which can be regarded as general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0012] See also Figures 1 to 3 The present invention provides a power management method based on a household energy storage power supply system, comprising the following steps: Step S1: Collect household electricity monitoring data streams, conduct device usage behavior preference mining and multi-scale state transition deep learning, and generate a user device behavior prediction model; Step S2: mining the electricity consumption behavior of each electricity terminal based on the household electricity consumption monitoring data stream, and performing multi-time point electricity demand prediction based on the user device behavior prediction model to construct a user's multi-time point electricity demand prediction map; Step S3: obtaining meteorological environment parameters of the current user's home area, and adjusting the electricity demand forecast graphs for the multiple time points based on the meteorological environment parameters to construct a climate disturbance electricity demand forecast graph; Step S4: Analyze the electricity demand forecast map for climate disturbances by electricity consumption period, perform adaptive energy storage capacity analysis, and construct an adaptive power supply energy storage strategy; Step S5: pre-allocate energy storage capacity based on the adaptive power storage strategy, perform multi-objective coordinated control, and build an intelligent energy storage power control engine; Step S6: Perform online optimization of strategy parameters and iterative decision optimization on the intelligent energy storage power supply control engine to build a power supply iterative optimization management model.

[0013] In the embodiment of the present invention, see Figure 1, is a schematic flow chart of the steps of a power management method based on a household energy storage power supply system of the present invention. In this example, the steps of the power management method based on a household energy storage power supply system include: Step S1: Collect household electricity monitoring data streams, conduct device usage behavior preference mining and multi-scale state transition deep learning, and generate a user device behavior prediction model; In this embodiment, after obtaining user authorization, multi-dimensional electricity usage data is continuously collected from smart meters and branch monitoring devices (such as smart sockets and smart circuit breakers) installed in households, generating a high-precision electricity usage data stream. The collected data should include each appliance's operating status (on / off), power consumption (in watts), continuous operating time, operating hours, and power usage frequency. A sampling period of 15 or 30 seconds is recommended to ensure that the start / stop status and power fluctuation behavior of most small and medium-sized household appliances are captured. In the experimental setup, a 30-second collection period allowed for 2,880 records to be collected per day, corresponding to the power variation trajectory of each device over a 24-hour period. After being uploaded to the local edge gateway device, the data is encrypted and cached and regularly synchronized to a cloud database. The data set should retain fields such as timestamps, device unique identifiers, measurement values ​​(power, voltage, current, etc.), and status flags. This data serves as the raw input for behavioral pattern recognition and model training, directly impacting the accuracy of the predictive model. After collecting a sufficient amount of electricity usage data (e.g., a continuous collection period exceeding two weeks, containing at least one million data points), the data needs to be preprocessed and analyzed for behavioral preferences. By analyzing information such as device operation time patterns, frequency characteristics, and daily startup time distribution, the system can construct user preference models for various devices. By statistically analyzing the probability of air conditioners starting up at different times, it can be found that users frequently start their devices between 6:00 PM and 10:00 PM on summer weekdays, which the system labels as "nighttime high-frequency load." Washing machines, on the other hand, may be concentrated during the daytime on weekends, thus being labeled as "weekend periodic load." Clustering algorithms (such as K-Means and DBSCAN) can be used at this stage to categorize behavioral patterns and map device usage time series data into multiple typical behavior clusters. Hidden Markov models (HMMs) are also used to identify transition probabilities between device usage states. For example, the probability of an air conditioner transitioning from standby to running is 0.78, and the average duration of a fan switching from running to off is 2.5 hours. This type of behavioral mining generates a set of user preference feature labels, providing semantic support for the input of subsequent state prediction models. The system can use "Device A has a 0.86 probability of starting up at 18:00 on Wednesday" as one of the model input features. After extracting behavioral preference features, the system needs to establish a deep prediction model with multi-timescale recognition capabilities to accurately predict future device usage probabilities and power loads. This model serves as the foundational prediction engine for energy storage scheduling and power optimization. In terms of model architecture design, a hybrid model combining convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) is recommended. CNNs are responsible for extracting local features from high-frequency time series, such as microscopic patterns in a device's daily power usage cycle; while LSTMs can handle dependencies in long time series, capturing the periodic startup behavior of devices within weeks and months. Model inputs include multi-scale historical power series from the past 24 hours to 7 days, user preference labels, and temporal context information (such as holiday markers, weather conditions, and temperature).The output is a predicted probability of whether a device will be powered on within a certain time period in the future, along with the corresponding power level prediction (e.g., predicting that a device will operate at an average power of 120W for 45 minutes at 7:30 PM on Thursday). Data augmentation is recommended to improve model robustness, such as time series perturbation of historical data and interpolation of power sampling points. Training data should be divided into a training set (70%), a validation set (15%), and a test set (15%). Classification accuracy can be evaluated using the cross-entropy loss function, and the accuracy of the regressed power can be evaluated using mean squared error. Through repeated training and model iteration, a behavior prediction model with device-level predictive capabilities is ultimately developed. The system can predict electricity usage trends for the next 30 minutes to 2 hours at any given time point, providing predictive support for subsequent energy storage scheduling. After model training, the prediction model should be deployed in a lightweight manner on a home energy gateway (such as an edge computing box) or the embedded chip in a smart inverter to achieve real-time inference. Model compression, such as parameter pruning and quantization conversion, can be performed before deployment to ensure efficiency on resource-constrained devices. After deployment, the system automatically calls the behavior prediction module every 30 minutes or hour to predict the startup probability and power demand of various major load devices in the future period. It then dynamically adjusts the energy storage system's charging and discharging strategies based on the current energy storage status and electricity price curve. For example, if it predicts that the air conditioner, water heater, and rice cooker will all turn on simultaneously at 7:30 PM, the system can enter charging mode at 6:30 PM to ensure that power is sufficient to cover the peak load. This prediction module serves as a source of prior knowledge for the intelligent energy storage system's control strategy, and its accuracy directly impacts the economic efficiency and stability of power scheduling. Experimental data shows that after deploying the prediction model, the system's prediction accuracy for device startup times has increased by an average of approximately 12.4%, and its overall peak power handling capability has improved by over 15%.

[0014] Step S2: mining the electricity consumption behavior of each electricity terminal based on the household electricity consumption monitoring data stream, and performing multi-time point electricity demand prediction based on the user device behavior prediction model to construct a user's multi-time point electricity demand prediction map; In this embodiment, the raw household electricity monitoring data stream is segmented by device to create independent operating trajectories for each appliance type. This data, sourced from a deployed branch-based intelligent energy metering system, can accurately distinguish the individual power usage curves of power users such as air conditioners, water heaters, washing machines, refrigerators, and rice cookers. The sampling period remains between 30 seconds and 1 minute. The system segments the continuous data stream by device number, generating multiple subsets, each dedicated to a single device. During this processing, device start / stop characteristics, power thresholds, and duration rules are combined to identify false triggers and spurious power readings. Power fluctuations less than 10W but lasting less than 2 minutes are identified as "false starts and stops" and are eliminated. This process also involves the use of a sliding window mechanism (e.g., a 1-hour window with a 30-minute step size) to generate a full-day (48-window) behavioral trajectory sequence for each device. This processing approach makes it easier for subsequent models to capture behavioral patterns and state transition trends between time periods. After segmenting the power usage data stream at the device level, the system uses the user device behavior prediction model trained in step S1 to generate power forecasts for each terminal device at multiple future time points. This model is essentially a deep time series prediction architecture based on multi-scale temporal features and behavioral preference labels, supporting power demand forecasts from minute to hour levels. In practical applications, prediction time points can be categorized by demand into short-term (15-minute forecasts), medium-term (1-hour forecasts), and long-term (6-hour or full-day forecasts). For short-term predictions, for example, the system generates a rolling 5-minute forecast of the device's power-on probability and power demand for the next 30 minutes. Long-term predictions, based on daily power usage patterns, predict high and low load periods throughout the day. Input features include the historical 24-hour power series, device type, operating status, user behavior preference labels (e.g., "high frequency use during weekday nights"), weather information (e.g., external temperature and humidity), and date type (e.g., holiday markings). The output is the power value and operating probability for each predicted time point. In experimental tests, the predicted power consumption of a water heater at 5:30 PM in the evening of winter was 1800W, with a predicted startup probability of 93%. The predicted power consumption of a washing machine at 12:00 PM on Sunday was 500W, with a predicted probability of 68%. Through batch parallel prediction, the system quickly obtains the power demand trend of each device over the next period of time. After completing individual predictions for all terminal devices, the system aggregates and integrates all predicted data along the time dimension to generate a comprehensive multi-point electricity demand forecast for each user. This forecast is presented as a time series, with each time point representing the total predicted power demand value for the user's household for a specific period in the future. During the integration process, the risk of concurrent startup and operational overlap between devices should be fully considered. To improve the real-world adaptability of the forecast, the system introduces a load concurrency probability matrix to estimate the likelihood of linkage between devices.A water heater and rice cooker may run simultaneously during dinner time, and their combined power peak may be as high as 3000W or more; while a refrigerator is a continuous load and should maintain a baseline power contribution (such as an average of 120W) throughout the entire period.

[0015] After the integration is complete, the forecast graph should show the following characteristics: There are obvious intraday cyclical trends (e.g., three load peaks at breakfast, lunch, and dinner); The equipment startup concentration area corresponds to a significant power surge; At night (midnight – 5:00 AM), power levels show a low-load plateau. Based on this chart, the system also generates auxiliary analytical charts, such as daily power peak forecasts, peak-to-valley differences, and power consumption fluctuation analysis, providing data support for energy storage system scheduling decisions. The resulting multi-point user power demand forecast not only demonstrates the data but also serves as a key input for intelligent control decisions made by the energy storage system. This information is fed into the energy storage scheduling module as feedforward information, which dynamically adjusts the energy storage system's charging and discharging strategies based on factors such as the current remaining battery charge, electricity price signals, and user preferences. If peak power is predicted between 6:00 PM and 8:30 PM, the system can initiate battery charging mode at 4:30 PM to ensure sufficient reserve power during peak periods. If the load reaches a low point after 8:30 PM, the system can terminate discharge and switch to backup mode to minimize energy loss. In addition, the prediction chart will also be used to provide interactive feedback to users. For example, through the home energy management app, users will be prompted that "the electricity load at 19:00 is too high, and it is recommended to use the electric water heater during off-peak hours", thereby improving users' energy awareness and system coordination capabilities.

[0016] Step S3: obtaining meteorological environment parameters of the current user's home area, and adjusting the electricity demand forecast graphs for the multiple time points based on the meteorological environment parameters to construct a climate disturbance electricity demand forecast graph; In this embodiment, high-frequency meteorological data of the region is collected and a climate parameter matrix highly correlated with electricity consumption behavior is constructed. The collection method can be through accessing the National Meteorological Observatory or third-party meteorological APIs (such as Hefeng Weather and OpenWeather), and obtains data including but not limited to the following: Current and next 24-hour outdoor temperature (°C); Relative humidity (%); Light intensity (Lux or W / m²); wind speed and direction; precipitation probability and precipitation amount; Other influencing variables such as air pressure, cloud cover, and sunshine duration; A sampling period of 1 hour is recommended to ensure synchronization between meteorological parameters and the time granularity of electricity demand forecasts. During data processing, it is necessary to analyze the correlation between historical meteorological and electricity demand data and eliminate climate variables that have a minimal impact on household electricity loads. Wind speed has a minimal impact on indoor loads and can be treated as a weak indicator, while temperature, humidity, and precipitation require more focused modeling. After obtaining meteorological data, these environmental variables need to be integrated with the original electricity demand forecast to construct a climate perturbation adjustment model. The goal is to identify the "perturbation effects" of different climate conditions on the operating behavior of various electrical devices and adjust the forecasts accordingly. The specific approach is to cross-model historical device operating data with meteorological variables. Using methods such as multivariate regression, random forest regression, or neural network mapping, the model learns how different devices operate under different meteorological conditions. For every 1°C drop in temperature, air conditioner operating time increases by approximately 12%, while when humidity exceeds 80%, the probability of dehumidifier activation increases by 23%. This type of climate perturbation model should be developed specifically for key load devices, such as air conditioners, water heaters, and electric heaters, which are climate-sensitive. In addition, by introducing a causal reasoning network based on a graph structure, the potential impact of meteorological conditions on the state transition path of the equipment can be identified, further improving the accuracy of disturbance modeling. Through the trained disturbance model, meteorological variables are used as input to dynamically adjust the original equipment power forecast sequence to generate a climate disturbance correction factor that fits the current environment. The original predicted power of the air conditioner at 18:00 was 1200W. In a high temperature and low humidity environment of 36°C, it was adjusted to 1450W to reflect the trend of more frequent start and stop of the air conditioner. After obtaining the climate disturbance correction value of each device, the system superimposes these disturbance items on the original power demand forecast map, and re-aggregates the predicted power value at each time point according to the time dimension to generate a climate-disturbed load forecast map. This map is a high-level result of "meteorological overlay" of the traditional behavior-based model prediction results, and has the following key features: It has stronger peak sensitivity in extreme weather conditions (such as high temperatures, cold waves, heavy rains, etc.); Power variation trends showing dynamic device-level response; It can respond to the induced effects of climate change on user behavior in real time.

[0017] Experimental verification, by comparing actual electricity consumption data, showed that the introduction of the climate perturbation prediction model reduced the forecast error for air conditioning loads by 18% and the overall peak load error by 12.6%. In particular, under scenarios with high temperatures (>35°C) and sudden increases in humidity (>90%), the perturbation prediction map demonstrated significantly better dynamic response than the static behavior model's prediction map.

[0018] Step S4: Analyze the electricity demand forecast map for climate disturbances by electricity consumption period, perform adaptive energy storage capacity analysis, and construct an adaptive power supply energy storage strategy; This embodiment uses time series statistics and load peak identification techniques to analyze power consumption trends over the next 24 hours, identifying potential high-load periods, periods of frequent load fluctuations, and periods of low load. Behavioral labels are then assigned to these different time periods. The method uses a sliding window approach (recommended window length: 60 minutes, step length: 15 minutes) to scan the forecast graph. Within each window, statistics are collected for: average power; peak-to-valley difference; sudden increase (i.e., instantaneous power jump); and duration of sustained peak power (duration exceeding 3000W). Combining these characteristic parameters, clustering algorithms (such as K-means and DBSCAN) are used to divide the 24-hour period into four categories: peak, sub-peak, stable, and low-peak. In experiments, for a sample Shanghai user, the system identified the peak period as 6:00 PM–9:00 PM, with an average predicted power of 3.7 kW and a maximum instantaneous peak of 4.2 kW. The low-peak period, on the other hand, is from 00:00 AM–05:00 AM, with an average power of approximately 0.6 kW. After completing the time period analysis, the system then performs a flexible matching analysis of the energy storage system's charge and discharge capacity based on the load characteristics of each period. The core of this analysis is to dynamically plan the optimal energy storage capacity ratio for each period based on the current battery capacity, battery charge and discharge efficiency, expected energy demand, and estimated photovoltaic energy availability. The method first calculates the cumulative energy demand (in kWh) for each period. This is then combined with the specifications of the current household energy storage system (e.g., battery capacity of 10 kWh, charging efficiency of 92%, and discharge efficiency of 90%) to derive the minimum guaranteed energy storage capacity for that period. A "safety buffer" factor (e.g., 20%) is added to mitigate unexpected load surges or forecast errors. For the aforementioned peak period of 6:00 PM–9:00 PM, the cumulative energy demand is predicted to be 10.2 kWh. After deducting the 2 kWh that the grid may support, the energy storage system must cover approximately 8.2 kWh. Multiplying this by the inverse of the discharge efficiency (approximately 1.11), the actual discharge demand is calculated to be 9.1 kWh. Considering a 20% safety margin, the recommended peak energy storage capacity is approximately 10.9 kWh, which is close to the system's upper limit. Energy storage strategies cannot be statically formulated based solely on current load demand; they should dynamically adjust target energy storage capacity based on long-term user behavior patterns and seasonal variations. The system then incorporates the user device behavior prediction model extracted in step S1, along with historical periodic load data (such as weekday / weekend differences and winter / summer electricity usage differences), to perform a long-term behavioral overlay analysis. A household may frequently use an electric heater in winter, increasing peak load demand, while in summer, total electricity consumption may decrease due to daytime outings. The system adjusts the current energy storage strategy target based on the energy storage usage trajectory and actual load response curve for similar dates over a one-month period using a sliding average method and a weighted correction factor (e.g., α = 0.7 for recent performance and 0.3 for historical performance).After completing time period analysis and capacity matching, the system generates an adaptive power storage strategy table based on the strategy matrix. This strategy table covers the following key elements: target energy storage capacity for each time period; charging and discharging priorities and restrictions (such as "no discharge at night"); dynamic dispatch coefficients (such as increasing safety margins when forecast errors are high); coupling plans with the photovoltaic system (such as prioritizing charging between 2:00 PM and 4:00 PM when expected sunlight is high); and user interaction feedback windows (such as suggesting times to use high-power equipment during off-peak hours). Once generated, the strategy table is automatically synchronized to the energy storage system's dispatch module, which makes real-time decisions based on system status (battery charge, electricity prices, weather conditions, etc.). The strategy supports rolling updates, with a recommended recalculation every two hours to ensure the system's adaptive control capabilities for "on-the-fly optimization."

[0019] Step S5: pre-allocate energy storage capacity based on the adaptive power storage strategy, perform multi-objective coordinated control, and build an intelligent energy storage power control engine; In this embodiment, an adaptive power storage strategy framework is established. A dynamic energy storage strategy model is designed based on household electricity demand and the characteristics of the energy storage system. This model considers multiple parameters, including electricity market prices, household electricity usage patterns, the charging and discharging efficiency of the energy storage device, and the battery's health. Through real-time monitoring and historical data analysis, a predictive model is established to identify peak and off-peak periods for household electricity consumption. Time series analysis methods are used to predict electricity demand over the next 24 hours, ensuring that the energy storage system can charge before peak demand and discharge during low demand periods. Energy storage capacity is pre-allocated. After determining electricity demand, energy storage capacity is rationally allocated based on the predicted electricity usage patterns. A priority strategy can be established, for example, prioritizing power supply to critical equipment during peak periods while dynamically adjusting the energy storage charging and discharging strategy based on electricity prices. Specifically, the system proactively charges when electricity prices are low, while prioritizing discharge when prices are high to reduce household electricity costs. Pre-allocation of energy storage capacity ensures that the energy storage device's charging and discharging capabilities can meet household electricity demand during different time periods. Multi-objective coordinated control is implemented. Establish a coordinated control algorithm to optimize the energy storage system's operating strategy. This algorithm must simultaneously consider multiple objectives, including economic benefits, energy utilization, and environmental impact. Use linear programming or genetic algorithms to maximize the economic benefits of the energy storage system while meeting household electricity needs. By monitoring the grid's status and market electricity prices in real time, dynamically adjust the energy storage system's charging and discharging plans to ensure optimal decisions are made in different scenarios. Build an intelligent energy storage power supply control engine. Integrate the above modules to form an intelligent control system that receives real-time electricity usage data and market electricity price information from household smart meters and automatically adjusts the charging and discharging strategies of the energy storage capacity. Provide a user-friendly interface with visual data display, allowing households to easily monitor and adjust the operating status of the energy storage system. Furthermore, the system should have self-learning capabilities, enabling it to continuously optimize its operating strategy based on historical electricity usage data and market changes.

[0020] Step S6: Perform online optimization of strategy parameters and iterative decision optimization on the intelligent energy storage power supply control engine to build a power supply iterative optimization management model.

[0021] In this embodiment, after the policy execution effect data is standardized, the system enters the "online tuning" stage, dynamically adjusting the key parameters of the current policy model through real-time feedback data to optimize the overall operation effect. The main adjustable parameters include: energy storage capacity allocation ratio (whether the proportion of high-priority equipment is too high); Scheduling time window width (whether the policy takes effect with a delay); Dynamic safety redundancy factor (whether actual load fluctuations are adequately covered); Response sensitivity coefficient (regulation response speed to sudden load);Priority sorting (whether it conforms to actual user behavior). Online parameter tuning usually adopts a hybrid model of rule-driven and machine learning. Initially, parameter correction is triggered by empirical rules. For example, if the policy deviation rate exceeds 15% for 30 consecutive minutes, the current power allocation value is reduced by 10%. After the system has been running for a long time, an adaptive adjustment algorithm can be introduced. For example, based on the "Policy Gradient" method in reinforcement learning, the policy parameter space is adjusted by gradient ascent / descent to gradually approach the optimal policy parameter group. Policy online tuning mainly fine-tunes the policy parameters of the current cycle. To achieve continuous evolution of the system and performance leaps, an iterative optimization mechanism must be introduced to enable the intelligent control engine to form long-term memory and cross-cycle policy migration capabilities through multiple rounds of learning. To this end, a reinforcement learning framework is introduced in this stage to construct a "power iterative optimization management model." The reinforcement learning model, based on a state-action-reward (SAR) structure, considers the operating state of the home energy storage system as the environmental state S (including battery capacity, current load forecast, and device on / off status), the policy parameter set as the selectable action A (e.g., different power allocation schemes, scheduling window combinations), and the objective function (e.g., "energy efficiency + comfort + cost savings") as the reward R. The system continuously tries different policies, obtains feedback, and updates the policy value function V(s) or Q(s,a). In the experimental platform, a model based on deep reinforcement learning (DQN) was used to select actions based on policy parameters. "Virtual execution" was simulated using historical operating data to evaluate the policy reward. Over a 7-day rolling learning window, the system iterated 120 policies, ultimately converging to an optimal policy combination with a 22% improvement in average benefit. This fully optimizes the previous policy in terms of energy efficiency, response latency, and load coverage. After completing the policy optimization iterations, a persistent storage and migration mechanism for the power management model is required to ensure long-term stable operation and cross-scenario adaptability of the control system. During this phase, the system versions and archives the optimized policy model and records the corresponding execution environment characteristic parameters (such as climate conditions, electricity price structure, user habits, etc.) for policy migration training. A certain optimization model performs well in the northern winter, characterized by concentrated loads during peak hours, water heaters and electric heaters as the main loads, and weak photovoltaic output. When migrating to the southern summer scenario, the system first compares the load structure and equipment composition of the two locations, automatically adjusts the policy model input weights, and achieves rapid adaptation through a "transfer learning + fine-tuning training" model. It only takes three days of running data to complete the policy reshaping. In addition, the system sets up a policy degradation detection mechanism. If the current policy continues to show abnormal behaviors such as increased deviation rate and frequent user intervention within a certain period of time (such as three days), it will automatically roll back to the previous policy version or trigger a new round of optimization training to ensure stable system operation.Ultimately, the management model forms the core neural network of the intelligent energy storage power system with a closed-loop structure of "control engine-optimization learning-strategy archiving-migration application", realizing an intelligent management system that transitions from passive response to active learning, and from fixed logic to dynamic evolution.

[0022] In this embodiment, refer to Figure 2 , is a flowchart of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Real-time high-frequency sampling based on household smart meters to collect household electricity monitoring data streams; Identify multiple household electrical appliances based on the household electricity monitoring data stream, perform non-intrusive multi-scale load decomposition, and extract the start-stop transient characteristics, steady-state operating power fingerprint, and harmonic spectrum characteristics of each electrical appliance; Performing device identity analysis on the start-stop transient characteristics to obtain type characteristics of each electrical device; Mining usage behavior preferences of different devices for the steady-state operation power fingerprint according to the type characteristics to generate user manipulation behavior patterns of different devices; Perform multi-scale state transition deep learning on the user control behavior pattern to generate a user device behavior prediction model.

[0023] In this embodiment, after obtaining user authorization, a smart meter performs high-frequency, real-time sampling of the household's total power signal. Smart meters must have high-resolution current and voltage sampling capabilities. Generally speaking, the sampling frequency of ordinary smart meters is approximately 1Hz to 1kHz. However, non-intrusive load identification requires capturing start-stop transient characteristics and harmonics, so it is recommended to increase the sampling frequency to 10kHz to 100kHz. In this experiment, TI's ADE9000 energy metering chip was used, which supports voltage and current waveform capture at 32kHz and above, sufficient for high-frequency analysis. During data acquisition, a synchronous sampling mechanism should be implemented to ensure alignment of current and voltage waveforms, facilitating subsequent extraction of features such as active power, reactive power, power factor, and harmonic spectrum. Furthermore, data must be accurately timestamped and stored and processed on a local edge computing device (such as a Raspberry Pi 4B) or uploaded to the cloud. The amount of data generated per second can reach hundreds of KB, necessitating data compression and local preliminary preprocessing. For example, window segmentation (one window every 100ms) and preliminary filtering of the raw waveforms are necessary to reduce subsequent processing complexity. The usage of different electrical appliances can be analyzed from the total load signal. This task is achieved using non-intrusive load decomposition (NILM) technology, specifically a deep learning approach based on multi-scale time-domain feature modeling. The collected voltage and current waveforms are first segmented and processed to extract transient features (such as current spikes and inrush current), steady-state power (such as mean active power), and frequency-domain features (such as THD, total harmonic distortion). For multi-scale decomposition, wavelet transforms (such as the Daubechies-4 wavelet) can be used to perform multi-resolution analysis on the current signal, isolating features in different frequency bands to capture both short-term (start-stop transients) and long-term (continuous operation) behavior. To handle nonlinear and non-stationary features, empirical mode decomposition (EMD) or variational mode decomposition (VMD) methods can be introduced to extract intrinsic mode functions (IMFs) from the signal. Regarding the model, a multi-task learning framework constructed by combining a deep residual network (ResNet) with an LSTM can simultaneously classify the status of multiple appliances. Experiments trained on typical devices such as electric kettles (1.2kW), refrigerators (120W), and washing machines (400-800W), achieving an accuracy rate exceeding 90%. Key features of various electrical devices were extracted for identification and behavior modeling. Device features were categorized into the following categories: Start-stop transient features: These include inrush current (typically lasting tens to hundreds of milliseconds) and sudden changes in current waveforms. The current surge during air conditioner startup can reach up to four times the normal operating current.

[0024] Steady-state power fingerprint: This refers to the active power and power factor of the device during stable operation. Taking an electric water heater as an example, its steady-state power characteristic is constant at around 1.2kW, with a small fluctuation range.

[0025] Harmonic spectrum characteristics: Obtain frequency domain energy distribution through FFT or STFT (short-time Fourier transform). Electronic devices such as switching power supplies often have 3rd, 5th, and 7th harmonics.

[0026] These features are extracted using a sliding window and uniformly normalized (z-score or min-max normalization) before being used as input for subsequent recognition and analysis models. In experiments, each feature vector is set to 256 dimensions, with a window length of 1 second and a 0.5-second overlap. Machine learning or deep learning models are used to map transient device features to the identities of specific appliances. A recommended approach is to build a multi-classification model, trained using, for example, XGBoost, SVM, or a 1D CNN model. When constructing a dataset, transient segments are labeled as the instants when a particular appliance is turned on or off, and these features are used as input. In one experiment, transient startup features for 10 appliance categories (such as microwave ovens, ovens, air conditioners, and refrigerators) were collected, with 1,000 samples per category. After training with a 1D CNN model, recognition accuracy reached 93.2%. Using an attention mechanism can also enhance the model's recognition capabilities for complex appliances, such as washing machines with multiple operating modes. Furthermore, this step can be combined with prior knowledge, such as by considering the device's power level or the time periods during which it is likely to start or stop, to form a Bayesian network-based discriminant model for more accurate device identification. This model then models the user's habits in operating these devices. This process requires long-term collection of device operation data, including statistics on startup time, operating duration, and daily / weekly periodicity, to generate device usage patterns. For example, air conditioners are typically used between 6:00 PM and 11:00 PM in the summer, with an average operating time of three hours. Refrigerators exhibit cyclical operation throughout the day, and electric kettles are used primarily in the mornings and evenings. Cluster analysis (such as K-means or DBSCAN) can identify similar device usage patterns. Association rule mining (such as the Apriori algorithm) can also uncover interoperable behaviors between devices, such as the probability of lights turning on when the TV is turned on increasing to over 70%. Behavioral patterns can be expressed as state transition matrices or time series, such as: [06:00 - 08:00] → Electric kettle turned on → Used for 3 minutes → Average power 1.2kW; [18:00 - 23:00] → Air conditioning on → Average temperature set to 26°C; Finally, to achieve predictive power management and energy-optimized scheduling, a model that can predict a user's next behavior is needed. This model is recommended to employ a multi-scale time series modeling architecture, such as a Transformer-based or Temporal Convolutional Network (TCN) framework. The model input is the time series of device usage over the past week (e.g., hourly operating status), and the output is the device start / stop probability over the next period. A sliding window (7 days, 24 hours per day) is used to construct the input sequence, and positional encoding (position embedding) is employed to enhance time perception. In experiments, the Transformer model predicted device usage over a 24-hour period with an accuracy of 87% and an F1-score of approximately 0.85. The model output can be combined with a home energy storage system (e.g., a 5kWh lithium battery) to adjust power policies. For example, charging in advance when peak load is predicted or discharging during off-peak periods can achieve energy savings and cost optimization.

[0027] In this embodiment, refer to Figure 3 , is a flowchart of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Mining the electricity consumption behavior of each electricity user terminal based on the household electricity consumption monitoring data stream to extract the electricity consumption behavior data of all electricity users; Extracting terminal power usage event parameters, power load level, and device activation frequency information of each user based on the power usage behavior data; Performing event density differential trend analysis on the power consumption event parameters to generate a user behavior rhythm curve; Evaluate the power performance rigidity of the power load level based on the behavior rhythm curve and generate the user behavior power rigidity index; Perform time series fitting of the device activation frequency information within a period to construct usage frequency curves for multiple devices; Conducting in-depth mining of behavioral demand volatility on the usage frequency curve and the user behavior power rigidity index to generate user electricity behavior demand fluctuation characteristics; Based on the user equipment behavior prediction model, the user's electricity demand fluctuation characteristics are used to predict the electricity demand at multiple time points, and a user's electricity demand prediction graph at multiple time points is constructed.

[0028] In this embodiment, the operating behavior of each household's electrical terminal device is identified from the total household electricity data, and behavioral event information for each device is extracted. Since the system does not install individual device meters, non-intrusive load monitoring (NILM) is used to extract device-by-device electricity usage behavior. The system relies on high-frequency smart meters (sampling frequency recommended to be above 10kHz) to obtain the household's total voltage and current waveforms. The acquisition time can be aggregated into windows ranging from minutes to hours. The data is segmented using a sliding window technique, and features such as current fluctuation trends, transient characteristics of load on / off, and steady-state power changes are extracted using short-time Fourier transforms (STFTs) or wavelet transforms (DWTs). These extracted features are then fed into a trained deep learning model (e.g., a multi-classification recognition model based on a CNN-LSTM architecture) to identify the behavior of different types of devices. The system can extract behavioral parameters such as the start / stop times, operating cycles, power levels, operating frequencies, and stability indicators for each device from the total load profile. After a week of data collection, the system successfully identified 10 typical device types (such as kettles, refrigerators, washing machines, and air conditioners) and constructed a list of behavioral events. Each record includes device type, start time, end time, peak power, stable power, and operating time. This behavioral data provides the foundation for subsequent user behavior modeling. Statistics are collected for each device's start and stop times, operating time, power fluctuation range, and on / off frequency. Event data is standardized to create a unified analysis format. Load level classification: Based on the average operating power, devices are divided into low-power devices (0-150W, such as routers and desk lamps), medium-power devices (150-1000W, such as TVs and washing machines), and high-power devices (>1000W, such as water heaters and rice cookers). This classification helps assess the contribution of different power loads to the overall electricity consumption structure and identify high-energy consumption points. Device activation frequency information: Statistics are collected for each device's average daily activation times, average operating time, and typical activation time period (e.g., kettles are concentrated between 7:00 and 8:00 AM). Frequency aggregation is also performed for this time period (by hour, day, or week). Activation frequency, as an important quantitative indicator of user usage habits, is an important input for subsequent rhythm modeling and predictive modeling. Statistics are collected for each device's start and stop time, operating time, power fluctuation range, switching frequency, etc. Event data is standardized to form a unified analysis format. Load level classification: Based on the average operating power of the device, it is divided into low-power devices (0-150W, such as routers and desk lamps), medium-power devices (150-1000W, such as TVs and washing machines), and high-power devices (>1000W, such as water heaters and rice cookers). This classification helps to assess the proportion of different power loads in the overall power consumption structure and reveal high energy consumption points.Device activation frequency information: Counts the average daily number of device activations, average operating time, and typical activation time periods (e.g., kettles are concentrated between 7:00 AM and 8:00 AM), and aggregates the activation frequency by hour, day, or week. Activation frequency, as an important quantitative indicator of user usage habits, serves as a crucial input for subsequent rhythmic and predictive modeling. "Energy rigidity" measures a user's dependence on specific electricity usage behaviors, specifically whether they must use a load at a specific time and in a specific manner. High rigidity indicates that the behavior is unschedulable and should be prioritized by the energy storage system; otherwise, it can be appropriately deferred. The standard deviation and coefficient of variance (CV) are calculated for the rhythmic curve of each load level to determine whether the behavior is highly time-dependent. The proportion of high-density areas in the behavior curve to the total density is assessed to determine whether high concentration exists. High-power devices are given a higher weight to reflect their greater impact on the energy system.

[0029] Electrical rigidity can be defined as follows: A water heater starts up at 7:00 AM and 9:00 PM daily, exhibiting extremely low volatility and high rigidity. A robot vacuum cleaner, however, has highly random usage times and low rigidity. This metric can be used as a basis for prioritizing energy storage response in energy scheduling. The activation time distribution of each device is statistically analyzed by day or week, such as the number of startups per 24-hour period per day, to form a basic frequency sequence. An appropriate fitting model is then selected, such as a Fourier transform to extract the dominant frequency and periodic variation patterns; trigonometric function fitting (such as the sin / cos curve) to express periodic variations; polynomial fitting or a temporal convolutional neural network (TCN) to handle complex fluctuation patterns. The fitted frequency curve is then used for subsequent load forecasting. For example, during peak frequency curve periods, the system should prioritize energy storage resources. In an experiment, washing machine usage exhibited weekend peak behavior, while refrigerator usage exhibited a typical 24-hour fluctuation curve. A volatility index (WBI) is generated based on the combination of the frequency curve's volatility (such as standard deviation and frequency jump rate) and rigidity. Behavior pattern clustering: User behavior is categorized into three types: high volatility (random usage), low volatility (regular usage), and mixed (partial regularity, part volatility), and scheduling strategies are developed for each. Equipment importance ranking: Equipment scheduling priority is assessed based on volatility and energy consumption levels. Equipment with high volatility often affects the forecast accuracy of energy storage scheduling, while equipment with low volatility can be used for load shifting and peak energy consumption reduction.

[0030] Finally, the system combines the existing user behavior model (such as the Transformer-based sequence prediction model) with the fluctuation characteristics extracted in step 6 to perform multi-point electricity demand forecasting.

[0031] Model inputs include: User's historical electricity usage behavior sequence (past N days); Frequency curve and behavioral rhythm curve of each device; Volatility index and rigidity index; The model outputs predicted power values ​​for each device or the total load at several future time points (e.g., hourly over the next 24 hours). The output is visualized through charts such as daily demand forecasts and equipment peak time period forecasts.

[0032] The prediction results can be directly used to formulate scheduling plans for home energy storage systems, achieving: Energy storage charging before high demand periods; Peak discharge during low demand periods; Delayed operation of adjustable loads to smooth out peaks and fill valleys; Experimental verification shows that the prediction accuracy of this type of model can reach 85~92%, significantly improving the operating economy and user experience of home energy storage systems.

[0033] In this embodiment, the detailed implementation steps of step S3 include: Obtain meteorological environment parameters of the current user's home area, including temperature, humidity, light intensity, wind speed, and air pressure change data; Analyze the time series meteorological situation changes of meteorological environmental parameters to generate meteorological situation change characteristics; Performing nonlinear correlation analysis on the electricity consumption behavior data of the electricity consumption terminal according to the characteristics of meteorological situation changes to generate a meteorological-electricity consumption behavior mapping law; Based on the weather-electricity behavior mapping law, the type characteristics of each electrical device are used to identify climate-sensitive electrical equipment, and the climate change response sensitivity is calculated to generate the climate response sensitivity coefficient of the electrical equipment; The electricity demand forecast graphs at the multiple time points are adjusted for climate disturbance demand according to the climate response sensitivity coefficient to construct a climate disturbance electricity demand forecast graph.

[0034] In this embodiment, data is collected by calling the regional meteorological bureau's data interface or utilizing local meteorological sensor devices. For example, OpenWeatherMap or the National Meteorological API can be used. For a laboratory verification environment, taking a residential area in Beijing as an example, data collection is performed hourly. Because meteorological data units vary (temperature in °C, humidity in%, wind speed in m / s), the timestamp format must be standardized and normalized for subsequent model input. The collected meteorological data is aligned with electricity usage data from smart meters to create a "weather-electricity usage" time series matrix with time as the primary axis. Time series trend modeling is performed for each meteorological parameter (such as temperature and humidity). An ARIMA model or exponential smoothing method can be used to extract trend lines from the temperature curve to identify states such as rising, falling, and stable temperatures. Time periods are divided into different meteorological states based on changing trends and thresholds, such as high temperature and high humidity (T>30°C, H>80%), sunny and dry weather, and periods of high wind speed disturbances. Use principal component analysis (PCA) or self-organizing map (SOM) to cluster the five-dimensional meteorological parameters and summarize several typical meteorological scenarios (such as "sunny and hot", "cloudy and cool", "strong wind and high pressure", etc.). Analyze the rate of change of each meteorological factor (i.e., the first-order derivative) to identify periods of rapid change (such as sudden temperature drop, high wind speed intrusion, etc.), which often have a strong impact on electricity consumption behavior. The output result is a set of situation change characteristics for multiple time periods and multiple meteorological combinations. In the experiment, the data modeling of 14 consecutive days in a certain place found that the meteorological situation changes can be effectively divided into 6 typical states. Each meteorological situation feature vector is correspondingly combined with the electricity consumption behavior data in the same time period (such as the power, operating frequency, usage time, etc. of each device) to construct a high-dimensional mapping sample. Support vector regression (SVR) can be used for nonlinear function fitting; Neural network models (such as MLP or RBF networks) can also be used to capture complex nonlinear relationships; If the sample size is large, a random forest regression model can be used for modeling. Through the feature importance scoring mechanism within the model, it is clear which meteorological factors have a dominant influence on the usage behavior of which devices. For example, it was found that: temperature is the main influencing factor of the change in the frequency of air conditioning activation; light intensity directly affects the charging efficiency of the photovoltaic system and the frequency of use of the curtain motor; wind speed changes indirectly affect the response logic of the indoor temperature control system. The final generated "meteorological-electricity consumption behavior mapping law" is a functional relationship group that can be used for system learning and prediction, with high timeliness and prediction capabilities. For each device, the rate of change of its electricity consumption behavior with meteorological changes is extracted; Determine its responsiveness to major meteorological variables; If the response is significant (e.g., correlation coefficient > 0.6), it is considered a climate-sensitive device. A response model Y = f(X) is established between the device's power consumption parameter Y and the meteorological parameter X. Find its local rate of change (partial derivative or gradient), that is, the magnitude of the change in electricity consumption caused by a unit of meteorological change; After standardization, the sensitivity coefficient S is obtained, for example: ; Indicates the sensitivity of the equipment power to temperature. Taking air conditioning as an example, under 15 consecutive days of temperature changes, its daily power consumption is strongly positively correlated with the temperature. The calculated temperature sensitivity coefficient is 0.82 (indicating that for every 1°C increase in temperature, the daily power consumption increases by 8.2%), making it a highly sensitive device. The system will generate a list of equipment sensitivities, among which highly sensitive equipment should be given priority in dispatching energy storage resources during periods of sudden meteorological changes to ensure their use. The future forecasted meteorological change ΔX (such as temperature changes, humidity mutations, etc.) is input into the established meteorological-power consumption behavior mapping function; Combined with the sensitivity coefficient S, the disturbance correction ΔP = S × ΔX is calculated.

[0035] Adjust the power forecast value of each device at the corresponding time point in the original forecast diagram; For example, if the temperature rises by 3°C between 3:00 PM and 5:00 PM in the 24-hour forecast, the predicted air conditioning load will increase by ΔP = 0.82 × 3 = 2.46 kWh. This generates a "climate disturbance electricity demand forecast map" with climate disturbance corrections, including the power demand adjustments for each device within the forecast period. The system also generates highly sensitive device dispatch warnings to prompt the energy storage system to pre-adjust capacity.

[0036] In this embodiment, step S4 includes the following steps: Analyze the power consumption period of the climate disturbance power demand forecast map, identify the peak power conflict time period, and mark the peak power consumption time period; Calculating the normalized power supply power of the power grid during peak power consumption time based on the household power consumption monitoring data stream to generate the normalized power supply power during peak hours; Predict the peak electricity demand during peak electricity consumption time periods to obtain the peak household electricity demand; The multi-period electricity demand deviation is calculated based on the household electricity demand peak and the normalized power supply power during the peak period to obtain the multi-period electricity demand deviation; Adaptive energy storage capacity analysis is performed based on electricity demand deviations in multiple time periods to build an adaptive power supply storage strategy.

[0037] In this embodiment, based on the "climate disturbance electricity demand forecast map" after integrating meteorological disturbance factors, the core goal of this step is to identify the time periods when users have obvious peak electricity consumption in the future, and mark the key load pressure intervals through power conflict analysis to provide a basis for subsequent energy storage configuration and power scheduling. Time period division: Based on the time axis, the forecast map is divided into several time windows in hourly units, such as 24 time periods per day. The total predicted load power value of each time period is extracted, and the sliding average trend line is calculated to identify the local power increase trend. The hourly predicted power change rate (ΔP) is used as the key indicator. If ΔP exceeds the set threshold (such as 30% of the average power in the previous hour), it is identified as a power conflict period.

[0038] Peak hour identification criteria are: Peak power exceeds the daily average power by more than 1.3 times; Multiple high-energy-consuming devices (such as air conditioners, water heaters, and dryers) are running at the same time; It highly overlaps with external meteorologically sensitive periods (such as hot periods triggering the superimposed activation of air conditioners).

[0039] The system outputs a list of peak hours, such as "5:00 PM - 7:00 PM" and "9:00 PM - 10:00 PM," and graphically illustrates the equipment operating density during these periods for energy storage strategy invocation. The user's predicted load reached 6.2 kW during the 5:00 PM - 7:00 PM period, while the daily average power consumption was 3.9 kW, meeting the peak identification criteria and thus marking it as a key power supply control period. After identifying the peak power consumption periods, the grid's normalized power supply capacity during these periods must be assessed. "Normalized power supply" refers to the average power level that the grid can stably supply under normal weather conditions. To achieve this, historical grid operation data is collected, eliminating abnormal fluctuations caused by equipment maintenance, inclement weather, and other factors, to establish a normalized power supply capacity model. The stable power supply capacity of a certain city's household grid during the summer weekday period from 6:00 PM to 9:00 PM is 3.5 kW per household (based on the average value for the same season over the past three years). On weekends, when the overall load is slightly lower, this value can reach 3.8 kW. By comparing this with the predicted peak-hour power consumption, we can identify time periods with insufficient power supply capacity. We also need to consider the transformer's load capacity across different zones. For example, if a transformer with a rated capacity of 250kVA serves 50 households, its average power supply capacity per household is 5kW. However, during peak hours, it typically only maintains a 70% load stability of 3.5kW. Ultimately, peak power demand periods are mapped to power supply capacity constraint intervals, and the normal power supply power (P_base(t)) for each time period is recorded for subsequent deviation calculations.

[0040] Based on historical load data, household behavior patterns (such as cooking time and equipment usage habits) and climate disturbance variables (high temperature and heavy rainfall days), machine learning methods are used for modeling and prediction.

[0041] Model selection can include XGBoost, LSTM network, LightGBM, etc., by inputting: Historical electricity consumption sequence (past 7 days) Meteorological parameters (temperature, humidity, wind speed, weather conditions) Time variables (weekday / weekend, hours) Device usage behavior data (smart socket feedback) The output is the predicted peak power value P_peak(t) for each time period. For example, between 6:00 PM and 7:00 PM, the household's peak power consumption is predicted to be 5.2 kW. The model training phase requires cross-validation using a large-scale dataset (e.g., 90 days of historical data, split into 70% training, 15% validation, and 15% testing). Evaluation metrics include MAE (mean absolute error) and RMSE (root mean square error) to ensure good generalization in peak load periods (RMSE < 0.5 kW is considered optimal). The difference between the predicted peak power consumption of the household and the grid's normalized power supply capacity during each peak period is calculated to obtain the precise power deviation D(t) required for power compensation. If ΔP > 0, there is a power shortage during that period, and energy storage is needed. If ΔP ≤ 0, the grid can independently handle the load. The hourly or 15-minute deviations are combined into a vector, ΔP(t), which can be used for peak shifting, energy storage allocation, and other applications.

[0042] Deviation Importance Classification: ΔP<0.5kW: slight fluctuation; ΔP ∈ [0.5, 2.0]kW: medium load pressure; ΔP>2.0kW: Immediate response is required, triggering a dispatch warning.

[0043] For example, if the predicted value from 18:00 to 19:00 on a certain day is 6.8kW and the normal power supply capacity is 4.6kW, then ΔP is 2.2kW, and the system will label this period as "medium-high risk".

[0044] In this embodiment, step S5 includes the following steps: Predicting the power demand of equipment according to the peak power consumption period to obtain the power demand of different equipment; Pre-allocating energy storage capacity based on the power demand based on the adaptive power storage strategy to generate an energy storage capacity quota for each device; generating a real-time power dispatch instruction according to the energy storage capacity quota; Based on real-time power dispatch instructions, accurate power output is carried out in different time periods, and power response monitoring parameters of the energy storage system are collected; Perform multi-objective coordinated control of the power response monitoring parameters of the energy storage system and build an intelligent energy storage power supply control engine.

[0045] In this embodiment, the overall peak load is broken down into the specific demands of each electrical device. This system identifies which devices will start, run, and stop during specific periods of time, and estimates their corresponding operating power. This prediction is based on historical operating data, user behavior, and patterns in device behavior influenced by climatic factors. In practice, the system first determines the device's power usage patterns based on historical monitoring data, such as whether the air conditioner always starts after 3 p.m. or whether washing machines are used primarily in the evening. It also considers external meteorological conditions, such as whether rising temperatures will increase air conditioner load. Using this historical data and forecast information, the system can predict the power consumption and duration distribution of key load devices, such as air conditioners, water heaters, and refrigerators, during a future peak power consumption period. The forecasting process typically references peak-period device operating data from the past one to two weeks to identify patterns and adjust the model based on weather forecasts (e.g., high temperatures or the onset of heavy rain). The system predicts that from 5:00 PM to 7:00 PM that afternoon, as the outdoor temperature rises, the air conditioner may run continuously at maximum power for 90 minutes. The water heater will also be activated for bathing during the same period, and the washing machine is expected to complete a wash cycle before 7:00 PM after dinner. Ultimately, the system outputs a list of device power requirements, accurate to every 15 minutes. After completing the device-level power forecast, the system allocates capacity to each device based on the remaining available capacity of the home energy storage battery. This process takes into account factors such as the limited energy storage resources, the criticality of the device's operation, user comfort requirements, and grid conditions. First, the system assesses the current state of the energy storage battery. For example, if its total capacity is 10 kWh and its current charge is 70%, 7 kWh is available to support the load. Next, it considers device priority. For example, medical equipment (such as ventilators) must be prioritized, followed by comfort equipment such as air conditioners and water heaters, and finally kitchen and lighting equipment. Next, based on the predicted operating power and operating duration of each device, the system appropriately allocates the energy available from the battery to each device. The system might decide to allocate 2.5 kWh to the air conditioner, 1.5 kWh to the water heater, 1.2 kWh to the refrigerator, and 1.8 kWh to lighting and other appliances combined. This pre-allocation ensures that critical equipment receives priority and stable power supply when energy storage is limited. The entire allocation plan is dynamically adjusted and optimized daily based on battery charging and discharging conditions and user behavior. This creates a list of energy storage capacity quotas, which serves as the basis for subsequent power scheduling. After determining the energy allocation for each device type, the system converts the energy storage capacity into specific power output instructions within specific time periods. The goal is to pre-plan and execute on demand the power requirements of each device at each point in time, the amount of power the energy storage system should output, and the time period during which it should be output. The generation of real-time power scheduling instructions breaks down the previous quota table into finer time granularity, such as scheduling intervals of 5 or 15 minutes, to generate a continuous scheduling list.Between 6:00 PM and 6:15 PM, the air conditioner needs to output 1 kW, the water heater needs to output 1.2 kW, and the refrigerator needs to maintain a normal operating power of 0.2 kW. When generating dispatch instructions, the system prioritizes avoiding overloading the energy storage system due to excessive instantaneous output power, while also preventing the battery from depleting before the peak period. Furthermore, the system dynamically adjusts the power output plan based on weather fluctuations. For example, during a sudden drop in temperature, air conditioning demand decreases, and the system automatically reduces air conditioning output, allocating power to lighting or other devices. Finally, the home energy management controller issues dispatch instructions to the energy storage control module, instructing it to output power according to the plan during each time period. During the dispatch execution phase, the system strictly outputs the required power according to the plan and simultaneously monitors the energy storage system's responsiveness. This includes real-time monitoring of multiple indicators, including power output stability, system response time, device load acceptance, and energy storage system status (such as power, voltage, current, and temperature).

[0046] During execution, the energy storage system converts stored energy into AC power through an inverter and delivers it to household loads. The system then checks whether the output power at each point in time meets the dispatch instructions. For example, if the system requires 1 kW of output to an air conditioner between 6:00 PM and 6:15 PM, but detects only 0.8 kW, the system will immediately record this discrepancy and identify the cause. This could be due to excessive battery temperature limiting output or changes in the load power of the device itself. Furthermore, the monitoring system continuously assesses the health of the energy storage batteries, including parameters such as charge and discharge efficiency, internal impedance changes, and charge cycle life. Any detection of slow system response or decreased output efficiency during certain periods of time will serve as important input for subsequent optimization of the control strategy. This entire monitoring process ensures that the dispatch plan is not only executed but also has controllable execution quality. Based on this, the system establishes a feedback loop, providing real-time data support for the next stage of the intelligent control engine. By comprehensively analyzing real-time monitoring data, an intelligent control engine with the capabilities to judge, adapt, optimize, and adjust is constructed, enabling the energy storage system to autonomously make optimal dispatch decisions based on constantly changing power demand and environmental conditions. During the actual construction process, the system uses collected power response data, device operating data, and battery status data as input. It compares scheduling command execution deviations, response latency, and operational stability, and establishes a multi-objective control model. This model dynamically coordinates four objectives: user comfort, battery life preservation, overall system efficiency, and power safety. The control engine automatically adjusts power allocation strategies when it detects a sharp increase in load on a device or abnormal battery status. This can include temporarily delaying power supply to secondary devices, reordering loads, or activating backup photovoltaic power supply paths. It also learns from past operational experience and autonomously optimizes future scheduling plans, forming a closed-loop prediction-execution-feedback-optimization mechanism. The system also features self-healing and scenario-shifting capabilities. For example, it automatically switches to energy storage-prioritized mode in the event of a temporary grid outage in a home, and reduces power supply to non-critical devices during peak electricity prices to save costs. In this embodiment, the specific steps of performing multi-objective coordinated control on the power response monitoring parameters of the energy storage system and building an intelligent energy storage power supply control engine are as follows: Calculating the actual power output time of the energy storage system power response monitoring parameter; Extracting an instruction generation timestamp of the real-time power scheduling instruction; Performing a response delay calculation on the instruction generation timestamp and the actual power output time to generate a power delay response value of the energy storage system; Performing adaptive response compensation based on the power delay response value to obtain an adaptive delay response compensation parameter; Analyze output power fluctuations of energy storage system power response monitoring parameters and extract output power waveforms; Perform multi-scale wavelet decomposition on the output power waveform to extract power harmonic features; Multi-objective coordinated control of adaptive delay response compensation parameters is performed according to power harmonic characteristics to build an intelligent energy storage power supply control engine.

[0047] In this embodiment, after the energy storage system receives the scheduling instruction, its actual power output is not completed immediately, but there is a physical response and control delay. Therefore, this step is intended to accurately record the moment when the energy storage system actually completes the power output change, that is, the "power response completion time". This parameter is usually collected by the monitoring module inside the energy storage system and uploaded to the home energy management platform through the data acquisition system. The system needs to set a high-frequency sampling mechanism, such as a power output data collection frequency of once every 500 milliseconds, to ensure that the complete details of the power change can be captured. The sampled data will include the output current, voltage and calculated output power at each time node. When the system completes a scheduling instruction, by analyzing the power change trend in the sampled data, find the time point when the actual output power reaches the set threshold (such as the target power ±5%) and remains stable. This point can be considered as the "actual power response time". If the system receives a command at 19:00:00 requesting a 1.5 kW output, and the sampled data shows a gradual increase in power from 0.3 kW to 1.45 kW at 19:03 and remaining stable, then 19:00:03 is the actual response time. This process provides basic data for subsequent response delay calculation and compensation analysis. To accurately compare the response speed of the energy storage system, it is necessary to extract the "timestamp" of each dispatch command issued by the system. This timestamp is typically generated by the home energy management system or local controller, recording the exact moment the dispatch command was created and issued to the energy storage inverter. This process is usually automated within the dispatch system, with each control command being stamped with a precise millisecond-level timestamp upon generation and recorded synchronously in the system log. If the controller generates a command requesting "increase the air conditioner power to 1.5 kW" at exactly 19:00:00, this time is the "command generation time." This time, along with the actual power output time, is used in subsequent latency calculations. During the experiment, it is recommended to use a unified system clock for time synchronization to avoid system time drift between different devices. Most smart energy management platforms use the Network Time Protocol (NTP) to maintain consistent time across control nodes, ensuring the accuracy of command timestamps. This timestamp is not only the basis for delay calculation but can also be used in various analytical scenarios, such as system fault diagnosis, command execution efficiency evaluation, and scheduling system performance optimization. By knowing the time when the scheduling command is generated and the actual power output time, the energy storage system's response delay to each power scheduling command can be calculated using a simple time difference. This parameter, called the "power delay response value," represents the energy storage system's dynamic response speed to control commands. In a scheduling process, if the command is generated at 19:00:00 and the system power reaches the target output at 19:00:03 and remains stable, the response delay is 3 seconds. In this way, the system can generate a set of "command-response delay" data for each scheduling command, forming a time series log for subsequent compensation and optimization.

[0048] In actual operation, response delay data spanning several consecutive hours or days can be aggregated and analyzed to identify patterns such as periodic increases in delay, abnormal delays in specific devices, or decreased response performance due to system load. Experiments have shown that in most home energy storage systems, the ideal response delay should be within the 2–4 second range. Consistent delays exceeding 5 seconds typically indicate system load pressure or communication bottlenecks. Once the system has acquired the response delay data for each dispatch instruction, the next step is adaptive compensation. This involves feedforward adjustments to future dispatch strategies or pre-setting instructions based on historical and current response delays to improve the real-time performance and stability of the overall control system. Compensation is typically implemented using a combination of two strategies: "response prediction," which predicts the likely response time of the current instruction by learning from historical delay patterns and sending control signals in advance; and "dynamic lag correction," which automatically adds time margin to the dispatch plan when certain devices frequently experience response delays. The system identified an average response delay of 4 seconds for water heaters, while the target accuracy requirement was less than 1 second. Therefore, the system will plan to advance the start-up time of the water heater to 4 seconds before the target demand time, so that the system can reach the target power output just when the user demand arrives. Compensation parameters usually include equipment response lag time, compensation correction value, system status assessment factor, etc. These compensation values ​​will be fed back to the dispatching system in real time and taken into account when generating instructions to achieve truly adaptive response control. By dynamically adjusting the compensation parameters, the system can continuously optimize control efficiency during operation and effectively reduce the risk of load gap caused by control lag. In addition to response delays, the energy storage system may have output fluctuations during power output, that is, the power cannot remain stable in a short period of time. The task of this step is to extract the power waveform from the high-frequency power data collected by the energy storage system for subsequent harmonic analysis and system stability evaluation.

[0049] Waveform acquisition is typically achieved using high-precision power sensors or voltage and current sensing devices. The system's power output is sampled at a frequency of 100 milliseconds or higher, and the data sequence is cached locally. By plotting this continuous sampling data as a power waveform, it is possible to observe whether there are periodic jitter, spikes, dips, or other abnormal power fluctuations. During an output schedule, the air conditioner's power should be stable at 1.5 kW. However, if the waveform shows several rapid fluctuations between 0.8 and 2.0 kW within the initial two seconds, further analysis is needed to determine whether the cause is a sudden load change, a slow inverter response, or changes in the battery's internal impedance. This fluctuation not only affects the user experience but may also cause equipment aging and reduce the lifespan of the energy storage system. Therefore, accurate power waveform extraction and analysis are crucial for the long-term stable operation of the energy storage system. This involves extracting hidden harmonic features. This is a signal analysis method that can decompose the details of the fluctuation at different time and frequency resolutions, identifying high-frequency interference, fundamental offset, and harmonic distortion within the waveform. In experimental applications, a suitable wavelet basis (such as db4 or sym5) is typically selected to perform wavelet decomposition of the power signal at five or more levels. Signal reconstruction at different scales is then used to analyze the high-frequency components for periodic pulses, short-period interference, or systematic fluctuations. This process can identify higher-order harmonics caused by load switching, electromagnetic interference, or instability in the inverter control algorithm. Analysis of a particular power waveform revealed a significant oscillation peak at 150 Hz in the fourth-level decomposition. The system identified this as third-harmonic interference, likely originating from the frequency conversion process of the air-conditioning compressor. Identifying these harmonic components facilitates subsequent control engine compensation, filtering, and scheduling adjustments during power output. This analysis, by introducing a frequency domain perspective, enhances understanding of the dynamic behavior of the energy storage system, moving beyond simply determining output and delving deeper into the "output quality" dimension. The previously collected response delay data, compensation parameters, power fluctuation characteristics, and harmonic analysis results are integrated into the control system of the intelligent energy storage power supply control engine to construct a multi-objective coordinated control model. This engine no longer focuses solely on whether it can "complete the scheduling", but instead focuses on the accuracy, real-time performance, stability and power quality of the scheduling.

[0050] The control engine is centered around multi-objective coordination: on the one hand, it minimizes response latency to ensure on-time device startup; on the other, it ensures a smooth output power waveform and minimizes harmonic pollution. Simultaneously, it ensures that the energy storage system operates within a safe range to avoid thermal runaway or shortened battery life due to frequent discharge or large power fluctuations. In terms of implementation, the system can incorporate fuzzy logic control, adaptive weighted regulation algorithms, or reinforcement learning-based strategy optimization modules to dynamically balance multiple objective parameters. If the system identifies a device with a three-second response delay and frequent output jitter, the control engine automatically reduces its instantaneous power load and advances its dispatch instructions by two seconds, comprehensively adjusting the scheduling plan. During operation, the control engine continuously learns from historical dispatch results and feedback to improve prediction accuracy and response quality. By forming a closed-loop monitoring-analysis-compensation-control system, the energy storage system evolves from static execution to intelligent optimized scheduling, which is the key to achieving efficient and intelligent home power management systems.

[0051] In this embodiment, step S6 includes the following steps: Collect user feedback information of energy storage power system; Performing an energy storage supply quality assessment based on the user feedback information to obtain a user energy storage control assessment report; Conduct deep reinforcement learning on user energy storage control evaluation reports to identify defects in user feedback; Online tuning of strategy parameters is performed based on defect directions reported by users, and iterative decision optimization is performed on the intelligent energy storage power supply control engine to build an iterative power supply optimization management model.

[0052] In this embodiment, during the actual operation of a home energy storage power system, user experience is a key criterion for evaluating the effectiveness of the system's control strategy. User feedback includes subjective evaluations of power supply stability, response speed, power satisfaction, system noise, energy consumption, device compatibility, and other aspects, as well as proactive fault reporting data. Therefore, establishing a comprehensive user feedback collection system is the first step in control optimization. During implementation, the system should integrate a user interface, such as a home energy app or touch panel, to collect user feedback through questionnaires, rating systems, and voice recognition. Furthermore, the system should integrate with home IoT devices to automatically collect user behavior characteristics through devices such as smart speakers and home terminals, such as implicit feedback behaviors such as frequent manual device on / off switching, complaints about delayed startup, and unstable charging. For example, if a user frequently reports "unstable water temperature" when using the water heater at night, or if the system records an unusual frequency of users manually adjusting the device's power mode, these behaviors can be converted into feedback data. For quantification, feedback data should be encoded into a standard structure, such as a power supply satisfaction score (0-10 points), a response satisfaction level (low / medium / high), and keywords for subjective suggestions. Feedback information forms a user feedback dataset for subsequent system performance evaluation and control strategy adjustments, serving as a key entry point for the system's "people-centric" approach. Collected user feedback information undergoes structured processing and, combined with system operating data, constructs a comprehensive energy storage supply quality assessment model. This model comprehensively considers subjective perceptions and objective data, using quantitative indicators to reflect the overall performance of the control system. The "User Energy Storage Control Assessment Report" generated through this process serves as the foundation for feedback-driven optimization. Specifically, the assessment model consists of multi-dimensional indicators, including but not limited to: power response time stability (unit: seconds), power fluctuation rate (unit: W / s), voltage stability (unit: V), power dispatch success rate (%), and user satisfaction ratings. Each indicator is algorithmically mapped to the relationship between user feedback and system monitoring data. If users rate "slow startup response" low, the actual average dispatch response time can be analyzed accordingly. If it exceeds a set threshold, such as 4 seconds, the assessment is considered "unsatisfactory."

[0053] Furthermore, the system should categorize and assess the power supply conditions of various types of equipment, such as air conditioners, washing machines, and electric vehicles, to identify differences in energy storage adaptability across different devices. Once generated, the report will be presented in the form of a visual dashboard or periodic reports for reference by users and system developers. Once standardized user feedback evaluation reports are generated, deep reinforcement learning methods can be introduced to identify patterns and identify flaws within the evaluation results. The core goal of this step is to enable the system to autonomously learn to identify weaknesses in system operation from user feedback, enabling intelligent optimization at the policy level. During this implementation, the system builds a state-action-reward model (RL framework) centered around the "control strategy-user evaluation score-feedback discrepancy" model. If a control strategy causes a decrease in user ratings, the system considers this a negative incentive; if the adjusted strategy results in an increase in ratings, this is considered a positive incentive. By continuously collecting user ratings for different strategies, a deep neural network reinforcement learning model is trained, allowing the system to gradually understand which policy parameter combinations best enhance the user experience. In the experimental parameters, the reinforcement learning model can be set to update daily, with each iteration lasting 24 hours. The system automatically adjusts the next cycle's strategy based on the previous day's feedback data. Taking air conditioning response control as an example, the system may discover that adjusting the response compensation time from 3 seconds to 2 seconds significantly improves user satisfaction scores (for example, from 6.8 to 8.4). This strategy will then be used as the direction for further policy optimization. After identifying the control defects in user feedback, the system must immediately perform online tuning of the policy parameters. This tuning process is not a simple fixed parameter adjustment, but rather an online adaptive mechanism that optimizes the policy parameters in small increments without affecting the current system's stable operation, thereby improving overall control accuracy and user satisfaction.

[0054] The key to online tuning lies in real-time performance and safety. If the system detects unstable output current of the energy storage system in a high-temperature environment, possibly due to a failure of the temperature compensation strategy, it can lower the inverter's power output limit by 5% and simultaneously adjust the operating curve of the battery temperature control module to ensure a balance between control accuracy and device lifespan. During this process, the control engine's core algorithm structure must feature adaptive weight adjustment and support multi-dimensional parameter linkage. For example, for water heater scheduling, the system can jointly adjust the three parameters of "response lead," "dispatching frequency," and "maximum discharge rate" to achieve the expected improvement in user ratings after optimization. The results of the policy optimization are fed back to the control engine's core module, forming a policy iteration. In the experimental setup, each policy adjustment can be validated through 72 hours of stable operation. If user feedback indicates continued improvement, the system automatically identifies the policy as effective and incorporates it into the long-term operation plan. Through continuous policy self-adjustment, performance verification, and user rating feedback, the system gradually builds an intelligent energy storage power supply control engine with evolutionary capabilities. The above processes of feedback analysis, learning identification, and policy optimization form a unified management framework, forming a "power supply iterative optimization management model." This model not only includes control logic but also integrates components such as a data collection module, a learning model library, a policy simulation platform, and a user evaluation feedback channel, forming a closed-loop iterative system. The core logic of the model design follows the following closed-loop mechanism: the system collects user feedback → generates an evaluation report → determines the direction of deep learning optimization → performs online policy tuning → updates the control engine → receives further user experience feedback. This closed-loop system completes a complete iteration cycle every 72 hours and supports extended features such as cross-season policy migration and personalized user preference adaptation. To ensure the stability and adaptability of the management model, the system should be configured with edge computing nodes for local learning and tuning, avoiding frequent reliance on cloud-based responses. Furthermore, model parameters are equipped with "experience retention weights" to prevent unexpected data anomalies from causing significant policy deviations. When user ratings fluctuate significantly within a short period of time (exceeding ±2 points), the system triggers a "prudent iteration mechanism," freezing parameters and conducting manual review to enhance the security of the control strategy.

[0055] In this embodiment, a power management system based on a home energy storage power supply system is provided, which is used to execute the power management method based on the home energy storage power supply system as described above, including: The behavior prediction module is used to collect household electricity monitoring data streams, conduct device usage behavior preference mining and multi-scale state transition deep learning, and generate user device behavior prediction models; An electricity demand forecasting module is used to mine the electricity consumption behavior of each electricity terminal based on the household electricity consumption monitoring data stream, and to forecast electricity demand at multiple time points based on the user device behavior prediction model, thereby constructing a user's electricity demand forecast graph at multiple time points; A climate disturbance analysis module is used to obtain meteorological environment parameters of the current user's home area, adjust the electricity demand forecast graphs for the multiple time points based on the meteorological environment parameters, and construct a climate disturbance electricity demand forecast graph; Adaptive energy storage module, used to analyze electricity demand forecasts for climate disturbances, analyze adaptive energy storage capacity, and build adaptive power storage strategies; The capacity pre-allocation module is used to pre-allocate energy storage capacity based on the adaptive power storage strategy, perform multi-objective coordinated control, and build an intelligent energy storage power control engine; The iterative optimization module is used to perform online tuning of strategy parameters and iterative decision optimization on the intelligent energy storage power supply control engine, and to build an iterative optimization management model for the power supply.

[0056] The present invention further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the power management method based on the home energy storage power system described above are implemented.

[0057] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0058] The foregoing description is intended only to provide specific embodiments of the present invention, which are intended to enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. A power management method based on a household energy storage power supply system, characterized in that: The following steps are involved: Step S1: Collect household electricity monitoring data streams, conduct device usage behavior preference mining and multi-scale state transition deep learning, and generate a user device behavior prediction model; Step S2: mining the electricity consumption behavior of each electricity terminal based on the household electricity consumption monitoring data stream, and performing multi-time point electricity demand prediction based on the user device behavior prediction model to construct a user's multi-time point electricity demand prediction map; Step S3: obtaining meteorological environment parameters of the current user's home area, and adjusting the electricity demand forecast graphs for the multiple time points based on the meteorological environment parameters to construct a climate disturbance electricity demand forecast graph; Step S4: Analyze the electricity demand forecast map for climate disturbances by electricity consumption period, perform adaptive energy storage capacity analysis, and construct an adaptive power supply energy storage strategy; Step S5: pre-allocate energy storage capacity based on the adaptive power storage strategy, perform multi-objective coordinated control, and build an intelligent energy storage power control engine; Step S6: Perform online optimization of strategy parameters and iterative decision optimization on the intelligent energy storage power supply control engine to build a power supply iterative optimization management model.

2. The power management method based on the home energy storage power supply system according to claim 1, characterized in that: The specific steps of step S1 are: Real-time high-frequency sampling based on household smart meters to collect household electricity monitoring data streams; Identify multiple household electrical appliances based on the household electricity monitoring data stream, perform non-intrusive multi-scale load decomposition, and extract the start-stop transient characteristics, steady-state operating power fingerprint, and harmonic spectrum characteristics of each electrical appliance; Performing device identity analysis on the start-stop transient characteristics to obtain type characteristics of each electrical device; Mining usage behavior preferences of different devices for the steady-state operation power fingerprint according to the type characteristics to generate user manipulation behavior patterns of different devices; Multi-scale state transition deep learning is performed on the user control behavior pattern to generate a user device behavior prediction model.

3. The power management method based on the household energy storage power supply system according to claim 1, characterized in that: The specific steps of step S2 are: Mining the electricity consumption behavior of each electricity user terminal based on the household electricity consumption monitoring data stream to extract the electricity consumption behavior data of all electricity users; Extracting terminal power usage event parameters, power load level, and device activation frequency information of each user based on the power usage behavior data; Performing event density differential trend analysis on the power consumption event parameters to generate a user behavior rhythm curve; Evaluate the power performance rigidity of the power load level based on the behavior rhythm curve and generate the user behavior power rigidity index; Perform time series fitting of the device activation frequency information within a period to construct usage frequency curves for multiple devices; Conducting in-depth mining of behavioral demand volatility on the usage frequency curve and the user behavior power rigidity index to generate user electricity behavior demand fluctuation characteristics; Based on the user equipment behavior prediction model, the user's electricity demand fluctuation characteristics are used to predict the electricity demand at multiple time points, and a user's electricity demand prediction graph at multiple time points is constructed.

4. The power management method based on the home energy storage power supply system according to claim 1, characterized in that: The specific steps of step S3 are: Obtain meteorological environment parameters of the current user's home area, including temperature, humidity, light intensity, wind speed, and air pressure change data; Analyze the time series meteorological situation changes of meteorological environmental parameters to generate meteorological situation change characteristics; Performing nonlinear correlation analysis on the electricity consumption behavior data of the electricity consumption terminal according to the characteristics of meteorological situation changes to generate a meteorological-electricity consumption behavior mapping law; Based on the weather-electricity behavior mapping law, the type characteristics of each electrical device are used to identify climate-sensitive electrical equipment, and the climate change response sensitivity is calculated to generate the climate response sensitivity coefficient of the electrical equipment; The electricity demand forecast graphs at the multiple time points are adjusted for climate disturbance demand according to the climate response sensitivity coefficient to construct a climate disturbance electricity demand forecast graph.

5. The power management method based on the household energy storage power supply system according to claim 1, characterized in that: The specific steps of step S4 are: Analyze the power consumption period of the climate disturbance power demand forecast map, identify the peak power conflict time period, and mark the peak power consumption time period; Calculating the normalized power supply power of the power grid during peak power consumption time based on the household power consumption monitoring data stream to generate the normalized power supply power during peak hours; Predict the peak electricity demand during peak electricity consumption time periods to obtain the peak household electricity demand; The multi-period electricity demand deviation is calculated based on the household electricity demand peak and the normalized power supply power during the peak period to obtain the multi-period electricity demand deviation; Adaptive energy storage capacity analysis is performed based on electricity demand deviations in multiple time periods to build an adaptive power supply storage strategy.

6. The power management method based on the household energy storage power supply system according to claim 1, characterized in that: The specific steps of step S5 are: Predicting the power demand of equipment according to the peak power consumption period to obtain the power demand of different equipment; Pre-allocating energy storage capacity based on the power demand based on the adaptive power storage strategy to generate an energy storage capacity quota for each device; generating a real-time power dispatch instruction according to the energy storage capacity quota; Based on real-time power dispatch instructions, accurate power output is carried out in different time periods, and power response monitoring parameters of the energy storage system are collected; Perform multi-objective coordinated control of the power response monitoring parameters of the energy storage system and build an intelligent energy storage power supply control engine.

7. The power management method based on the household energy storage power supply system according to claim 6, characterized in that: The specific steps of performing multi-objective coordinated control of the energy storage system power response monitoring parameters and building an intelligent energy storage power supply control engine are as follows: Calculating the actual power output time of the energy storage system power response monitoring parameter; Extracting an instruction generation timestamp of the real-time power scheduling instruction; Performing a response delay calculation on the instruction generation timestamp and the actual power output time to generate a power delay response value of the energy storage system; Performing adaptive response compensation based on the power delay response value to obtain an adaptive delay response compensation parameter; Analyze output power fluctuations of energy storage system power response monitoring parameters and extract output power waveforms; Perform multi-scale wavelet decomposition on the output power waveform to extract power harmonic features; Multi-objective coordinated control of adaptive delay response compensation parameters is performed according to power harmonic characteristics to build an intelligent energy storage power supply control engine.

8. The power management method based on the household energy storage power supply system according to claim 1, characterized in that: The specific steps of step S6 are: Collect user feedback information of energy storage power system; Performing an energy storage supply quality assessment based on the user feedback information to obtain a user energy storage control assessment report; Conduct deep reinforcement learning on user energy storage control evaluation reports to identify defects in user feedback; Online tuning of strategy parameters is performed based on defect directions reported by users, and iterative decision optimization is performed on the intelligent energy storage power supply control engine to build an iterative power supply optimization management model.

9. A power management system based on a household energy storage power system, characterized in that: The method for executing the power management method based on the home energy storage power supply system according to claim 1 comprises: The behavior prediction module is used to collect household electricity monitoring data streams, conduct device usage behavior preference mining and multi-scale state transition deep learning, and generate user device behavior prediction models; An electricity demand forecasting module is used to mine the electricity consumption behavior of each electricity terminal based on the household electricity consumption monitoring data stream, and to forecast electricity demand at multiple time points based on the user device behavior prediction model, thereby constructing a user's electricity demand forecast graph at multiple time points; A climate disturbance analysis module is used to obtain meteorological environment parameters of the current user's home area, adjust the electricity demand forecast graphs for the multiple time points based on the meteorological environment parameters, and construct a climate disturbance electricity demand forecast graph; Adaptive energy storage module, used to analyze electricity demand forecasts for climate disturbances, analyze adaptive energy storage capacity, and build adaptive power storage strategies; The capacity pre-allocation module is used to pre-allocate energy storage capacity based on the adaptive power storage strategy, perform multi-objective coordinated control, and build an intelligent energy storage power control engine; The iterative optimization module is used to perform online tuning of strategy parameters and iterative decision optimization on the intelligent energy storage power supply control engine, and to build an iterative optimization management model for the power supply.

10. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the power management method based on the home energy storage power supply system according to any one of claims 1 to 8 are implemented.

Images