Power conversion control method of household energy storage system and magnetic component

Abstract

The invention discloses a power conversion control method of a household energy storage system and a magnetic component, and the method comprises the steps: collecting an input voltage range, an output power demand and a frequency range according to the working characteristics of the system, selecting a nanocrystalline magnetic core material, and designing a multi-layer distributed winding structure to construct the magnetic component; acquiring operation monitoring data through a multi-dimensional data acquisition network, and generating a working condition description set through feature extraction and classification identification; a multi-target optimization model is constructed through bottleneck point identification and capacity estimation, and a power optimization distribution scheme is generated; then a control strategy selection space is constructed, an adaptive decision engine is operated, a PWM control strategy is selected, multiple PWM generation algorithms are realized, parameters are adjusted in real time, and a control signal is output; and finally, realizing system security monitoring and self-adaptive protection response through differential privacy protection processing and an anomaly detection model. According to the invention, the adaptability of the system in a wide voltage input range can be improved, and the loss of magnetic components under different load conditions is reduced.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, specifically to a power conversion control method and magnetic components for a home energy storage system. Background Technology

[0002] Home energy storage systems are an important component of modern smart grids, enabling energy storage, peak-valley regulation, and emergency backup functions. They are crucial for improving energy efficiency and ensuring household electricity safety. This field primarily encompasses core technologies such as energy storage batteries, power converters, control systems, and magnetic components.

[0003] Traditional home energy storage systems typically use magnetic components made of silicon steel sheets or ferrite materials, combined with basic PWM control algorithms to achieve power conversion. For example, common solutions employ simple SPWM modulation technology and PI-based closed-loop feedback systems, achieving a conversion efficiency of 80%-90%, but exhibiting poor performance consistency across a wide voltage input range.

[0004] Existing advanced energy storage systems employ fixed-parameter digital control algorithms for power conversion control, combined with optimized magnetic materials, achieving high conversion efficiencies. However, these systems lack real-time adaptive capabilities when facing complex and variable household power environments, resulting in significant performance fluctuations under different operating conditions. Particularly during battery charging and discharging, voltage variations cause converter operating point drift, making it difficult for magnetic components to maintain optimal performance across the entire operating range.

[0005] The main problems with existing technologies include: first, the power converter has insufficient adaptability over a wide input voltage range, resulting in efficiency fluctuations; second, the design of magnetic components lacks targeted optimization, resulting in significant losses and severe heat generation under different load conditions; and third, the system lacks an intelligent data-driven control mechanism, which cannot optimize and adjust according to actual power consumption patterns and battery characteristics, thus limiting the improvement of overall performance. Summary of the Invention

[0006] The purpose of this invention is to provide a power conversion control method and magnetic components for a home energy storage system, in order to solve the problems of insufficient adaptability of power converters in a wide input voltage range, large losses of magnetic components under different load conditions, and lack of intelligent data-driven control mechanism in the existing technology.

[0007] To achieve the above objectives, the technical solution provided by the present invention is as follows: A power conversion control method for a home energy storage system includes: Based on the working characteristics of home energy storage systems, the system's input voltage range, output power requirements, and frequency range are collected. Magnetic components are constructed by selecting nanocrystalline magnetic core materials and designing a multi-layer distributed winding structure. Based on the magnetic components, the operation monitoring data of battery status, load power and power grid status are collected in real time through a multi-dimensional data acquisition network, and feature extraction and classification are performed to generate a set of operating condition descriptions containing operating condition types and parameters. Based on the operating condition description set and the preset system physical model, a multi-objective optimization model is constructed through bottleneck point identification and capacity estimation to generate a power optimization allocation scheme. Based on the power optimization allocation scheme and real-time system status, a PWM control strategy is selected by constructing a control strategy selection space and running an adaptive decision engine. Based on the PWM control strategy, multiple PWM generation algorithms such as SPWM, SVPWM and DPWM are implemented and real-time parameters are adjusted to output a power conversion control signal. Based on system operation status data, differential privacy protection processing and anomaly detection models are used to achieve system security monitoring and adaptive protection response.

[0008] Preferably, the method for constructing magnetic components based on the operating characteristics of the home energy storage system, by collecting the system's input voltage range, output power demand, and frequency range, and by selecting nanocrystalline magnetic core materials and designing a multi-layer distributed winding structure, includes: Based on the working characteristics of home energy storage systems, the magnetic permeability, saturation magnetic induction intensity and loss characteristics of nanocrystalline materials were analyzed, and the working magnetic flux density range of 0.3T-0.8T was determined. Based on the aforementioned operating magnetic flux density range, a multi-layer distributed winding structure was designed and the conductor cross-section and arrangement were optimized to obtain an optimized winding structure with low skin effect and proximity effect. Based on the optimized winding structure with low skin effect and proximity effect, the magnetic components were obtained by simulating the temperature rise and loss distribution at different operating points using finite element analysis software.

[0009] Preferably, based on the magnetic components, the operation monitoring data of battery status, load power, and power grid status are collected in real time through a multi-dimensional data acquisition network, and feature extraction and classification are performed to generate an operating condition description set containing operating condition types and parameters, including: Based on the magnetic components and system operation data, the system collects operating monitoring data of battery SOC, voltage, temperature and load power through a multi-dimensional data acquisition network to generate a raw dataset reflecting the system's operating conditions. The original dataset is filtered, normalized, and subjected to preliminary feature extraction to generate a structured preprocessed dataset. Based on the structured preprocessed dataset, voltage range distribution, power fluctuation rate, load periodicity, and battery charge / discharge state characteristic parameters are calculated to form an operating condition feature vector. Based on the operating condition feature vector, operating condition identification is performed using a finite state machine and a lightweight XGBoost classifier to obtain the operating condition description set containing operating condition types and parameters.

[0010] Preferably, the step of constructing a multi-objective optimization model based on the operating condition description set and the preset system physical model, and generating a power optimization allocation scheme through bottleneck point identification and capacity estimation, includes: Based on the operating condition description set and the preset system physical model, an energy flow diagram is established under the current operating state, forming an energy transmission network topology. Based on the energy transmission network topology, the bottleneck analysis method is used to calculate the load rate, influence coefficient and sensitivity of each node, and identify the power bottleneck point of the system. Capacity estimation is performed on the power bottleneck points of the system and a bottleneck model is established to quantify the impact of each bottleneck point on the overall system performance. Based on the impact of each bottleneck on the overall system performance, a multi-objective optimization model that comprehensively considers efficiency, temperature rise, and system lifetime is constructed to obtain the power optimization allocation scheme.

[0011] Preferably, the step of selecting a PWM control strategy based on the power optimization allocation scheme and real-time system status by constructing a control strategy selection space and running an adaptive decision engine includes: Based on the power optimization allocation scheme and the real-time system status, a control strategy selection space containing multiple PWM modulation strategies and parameter combinations is constructed to form a candidate control strategy set. Based on the candidate control strategy set, feature parameters of control complexity, response time requirement and control accuracy requirement are extracted to generate control strategy feature vectors; Based on the feature vector of the control strategy, an adaptive decision engine is constructed using a finite state machine and a lightweight decision tree classifier to determine the control mode and strategy; Based on the control mode and strategy, smooth transition and parameter optimization are performed to obtain the PWM control strategy.

[0012] Preferably, the step of outputting a power conversion control signal based on the PWM control strategy, by implementing multiple PWM generation algorithms such as SPWM, SVPWM, and DPWM and adjusting parameters in real time, includes: Based on the PWM control strategy, a sinusoidal pulse width modulation (SPWM) algorithm is implemented and the carrier frequency is adaptively adjusted to generate a basic PWM control signal. Based on the aforementioned basic PWM control signal, the Space Vector Pulse Width Modulation (SVPWM) algorithm is implemented to perform sector judgment and vector decomposition, generating a high-utilization PWM control signal. Based on the high-utilization PWM control signal, the discontinuous pulse width modulation (DPWM) algorithm is implemented to reduce the number of switching operations and generate a low-loss PWM control signal. Based on the low-loss PWM control signal, real-time parameter fine-tuning of modulation depth, switching frequency, and dead time is performed to generate an optimized PWM control signal. Based on the optimized PWM control signal, the power conversion control signal is obtained by performance evaluation and adaptive adjustment through a feedback monitoring network.

[0013] Preferably, the step of achieving system security monitoring and adaptive protection response based on system operating status data through differential privacy protection processing and anomaly detection models includes: Based on system operation status data, electrical parameters, temperature parameters and control parameters are collected to construct a system status dataset, forming the foundation of monitoring data; Based on the aforementioned monitoring data, a differential privacy algorithm is applied to perform sensitive data protection processing to generate privacy-protected monitoring data; Based on the privacy-protected monitoring data, an anomaly detection model is constructed to identify potential risks of over-temperature, over-current, and short circuit, and anomaly detection results and protection strategies are generated. Based on the anomaly detection results and protection strategies, corresponding levels of protection measures are automatically triggered according to the severity of the anomaly, resulting in the system security monitoring and adaptive protection response.

[0014] Preferably, the step of implementing the Space Vector Pulse Width Modulation (SVPWM) algorithm based on the basic PWM control signal to perform sector determination and vector decomposition, and generate a high-utilization PWM control signal, includes: Based on the basic PWM control signal and three-phase voltage command, the three-phase system is converted into a vector space representation through coordinate transformation to establish a voltage vector coordinate system; Based on the voltage vector coordinate system, sector identification is performed to determine the six sector positions where the current vector is located, and sector identification results are obtained. Based on the sector identification results, the duration of action of adjacent basic vectors is calculated and vector synthesis is performed to determine the PWM switching timing. Based on the PWM switching timing, dead time compensation and carrier arrangement optimization are performed to generate switching control pulses; Based on the switching control pulse, the high-utilization PWM control signal is obtained through output filtering and harmonic suppression.

[0015] Preferably, the step of constructing a multi-objective optimization model that comprehensively considers efficiency, temperature rise, and system lifetime based on the impact of each bottleneck on the overall system performance, to obtain the power optimization allocation scheme, includes: Based on the impact of each bottleneck on the overall system performance, an objective function for maximizing energy conversion efficiency is established, thus forming an efficiency optimization objective. Based on the efficiency optimization objective, objective functions for minimizing component temperature rise and maximizing battery cycle life are established, and a set of multi-objective functions is constructed. Based on the aforementioned multi-objective function set, power balance constraints, battery SOC constraints, and temperature constraints are set to establish a constrained optimization problem. Based on the aforementioned constrained optimization problem, a weighted sum method combined with gradient descent algorithm is used to solve it, and a solution set is obtained. Based on the solution set, the power setpoint, operating frequency range, and current limiting control parameters of each stage of the converter are converted to obtain the power optimization allocation scheme.

[0016] Preferably, the step of extracting feature parameters of control complexity, response time requirement, and control accuracy requirement based on the candidate control strategy set, and generating a control strategy feature vector, includes: Based on the set of candidate control strategies, the power change rate, load characteristics, and operating point location are analyzed to calculate the control complexity index. Based on the aforementioned control complexity index, the load change rate, transient requirements, and operating condition transition types are analyzed to quantify the response time requirement parameters. Based on the aforementioned response time requirement parameters, load sensitivity, efficiency gradient, and safety margin are analyzed to determine the control accuracy requirements. Based on the aforementioned control accuracy requirements, the electrical signal noise level, harmonic content, and external interference intensity are evaluated to obtain a noise interference assessment. Based on the noise interference assessment, the data is standardized and mapped to the 0-1 range to obtain the feature vector of the control strategy.

[0017] The present invention also provides a magnetic component for implementing the power conversion control method of the above-mentioned home energy storage system.

[0018] The beneficial effects of this invention are: 1. By adopting a combination design of nanocrystalline materials and multilayer winding structure, a high-efficiency and low-loss magnetic component is constructed, realizing high-efficiency energy transmission over a wide operating range, increasing power density by 30%, and achieving a conversion efficiency of over 96%; 2. By adopting the bottleneck capacity estimation method based on the Vantage point selection algorithm, the power bottleneck point in the system is identified and its capacity is estimated, so as to realize the accurate analysis and optimized control of energy flow and enable the system to obtain the best performance under different operating conditions. 3. By using a PWM control strategy selection mechanism similar to an adaptive hybrid sorting paradigm, the optimal control algorithm is automatically selected based on the operating conditions, and the control parameters are fine-tuned in real time to achieve efficient and stable operation of the power converter across the entire operating range; 4. By using a system monitoring mechanism based on a Local Edge differential privacy algorithm, data privacy can be protected while accurately detecting system anomalies and providing tiered protection responses, thus ensuring the safe and stable operation of the energy storage system; 5. By integrating feature extraction and decision engine into the operating condition identification system, the system operating condition can be identified in real time and accurately, providing a decision basis for adaptive control and adapting to the diverse needs of household electricity peak-valley regulation and emergency backup power. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a power conversion control method for a home energy storage system provided in an embodiment of the present invention; Figure 2 A flowchart for bottleneck identification and capacity estimation provided in an embodiment of the present invention; Figure 3 The flowchart illustrates the implementation of the PWM generation algorithm provided in this embodiment of the invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0022] like Figure 1 As shown, this embodiment of the invention provides a power conversion control method for a home energy storage system, including the following steps: Step S1: Based on the working characteristics of the home energy storage system, collect the system input voltage range, output power requirements and frequency range, and construct magnetic components by selecting nanocrystalline magnetic core materials and designing a multi-layer distributed winding structure.

[0023] In this step, based on the operating characteristics of the home energy storage system, key parameters such as the system input voltage range (100V-450V DC), output power requirement (6kW), and frequency range (20kHz-100kHz) are collected to establish a performance requirement model for magnetic components. Based on this model, nanocrystalline core materials are selected, and their permeability (μ>15000), saturation magnetic induction (Bs>1.2T), and loss characteristics (<3W / kg@100kHz) are analyzed to determine the optimal operating magnetic flux density range (0.3T-0.8T). Based on the characteristics of the nanocrystalline material, a multi-layer distributed winding structure is designed, optimizing the conductor cross-section and arrangement to reduce skin effect and proximity effect, thereby lowering high-frequency copper losses. The multi-layer distributed winding structure uses an interleaved arrangement, ensuring that the current directions of adjacent conductor layers are opposite, effectively canceling the magnetic field and reducing proximity effect losses. The conductor cross-section adopts a flattened design, with a thickness less than the skin depth and an appropriately increased width to maintain sufficient cross-sectional area, thus reducing high-frequency losses caused by the skin effect. Finite element analysis software was used to simulate the temperature rise and loss distribution of magnetic components at different operating points, optimizing the core shape and winding topology to achieve efficient energy transfer over a wide operating range. The simulation analysis considered the effects of current harmonics, temperature variations, and frequency. Through iterative optimization of design parameters, it was ensured that the magnetic components maintained high efficiency and low temperature rise across a wide voltage range of 100V-450V. Samples of the optimized magnetic components were fabricated and tested. Temperature rise tests, efficiency tests, and EMI tests confirmed that their performance indicators met the design requirements across the entire operating range.

[0024] Step S2: Based on the magnetic components, real-time monitoring data of battery status, load power and power grid status are collected through a multi-dimensional data acquisition network, and feature extraction and classification are performed to generate a set of operating condition descriptions containing operating condition types and parameters.

[0025] In this step, based on the workflow of a home energy storage system, a multi-dimensional data acquisition network is designed to collect key parameters such as battery status (SOC, voltage, temperature), load power, and grid status in real time. This network adopts a layered architecture: the bottom layer is the sensor interface layer, the middle layer is the data preprocessing layer, and the top layer is the data integration layer. Battery SOC is calculated in real time using a high-precision coulomb counter with an accuracy of ±1%; battery terminal voltage is acquired through a differential sampling circuit with a sampling rate of 10kHz and a resolution of 12 bits; battery temperature is measured with an NTC thermistor array placed at key locations in the battery module, achieving a measurement accuracy of ±0.5℃. Load power monitoring is achieved through a bidirectional power metering module, employing an isolated Hall current sensor and a high-precision voltage sampling circuit, capable of simultaneously monitoring active and reactive power, with a sampling frequency of 20kHz and a calculation accuracy better than 0.5%. The acquired multi-dimensional data undergoes preprocessing, including filtering, normalization, and feature extraction, to generate a dataset reflecting the system's operating characteristics. Differentiated filtering strategies are employed for different types of sensor data: For high-frequency sampled data such as voltage and current, a Butterworth low-pass filter is used to remove high-frequency noise, with the cutoff frequency set to 1 / 10 of the sampling frequency; for parameters with slow changes such as temperature, a moving average filter is used, with a window size of 16-32 sampling points; for parameters important for sudden change detection (such as short-circuit current), a median filter is used to remove spike noise while preserving signal edge characteristics. A feature extraction module is designed to calculate key feature parameters of the dataset, such as voltage range distribution, power fluctuation rate, load periodicity, and battery charge / discharge state. Voltage range distribution features are obtained through voltage distribution histogram analysis, dividing the voltage range into multiple intervals (such as 100V-200V, 200V-300V, etc.), and statistically analyzing the voltage distribution probability of each interval within different time windows to form a voltage distribution fingerprint. Based on the extracted feature parameters, a working condition classification engine is constructed, combining a finite state machine and a lightweight XGBoost classifier to achieve fast and accurate identification of system working conditions. Finite state machines (FSMs) are used to establish state transition models for system operating conditions, defining several main operating states of a home energy storage system: grid-powered mode, battery discharging mode, battery charging mode, hybrid power supply mode, and fault protection mode. Based on the identification results, an operating condition parameter mapping table is established, generating an operating condition description set containing operating condition types and key parameters, providing a decision-making basis for subsequent control strategies.

[0026] Step S3: Based on the operating condition description set and the preset system physical model, construct a multi-objective optimization model through bottleneck point identification and capacity estimation, and generate a power optimization allocation scheme.

[0027] In this step, the operating condition description set output by the operating condition identification system is received, and combined with the preset system physical model, an energy flow graph for the current operating state is established. The system physical model is a pre-established mathematical model containing all energy transmission paths and conversion nodes, accurately describing the various functional modules in the home energy storage system and their interrelationships, including battery modules, DC / DC converters, DC / AC inverters, grid interfaces, and household loads. Each module has a corresponding mathematical expression describing its energy conversion characteristics; for example, the converter's efficiency curve η=f(Pin,Vin,Vout,T) indicates that efficiency is a function of input power, input voltage, output voltage, and temperature. The energy flow graph uses graph theory, treating each functional module as a node and the energy transmission path as a directed edge, with each edge labeled with attributes such as power magnitude, efficiency, and constraints. Based on the energy flow graph, a bottleneck analysis method similar to the Vantage point selection algorithm is introduced to identify power bottlenecks in the system, such as converter thermal limitations, battery power limitations, or line transmission limitations. The system first calculates the current load factor of each node (such as converters, batteries, lines, etc.), which is the ratio of current power to rated power. Simultaneously, three key indicators are calculated: influence coefficient, sensitivity, and margin. The influence coefficient measures the degree of impact of the node on the overall system performance, derived through topological analysis in the energy flow graph; sensitivity represents the response speed of the system to changes in node performance; and margin represents the remaining capacity of the node before reaching its performance limit. Capacity estimation is performed on identified bottlenecks, establishing bottleneck models and quantifying the impact and limiting factors of each bottleneck on the overall system performance. Differentiated methods are used for capacity estimation depending on the type of bottleneck. For thermally limited bottlenecks (such as converter heat dissipation), the maximum sustainable power is estimated using a thermal model. A second-order thermal equivalent circuit model is used, analogizing heat flow to current and temperature to voltage. By measuring the temperature of key points and the ambient temperature, combined with pre-calibrated thermal resistance and thermal capacity parameters, the maximum power that the component can withstand long-term under conditions not exceeding the temperature limit is calculated. Based on the bottleneck capacity estimation results, a multi-objective optimization model is constructed, comprehensively considering factors such as efficiency, temperature rise, and system lifetime, to calculate the optimal power allocation scheme. Several optimization objectives were defined, including maximizing energy conversion efficiency, minimizing temperature rise of key components, maximizing battery cycle life, optimizing response speed, and maximizing system stability. An improved weighted sum method was used to solve this multi-objective optimization problem: F = w1×f1 + w2×f2 + ... + wn×fn, where fi are the normalized objective functions and wi are the corresponding weights. The optimized power allocation scheme was then converted into specific control target parameters, including the power setpoints, operating frequency ranges, and current limits of each stage of the converter, forming the optimized power allocation scheme.

[0028] Step S4: Based on the power optimization allocation scheme and real-time system status, select the PWM control strategy by constructing a control strategy selection space and running an adaptive decision engine.

[0029] In this step, the received power optimization allocation scheme, combined with the real-time system status, constructs a control strategy selection space, including various PWM modulation strategies and parameter combinations. The system pre-sets multiple PWM modulation strategies, forming a control strategy library, including: basic sinusoidal pulse width modulation (SPWM), space vector pulse width modulation (SVPWM), harmonic injection pulse width modulation (HIPWM), discontinuous pulse width modulation (DPWM), and optimized hybrid modulation strategies. For each modulation strategy, the system designs multiple sets of parameter combinations, forming a parameter space. Key parameters include: carrier frequency (affecting harmonic distribution and switching losses), dead time (affecting shoot-through short-circuit risk and output distortion), modulation depth (affecting DC bus utilization), and overmodulation strategy (how to handle high output demands). The system cross-combines modulation strategies and parameter combinations to form a complete control strategy selection space, reducing the theoretically large combination space to dozens of practically feasible candidate schemes through a pre-screening mechanism. A control strategy feature extraction module is designed to calculate key control-related parameters in real time, such as control complexity, response time requirements, and control accuracy requirements. Control complexity is a key characteristic, reflecting the requirements for control accuracy and flexibility under the current operating conditions. The system calculates the control complexity index by analyzing factors such as power change rate, load characteristics, and operating point location. Response time requirement is another important characteristic. The system quantifies the response time requirement by analyzing load change rate, transient requirements, and operating condition transition types. Based on feature parameters, a decision engine similar to an adaptive hybrid ranking paradigm is constructed, combining a finite state machine (FSM) and a lightweight decision tree to automatically select the optimal control strategy based on the characteristics of the current operating conditions. The FSM handles the high-level logic and state transitions for control strategy selection, defining several main control modes: high-efficiency mode, fast response mode, stable operation mode, low-noise mode, and fault protection mode. Each mode corresponds to different control strategy preferences and parameter ranges. The lightweight decision tree classifier is responsible for selecting specific control strategies and parameter combinations within the mode range defined by the FSM, based on detailed feature parameters. The decision engine has adaptive learning capabilities. The system records the actual performance indicators of different strategies under various operating conditions, such as efficiency, temperature rise, response speed, and stability. By analyzing this historical data, the decision rules are continuously adjusted, and the strategy selection logic is gradually optimized. Based on the control mode and strategy selected by the decision engine, smooth transition and parameter optimization are performed to obtain the final PWM control strategy.

[0030] Step S5: Based on the PWM control strategy, various PWM generation algorithms such as SPWM, SVPWM and DPWM are implemented and real-time parameters are adjusted to output a power conversion control signal.

[0031] In this step, based on the PWM control strategy, a sinusoidal pulse width modulation (SPWM) algorithm is implemented and the carrier frequency is adaptively adjusted to generate a basic PWM control signal. SPWM is the most basic implementation, generating a PWM signal by comparing a sinusoidal modulated wave and a triangular carrier. The SPWM is improved by introducing an adaptive mechanism for modulator depth and carrier frequency, providing a stable and reliable control foundation under simple operating conditions. The carrier frequency is dynamically adjusted according to load conditions and efficiency requirements. Under light load conditions, the switching frequency is appropriately increased to improve dynamic response and output quality, while under heavy load conditions, the switching frequency is decreased to reduce switching losses and heat generation. The frequency adjustment range is 20kHz-100kHz, varying in a step manner to avoid generating unpredictable subharmonics. Based on the basic PWM control signal, a space vector pulse width modulation (SVPWM) algorithm is implemented to perform sector judgment and vector decomposition, generating a high-utilization PWM control signal. SVPWM is a more advanced modulation technique that treats the three phases as a vector space, generating a PWM signal by rationally arranging the application time of the basic voltage vector. Optimized for home energy storage applications, the sector determination algorithm is simplified, a lookup table method is used to accelerate time calculation, and dynamic dead time adjustment is introduced, significantly improving computational efficiency and control accuracy. SVPWM has high DC bus utilization (up to 0.907) and good harmonic characteristics, making it suitable for operating conditions with high output quality requirements. Based on the high-utilization PWM control signal, a discontinuous pulse width modulation (DPWM) algorithm is implemented to reduce the number of switching operations and generate a low-loss PWM control signal. DPWM significantly reduces the number of switching operations and switching losses by pausing the switching action of one bridge arm in each cycle. Several DPWM variants are implemented, including DPWM1 (60° non-conducting), DPWM2 (30° non-conducting), and DPWM3 (synchronous non-conducting), and the optimal variant is automatically selected based on power factor and load characteristics. DPWM is particularly suitable for high-power operating conditions, reducing switching losses by 15%-30%, but with a slight sacrifice in output harmonic performance. Based on the low-loss PWM control signal, real-time parameter fine-tuning of modulation depth, switching frequency, and dead time is performed to generate an optimized PWM control signal. Real-time parameter fine-tuning is performed on the selected control strategy, optimizing key parameters such as modulation depth, switching frequency, and dead time based on feedback data. Modulation depth is the primary adjustment parameter, directly affecting the output voltage amplitude and DC bus utilization. Closed-loop control automatically adjusts the modulation depth to ensure accurate output voltage tracking of the setpoint while avoiding nonlinear distortion caused by overmodulation. Based on the optimized PWM control signal, performance evaluation and adaptive adjustment are performed through a feedback monitoring network to obtain the power conversion control signal. A complete feedback monitoring network is established to collect key parameters in real time during the power conversion process, including input / output voltage and current, converter temperature, switching waveforms, and output harmonics. After preprocessing, this feedback data is compared with expected performance indicators to generate performance deviation indicators.Performance metrics include efficiency (the difference between actual measured values ​​and theoretical expectations), harmonic distortion (the difference between measured THD values ​​and target values), dynamic response (the difference between actual response time and target time), and temperature rise rate (the difference between actual temperature rise and prediction model).

[0032] Step S6: Based on system operating status data, through differential privacy protection processing and anomaly detection models, realize system security monitoring and adaptive protection response.

[0033] In this step, operating status data from each node of the system is received, including electrical parameters, temperature parameters, and control parameters, to construct a system status dataset. Electrical parameters include voltage, current, power, and frequency of each node; temperature parameters include battery temperature, converter temperature, radiator temperature, and ambient temperature; control parameters include PWM parameters, control loop status, and system operating mode. A distributed data acquisition architecture is adopted, aggregating data from each node to the central processing unit via a high-speed data bus. The sampling frequency is dynamically adjusted according to the parameter change rate, using a high sampling rate for rapidly changing parameters (such as current) and a low sampling rate for slowly changing parameters (such as temperature) to optimize data acquisition efficiency. Based on a concept similar to the Local Edge differential privacy algorithm, sensitive data is processed for privacy protection, balancing data availability and privacy protection requirements. Differential privacy is a data protection technology that adds carefully designed noise to the raw data to ensure that the data analysis results do not reveal individual information. The system employs differentiated privacy protection strategies for different types of data: For highly sensitive data (such as user power consumption patterns), larger noise is added; for low-sensitivity data (such as system temperature), smaller noise is added; and for critical safety parameters (such as overcurrent protection thresholds), no noise is added to ensure safety. A hierarchical protection mechanism is designed, setting different protection strategies for parameters of different importance levels to ensure real-time monitoring and alarms for critical parameters (such as battery temperature and current overload). The system divides protection parameters into three levels: Level 1 parameters (such as short-circuit current and battery overtemperature) relate to system safety and require millisecond-level response; Level 2 parameters (such as overload and slight overtemperature) affect system performance and require second-level response; and Level 3 parameters (such as efficiency degradation and performance deterioration) affect system lifespan and require minute-level response. An anomaly detection model is constructed to identify potential system anomalies, such as overtemperature, overcurrent, and short-circuit risks, based on processed data, and generates corresponding protection strategies. The system uses a combination of multiple anomaly detection algorithms, including rule-based threshold detection, statistical outlier detection, and machine learning-based pattern recognition. An adaptive protection response mechanism is implemented, automatically triggering different levels of protection measures based on the severity of the anomaly, ranging from parameter adjustments to emergency power cuts, to ensure the safe and stable operation of the system. For minor anomalies, control parameters may be adjusted to restore normal operation; for moderate anomalies, power may be limited or the operating mode may be switched; for severe anomalies, emergency protection actions such as rapid power cuts or isolation of faulty components will be executed.

[0034] Step S1 specifically includes: Step S11: Based on the working characteristics of the home energy storage system, analyze the magnetic permeability, saturation magnetic induction intensity and loss characteristics of the nanocrystalline material, and determine the working magnetic flux density range of 0.3T-0.8T; Step S12: Based on the working magnetic flux density range, design a multi-layer distributed winding structure and optimize the conductor cross-section and arrangement to obtain an optimized winding structure with low skin effect and proximity effect; Step S13: Based on the optimized winding structure with low skin effect and proximity effect, the temperature rise and loss distribution at different operating points are simulated using finite element analysis software to obtain the magnetic component.

[0035] In step S11, based on the operating characteristics of the home energy storage system, key parameters such as the system input voltage range (100V-450VDC), output power requirement (6kW), and frequency range (20kHz-100kHz) are collected to establish a performance requirement model for the magnetic components. Based on the performance requirement model, nanocrystalline magnetic core materials are selected, and their permeability (μ>15000), saturation magnetic induction (Bs>1.2T), and loss characteristics (<3W / kg@100kHz) are analyzed to determine the optimal operating magnetic flux density range (0.3T-0.8T).

[0036] In step S12, based on the properties of nanocrystalline materials, a multi-layer distributed winding structure is designed, optimizing the conductor cross-section and arrangement to reduce skin effect and proximity effect, thereby lowering high-frequency copper losses. The multi-layer distributed winding structure adopts an interleaved arrangement, ensuring that the current directions of adjacent conductor layers are opposite, effectively canceling the magnetic field and reducing proximity effect losses. The conductor cross-section adopts a flattened design, with a thickness less than the skin depth and a width appropriately increased to maintain sufficient cross-sectional area, thus reducing high-frequency losses caused by the skin effect.

[0037] In step S13, finite element analysis software is used to simulate the temperature rise and loss distribution of the magnetic components at different operating points, optimizing the core shape and winding topology to achieve efficient energy transfer over a wide operating range. The simulation analysis considers current harmonics, temperature variations, and frequency effects. Through iterative optimization of design parameters, it ensures that the magnetic components maintain high efficiency and low temperature rise across a wide voltage range of 100V-450V. Samples of the optimized magnetic components are fabricated and tested. Temperature rise tests, efficiency tests, and EMI tests confirm that their performance indicators meet the design requirements across the entire operating range.

[0038] Specifically, the optimized magnetic components include the following key structural features: the magnetic core adopts an E-type or U-type nanocrystalline laminate structure, with the core column cross-sectional area increased by 15-20% compared to traditional designs, to reduce operating magnetic flux density and saturation risk; the air gap adopts a distributed design, decomposing the traditional single large air gap into 3-5 small air gaps evenly distributed in the magnetic circuit, reducing leakage flux and edge effect losses by approximately 25%; the winding adopts a PCB structure or flat copper foil structure, with the thickness controlled within the skin depth (δ) range, and for a 50kHz operating frequency, the copper foil thickness is controlled at 0.2-0.3mm; the winding adopts an interleaved parallel connection technology, decomposing the single-turn high-current winding into a multi-turn parallel structure, with each layer of winding accommodating a maximum of 2-3 turns, and the current directions of adjacent layers being opposite, significantly reducing proximity effect losses by approximately 30-40%. This optimized magnetic component design can maintain an efficiency of over 95% within a wide voltage range of 100V-450V, with temperature rise controlled below 40K, and a power density increase of 30% compared to traditional designs, reaching 5-6kW / L.

[0039] Step S2 specifically includes: Step S21: Based on the magnetic components and system operation data, collect the battery SOC, voltage, temperature and load power operation monitoring data through a multi-dimensional data acquisition network to generate an original dataset reflecting the system operating characteristics; Step S22: Filter, normalize, and perform preliminary feature extraction on the original dataset to generate a structured preprocessed dataset; Step S23: Based on the structured preprocessed dataset, calculate the voltage range distribution, power fluctuation rate, load periodicity, and battery charge / discharge state characteristic parameters to form an operating condition feature vector; Step S24: Based on the operating condition feature vector, the operating condition is identified using a finite state machine and a lightweight XGBoost classifier to obtain the operating condition description set containing the operating condition type and parameters.

[0040] In step S21, based on the workflow of the home energy storage system, a multi-dimensional data acquisition network is designed to collect key parameters such as battery status (SOC, voltage, temperature), load power, and grid status in real time. This network adopts a layered architecture: the bottom layer is the sensor interface layer, the middle layer is the data preprocessing layer, and the top layer is the data integration layer. Battery SOC is calculated in real time using a high-precision coulomb counter, achieving an accuracy of ±1%; battery terminal voltage is acquired through a differential sampling circuit with a sampling rate of 10kHz and a resolution of 12 bits; battery temperature is measured with an NTC thermistor array placed at key locations within the battery module, achieving a measurement accuracy of ±0.5℃. Load power monitoring is achieved through a bidirectional power metering module, employing an isolated Hall current sensor and a high-precision voltage sampling circuit, capable of simultaneously monitoring active and reactive power, with a sampling frequency of 20kHz and a calculation accuracy better than 0.5%.

[0041] In step S22, the collected multidimensional data is preprocessed, including filtering, normalization, and feature extraction, to generate a dataset reflecting the system's operating characteristics. The filtering process employs differentiated strategies for different types of sensor data: for high-frequency sampling data such as voltage and current, a Butterworth low-pass filter is applied to remove high-frequency noise, with the cutoff frequency set to 1 / 10 of the sampling frequency; for parameters that change slowly, such as temperature, a moving average filter is used, with a window size of 16-32 sampling points; for parameters important for sudden change detection (such as short-circuit current), a median filter is used to remove spike noise while preserving signal edge characteristics. The normalization process unifies parameters of different dimensions to a standard range, using a nonlinear normalization method, designing different normalization functions for different parameters based on their physical characteristics.

[0042] In step S23, a feature extraction module is designed to calculate key feature parameters of the dataset, such as voltage range distribution, power fluctuation rate, load periodicity, and battery charge / discharge state. Voltage range distribution features are obtained through voltage distribution histogram analysis, dividing the voltage range into multiple intervals (e.g., 100V-200V, 200V-300V, etc.), and statistically analyzing the voltage distribution probability of each interval within different time windows to form a voltage distribution fingerprint. Power fluctuation rate features are calculated by analyzing the time-domain and frequency-domain characteristics of the load power. In the time domain, the short-time power change rate and power jump frequency are calculated, and in the frequency domain, the main frequency components of power fluctuation are extracted using Fast Fourier Transform, and the power spectral entropy value is calculated to quantify the complexity of the fluctuation. Load periodicity features are extracted through autocorrelation analysis and wavelet transform. The autocorrelation function is used to detect repetitive patterns in the load, while the wavelet transform can identify periodic components at different time scales.

[0043] In step S24, based on the extracted feature parameters, a working condition classification engine is constructed. Combining a finite state machine (FSM) and a lightweight XGBoost classifier, rapid and accurate identification of system working conditions is achieved. The FSM is used to establish the state transition model of the system working conditions, defining several main operating states of the home energy storage system: grid power supply mode, battery discharge mode, battery charging mode, hybrid power supply mode, and fault protection mode. The lightweight XGBoost classifier is used to handle finer-grained working condition identification. This classifier uses tree depth constraints (maximum depth 6) and feature subsampling techniques to reduce model complexity while maintaining high recognition accuracy. The combination of the two techniques employs a hierarchical identification strategy: the FSM first determines the major state categories of the system, limiting the possible range of working conditions; then, XGBoost performs fine-grained classification within this limited range, outputting specific working condition types. Based on the identification results, a working condition parameter mapping table is established, generating a working condition description set containing working condition types and key parameters, providing a decision-making basis for subsequent control strategies.

[0044] The operating condition description set is a data structure containing multi-dimensional information, specifically including the following core components: operating condition type identifiers (such as peak discharge, off-peak charging, equalization float charging, etc.); estimated operating condition duration, derived based on historical data and current trend analysis; power demand parameters, including average power, peak power, and power change rate; battery parameter set, including current SOC, available capacity, temperature distribution, and health status; grid parameter set, including grid voltage, frequency stability, and available capacity; and priority indicators, indicating the performance objectives that the system should prioritize (such as efficiency, lifetime, or response speed).

[0045] The operating condition description set is transmitted between system modules via a standardized JSON format, employing an incremental update mechanism. A new description set is only generated when the operating condition changes significantly (parameter changes exceed a preset threshold), reducing the system communication burden. To address the uncertainty of operating condition identification, the description set also includes a confidence score (0-100%) for operating condition identification. When the confidence score is below 70%, multiple alternative operating condition descriptions are generated simultaneously, and a weight coefficient is assigned to each description for robust decision-making by subsequent modules.

[0046] Operating condition description sets demonstrate significant advantages in practical applications, especially under boundary conditions such as power grid fluctuations or load abrupt changes. Accurate operating condition descriptions enable the system to adjust control strategies 3-5 seconds in advance, avoiding over-response and oscillation problems common in traditional systems, and improving system stability and user experience.

[0047] The lightweight XGBoost classifier is constructed and trained as follows: First, a training set is built based on historical operating data, including feature vector samples labeled with different operating conditions. The importance of features is evaluated and dimensionality is reduced, selecting the most significant subset of features. Classifier parameters are set, including a maximum tree depth limit of 6 layers, a learning rate of 0.1, a feature sampling ratio of 0.8, and an L1 regularization coefficient of 0.5, to balance model complexity and prediction accuracy. A 5-fold cross-validation method is used to train the model and evaluate its accuracy, ensuring an accuracy of over 95% under different operating conditions. Finally, the trained model is deployed to the control system, and a periodic incremental learning mechanism is designed to continuously optimize model performance using newly collected operating data. This classifier can identify 10 typical operating conditions, including normal charging, fast charging, normal discharging, peak discharge support, grid fault emergency, and low-power standby, providing accurate operating condition judgment for subsequent control strategy selection.

[0048] like Figure 2 As shown, step S3 specifically includes: Step S31: Based on the operating condition description set and the preset system physical model, establish an energy flow diagram under the current operating state to form an energy transmission network topology; Step S32: Based on the energy transmission network topology, use the bottleneck analysis method to calculate the load rate, influence coefficient and sensitivity of each node, and identify the system power bottleneck point; Step S33: Estimate the capacity of the system power bottleneck and establish a bottleneck model to quantify the impact of each bottleneck on the overall system performance; Step S34: Based on the impact of each bottleneck on the overall system performance, construct a multi-objective optimization model that comprehensively considers efficiency, temperature rise and system lifetime to obtain the power optimization allocation scheme.

[0049] In step S31, the operating condition description set output by the operating condition identification system is received, and an energy flow graph under the current operating state is established by combining it with the preset system physical model. The system physical model is a pre-established mathematical model that includes all energy transmission paths and conversion nodes, accurately describing the various functional modules in the home energy storage system and their interrelationships, including battery modules, DC / DC converters, DC / AC inverters, grid interfaces, and home loads. Each module has a corresponding mathematical expression describing its energy conversion characteristics. For example, the efficiency curve of the converter, η=f(Pin,Vin,Vout,T), indicates that the efficiency is a function of input power, input voltage, output voltage, and temperature. The energy flow graph uses graph theory representation, treating each functional module as a node and the energy transmission path as a directed edge, with each edge labeled with attributes such as power magnitude, efficiency, and constraints.

[0050] The system physical model is a multi-level mathematical model library that covers the static characteristics and dynamic behavior of various functional modules of a home energy storage system. Its establishment process includes the following steps: First, characteristic data of each component (converter, battery, wiring) at different operating points are obtained through precise measurements; then, component characteristic models are established using piecewise polynomial fitting or neural network techniques, such as the converter efficiency model η = f(Pin, Vin, Vout, T), which accurately describes the relationship between efficiency and input power, input voltage, output voltage, and temperature; next, the topological connections between components are established to form a complete system network structure; finally, the model is calibrated and verified through numerical simulation and actual testing to ensure that the model prediction error is controlled within ±2%.

[0051] The model library contains the following core sub-models: power electronic converter models (including switching loss models, conduction loss models, and magnetic component loss models); battery models (including electrochemical dynamic models, thermal behavior models, and aging models); thermal models (using second-order RC equivalent circuits to represent the thermal characteristics of each component); and load models (describing the power characteristics and dynamic response of typical household loads).

[0052] The system physical model supports two operating modes: offline analysis mode, used for system design and parameter optimization, employing a high-precision, complex model; and online operation mode, used for real-time control decision-making, employing a simplified model to meet real-time computing requirements. The simplified model, through principal component analysis and sensitivity analysis, retains the factors most significantly affecting system performance, achieving a computation speed 5-10 times faster than the full model while maintaining a prediction accuracy of over 95%.

[0053] In step S32, based on the energy flow graph, a bottleneck analysis method similar to the Vantage point selection algorithm is introduced to identify power bottlenecks in the system, such as converter thermal limiting points, battery power limiting points, or line transmission limiting points. The system first calculates the current load rate of each node (e.g., converter, battery, line, etc.), which is the ratio of current power to rated power. Simultaneously, three key indicators are calculated: influence coefficient, sensitivity, and margin. The influence coefficient measures the degree of impact of the node on the overall system performance, derived through topology analysis in the energy flow graph; sensitivity represents the response speed of the system to changes in node performance; and margin represents the remaining distance of the node from its performance limit. Based on these indicators, the system uses a weighted scoring method to calculate the bottleneck index of each node: BI = α × load rate + β × influence coefficient + γ × sensitivity - δ × margin, where α, β, γ, and δ are preset weight coefficients based on system characteristics. These weight coefficients are determined based on system component characteristics and operating priorities, with a typical configuration of α=0.4, β=0.3, γ=0.2, and δ=0.1. For power electronic devices with limited heat dissipation, the β value is appropriately increased to 0.4-0.5; for battery systems, the δ value is increased to 0.2-0.3 to place greater emphasis on safety margins. The system also automatically adjusts these weighting coefficients every 3-6 months based on statistical analysis of operational data to adapt to component aging and changes in usage patterns. In practical applications, bottlenecks identified using this method can predict potential bottlenecks 15-30 minutes earlier than traditional methods based solely on load rate, significantly improving system operational stability. The node with the highest bottleneck index is identified as the primary bottleneck, and the system also identifies secondary bottlenecks as alternative optimization targets.

[0054] In step S33, capacity estimation is performed on the identified bottlenecks, a bottleneck model is established, and the impact and limiting factors of each bottleneck on the overall system performance are quantified. Different methods are used for capacity estimation depending on the type of bottleneck. For thermally limited bottlenecks (such as converter heat dissipation), the system estimates the maximum sustainable power using a thermal model. A second-order thermal equivalent circuit model is used, analogizing heat flow to current and temperature to voltage. By measuring the temperature at key points and the ambient temperature, combined with pre-calibrated thermal resistance and thermal capacity parameters, the maximum power that the component can carry over the long term under conditions not exceeding the temperature limit is calculated. For electrically limited bottlenecks (such as the maximum battery discharge current), a nonlinear model is established based on the battery chemical characteristics and SOC state to estimate the actual current carrying capacity under current conditions. A dynamic bottleneck model is also established, considering not only instantaneous capacity but also the time dimension. This relationship is described by time-varying parameters: C(t) = Cmax × f(t, Tenv,history), where C(t) is the capacity estimate at time t, Cmax is the theoretical maximum capacity, and f() is an adjustment function considering time, ambient temperature, and historical load.

[0055] In step S34, based on the bottleneck capacity estimation results, a multi-objective optimization model is constructed, comprehensively considering factors such as efficiency, temperature rise, and system lifetime to calculate the optimal power allocation scheme. The system defines multiple optimization objectives, including: maximizing energy conversion efficiency, minimizing the temperature rise of key components, maximizing battery cycle life, optimizing response speed, and maximizing system stability. The system establishes detailed mathematical models to describe each objective function. For example, the energy conversion efficiency objective function is: η_total = Pout / Pin = Pout / (Pout + Ploss), where Ploss is the total loss, composed of converter losses, line losses, etc., and each loss is a function of multiple variables such as power, voltage, and temperature. The system sets a series of constraints to ensure that the optimization results meet physical and safety limitations, including power balance constraints, battery SOC constraints, temperature constraints, and current constraints. An improved weighted sum method is used to solve this multi-objective optimization problem: F = w1×f1 + w2×f2 + ... + wn×fn, where fi are the normalized objective functions, and wi are the corresponding weights. The solution process employs an improved gradient descent algorithm combined with simulated annealing to avoid getting trapped in local optima. The optimized power allocation scheme is then converted into specific control target parameters, including the power setpoints, operating frequency ranges, and current limits of each stage of the converter, forming the optimized power allocation scheme.

[0056] Step S34 specifically includes: Step S341: Based on the impact of each bottleneck on the overall system performance, establish an objective function to maximize energy conversion efficiency, thus forming an efficiency optimization objective; Step S342: Based on the efficiency optimization objective, establish objective functions for minimizing component temperature rise and maximizing battery cycle life, and construct a set of multi-objective functions; Step S343: Based on the multi-objective function set, set power balance constraints, battery SOC constraints, and temperature constraints to establish a constrained optimization problem; Step S344: Based on the constrained optimization problem, the weighted sum method combined with the gradient descent algorithm is used to solve the problem and obtain the solution set; Step S345: Based on the solution set, convert it into the power setpoint, operating frequency range and current limiting control parameters of each stage of the converter to obtain the power optimization allocation scheme.

[0057] In step S341, an objective function for maximizing energy conversion efficiency is established based on the impact of each bottleneck on the overall system performance. This objective function considers the efficiency of all energy conversion stages in the system, including the efficiency of the DC / DC converter, the DC / AC inverter, and the line transmission efficiency. The efficiency objective function is expressed as: η_total = Pout / Pin = Pout / (Pout + Ploss), where Ploss is the total loss, composed of the losses of each stage. The loss of each stage is a function of multiple variables; for example, converter losses include switching losses, conduction losses, and magnetic component losses, which are closely related to the operating point (voltage, current, frequency).

[0058] In step S342, based on the efficiency optimization objective, objective functions for minimizing component temperature rise and maximizing battery cycle life are further established, constructing a set of multi-objective functions. The temperature rise objective function, based on a thermal model, predicts the temperature rise curves at key points under a given power allocation scheme, in the form: ΔT = f(P, t, Tamb), where P is power, t is time, and Tamb is ambient temperature. The battery life objective function, based on the rainflow counting method and a life model, calculates the impact of different discharge strategies on battery cycle life, expressed as: L = g(DOD, I, T, cycles), where DOD is depth of discharge, I is charge / discharge current, T is battery temperature, and cycles is the number of cycles.

[0059] In step S343, based on a multi-objective function set, power balance constraints, battery SOC constraints, and temperature constraints are set to establish a constrained optimization problem. The power balance constraint ensures that the system input power equals the output power plus losses: Pin = Pout + Ploss; the battery SOC constraint ensures that the battery operates within a safe range: SOCmin ≤ SOC ≤ SOCmax; the temperature constraint ensures that the temperature of each component does not exceed the safe limit: T ≤ Tmax; and the current constraint ensures that the current of each component does not exceed the rated value: I ≤ Irated. Furthermore, the bottleneck capacity identified in the previous step is used as a core constraint to ensure that the optimization result does not exceed the actual carrying capacity of the system.

[0060] In step S344, based on the constrained optimization problem, a weighted sum method combined with gradient descent is used to solve the problem and obtain the solution set. The weighted sum method transforms the multi-objective optimization problem into a single-objective problem: F = w1×f1 + w2×f2 + ... + wn×fn, where fi are the normalized objective functions and wi are the corresponding weights. The weights are set using an adaptive method, dynamically adjusted according to the current operating conditions and user preferences. The solution process uses an improved gradient descent algorithm combined with simulated annealing to avoid getting trapped in local optima. Considering the computational resource limitations of the embedded system, the algorithm is designed with an early termination condition, outputting the results promptly when a sufficiently good solution is reached or the computation time limit is exceeded, ensuring the real-time performance of the control.

[0061] The weighted sum method integrates multiple objective functions into a single objective function through weighting coefficients: F = w1×f_efficiency + w2×f_temperature + w3×f_battery_life + w4×f_response, where the weighting coefficients are automatically adjusted according to the current operating conditions. For example, when the battery SOC is low (<30%), w3 increases to 0.4-0.5; when the ambient temperature is high (>35°C), w2 increases to 0.3-0.4; and when the load fluctuates rapidly, w4 increases to 0.2-0.3.

[0062] The optimization problem is solved using an improved gradient descent algorithm combined with simulated annealing to avoid getting trapped in local optima. The solution process includes: initialization—randomly generating 10 initial solutions that satisfy the constraints; iterative optimization—performing gradient descent iterations on each initial solution, with the iteration step size dynamically adjusted according to the rate of change of the objective function, and a maximum of 100 iterations; simulated annealing—performing a random perturbation every 10 iterations, with the perturbation amplitude gradually decreasing as the iterations progress, and an initial acceptance probability of 0.3; convergence judgment—stopping the computation when the objective function improves by less than 0.1% after 5 consecutive iterations or when the maximum number of iterations is reached.

[0063] To meet real-time control requirements, the system also implements an early termination mechanism, setting a maximum calculation time of 200ms to ensure timely adjustment of the power allocation strategy in the event of sudden changes in battery status or load. This multi-objective optimization method achieves a good balance between battery cycle life, system efficiency, and response speed. Compared with single-objective optimization, battery cycle life is improved by 15-20%, while keeping the system efficiency loss below 0.5%.

[0064] In step S345, based on the solution set obtained from the solution, the abstract optimization results are converted into specific control parameters such as the power setpoint, operating frequency range, and current limit of each stage of the converter, forming a power optimization allocation scheme. For DC / DC bidirectional converters, the control target parameters include: target output voltage (accurate to 0.1V), current limit value (including the instantaneous maximum value and continuous steady-state value during the dynamic response process), power ramp rate (i.e., the maximum rate of power change to prevent system instability caused by sudden changes), and operating frequency range (selecting the optimal switching frequency based on the efficiency curve). For DC / AC inverters, these include output voltage amplitude, frequency, phase (synchronized with the grid in grid-connected mode), harmonic control parameters, and target power factor value. The system also generates detailed transient process control parameters to ensure a smooth transition when the operating conditions change and a switch from one power allocation scheme to another is required.

[0065] Step S4 specifically includes: Step S41: Based on the power optimization allocation scheme and the real-time system status, construct a control strategy selection space containing multiple PWM modulation strategies and parameter combinations to form a candidate control strategy set; Step S42: Based on the candidate control strategy set, extract the feature parameters of control complexity, response time requirement and control accuracy requirement, and generate control strategy feature vector; Step S43: Based on the control strategy feature vector, construct an adaptive decision engine using a finite state machine and a lightweight decision tree classifier to determine the control mode and strategy; Step S44: Based on the control mode and strategy, perform smooth transition and parameter optimization processing to obtain the PWM control strategy.

[0066] In step S41, the received power optimization allocation scheme, combined with the real-time system status, constructs a control strategy selection space, which includes various PWM modulation strategies and parameter combinations. Multiple PWM modulation strategies are preset, forming a control strategy library, including: basic sinusoidal pulse width modulation (SPWM), space vector pulse width modulation (SVPWM), harmonic injection pulse width modulation (HIPWM), discontinuous pulse width modulation (DPWM), and optimized hybrid modulation strategies. For each modulation strategy, multiple parameter combinations are designed, forming a parameter space. Key parameters include: carrier frequency (affecting harmonic distribution and switching losses), dead time (affecting shoot-through short-circuit risk and output distortion), modulation depth (affecting DC bus utilization), and overmodulation strategy (how to handle high output demands). The system cross-combines modulation strategies and parameter combinations to form a complete control strategy selection space, and through a pre-screening mechanism, reduces the theoretically large combination space to dozens of practically feasible candidate schemes.

[0067] In step S42, a control strategy feature extraction module is designed to calculate key control-related parameters in real time, such as control complexity, response time requirements, and control accuracy requirements. Control complexity is a key feature that reflects the requirements for control accuracy and flexibility under the current operating conditions. The system calculates the control complexity index by analyzing factors such as power change rate, load characteristics, and operating point location.

[0068] Response time requirements are another important characteristic. These requirements are quantified by analyzing the rate of load change, transient requirements, and types of operating condition transitions. Control accuracy requirements reflect the system's need for stability and accuracy of output parameters. The system determines control accuracy requirements by analyzing factors such as load sensitivity, efficiency gradient at the current operating point, and safety margin.

[0069] It also analyzes the noise level, harmonic content, and external interference intensity of electrical signals to assess the noise conditions of the current environment. These characteristic parameters are standardized and mapped to the 0-1 range to form a feature vector that comprehensively describes the characteristics of the current control task.

[0070] Step S42 specifically includes: Step S421: Based on the candidate control strategy set, analyze the power change rate, load characteristics and operating point location, and calculate the control complexity index; Step S422: Based on the control complexity index, analyze the load change rate, transient requirements, and operating condition transition type, and quantify the response time requirement parameters; Step S423: Based on the response time requirement parameters, analyze the load sensitivity, efficiency gradient, and safety margin to determine the control accuracy requirement indicators; Step S424: Based on the control accuracy requirements, evaluate the electrical signal noise level, harmonic content, and external interference intensity to obtain a noise interference assessment. Step S425: Based on the noise interference assessment, perform standardization processing and map to the 0-1 range to obtain the feature vector of the control strategy.

[0071] In step S421, based on the candidate control strategy set, the power change rate, load characteristics, and operating point location are analyzed to calculate the control complexity index. The power change rate is obtained by calculating the percentage change in power per unit time, reflecting the dynamic characteristics of the system load. Load characteristics include load type (such as resistive, inductive, or nonlinear load) and load balance, determined by analyzing the current waveform and power factor. The operating point location represents the relative position of the current system operating point on the efficiency curve or other performance curves, calculated by comparing the deviation between the current operating parameters and the optimal parameters. These factors combined form the control complexity index; the higher the index, the higher the requirements on the control system under the current operating condition.

[0072] In step S422, based on the control complexity index, the load change rate, transient requirements, and operating condition transition types are further analyzed to quantify the response time requirement parameters. The load change rate is obtained by calculating the first and second derivatives of the power change, reflecting the speed and acceleration of the load change. Transient requirements include voltage / current overshoot limits and settling time requirements, which are determined by analyzing system performance indicators. The operating condition transition type describes how the system switches from one operating state to another, such as a smooth transition or a rapid switch. These factors together form the response time requirement parameter; the higher this parameter, the faster the system's control response speed is required.

[0073] In step S423, based on the response time requirement parameters, load sensitivity, efficiency gradient, and safety margin are analyzed to determine the control accuracy requirement index. Load sensitivity represents the load's sensitivity to power quality, determined by analyzing the load type and importance; the efficiency gradient represents the slope of the efficiency curve at the current operating point, obtained by calculating the partial derivative of efficiency with respect to the control parameters; the safety margin represents the distance between the system's current operating point and the safety boundary, calculated by comparing the current parameters with the limit values. These factors combined form the control accuracy requirement index; the higher the index, the more precise the control strategy required for the system.

[0074] In step S424, based on the control accuracy requirements, the electrical signal noise level, harmonic content, and external interference intensity are evaluated to obtain a noise interference status assessment. The electrical signal noise level is obtained by calculating the signal-to-noise ratio; the harmonic content is calculated using Fourier analysis to determine the total harmonic distortion (THD); and the external interference intensity is determined by monitoring external factors such as power grid fluctuations and load abrupt changes. These factors combined form the noise interference status assessment. A higher assessment indicates more severe environmental noise interference, requiring a stronger anti-interference capability from the system.

[0075] In step S425, based on the noise interference assessment, all feature parameters are standardized and mapped to the 0-1 range to obtain the control strategy feature vector. The standardization process uses a nonlinear mapping function to ensure the comparability of different feature parameters while preserving their original distribution characteristics. The final feature vector includes dimensions such as control complexity, response time requirements, control accuracy requirements, and noise interference conditions, comprehensively describing the characteristics of the current control task and providing a decision-making basis for subsequent control strategy selection.

[0076] In step S43, based on feature parameters, a decision engine similar to an adaptive hybrid ranking paradigm is constructed. Combining a finite state machine (FSM) and a lightweight decision tree, it automatically selects the optimal control strategy according to the characteristics of the current operating conditions. The FSM handles the high-level logic and state transitions for control strategy selection, defining several main control modes: high-efficiency mode, fast-response mode, stable operation mode, low-noise mode, and fault protection mode. Each mode corresponds to different control strategy preferences and parameter ranges. The lightweight decision tree classifier is responsible for selecting specific control strategies and parameter combinations within the mode range defined by the FSM, based on detailed feature parameters. The decision engine has adaptive learning capabilities; the system records the actual performance indicators of different strategies under various operating conditions, such as efficiency, temperature rise, response speed, and stability. By analyzing this historical data, the decision rules are continuously adjusted, and the strategy selection logic is gradually optimized.

[0077] The adaptive learning mechanism of the decision engine is based on reinforcement learning. The system records the performance evaluation results after each control strategy selection, including efficiency, dynamic response, and stability metrics, and constructs a strategy-performance mapping database. This database initially contains 1000 records and uses a sliding window update strategy, retaining the latest 800 records and the 200 most representative historical records. Decision rules are optimized every 200 hours of operation, using a weighted performance scoring method to adjust the splitting threshold of the decision tree and the strategy selection of leaf nodes. The weighting coefficients are dynamically adjusted according to user preferences; for example, in high-efficiency mode, the efficiency metric weight is 0.6, stability is 0.3, and dynamic response is 0.1; while in fast-response mode, they are 0.2, 0.3, and 0.5 respectively. Through this adaptive learning mechanism, after 3-6 months of operation, a personalized control strategy library tailored to specific hardware characteristics and user habits can be formed, improving efficiency by an average of 2-3 percentage points compared to the initial strategy and significantly improving dynamic response performance.

[0078] In step S44, based on the control mode and strategy selected by the decision engine, a smooth transition and parameter optimization process is performed to obtain the final PWM control strategy. The decision engine implements a smooth transition mechanism to avoid frequent switching of control strategies. When changes in operating conditions cause changes in the optimal strategy, the switching cost and expected benefits are evaluated, and a switch is only executed when the benefits significantly outweigh the costs. When deciding on a switching strategy, the system plans a smooth transition path to ensure gradual changes in control parameters and avoid system instability caused by abrupt changes. Furthermore, the selected control strategy is optimized by fine-tuning control parameters such as modulation depth, carrier frequency, and dead time according to the specific characteristics of the current operating conditions, making the control strategy more adaptable to the current operating conditions and improving system performance.

[0079] The smooth transition mechanism employs a gradual parameter adjustment algorithm to avoid system instability caused by control strategy switching. The specific implementation method is as follows: First, assess the parameter differences between the current strategy and the target strategy, and calculate the parameter difference index D = ∑|Pi - Pi'| / Pi_max, where Pi is the current parameter, Pi' is the target parameter, and Pi_max is the maximum allowable parameter change. Then, design a transition curve based on the difference index D. For small adjustments with D < 0.2, a linear transition curve is used, with a transition time of 100-300ms; for medium adjustments with 0.2 ≤ D < 0.5, an S-shaped transition curve is used, with a transition time of 300-800ms; for large adjustments with D ≥ 0.5, a segmented transition strategy is used, first quickly adjusting to the intermediate state and stabilizing for 100ms, then completing the remaining adjustments, with a total transition time of 800-2000ms. It also sets limits on the rate of change of parameters, such as the rate of change of modulation depth not exceeding 0.1 / ms and the rate of change of switching frequency not exceeding 5kHz / s, to ensure that the system parameters change smoothly during any operating condition transition, effectively avoiding the output fluctuation and overshoot problems common in traditional control strategy switching.

[0080] like Figure 3 As shown, step S5 specifically includes: Step S51: Based on the PWM control strategy, implement the sinusoidal pulse width modulation (SPWM) algorithm and perform adaptive adjustment of the carrier frequency to generate the basic PWM control signal; Step S52: Based on the basic PWM control signal, implement the space vector pulse width modulation (SVPWM) algorithm to perform sector judgment and vector decomposition, and generate a high-utilization PWM control signal; Step S53: Based on the high-utilization PWM control signal, implement the discontinuous pulse width modulation (DPWM) algorithm to reduce the number of switching operations and generate a low-loss PWM control signal; Step S54: Based on the low-loss PWM control signal, perform real-time parameter fine-tuning of modulation depth, switching frequency and dead time to generate an optimized PWM control signal; Step S55: Based on the optimized PWM control signal, the power conversion control signal is obtained by performance evaluation and adaptive adjustment through a feedback monitoring network.

[0081] In step S51, based on the PWM control strategy, a sinusoidal pulse width modulation (SPWM) algorithm is implemented and the carrier frequency is adaptively adjusted to generate a basic PWM control signal. SPWM is the most basic implementation, generating a PWM signal by comparing a sinusoidal modulated wave and a triangular carrier wave. The SPWM has been improved by introducing an adaptive mechanism for modulated depth and carrier frequency, providing a stable and reliable control foundation under simple operating conditions. The carrier frequency is dynamically adjusted according to load conditions and efficiency requirements. Under light load conditions, the switching frequency is appropriately increased to improve dynamic response and output quality, while under heavy load conditions, the switching frequency is decreased to reduce switching losses and heat generation. The frequency adjustment range is 20kHz-100kHz, varying in a step-like manner to avoid generating unpredictable subharmonics.

[0082] In step S52, based on the basic PWM control signal, the Space Vector Pulse Width Modulation (SVPWM) algorithm is implemented to perform sector determination and vector decomposition, generating a high-utilization PWM control signal. SVPWM is a more advanced modulation technique that treats the three-phase system as a vector space, generating the PWM signal by rationally arranging the action time of the basic voltage vector. It has been optimized for home energy storage applications, simplifying the sector determination algorithm, using a lookup table method to accelerate time calculation, and introducing dynamic dead-time adjustment, significantly improving computational efficiency and control accuracy. SVPWM has high DC bus utilization (up to 0.907) and good harmonic characteristics, making it suitable for operating conditions with high output quality requirements.

[0083] Step S52 specifically includes: Step S521: Based on the basic PWM control signal and three-phase voltage command, the three-phase system is converted into a vector space representation through coordinate transformation to establish a voltage vector coordinate system; Step S522: Based on the voltage vector coordinate system, determine the six sector positions where the current vector is located by sector identification, and obtain the sector identification result; Step S523: Based on the sector identification results, calculate the duration of action of adjacent basic vectors and perform vector synthesis to determine the PWM switching timing; Step S524: Based on the PWM switching timing, perform dead time compensation and carrier arrangement optimization to generate switching control pulses; Step S525: Based on the switching control pulse, the high-utilization PWM control signal is obtained through output filtering and harmonic suppression processing.

[0084] In step S521, based on the basic PWM control signal and the three-phase voltage command, the three-phase system is converted into a vector space representation through coordinate transformation, establishing a voltage vector coordinate system. This step first converts the three-phase voltage command (Va, Vb, Vc) into α-β components in the stationary coordinate system using Clarke transformation: Vα = Va - (Vb + Vc) / 2, Vβ = (Vb -Vc) / 2. The α-β components are then represented as a voltage vector V = Vα + jVβ on the complex plane, where the magnitude and angle of this vector represent the magnitude and phase of the output voltage, respectively. This vector representation makes the control of three-phase systems more intuitive and efficient.

[0085] In step S522, based on the voltage vector coordinate system, sector identification is performed to determine the location of the current vector within six sectors, obtaining the sector identification result. SVPWM divides the vector space into six 60° sectors, each defined by two adjacent basic voltage vectors. The system calculates the angle of the voltage vector θ = arctan(Vβ / Vα) and maps it to the range of 0-360°, then determines the sector number N = ⌊θ / 60°⌋ + 1 based on the angle. To improve computational efficiency, the system also implements a fast sector identification algorithm based on a lookup table, directly determining the sector by comparing the magnitudes of the three-phase voltages, avoiding complex trigonometric function calculations.

[0086] In step S523, based on the sector identification results, the duration of action of adjacent basic vectors is calculated and vector synthesis is performed to determine the PWM switching timing. Within the determined sector, the target voltage vector can be achieved through a linear combination of two adjacent basic voltage vectors and a zero vector. The duration of action of each vector is calculated as follows: T1 = m·Ts·sin(N·π / 3 - θ)·sin(π / 3), T2 = m·Ts·sin(θ - (N-1)·π / 3)·sin(π / 3), T0 = Ts - T1 - T2, where m is the modulation index, Ts is the switching period, N is the sector number, and θ is the voltage vector angle. Based on the calculated duration, the turn-on and turn-off timing of each switch is determined to achieve the synthesis of the target voltage vector.

[0087] In step S524, dead-time compensation and carrier arrangement optimization are performed based on the PWM switching timing to generate switching control pulses. Dead time is a key parameter to prevent bridge arm shoot-through short circuits. An adaptive dead-time control algorithm is implemented, which accurately calculates the minimum safe dead time by monitoring the switching waveform and current zero-crossing point, with a typical value in the range of 100ns-500ns. Simultaneously, the carrier arrangement is optimized by adopting a center-aligned PWM mode to ensure uniform distribution of switching events, reducing harmonic content and electromagnetic interference. A dead-time compensation algorithm is also implemented, which reduces output voltage distortion caused by dead time by pre-considering the dead-time effect in the PWM waveform.

[0088] In step S525, based on the switching control pulse, a high-utilization PWM control signal is obtained through output filtering and harmonic suppression processing. The system designs a multi-stage filtering network, including a hardware LC filter and a software harmonic suppression algorithm, to reduce high-frequency harmonics generated by the PWM switching. The harmonic suppression algorithm uses selective harmonic cancellation technology to target specific harmonics, significantly improving the output waveform quality. Adaptive filter parameter adjustment is also implemented, dynamically adjusting filter parameters according to load characteristics and harmonic distribution, minimizing filter loss and phase delay while maintaining good filtering performance.

[0089] In step S53, based on the high-utilization PWM control signal, a discontinuous pulse width modulation (DPWM) algorithm is implemented to reduce the number of switching operations and generate a low-loss PWM control signal. DPWM significantly reduces the number of switching operations and switching losses by pausing the switching action of one bridge arm in each cycle. Several DPWM variants are implemented, including DPWM1 (60° non-conducting), DPWM2 (30° non-conducting), and DPWM3 (synchronous non-conducting), and the optimal variant is automatically selected based on the power factor and load characteristics. DPWM is particularly suitable for high-power applications, reducing switching losses by 15%-30%, but with a slight sacrifice in output harmonic performance. A hybrid modulation strategy is also implemented, seamlessly switching between different PWM algorithms in different operating ranges. For example, SVPWM is used in light-load areas to obtain the best output quality, HIPWM is used in medium-load areas to balance efficiency and quality, and DPWM is used in heavy-load areas to maximize efficiency.

[0090] Among them, the specific implementation of the DPWM algorithm includes: First, select the optimal discontinuous region position according to the system power factor cos φ. When cos φ > 0.866, select the DPWM1 mode (the 60° non-conduction region is located at ±30° of the voltage vector angle); when 0.5 < cos φ ≤ 0.866, select the DPWM2 mode (the 60° non-conduction region is located at 0° and 180° of the voltage vector angle); when cos φ ≤ 0.5, select the DPWM3 mode (the 60° non-conduction region is located at ±90° of the voltage vector angle). The voltage vectors in the non-conduction region are synthesized by two adjacent effective vectors, and the zero vectors are no longer used.

[0091] Under the premise of maintaining sufficient output voltage quality, the DPWM algorithm reduces the number of switching times by 1 / 3 in each switching cycle, significantly reducing the switching loss. Practical tests show that under the full-load condition of 6 kW, using the DPWM algorithm can reduce the junction temperature of the power transistor by 15 - 18°C, reduce the switching loss by about 28%, and increase the comprehensive efficiency by 1.2 - 1.5 percentage points. To ensure the output quality, an adaptive carrier frequency adjustment mechanism is designed. When the DPWM algorithm is used, the carrier frequency is increased by 10 - 15% to partially compensate for the decline in harmonic performance, and the output voltage THD is controlled within 2.5%, meeting the requirements of household electrical appliances.

[0092] In step S54, based on the low-loss PWM control signal, real-time parameter fine-tuning of the modulation depth, switching frequency, and dead time is performed to generate an optimized PWM control signal. Real-time parameter fine-tuning is performed on the selected control strategy, and key parameters such as the modulation depth, switching frequency, and dead time are optimized based on the feedback data. The modulation depth is the primary adjustment parameter, which directly affects the output voltage amplitude and the utilization rate of the DC bus. The system automatically adjusts the modulation depth through closed-loop control to ensure that the output voltage accurately tracks the set value and avoid nonlinear distortion caused by overmodulation. The switching frequency adjustment is the key to balancing efficiency and control accuracy. A dynamic frequency adjustment strategy is designed to optimize the switching frequency under different load conditions. The dead time is a key parameter to prevent the bridge arm from short-circuiting directly. An adaptive dead-time control algorithm is implemented, and the minimum safe dead time is accurately calculated by monitoring the switching waveform and the current zero-crossing point.

[0093] In step S55, based on the optimized PWM control signal, performance evaluation and adaptive adjustment are performed through a feedback monitoring network to obtain the power conversion control signal. A complete feedback monitoring network is established to collect key parameters in real time during the power conversion process, including input / output voltage and current, converter temperature, switching waveforms, and output harmonics. After preprocessing, this feedback data is compared with expected performance indicators to generate performance deviation indicators. Performance indicators include efficiency (the difference between actual measured values ​​and theoretical expectations), harmonic distortion (the difference between measured THD values ​​and target values), dynamic response (the difference between actual response time and target time), and temperature rise rate (the difference between actual temperature rise and the predicted model). Model-based predictive feedforward control is also implemented, adjusting control parameters in advance according to load change trends to improve the system's dynamic response capability. All parameter adjustments are subject to safety constraints to ensure that performance optimization does not exceed the safe operating range of components.

[0094] The feedback monitoring network adopts a distributed architecture, consisting of multiple sampling nodes and a central processing unit. Sampling nodes are located at key system positions, including: power nodes (monitoring voltage, current, and power), temperature nodes (monitoring the temperature of key components), switching nodes (monitoring PWM waveforms and switching states), and output nodes (monitoring harmonics and power quality). Sampling data is transmitted to the central processing unit via a high-speed CAN bus (1Mbps) or SPI interface (10Mbps) to ensure real-time data transmission.

[0095] The feedback data undergoes three levels of processing: the first level is raw data preprocessing, including filtering, calibration, and outlier detection; the second level is feature extraction, which calculates key performance indicators such as efficiency (deviation between measured and expected values), harmonic distortion (deviation between measured THD and target value), dynamic response (deviation between rise time, settling time, and target value), and temperature rise rate (deviation between measured and model predicted values); the third level is performance evaluation, which integrates various indicators to generate a control strategy performance score.

[0096] Based on the evaluation results, two types of adaptive adjustments are implemented: short-term fine-tuning—parameter optimization for the current control strategy, such as modulation depth fine-tuning (within ±5%), carrier frequency fine-tuning (within ±10%), and dead time fine-tuning (within ±50ns), with an adjustment period of 10-50ms; long-term learning—recording the performance of different strategies under various operating conditions and updating the strategy selection rules of the decision engine, with an adjustment period of 24-72 hours.

[0097] The feedback monitoring network employs a redundant design, with key parameters monitored simultaneously by multiple sensors. Multi-source data is fused using methods such as weighted averaging or median filtering to improve measurement reliability. When a sensor failure is detected, the system can automatically switch to a backup sensor or estimate parameter values ​​indirectly, ensuring the continuous and reliable operation of the control system.

[0098] Step S6 specifically includes: Step S61: Based on the system operating status data, collect electrical parameters, temperature parameters and control parameters to construct a system status dataset, forming the foundation of monitoring data; Step S62: Apply a differential privacy algorithm to the monitoring data base to perform sensitive data protection processing, and generate privacy-protected monitoring data; Step S63: Based on the privacy-protected monitoring data, construct an anomaly detection model to identify potential risks of over-temperature, over-current, and short circuit, and generate anomaly detection results and protection strategies; Step S64: Based on the anomaly detection results and protection strategy, automatically trigger corresponding level of protection measures according to the severity of the anomaly to obtain the system security monitoring and adaptive protection response.

[0099] In step S61, the system receives operating status data from each node, including electrical parameters, temperature parameters, and control parameters, and constructs a system status dataset. Electrical parameters include voltage, current, power, and frequency of each node; temperature parameters include battery temperature, converter temperature, radiator temperature, and ambient temperature; control parameters include PWM parameters, control loop status, and system operating mode. The system adopts a distributed data acquisition architecture, aggregating data from each node to the central processing unit via a high-speed data bus. The sampling frequency is dynamically adjusted according to the parameter change rate; fast-changing parameters (such as current) use a high sampling rate, while slow-changing parameters (such as temperature) use a low sampling rate, optimizing data acquisition efficiency.

[0100] In step S62, based on the idea of ​​a Local Edge differential privacy algorithm, sensitive data undergoes privacy protection processing to balance data availability and privacy protection needs. Differential privacy is a data protection technology that adds carefully designed noise to the original data to ensure that the data analysis results do not reveal individual information. The system employs differentiated privacy protection strategies for different types of data: for highly sensitive data (such as user power consumption patterns), larger noise is added; for low-sensitivity data (such as system temperature), smaller noise is added; and for critical safety parameters (such as overcurrent protection thresholds), no noise is added to ensure security. A local differential privacy mechanism is also implemented, performing privacy processing before the data leaves the local device to maximize user privacy protection.

[0101] The specific implementation method of differential privacy protection is as follows: First, the system data is classified into sensitivity levels, and the data is divided into high-sensitivity data (such as user electricity consumption behavior characteristics and battery charging and discharging modes), medium-sensitivity data (such as system efficiency and power allocation), and low-sensitivity data (such as ambient temperature and system operating status). Then, different Laplace noise mechanisms with different parameters are used for different levels of data: noise with parameter ε=0.1 is added to high-sensitivity data, noise with parameter ε=0.5 is added to medium-sensitivity data, and noise with parameter ε=2.0 or no noise is added to low-sensitivity data.

[0102] A data partitioning strategy is adopted, dividing time-series data into 10-minute segments for processing, reducing the risk of privacy leaks in long-term data analysis. To ensure the monitoring accuracy of key security parameters, the system employs special processing for security-related parameters such as temperature and current: differential privacy is applied only to historical and trend analysis data, while real-time monitoring data maintains its original accuracy, ensuring that security monitoring functions are not affected.

[0103] This differential privacy protection mechanism effectively prevents the possibility of inferring user behavior patterns and usage habits from system operation data, while maintaining sufficient data availability to support system optimization and anomaly detection. Testing and verification show that while providing ε=0.1 level privacy protection, the anomaly detection accuracy remains above 92%, only about 3% lower than using the original data.

[0104] In step S63, a hierarchical protection mechanism is designed, setting different protection strategies for parameters of different importance levels to ensure real-time monitoring and alarm of critical parameters (such as battery temperature and current overload). The system divides the protection parameters into three levels: Level 1 parameters (such as short-circuit current and battery overtemperature) are related to system safety and require millisecond-level response; Level 2 parameters (such as overload and slight overtemperature) affect system performance and require second-level response; Level 3 parameters (such as efficiency degradation and performance deterioration) affect system lifespan and require minute-level response. For each level of parameter, the system sets different monitoring frequencies, alarm thresholds, and response strategies, forming a complete hierarchical protection mechanism.

[0105] In step S64, an anomaly detection model is constructed to identify potential system anomalies, such as over-temperature, over-current, and short-circuit risks, based on processed data, and corresponding protection strategies are generated. A combination of multiple anomaly detection algorithms is employed, including rule-based threshold detection, statistical outlier detection, and machine learning-based pattern recognition. Threshold detection is used for explicit physical limitations, such as maximum current and maximum temperature; outlier detection is used to identify parameter fluctuations deviating from normal ranges; and pattern recognition is used to detect complex anomaly patterns, such as battery aging and component performance degradation. Predictive anomaly detection is also implemented, predicting potential anomalies by analyzing parameter change trends and taking preventative measures in advance. An adaptive protection response mechanism is implemented, automatically triggering different levels of protection measures based on the severity of the anomaly, ranging from parameter adjustment to emergency power cut-off, to ensure the safe and stable operation of the system. For minor anomalies, control parameters may be adjusted to restore normal operation; for moderate anomalies, the system may limit power or switch operating modes; for severe anomalies, the system will execute emergency protection actions, such as rapid power cut-off or isolation of faulty components.

[0106] The anomaly detection model based on machine learning is constructed using a semi-supervised learning method, mainly including the following steps: First, an Isolation Forest model is trained using normal operating data, with 100 trees, a subsampling ratio of 256 points, and a contamination parameter of 0.05. Then, a One-Class Support Vector Machine (SVM) is used as the validation model, with RBF kernel function and an anomaly ratio of 0.01. The detection results of the two models are fused through a weighted voting mechanism, assigning the Isolation Forest a weight of 0.7 and the One-Class SVM a weight of 0.3. After the model training is completed, the system saves the operating data of the most recent 30 days and updates the model parameters weekly with incremental data to ensure that the anomaly detection capability evolves in sync with the aging characteristics of the system. This anomaly detection model has a detection rate of over 95% for faults such as abnormal temperature, current fluctuations, and efficiency decline, and provides an average early warning of 2-5 minutes, giving the system sufficient time to take preventative protective measures.

[0107] This invention also provides a magnetic component for implementing the power conversion control method of the aforementioned home energy storage system. This magnetic component uses a nanocrystalline material to fabricate the magnetic core, exhibiting high permeability (μ>15000), high saturation magnetic induction (Bs>1.2T), and low loss characteristics (<3W / kg@100kHz). The core shape is optimized, employing an E-type or U-type structure with no air gap or only a tiny control air gap in the magnetic circuit to fully utilize the high permeability of the nanocrystalline material. The windings employ a multi-layer distributed structure, with conductors being flat copper foil or Litz wire, and the cross-section optimized to reduce skin effect and proximity effect. The winding arrangement uses an interleaved design, with current directions opposite in adjacent layers, effectively canceling magnetic fields and reducing high-frequency losses. The entire magnetic component is optimized through finite element analysis to ensure high efficiency and low temperature rise across a wide voltage range of 100V-450V and a frequency range of 20kHz-100kHz.

[0108] In one embodiment, the operating characteristics of the home energy storage system are as follows: input voltage range of 100V-450V DC, output power requirement of 6kW, and frequency range of 20kHz-100kHz. By selecting a nanocrystalline magnetic core material with permeability μ>15000, saturation magnetic induction Bs>1.2T, and loss characteristics <3W / kg@100kHz, the operating magnetic flux density range is determined to be 0.3T-0.8T. When designing the multi-layer distributed winding structure, the conductor cross-section and arrangement are optimized to reduce skin effect and proximity effect, thereby lowering high-frequency copper losses. Finite element analysis software is used to simulate temperature rise and loss distribution at different operating points, optimizing the core shape and winding topology to achieve efficient energy transfer over a wide operating range.

[0109] In a multi-dimensional data acquisition network, key parameters such as battery SOC, voltage, temperature, load power, and grid status are collected in real time. The collected data undergoes filtering, normalization, and feature extraction processing to calculate characteristic parameters such as voltage range distribution, power fluctuation rate, load periodicity, and battery charge / discharge status. Operating condition identification is performed using a finite state machine and a lightweight XGBoost classifier, generating an operating condition description set containing the operating condition type and key parameters.

[0110] In bottleneck identification and capacity estimation, a bottleneck analysis method similar to the Vantage point selection algorithm is employed based on the energy flow graph to calculate the load rate, influence coefficient, and sensitivity of each node, thereby identifying power bottlenecks in the system. Capacity estimation is performed on these bottlenecks, a bottleneck model is established, and the impact of each bottleneck on the overall system performance is quantified. A multi-objective optimization model is constructed, comprehensively considering factors such as efficiency, temperature rise, and system lifetime, to calculate the optimal power allocation scheme.

[0111] In the control strategy selection, a control strategy selection space containing multiple PWM modulation strategies and parameter combinations is constructed. Characteristic parameters such as control complexity, response time requirements, and control accuracy requirements are extracted. An adaptive decision engine is built using a finite state machine and a lightweight decision tree to automatically select the optimal control strategy based on the operating conditions.

[0112] In the PWM generation algorithm implementation, multiple PWM generation algorithms such as SPWM, SVPWM, and DPWM are implemented and dynamically switched according to the output of the decision engine. Real-time parameter fine-tuning is performed on the selected control strategy to optimize key parameters such as modulation depth, switching frequency, and dead time, achieving efficient and precise control of power conversion.

[0113] In system security monitoring, differential privacy protection processing is performed on sensitive data, anomaly detection models are built to identify potential system anomalies, and an adaptive protection response mechanism is implemented to ensure the safe and stable operation of the system.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A power conversion control method for a home energy storage system, characterized in that, include: Based on the working characteristics of home energy storage systems, the system's input voltage range, output power requirements, and frequency range are collected. Magnetic components are constructed by selecting nanocrystalline magnetic core materials and designing a multi-layer distributed winding structure. Based on the magnetic components, the operation monitoring data of battery status, load power and power grid status are collected in real time through a multi-dimensional data acquisition network, and feature extraction and classification are performed to generate a set of operating condition descriptions containing operating condition types and parameters. Based on the operating condition description set and the preset system physical model, a multi-objective optimization model is constructed through bottleneck point identification and capacity estimation to generate a power optimization allocation scheme. Based on the power optimization allocation scheme and real-time system status, a PWM control strategy is selected by constructing a control strategy selection space and running an adaptive decision engine. Based on the PWM control strategy, multiple PWM generation algorithms such as SPWM, SVPWM and DPWM are implemented and real-time parameters are adjusted to output a power conversion control signal. Based on system operation status data, differential privacy protection processing and anomaly detection models are used to achieve system security monitoring and adaptive protection response.

2. The method according to claim 1, characterized in that, Based on the operating characteristics of the home energy storage system, the system's input voltage range, output power demand, and frequency range are collected. Magnetic components are constructed by selecting nanocrystalline magnetic core materials and designing a multi-layer distributed winding structure, including: Based on the working characteristics of home energy storage systems, the magnetic permeability, saturation magnetic induction intensity and loss characteristics of nanocrystalline materials were analyzed, and the working magnetic flux density range of 0.3T-0.8T was determined. Based on the aforementioned operating magnetic flux density range, a multi-layer distributed winding structure was designed and the conductor cross-section and arrangement were optimized to obtain an optimized winding structure with low skin effect and proximity effect. Based on the optimized winding structure with low skin effect and proximity effect, the magnetic components were obtained by simulating the temperature rise and loss distribution at different operating points using finite element analysis software.

3. The method according to claim 1, characterized in that, Based on the magnetic components, the system collects real-time operational monitoring data on battery status, load power, and grid status through a multi-dimensional data acquisition network, performs feature extraction and classification, and generates an operational condition description set containing operational condition types and parameters, including: Based on the magnetic components and system operation data, the system collects operating monitoring data of battery SOC, voltage, temperature and load power through a multi-dimensional data acquisition network to generate a raw dataset reflecting the system's operating conditions. The original dataset is filtered, normalized, and subjected to preliminary feature extraction to generate a structured preprocessed dataset. Based on the structured preprocessed dataset, voltage range distribution, power fluctuation rate, load periodicity, and battery charge / discharge state characteristic parameters are calculated to form an operating condition feature vector. Based on the operating condition feature vector, operating condition identification is performed using a finite state machine and a lightweight XGBoost classifier to obtain the operating condition description set containing operating condition types and parameters.

4. The method according to claim 1, characterized in that, The process of constructing a multi-objective optimization model based on the operating condition description set and the preset system physical model, through bottleneck point identification and capacity estimation, and generating a power optimization allocation scheme includes: Based on the operating condition description set and the preset system physical model, an energy flow diagram is established under the current operating state, forming an energy transmission network topology. Based on the energy transmission network topology, the bottleneck analysis method is used to calculate the load rate, influence coefficient and sensitivity of each node, and identify the power bottleneck point of the system. Capacity estimation is performed on the power bottleneck points of the system and a bottleneck model is established to quantify the impact of each bottleneck point on the overall system performance. Based on the impact of each bottleneck on the overall system performance, a multi-objective optimization model that comprehensively considers efficiency, temperature rise, and system lifetime is constructed to obtain the power optimization allocation scheme.

5. The method according to claim 1, characterized in that, Based on the power optimization allocation scheme and real-time system status, the step of selecting a PWM control strategy by constructing a control strategy selection space and running an adaptive decision engine includes: Based on the power optimization allocation scheme and the real-time system status, a control strategy selection space containing multiple PWM modulation strategies and parameter combinations is constructed to form a candidate control strategy set. Based on the candidate control strategy set, feature parameters of control complexity, response time requirement and control accuracy requirement are extracted to generate control strategy feature vectors; Based on the feature vector of the control strategy, an adaptive decision engine is constructed using a finite state machine and a lightweight decision tree classifier to determine the control mode and strategy; Based on the control mode and strategy, smooth transition and parameter optimization are performed to obtain the PWM control strategy.

6. The method according to claim 1, characterized in that, The method based on the PWM control strategy, by implementing multiple PWM generation algorithms such as SPWM, SVPWM, and DPWM and adjusting parameters in real time, outputs a power conversion control signal, including: Based on the PWM control strategy, a sinusoidal pulse width modulation (SPWM) algorithm is implemented and the carrier frequency is adaptively adjusted to generate a basic PWM control signal. Based on the aforementioned basic PWM control signal, the Space Vector Pulse Width Modulation (SVPWM) algorithm is implemented to perform sector judgment and vector decomposition, generating a high-utilization PWM control signal. Based on the high-utilization PWM control signal, the discontinuous pulse width modulation (DPWM) algorithm is implemented to reduce the number of switching operations and generate a low-loss PWM control signal. Based on the low-loss PWM control signal, real-time parameter fine-tuning of modulation depth, switching frequency, and dead time is performed to generate an optimized PWM control signal. Based on the optimized PWM control signal, the power conversion control signal is obtained by performance evaluation and adaptive adjustment through a feedback monitoring network.

7. The method according to claim 1, characterized in that, The system security monitoring and adaptive protection response based on system operation status data, through differential privacy protection processing and anomaly detection models, includes: Based on system operation status data, electrical parameters, temperature parameters and control parameters are collected to construct a system status dataset, forming the foundation of monitoring data; Based on the aforementioned monitoring data, a differential privacy algorithm is applied to perform sensitive data protection processing to generate privacy-protected monitoring data; Based on the privacy-protected monitoring data, an anomaly detection model is constructed to identify potential risks of over-temperature, over-current, and short circuit, and anomaly detection results and protection strategies are generated. Based on the anomaly detection results and protection strategies, corresponding levels of protection measures are automatically triggered according to the severity of the anomaly, resulting in the system security monitoring and adaptive protection response.

8. The method according to claim 6, characterized in that, The process of implementing the Space Vector Pulse Width Modulation (SVPWM) algorithm based on the fundamental PWM control signal to perform sector determination and vector decomposition, and to generate a high-utilization PWM control signal, includes: Based on the basic PWM control signal and three-phase voltage command, the three-phase system is converted into a vector space representation through coordinate transformation to establish a voltage vector coordinate system; Based on the voltage vector coordinate system, sector identification is performed to determine the six sector positions where the current vector is located, and sector identification results are obtained. Based on the sector identification results, the duration of action of adjacent basic vectors is calculated and vector synthesis is performed to determine the PWM switching timing. Based on the PWM switching timing, dead time compensation and carrier arrangement optimization are performed to generate switching control pulses; Based on the switching control pulse, the high-utilization PWM control signal is obtained through output filtering and harmonic suppression.

9. The method according to claim 4, characterized in that, Based on the impact of each bottleneck on the overall system performance, a multi-objective optimization model is constructed that comprehensively considers efficiency, temperature rise, and system lifetime, resulting in the power optimization allocation scheme, including: Based on the impact of each bottleneck on the overall system performance, an objective function for maximizing energy conversion efficiency is established, thus forming an efficiency optimization objective. Based on the efficiency optimization objective, objective functions for minimizing component temperature rise and maximizing battery cycle life are established, and a set of multi-objective functions is constructed. Based on the aforementioned multi-objective function set, power balance constraints, battery SOC constraints, and temperature constraints are set to establish a constrained optimization problem. Based on the aforementioned constrained optimization problem, a weighted sum method combined with gradient descent algorithm is used to solve it, and a solution set is obtained. Based on the solution set, the power setpoint, operating frequency range, and current limiting control parameters of each stage of the converter are converted to obtain the power optimization allocation scheme.

10. The method according to claim 5, characterized in that, The step of extracting feature parameters of control complexity, response time requirements, and control accuracy requirements based on the candidate control strategy set, and generating a control strategy feature vector, includes: Based on the set of candidate control strategies, the power change rate, load characteristics, and operating point location are analyzed to calculate the control complexity index. Based on the aforementioned control complexity index, the load change rate, transient requirements, and operating condition transition types are analyzed to quantify the response time requirement parameters. Based on the aforementioned response time requirement parameters, load sensitivity, efficiency gradient, and safety margin are analyzed to determine the control accuracy requirements. Based on the aforementioned control accuracy requirements, the electrical signal noise level, harmonic content, and external interference intensity are evaluated to obtain a noise interference assessment. Based on the noise interference assessment, the data is standardized and mapped to the 0-1 range to obtain the feature vector of the control strategy.

11. A magnetic component, characterized in that, A power conversion control method for implementing the home energy storage system according to any one of claims 1-10.

Images