Inverter efficient management system and method based on adaptive control and multi-mode switching

Abstract

The invention relates to the technical field of inverter control, in particular to an inverter efficient management system and method based on adaptive control and multi-mode switching, and the system comprises a sensor module which is used for collecting multi-dimensional data such as input voltage / current, output voltage / current, load type, switching device temperature, power grid state and predictive information; the modal management module comprises a feature fusion unit, a lightweight machine learning model unit and a predictive switching unit, the feature fusion unit is used for fusing multi-dimensional features collected by the sensor in real time, firstly, fine modal division and accurate switching are achieved through multi-dimensional fusion and an intelligent algorithm, and stability and response speed are improved; adaptive control and mode switching are deeply coupled, the performance of each mode is optimized, and the switching stability is improved; 3, full-load efficient operation is realized through system-level optimization; fourthly, the robustness and reliability are enhanced, and various abnormities can be handled; and 5, realizing continuous optimization and flexible deployment by means of cloud edge collaboration.

Description

Technical Field

[0001] The present invention relates to the technical field of inverter control, and more particularly to an efficient inverter management system and method based on adaptive control and multi-mode switching. Background Art

[0002] Inverters, as core devices for power conversion, are widely used in photovoltaic power generation, new energy vehicles, uninterruptible power supplies (UPS), microgrids, and other fields. Their operating efficiency, stability, and robustness directly impact system performance. Existing inverter management technologies primarily rely on modal division and fixed control strategies based on simple thresholds (such as load power and input voltage range). These technologies struggle to adapt to complex and changing operating conditions and suffer from the following significant deficiencies:

[0003] In terms of modal division and switching, traditional methods only rely on fixed thresholds and cannot adapt to complex working conditions such as load nonlinearity, device temperature drift, and grid fluctuations. This results in rough modal division and rigid switching logic, which easily leads to unnecessary frequent switching or delayed switching.

[0004] Adaptive control and modal management are separated from each other. The controller parameters under different modes are not optimized according to the operating conditions. The switching process lacks transient adaptive compensation, resulting in slow convergence speed, insufficient control accuracy, and easy to cause output voltage overshoot or oscillation.

[0005] In terms of efficiency optimization, existing technologies lack a global perspective, have not established a detailed efficiency model, and have unreasonable mode selection. It is difficult to balance low power consumption and dynamic response under light load or standby conditions, resulting in low efficiency in some load ranges.

[0006] Faced with component aging, sensor noise or grid harmonic interference, traditional systems lack robustness and fault tolerance. Control strategies based on fixed parameters are sensitive to changes in device parameters, lack fault mode reconstruction mechanisms, and have poor reliability.

[0007] In addition, traditional inverters rely on offline debugging, lack real-time self-learning and remote optimization capabilities, cannot adapt to different application scenarios, and are unable to meet the demand for efficient and highly reliable power supplies in the new energy field.

[0008] Based on this, the present invention proposes an efficient inverter management system and method based on adaptive control and multi-mode switching to solve the problems existing in the above-mentioned background technology. Summary of the Invention

[0009] The purpose of the present invention is to provide an efficient inverter management system and method based on adaptive control and multi-mode switching, which solves the technical problems raised in the background technology.

[0010] The purpose of the present invention can be achieved through the following technical solutions:

[0011] An efficient inverter management system based on adaptive control and multi-mode switching, including:

[0012] Sensor modules are used to collect multi-dimensional data such as input voltage / current, output voltage / current, load type, switch device temperature, grid status, and predictive information;

[0013] The modal management module includes a feature fusion unit, a lightweight machine learning model unit, and a predictive switching unit. The feature fusion unit is used to fuse the multi-dimensional features collected by sensors in real time. The lightweight machine learning model unit uses a decision tree, SVM, or a small neural network to analyze the fusion features online and intelligently decide the current optimal operating mode and switching timing. The model can also continuously learn and optimize online. The predictive switching unit is used to analyze the current state change trend, predict the most likely operating conditions in the short term, and trigger modal switching preparation in advance or directly and smoothly switch to the predicted mode.

[0014] An adaptive control module equips each operating mode with a dedicated adaptive controller. The dedicated adaptive controller adopts model reference adaptive control (MRAC), self-tuning control (STC), or sliding mode adaptive control strategies, and its parameter update rules and convergence targets are customized for the characteristics of the mode. During the transient process of mode switching, an adaptive mechanism is introduced to smooth the transition. The adaptive mechanism includes adaptive adjustment of the switching ramp time, using an adaptive observer to estimate the disturbance and inject compensation, and allowing partial parameter migration or sharing between related modes under safe and controllable conditions.

[0015] The efficiency optimization module establishes a detailed efficiency model that includes switching loss, conduction loss, iron loss, copper loss, etc., calculates or predicts the theoretical efficiency of each candidate mode in real time through table lookup, and selects the mode with the highest efficiency. In the selected mode, the switching frequency is dynamically optimized in real time based on load current, temperature, efficiency model, etc., and combined with adaptive observer dynamic optimization and dead time compensation, an ultra-low power mode is designed for extremely light load or standby state, and the entry and exit thresholds are accurately determined.

[0016] A fault management module integrates an adaptive observer-based fault diagnosis algorithm to detect performance degradation or failure of key components, triggering modal reconstruction, switching to a degraded mode and activating a dedicated adaptive controller, or switching to a mode with strong anti-disturbance capabilities when strong external interference is detected;

[0017] The digital twin and cloud-edge collaboration module establishes a high-fidelity digital twin model, performs offline simulation optimization on the cloud or edge server, sends the optimized parameters and strategies to the device side, and aggregates device operation data for group learning and continuous optimization.

[0018] As a further solution of the present invention: the lightweight machine learning model uses the following decision function to make modal decisions:

[0019]

[0020] Among them, M is the operating mode of the final decision, f(X) is the decision function, X is the multidimensional feature vector collected by the sensor and processed by the feature fusion unit, and M set is the set of all candidate modes, p(m|X) is the probability that the system is in mode m under the condition of feature vector X, and the probability distribution is obtained by training the model on historical operation data.

[0021] As a further solution of the present invention: when the dedicated adaptive controller adopts the model reference adaptive control (MRAC) strategy, its parameter update formula is:

[0022]

[0023] in, is the adaptive controller parameter is the update rate of the system, Γ is the positive definite adaptive gain matrix, φ(t) is the regression vector related to the system output, and e(t) is the error between the reference model output and the actual system output.

[0024] As a further solution of the present invention: an efficient inverter management method based on adaptive control and multi-mode switching includes the following steps:

[0025] Data collection and feature fusion: The sensor module collects multi-dimensional operating data in real time and inputs it into the modal management module for feature fusion to obtain a comprehensive feature vector containing the current operating status;

[0026] Mode decision and switching: Utilizes a trained lightweight machine learning model to analyze fusion features and, combined with a predictive switching algorithm, determines the optimal operating mode and switching timing. If future operating condition changes are predicted, switching preparations can be made in advance or the system can be switched directly.

[0027] Adaptive control implementation: Based on the current mode, a dedicated adaptive controller is called to enable adaptive auxiliary mechanisms during the switching transient process, such as adjusting switching parameters and compensating for disturbances. It also allows parameter migration between related modes to accelerate adaptive convergence.

[0028] Efficiency optimization: After selecting a mode based on the efficiency model, the switching frequency and dead time are dynamically optimized in that mode, entering ultra-high efficiency mode for light load / standby conditions to improve overall system efficiency.

[0029] Fault handling and enhanced robustness: Real-time monitoring of component status and external interference. When a fault or strong interference is detected, the system switches to the corresponding degraded or anti-disturbance mode, enabling dedicated control strategies to maintain system operation or shut down safely.

[0030] Offline optimization and online deployment: Use the digital twin model to perform offline simulation in the cloud to optimize modal division, switching logic, and controller parameters. The optimization results are sent to the device end, and the model is updated with device operation data to achieve continuous optimization.

[0031] As a further solution of the present invention: the predictive switching step includes analyzing the state change trend, predicting the future operating conditions by constructing a state transition prediction model and triggering switching preparation or direct switching in advance, and the state transition prediction model is trained based on historical operating data and current state characteristics.

[0032] As a further solution of the present invention: the efficiency optimization step includes establishing an efficiency model including switching loss, conduction loss, iron loss, and copper loss, and calculating the theoretical efficiency of each mode in real time using the following formula:

[0033]

[0034] Where, η is the inverter efficiency, P out is the output power, P in is the input power P switch is the switching loss, P conduction is the conduction loss, P iron is the iron loss, P copper is the copper loss, and the mode with the highest efficiency is selected based on the calculation results.

[0035] Beneficial effects of the present invention:

[0036] (1) Intelligent modal division and precise switching: Breaking through the traditional modal division method based on simple thresholds, this system achieves more refined modal division and smarter switching decisions by integrating multi-dimensional real-time operating status information and combining machine learning and prediction algorithms. It can accurately adapt to complex and changing working conditions, avoid unnecessary frequent switching and delayed switching, and significantly improve system operation stability and response speed.

[0037] (2) Deeply coupled adaptive control: The adaptive control algorithm is deeply embedded in the modal management and switching process, and a dedicated adaptive controller and parameter update rules are customized for each mode. At the same time, an adaptive auxiliary mechanism is introduced during the switching transient process, and inter-modal parameter migration is supported. This optimizes the control performance of each mode, significantly improves the voltage and current stability during the switching process, and reduces waveform distortion.

[0038] (3) System-level efficient operation optimization: With the core goal of maximizing the overall system efficiency, a refined efficiency model is established to drive mode selection. Combined with dynamic optimization of switching frequency, adaptive dead time compensation, and ultra-high efficiency light-load mode design, efficient operation is achieved within the full load range, effectively solving the problem of low efficiency of traditional technologies under partial load and light load conditions.

[0039] (4) Strong robustness and high reliability: Utilizing the synergy of multimodality and adaptive control, the system's ability to cope with parameter perturbations, component aging, external interference, and faults is enhanced. Integrated fault diagnosis and modal reconstruction mechanisms enable the system to quickly switch to the appropriate mode and activate targeted control strategies when an anomaly is detected, maintaining basic functions or safely shutting down, significantly improving system reliability and service life.

[0040] (5) Continuous optimization and flexible deployment: Leveraging digital twins and cloud-edge collaborative architecture, the system achieves offline simulation optimization and online parameter updates. By aggregating operating data from multiple devices for group learning, the system overcomes the limitations of single-device data and achieves global performance optimization. This facilitates unified management and upgrades after large-scale deployment, enabling continuous adaptation to changing requirements in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below with reference to the accompanying drawings.

[0042] Figure 1 This is a system block diagram of the inverter efficient management system and method based on adaptive control and multi-mode switching of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] According to the attached Figure 1 , an efficient inverter management system based on adaptive control and multi-mode switching, including:

[0045] Sensor modules are used to collect multi-dimensional data such as input voltage / current, output voltage / current, load type, switch device temperature, grid status, and predictive information;

[0046] In practical applications, such as photovoltaic inverter scenarios, voltage / current sensors use high-precision Hall sensors, which are connected in series or in parallel on the input side of the photovoltaic array and the output side of the grid to collect the input side voltage V in real time. i , current li The output side voltages V0 and l0 are also measured. The NTC thermistor temperature sensor is tightly attached to the surface of the switching device (such as the IGBT module), and thermal conductivity is enhanced by thermal grease to accurately monitor the temperature T8. The load type identification module analyzes the output current harmonics based on the Fast Fourier Transform (FFT) algorithm and calculates the load nonlinearity index γ. When connected to the grid, the grid monitoring unit obtains the grid frequency f through synchronous sampling technology. g , voltage phase θ g In photovoltaic applications, access the irradiance prediction interface provided by the meteorological department or dedicated prediction equipment to obtain the irradiance prediction value G pred .

[0047] The modal management module includes a feature fusion unit, a lightweight machine learning model unit, and a predictive switching unit. The feature fusion unit is used to fuse the multi-dimensional features collected by sensors in real time. The lightweight machine learning model unit uses a decision tree, SVM, or a small neural network to analyze the fusion features online and intelligently decide the current optimal operating mode and switching timing. The model can also continuously learn and optimize online. The predictive switching unit is used to analyze the current state change trend, predict the most likely operating conditions in the short term, and trigger modal switching preparation in advance or directly and smoothly switch to the predicted mode.

[0048] In actual applications, the feature fusion unit normalizes the various types of collected data, converts them into the same dimension and value range, and then combines them into a multi-dimensional feature vector. The lightweight machine learning model unit uses a small convolutional neural network (CNN). Before deployment, it is trained using a large amount of actual operating condition data. The data covers different load types, input voltage fluctuations, temperature changes and other scenarios. After training, the model is solidified into the controller, and the feature vectors are received in real time. The probability value of each mode is calculated through forward propagation, and the mode with the highest probability is selected as the current optimal operating mode. The predictive switching unit learns the historical feature vector sequence based on the long short-term memory network (LSTM) to predict the changes in operating conditions within the next 5ms. Once it is predicted that the feature vector will cross the modal threshold, the mode switching preparation program is immediately triggered.

[0049] An adaptive control module equips each operating mode with a dedicated adaptive controller. The dedicated adaptive controller adopts model reference adaptive control (MRAC), self-tuning control (STC), or sliding mode adaptive control strategies, and its parameter update rules and convergence targets are customized for the characteristics of the mode. During the transient process of mode switching, an adaptive mechanism is introduced to smooth the transition. The adaptive mechanism includes adaptive adjustment of the switching ramp time, using an adaptive observer to estimate the disturbance and inject compensation, and allowing partial parameter migration or sharing between related modes under safe and controllable conditions.

[0050] In practical applications, a dedicated adaptive controller is designed for each mode. In light-load pulse mode, a pulse skipping control strategy is adopted, combined with a self-tuning controller (STC), by adjusting the proportional parameters.

[0051] Control the output voltage. The medium-load SVPWM mode uses a model reference adaptive controller (MRAC) to adjust the estimated values ​​of system parameters such as inductance and capacitance in real time according to the parameter update formula in claim 3. The heavy-load sliding mode mode constructs a sliding surface function and calculates the equivalent control quantity through an adaptive observer to achieve stable control of the inverter. During the transient process of mode switching, the parameters of the target mode controller are smoothly transitioned and disturbance compensated using the adaptively adjusted ramp time and adaptive observer. At the same time, an inter-modal parameter migration rule is established. When switching from one mode to another related mode, some of the learned effective parameters are migrated under safe and controllable conditions to accelerate the adaptive convergence speed in the new mode.

[0052] The efficiency optimization module establishes a detailed efficiency model that includes switching loss, conduction loss, iron loss, copper loss, etc., calculates or predicts the theoretical efficiency of each candidate mode in real time through table lookup, and selects the mode with the highest efficiency. In the selected mode, the switching frequency is dynamically optimized in real time based on load current, temperature, efficiency model, etc., and combined with adaptive observer dynamic optimization and dead time compensation, an ultra-low power mode is designed for extremely light load or standby state, and the entry and exit thresholds are accurately determined.

[0053] In practical applications, a detailed efficiency model is developed, comprehensively considering factors such as switching loss, conduction loss, iron loss, and copper loss. For switching loss, switching energy parameters are obtained from the switching device datasheet and calculated based on the actual switching frequency and number of switches per cycle. Conduction loss is calculated based on the device's on-resistance and the RMS current. Every 20ms, the efficiency model is used to calculate the theoretical efficiency of each candidate mode, and the most efficient mode is selected. In the selected mode, such as the SVPWM mode, a preset frequency adjustment strategy is used to dynamically optimize the switching frequency based on parameters such as load current, temperature, and the efficiency model, minimizing switching losses while meeting dynamic response and THD requirements. Furthermore, an adaptive observer monitors changes in switching device characteristics to dynamically optimize and compensate for dead-time, reducing losses and waveform distortion caused by dead-time effects. For extremely light loads or standby conditions, an ultra-low-power pulse-skipping or burst mode is designed. Experiments and data analysis are used to determine the precise entry and exit thresholds for these modes to avoid frequent switching or false triggering.

[0054] A fault management module integrates an adaptive observer-based fault diagnosis algorithm to detect performance degradation or failure of key components, triggering modal reconstruction, switching to a degraded mode and activating a dedicated adaptive controller, or switching to a mode with strong anti-disturbance capabilities when strong external interference is detected;

[0055] In practical applications, the fault diagnosis algorithm based on the adaptive observer monitors the parameters of key components in real time. Taking power devices as an example, the performance degradation of the device is judged by calculating the residual ΔR of the device's on-resistance. When ΔR>20%, the device is deemed to have failed. Once a fault is detected, the system intelligently switches to the corresponding degraded mode according to the type and severity of the fault. For example, when a single-phase fault occurs in a three-phase inverter, it switches to a two-phase H-bridge mode, and starts a dedicated adaptive controller optimized for the degraded mode to limit the output power. At the same time, the fault code and status information are sent to the remote monitoring center through the communication module. When strong external interference is detected (such as sudden rise and fall of grid voltage, and drastic load jumps), it actively switches to a mode with stronger anti-interference ability, such as a mode using sliding mode control, and cooperates with the adaptive law to quickly suppress the disturbance.

[0056] The digital twin and cloud-edge collaboration module builds a high-fidelity digital twin model, performs offline simulation optimization on the cloud or edge server, sends the optimized parameters and strategies to the device side, and aggregates device operation data for group learning and continuous optimization.

[0057] In practical applications, a digital twin model highly consistent with the actual inverter system is constructed in the cloud using the professional simulation software PLECS and the Python programming environment. This model covers multiple aspects, including circuit topology, thermal model, and control logic. On edge servers or in the cloud, historical operating data and massive amounts of complex operating condition data generated by simulations are used to perform offline simulation optimization of the digital twin model, adjusting the adaptive controller parameters, modal partitioning thresholds, and switching logic parameters for each mode. Once optimized, the optimal parameter set and improved strategy rules are distributed to the actual device via a secure data transmission protocol. During actual device operation, real-time operating data (including sensor data, modal operating status, efficiency values, etc.) is uploaded to the cloud for updating the digital twin model and further optimization. Simultaneously, data uploaded by multiple similar devices is aggregated and analyzed for big data to identify optimal operating modes and adaptive rules, which are then rolled out to all devices, achieving continuous system optimization and performance improvement.

[0058] As a further solution of the present invention: the lightweight machine learning model uses the following decision function to make modal decisions:

[0059]

[0060] Among them, M is the operating mode of the final decision, f(X) is the decision function, X is the multidimensional feature vector collected by the sensor and processed by the feature fusion unit, and M set is the set of all candidate modes, P(m|X) is the probability that the system is in mode m under the condition of feature vector X, and the probability distribution is obtained by training the model on historical operation data.

[0061] In the modal decision process, the lightweight machine learning model uses a small convolutional neural network (CNN). First, the collected multi-dimensional feature vectors are preprocessed and converted into a tensor form suitable for CNN input, such as expanding a 10-dimensional feature vector into a two-dimensional matrix of a specific size. During the model training phase, a large amount of inverter operation data under different working conditions is collected, including normal operation, load changes, input voltage fluctuations, temperature changes, etc. The data is labeled to determine the optimal operating mode corresponding to each working condition. The labeled data is used to train CNN, and the cross entropy loss function is used as the optimization target. The model parameters are adjusted through the stochastic gradient descent (SGD) algorithm so that the model can accurately predict the probability value of each mode under different feature vector conditions. After training, the model is deployed to the controller of the inverter. During real-time operation, the controller inputs the feature vectors collected and processed in real time into CNN, and after calculations of the convolution layer, pooling layer and fully connected layer, the probability distribution P(m|X) of each mode is output, and then according to

[0062] The mode with the highest probability is selected as the operating mode M of the current inverter to realize intelligent mode decision-making based on probability prediction.

[0063] As a further solution of the present invention: when the dedicated adaptive controller adopts the model reference adaptive control (MRAC) strategy, its parameter update formula is:

[0064]

[0065] in, is the adaptive controller parameter is the update rate of the system, Γ is the positive definite adaptive gain matrix, φ(t) is the regression vector related to the system output, and e(t) is the error between the reference model output and the actual system output.

[0066] When the dedicated adaptive controller adopts the model reference adaptive control (MRAC) strategy, the medium load SVPWM mode is used as an example for implementation. First, determine the reference model Where L is the inductance value, C is the capacitance value, and R is the resistance value. The initial nominal values ​​of these parameters are determined according to the circuit design of the inverter and the actual device selection. During operation, the output voltage and current signals of the inverter are collected in real time. After filtering and sampling, the regression vector φ(t) related to the system output is constructed. At the same time, the error e(t) between the reference model output and the actual system output is calculated. The positive definite adaptive gain matrix Γ is designed according to the dynamic performance requirements and stability conditions of the system. For example, it is set to a diagonal matrix diag[0.1,0.1]. Then, according to

[0067] Real-time calculation of adaptive controller parameters The update rate is constantly updated through integral operation. Realize online estimation and adjustment of system parameters such as inductance and capacitance, so that the output of the actual system can quickly track the output of the reference model, and improve the control performance and stability of the inverter in medium-load SVPWM mode.

[0068] As a further solution of the present invention: an efficient inverter management method based on adaptive control and multi-mode switching includes the following steps:

[0069] Data collection and feature fusion: The sensor module collects multi-dimensional operating data in real time and inputs it into the modal management module for feature fusion to obtain a comprehensive feature vector containing the current operating status;

[0070] During the operation of the inverter, the sensor module collects multi-dimensional data such as input voltage / current, output voltage / current, load type, switching device temperature, grid status and predictive information in real time at a sampling frequency of 20kHz. The collected data is first low-pass filtered to remove high-frequency noise and then transmitted to the controller. Within the controller, the feature fusion unit normalizes the data and converts data of different dimensions and value ranges into a unified format. For example, the voltage data is normalized to the [0,1] interval and the current data is normalized to the [-1,1] interval. Finally, the processed data is combined into a comprehensive feature vector containing the current operating status for subsequent modal decision-making and control.

[0071] Mode decision and switching: Utilizes a trained lightweight machine learning model to analyze fusion features and, combined with a predictive switching algorithm, determines the optimal operating mode and switching timing. If future operating condition changes are predicted, switching preparations can be made in advance or the system can be switched directly.

[0072] In practical applications, a trained lightweight machine learning model (such as a small CNN) is solidified in the controller, and the controller inputs the fused feature vector into the model. The model calculates the probability value of each mode through forward propagation, and selects the current optimal operating mode according to the decision function in claim 2. At the same time, the predictive switching unit analyzes the historical feature vector sequence based on the LSTM network to predict future changes in working conditions. When it is predicted that the working conditions will change within the next 5ms and the current mode is no longer optimal, the controller initialization program of the target mode is activated in advance, including loading the control parameters of the target mode, configuring related registers, etc. When the conditions set in the mode switching criterion matrix C are met, the mode switching operation is executed, and adaptive mechanisms such as slope compensation are enabled during the switching process to make the control parameters transition smoothly and reduce the voltage and current shocks during the switching process.

[0073] Adaptive control implementation: Based on the current mode, a dedicated adaptive controller is called to enable adaptive auxiliary mechanisms during the switching transient process, such as adjusting switching parameters and compensating for disturbances. It also allows parameter migration between related modes to accelerate adaptive convergence.

[0074] In practical applications, the corresponding dedicated adaptive controller is called based on the current mode. During the transient process of mode switching, an adaptive auxiliary mechanism is enabled. For example, an adaptive observer is used to estimate the disturbance or parameter change at the switching moment in real time, and the estimation result is fed back to the target mode controller for compensation. At the same time, the ramp time of the switching process is adaptively adjusted to ensure a smooth switching process. When operating in different modes, partial parameter migration between related modes is allowed under safe and controllable conditions. For example, when switching from SVPWM mode to sliding mode mode, the inductance value identified in the SVPWM mode is used as the initial value of the sliding mode controller and adjusted according to the preset fault tolerance range. This accelerates the adaptive convergence speed in the new mode and improves the overall control performance of the system.

[0075] Efficiency optimization: After selecting a mode based on the efficiency model, the switching frequency and dead time are dynamically optimized in that mode, entering ultra-high efficiency mode for light load / standby conditions to improve overall system efficiency.

[0076] In practical applications, based on the established refined efficiency model, the theoretical efficiency of each candidate mode is calculated every 20ms. In the selected mode, such as SVPWM mode, the switching frequency is adjusted in real time according to the preset dynamic frequency optimization strategy based on parameters such as load current, temperature, and efficiency model. At the same time, combined with the monitoring results of the switching device characteristic changes by the adaptive observer, the dead time is dynamically optimized and compensated. For extremely light load or standby state, when the load current is detected to be lower than the set light load threshold (such as I o <0.12I rated) and lasts for more than a certain time (such as 500ms), it enters ultra-low power consumption pulse skipping or burst mode; when the load current rises and exceeds the exit threshold (such as I o >0.15I rated ), exit this mode and return to normal operation mode to achieve efficient operation within the full load range.

[0077] Fault handling and enhanced robustness: Real-time monitoring of component status and external interference. When a fault or strong interference is detected, the system switches to the corresponding degraded or anti-disturbance mode, enabling dedicated control strategies to maintain system operation or shut down safely.

[0078] In actual applications, the fault management module monitors component status and external interference in real time. The parameter residuals of key components (such as power devices and sensors) are calculated through an adaptive observer. When the residual exceeds the set threshold (such as the residual on-resistance of the power device ΔR>20%), the component performance is determined to be degraded or faulty, triggering modal reconstruction, switching to the corresponding degraded mode, and starting a dedicated adaptive controller optimized for the degraded mode to limit the output power. At the same time, the fault code and status information are sent to the remote monitoring center through the communication module. When strong external interference is detected (such as a sudden increase or decrease in the grid voltage exceeding 10% of the rated value, or a sudden change in the load current exceeding 20% ​​of the rated value), it actively switches to a mode with stronger anti-interference ability (such as a sliding mode control mode) and cooperates with the adaptive law to quickly suppress the disturbance, maintain the stable operation of the system or shut down safely, and improve the robustness and reliability of the system.

[0079] Offline optimization and online deployment: Use digital twin models to conduct offline simulations in the cloud to optimize modal division, switching logic, and controller parameters. The optimization results are sent to the device side, and the model is updated based on device operation data to achieve continuous optimization.

[0080] In actual applications, digital twin models are used for offline simulation optimization in the cloud or edge server. First, the operating data of the actual equipment and the complex working condition data generated by simulation are input into the digital twin model, and the adaptive controller parameters, modal division thresholds, switching logic parameters, etc. under each mode are adjusted. Through multiple simulations and optimization algorithms (such as genetic algorithms and particle swarm optimization algorithms), the optimal parameter combination and control strategy are found. After the optimization is completed, the optimized parameter set and strategy rules are sent to the actual device through a secure communication protocol (such as SSL / TLS encryption). During the operation of the actual device, the real-time operating data (including sensor acquisition data, modal operating status, efficiency value, fault information, etc.) is uploaded to the cloud. The cloud analyzes and processes the data, updates the digital twin model, and further optimizes the control strategy to achieve continuous optimization and performance improvement of the system, while facilitating unified management and remote upgrades of large-scale equipment.

[0081] As a further solution of the present invention: the predictive switching step includes analyzing state change trends, predicting future operating conditions by building a state transition prediction model and triggering switching preparation or direct switching in advance, wherein the state transition prediction model is trained based on historical operating data and current state characteristics;

[0082] In the predictive switching step, the predictive switching unit constructs a state transition prediction model based on a long short-term memory (LSTM) network. First, a large amount of historical operating data is collected, including multidimensional feature vectors X and corresponding operating modes M at different time points, and the data is divided into a training set and a test set. The LSTM network is trained using the training set, and appropriate network parameters such as the number of hidden layer neurons, learning rate, and number of training iterations are set. During training, the historical feature vector sequence is used as input to predict the feature vector change trend and possible operating conditions within the next 5ms. During real-time operation, the controller inputs the feature vectors of the current and historical period (such as the past 100 sampling cycles) into the trained LSTM model. The model predicts future operating condition changes by learning and analyzing the sequence data. When it is predicted that the future operating conditions will cause the feature vector to cross the modal threshold, the modal switching preparation program is triggered in advance, including initializing the controller parameters of the target mode and configuring related hardware resources. At the same time, based on the prediction results, the relevant parameters of the switching process, such as the ramp time and the gain of the switching function, are adaptively adjusted to ensure that the target mode can be switched quickly and smoothly when the actual working conditions change, reducing the performance loss caused by switching delays and improving the dynamic response capability and stability of the system.

[0083] As a further solution of the present invention: the efficiency optimization step includes establishing an efficiency model including switching loss, conduction loss, iron loss, and copper loss, and calculating the theoretical efficiency of each mode in real time using the following formula:

[0084]

[0085] Where, η is the inverter efficiency, P out is the output power, P in is the input power P switch is the switching loss, P conduction is the conduction loss, P iron is the iron loss, P copper is the copper loss, and the mode with the highest efficiency is selected based on the calculation results;

[0086] In the efficiency optimization step, a detailed efficiency model including switching loss, conduction loss, iron loss, and copper loss is established. switch By consulting the data sheet of the switching device (such as IGBT), we can obtain its turn-on energy E under different voltage and current conditions. onand the turn-off energy E off , combined with the actual operating switching frequency f s and the number of switches per cycle N sw , according to the formula P switch =f s ·(E on +E off )·N sw Calculate the conduction loss P conduction According to the device on-resistance R ds(on) (can be obtained from the device data sheet or online measurement) and the effective current value I rms , using the formula Calculation. Iron loss and copper loss are calculated based on the inverter's magnetic circuit and circuit design parameters, combined with the actual operating voltage, current, and frequency, using empirical formulas or finite element analysis methods. Every 20ms, use the efficiency calculation formula:

[0087]

[0088] Calculate the theoretical efficiency of each candidate mode, where P out The output power can be calculated as the product of the output voltage V0 and the output current l0. The theoretical efficiency values ​​of each mode are compared, and the mode with the highest efficiency is selected as the current operating mode, implementing mode selection driven by the efficiency model. In the selected mode, such as the SVPWM mode, the switching frequency is dynamically optimized in real time based on parameters such as the load current l0, temperature, and efficiency model. For example, according to a preset piecewise function adjustment strategy, different switching frequencies are used in different load ranges. This minimizes switching losses while meeting dynamic response and THD requirements, further improving the overall system efficiency.

[0089] It should be noted that all data collected in this application is collected with the consent and authorization of the user, and the use of the data is legal and compliant, and the use and processing of the data complies with the relevant laws, regulations and standards of the relevant regions. The above formulas are all dimensionless and numerical calculations. The formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formula are set by those skilled in the art based on actual conditions.

[0090] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An efficient inverter management system based on adaptive control and multi-mode switching, characterized by: include: Sensor modules are used to collect multi-dimensional data such as input voltage / current, output voltage / current, load type, switch device temperature, grid status, and predictive information; The modal management module includes a feature fusion unit, a lightweight machine learning model unit, and a predictive switching unit. The feature fusion unit is used to fuse the multi-dimensional features collected by sensors in real time. The lightweight machine learning model unit uses a decision tree, SVM, or a small neural network to analyze the fusion features online and intelligently decide the current optimal operating mode and switching timing. The model can also continuously learn and optimize online. The predictive switching unit is used to analyze the current state change trend, predict the most likely operating conditions in the short term, and trigger modal switching preparation in advance or directly and smoothly switch to the predicted mode. An adaptive control module equips each operating mode with a dedicated adaptive controller. The dedicated adaptive controller adopts model reference adaptive control (MRAC), self-tuning control (STC), or sliding mode adaptive control strategies, and its parameter update rules and convergence targets are customized for the characteristics of the mode. During the transient process of mode switching, an adaptive mechanism is introduced to smooth the transition. The adaptive mechanism includes adaptive adjustment of the switching ramp time, using an adaptive observer to estimate the disturbance and inject compensation, and allowing partial parameter migration or sharing between related modes under safe and controllable conditions. The efficiency optimization module establishes a detailed efficiency model that includes switching loss, conduction loss, iron loss, copper loss, etc., calculates or predicts the theoretical efficiency of each candidate mode in real time through table lookup, and selects the mode with the highest efficiency. In the selected mode, the switching frequency is dynamically optimized in real time based on load current, temperature, efficiency model, etc., and combined with adaptive observer dynamic optimization and dead time compensation, an ultra-low power mode is designed for extremely light load or standby state, and the entry and exit thresholds are accurately determined. A fault management module integrates an adaptive observer-based fault diagnosis algorithm to detect performance degradation or failure of key components, triggering modal reconstruction, switching to a degraded mode and activating a dedicated adaptive controller, or switching to a mode with strong anti-disturbance capabilities when strong external interference is detected; The digital twin and cloud-edge collaboration module establishes a high-fidelity digital twin model, performs offline simulation optimization on the cloud or edge server, sends the optimized parameters and strategies to the device side, and aggregates device operation data for group learning and continuous optimization.

2. The inverter efficient management system based on adaptive control and multi-mode switching according to claim 1, characterized in that: The lightweight machine learning model uses the following decision function to make modal decisions: Among them, M is the operating mode of the final decision, f(X) is the decision function, X is the multidimensional feature vector collected by the sensor and processed by the feature fusion unit, and M set is the set of all candidate modes, P(m|X) is the probability that the system is in mode m under the condition of feature vector X, and the probability distribution is obtained by training the model on historical operation data.

3. The inverter efficient management system based on adaptive control and multi-mode switching according to claim 1, characterized in that: When the dedicated adaptive controller adopts the model reference adaptive control (MRAC) strategy, its parameter update formula is: in, is the adaptive controller parameter is the update rate of the system, Γ is the positive definite adaptive gain matrix, φ(t) is the regression vector related to the system output, and e(t) is the error between the reference model output and the actual system output.

4. An efficient inverter management method based on adaptive control and multi-mode switching, characterized in that: The following steps are involved: Data collection and feature fusion: The sensor module collects multi-dimensional operating data in real time and inputs it into the modal management module for feature fusion to obtain a comprehensive feature vector containing the current operating status; Mode decision and switching: Utilizes a trained lightweight machine learning model to analyze fusion features and, combined with a predictive switching algorithm, determines the optimal operating mode and switching timing. If future operating condition changes are predicted, switching preparations can be made in advance or the system can be switched directly. Adaptive control implementation: Based on the current mode, a dedicated adaptive controller is called to enable adaptive auxiliary mechanisms during the switching transient process, such as adjusting switching parameters and compensating for disturbances. It also allows parameter migration between related modes to accelerate adaptive convergence. Efficiency optimization: After selecting a mode based on the efficiency model, the switching frequency and dead time are dynamically optimized in that mode, entering ultra-high efficiency mode for light load / standby conditions to improve overall system efficiency. Fault handling and enhanced robustness: Real-time monitoring of component status and external interference. When a fault or strong interference is detected, the system switches to the corresponding degraded or anti-disturbance mode, enabling dedicated control strategies to maintain system operation or shut down safely. Offline optimization and online deployment: Use the digital twin model to perform offline simulation in the cloud to optimize modal division, switching logic, and controller parameters. The optimization results are sent to the device end, and the model is updated with device operation data to achieve continuous optimization.

5. The method for efficient inverter management based on adaptive control and multi-mode switching according to claim 4, characterized in that: The predictive switching step includes analyzing the state change trend, predicting the future working conditions by building a state transition prediction model and triggering switching preparation or direct switching in advance. The state transition prediction model is trained based on historical operating data and current state characteristics.

6. The method for efficient inverter management based on adaptive control and multi-mode switching according to claim 4, characterized in that: The efficiency optimization step includes establishing an efficiency model that includes switching loss, conduction loss, iron loss, and copper loss, and calculating the theoretical efficiency of each mode in real time using the following formula: Where, η is the inverter efficiency, P out is the output power, P in is the input power P switch is the switching loss, P conduction is the conduction loss, P iron is the iron loss, P copper is the copper loss, and the mode with the highest efficiency is selected based on the calculation results.

Images