An adaptive switching system and method for power supply and distribution

Abstract

The application discloses a power supply and distribution adaptive switching system and method, which comprises a data acquisition module, an analysis module, a modeling module, a decision module, an execution module and a warning module. The data acquisition module generates real-time monitoring signals. The analysis module performs dynamic weighted evaluation according to the real-time monitoring signals and generates evaluation signals. The modeling module constructs a virtual model of a power supply system and generates simulation signals. The decision module generates control signals containing path selection instructions. The execution module controls the switching state of semiconductor devices according to the control signals and generates switch feedback signals. The warning module forms equipment health evaluation results according to the switch feedback signals and abnormal arc signals when the switching connection state of the execution module and transmits the results to the decision module. The application can solve the problem that the traditional power supply and distribution switching system relies on a single threshold to judge the power supply state and cannot comprehensively evaluate the power quality.

Description

Technical Field

[0001] This invention relates to the field of power system automation control, and specifically to an adaptive switching system and method for power supply and distribution. Background Technology

[0002] Traditional power supply and distribution switching technology mainly relies on automatic transfer switches (ATS), which achieve power conversion based on the simple criterion of voltage presence or absence. However, it does not consider power quality issues such as harmonic distortion and frequency fluctuations, which may cause precision equipment to be damaged after switching due to substandard power quality.

[0003] With the increasing proportion of renewable energy grid connection, the demand for coordinated switching of multiple power sources in microgrid scenarios is becoming more prominent. Existing technologies, which employ fixed threshold strategies, cannot adapt to grid parameter changes caused by fluctuations in wind, solar, and energy storage output, often leading to power oscillations or overload tripping. In the industrial sector, the back electromotive force generated during the switching of large-capacity inductive loads can cause transient current surges. Traditional mechanical contactors have limited arc-extinguishing capabilities, and frequent operation can easily lead to contact adhesion faults. Existing monitoring systems mostly adopt remote IoT architectures, and signal transmission delays cause switching commands to lag, making it difficult to meet millisecond-level real-time requirements. In terms of fault early warning, conventional methods rely on threshold alarm mechanisms, which can only provide passive protection after a fault occurs and lack the ability to predict the aging state of switching devices. Furthermore, multi-power systems in commercial buildings lack economic optimization models and cannot dynamically select power sources in conjunction with time-of-use pricing, resulting in wasted energy costs.

[0004] While digital twin technology has seen some application in power grid simulation in recent years, existing solutions have failed to achieve closed-loop verification between the virtual and physical systems. Model accuracy gradually declines as equipment ages, leading to accumulated discrepancies between simulation results and actual conditions. Edge computing technology in power systems is largely limited to data acquisition and has not yet formed a strategy optimization mechanism for deep collaboration with the cloud. These technological shortcomings collectively result in significant deficiencies in the security, economy, and intelligence of existing power supply and distribution switching systems. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide an adaptive switching system and method for power supply and distribution, which solves the problem that traditional power supply and distribution switching systems rely on a single threshold to judge the power supply status and cannot comprehensively evaluate power quality. This invention collects voltage, current, and load impedance signals in real time using multiple types of sensors, constructs an analysis module including a dynamic weight evaluation model, and comprehensively calculates power quality indicators; it uses digital twin technology to create a virtual power supply system, simulates transient responses under different switching paths, and selects the optimal solution through reinforcement learning algorithms; it adopts a hybrid actuator of semiconductor switches and mechanical contactors in parallel, prioritizing fast switching through solid-state devices to reduce arc impact; it establishes a closed-loop verification mechanism to compare actual switching parameters with simulation results and dynamically correct the model accuracy; it introduces a high-frequency arc feature library to identify abnormal signals and triggers preventative switching based on equipment lifespan prediction; and it achieves a dynamic balance between power supply cost and security through edge computing and cloud-based collaborative optimization strategies.

[0006] This invention provides an adaptive switching system for power supply and distribution, comprising: The data acquisition module collects voltage waveform signals, current phase signals, and load impedance spectrum signals, and generates real-time monitoring signals. The analysis module dynamically weights and evaluates voltage stability, harmonic distortion rate, frequency deviation, load matching degree, and power supply cost based on real-time monitoring signals, and generates evaluation signals. The modeling module constructs a virtual model of the power supply system and inputs the evaluation signal into the virtual model to generate a simulated signal that includes impact prediction and recovery time. The decision module sorts and processes the analog signals using an algorithm and generates control signals containing path selection instructions. The execution module controls the semiconductor devices to switch the physical connection state of the power supply according to the control signal, and at the same time generates a switch feedback signal; The early warning module generates equipment health assessment results based on switch feedback signals and abnormal arc signals when the execution module switches connection states, and transmits them to the decision module.

[0007] In one embodiment of the present invention, the data acquisition module includes a wideband impedance detection unit. The wideband impedance detection unit acquires the load impedance spectrum signal by injecting a test signal of a specific frequency and is linked with the load type identification unit. When a change in load type is detected, the sampling frequency range of the sensor array is dynamically adjusted. The analysis results of the load impedance spectrum signal are synchronously transmitted to the modeling module to correct the equivalent impedance parameters of the load end in the virtual model. A signal compensation mechanism is established between the wideband impedance detection unit and the current phase signal acquisition channel to eliminate the interference of the test signal injection on the original current phase measurement.

[0008] In one embodiment of the present invention, the weight coefficients of the dynamic weighted evaluation in the analysis module are automatically adjusted according to the power supply scenario mode. When the system is in the economic priority mode, the weight ratio of the power supply cost index is increased. When the system is in the safety priority mode, the comprehensive weight of voltage stability and harmonic distortion rate is strengthened. The weight adjustment process realizes pattern recognition through feature extraction of historical operating data and keeps in linkage with the virtual model parameter update of the modeling module to form a closed-loop optimization mechanism for the evaluation strategy.

[0009] In one embodiment of the present invention, the virtual model of the modeling module includes a dynamic reconstruction function of the power supply network topology. When a distributed power source access or load node change is detected, the virtual model automatically generates a new equivalent circuit model and verifies its convergence. A time-domain synchronization channel is established between the virtual model and the physical system. The time scale deviation caused by signal transmission delay is compensated by the interpolation algorithm to ensure that the impact prediction result is aligned with the time-domain characteristics of the actual transient process.

[0010] In one embodiment of the present invention, the algorithm of the decision module includes a multi-objective optimization engine, which quantifies the power supply reliability index, equipment loss cost and switching operation risk into multi-dimensional constraints, filters the non-dominated solution set through Pareto front analysis, and introduces a fuzzy comprehensive evaluation model to prioritize the candidate schemes. The output results of the multi-objective optimization engine are cross-validated with the equipment health assessment results of the early warning module to form a quantitative scoring mechanism for decision credibility.

[0011] In one embodiment of the present invention, the execution module includes a fast semiconductor switch group and a mechanical contactor group arranged in parallel. The semiconductor switch group is configured to prioritize responding to high-frequency switching commands and undertake transient current transfer. The mechanical contactor group automatically closes after the semiconductor switch completes conduction to maintain steady-state current carrying. An action timing coordination unit is provided between the semiconductor switch group and the mechanical contactor group to ensure that the opening and closing operations of the two groups of devices meet the preset phase synchronization requirements.

[0012] In one embodiment of the present invention, the early warning module includes a high-frequency signal feature library, which stores waveform pattern features of typical fault arcs. The module performs continuous spectrum analysis on the switch feedback signal through a sliding time window. When an abnormal signal with a matching degree exceeding a set threshold is detected, an early warning command containing fault location information is automatically generated. The early warning command triggers the modeling module to start simulation and deduction of a specific fault scenario, and the simulation results are compared with real-time monitoring data to confirm the fault type.

[0013] In one embodiment of the present invention, an edge computing node is included. The edge computing node is deployed between the data acquisition module and the analysis module and is configured to perform localized preprocessing on real-time monitoring signals, including signal noise reduction, feature extraction and data compression. The edge computing node establishes a secure communication link with the cloud management platform, periodically uploads de-identified running data and receives updated algorithm models, so as to realize remote iterative optimization of the control strategy.

[0014] In one embodiment of the present invention, a closed-loop verification channel is established between the modeling module and the execution module. When the deviation between the measured transient impact peak value and the simulated predicted value exceeds the tolerance range during the actual switching process, a model correction instruction is automatically generated and fed back to the virtual model. The correction instruction includes parameter adjustment coefficients and topology update flags, triggering the virtual model to perform adaptive calibration of structural parameters. At the same time, the calibration results are synchronized to the evaluation weight calculation process of the analysis module.

[0015] The present invention also provides an adaptive switching method for power supply and distribution, comprising: S1: Collect voltage waveform signals, current phase signals, and load impedance spectrum signals, and generate real-time monitoring signals; S2: Dynamically weighted evaluation of voltage stability, harmonic distortion rate, frequency deviation, load matching degree and power supply cost based on real-time monitoring signals, and generate evaluation signals; S3: Construct a virtual model of the power supply system and input the evaluation signal into the virtual model to generate a simulated signal that includes impact prediction and recovery time; S4: The analog signals are sorted and processed using an algorithm to generate control signals containing path selection instructions; S5: Controls the semiconductor device to switch the physical connection state of the power supply according to the control signal, and generates a switching feedback signal at the same time; S6: Based on the switch feedback signal and the abnormal arc signal when the execution module switches the connection state, the equipment health assessment result is generated and transmitted to the decision module.

[0016] This invention provides an adaptive switching system and method for power supply and distribution. It collects voltage, current, and load impedance signals in real time using multiple types of sensors, constructs an analysis module including a dynamic weighted evaluation model, and comprehensively calculates power quality indicators. It utilizes digital twin technology to create a virtual power supply system, simulating transient responses under different switching paths, and selects the optimal solution through reinforcement learning algorithms. It employs a hybrid actuator combining semiconductor switches and mechanical contactors in parallel, prioritizing rapid switching using solid-state devices to reduce arc impact. A closed-loop verification mechanism is established to compare actual switching parameters with simulation results and dynamically correct model accuracy. A high-frequency arc feature library is introduced to identify abnormal signals, and preventative switching is triggered by equipment lifespan prediction. Through edge computing and cloud-based collaborative optimization strategies, a dynamic balance between power supply cost and security is achieved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A system architecture diagram of an adaptive switching system for power supply and distribution; Figure 2 This is a flowchart of an adaptive switching method for power supply and distribution. Detailed Implementation

[0019] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0020] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0021] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0022] Please see Figure 1-2The diagram illustrates an adaptive switching system and method for power supply and distribution according to the present invention. The adaptive switching system for power supply and distribution includes a data acquisition module, an analysis module, a modeling module, a decision-making module, an execution module, and an early warning module. The data acquisition module collects voltage waveform signals, current phase signals, and load impedance spectrum signals, and generates real-time monitoring signals. The analysis module dynamically weights and evaluates voltage stability, harmonic distortion rate, frequency deviation, load matching degree, and power supply cost based on the real-time monitoring signals, and generates evaluation signals. The modeling module constructs a virtual model of the power supply system and inputs the evaluation signals into the virtual model, generating simulated signals including impact prediction and recovery time. The decision-making module sorts the simulated signals using an algorithm and generates control signals including path selection instructions. The execution module controls semiconductor devices to switch the physical connection state of the power supply according to the control signals, and simultaneously generates switching feedback signals. The early warning module forms a device health assessment result based on the switching feedback signals and abnormal arc signals when the execution module switches the connection state, and transmits this result to the decision-making module.

[0023] like Figure 1As shown, this invention relates to an adaptive switching system for power supply and distribution. The core innovation of the data acquisition module lies in the design and collaborative working mechanism of its wideband impedance detection unit. This unit employs active signal injection technology, injecting test signals within a specific frequency range into the power supply circuit to acquire full-band characteristic data of the load impedance spectrum in real time. The frequency range of the test signal is automatically adjusted according to the load type. For example, for inductive loads such as motors, the focus is on capturing impedance characteristics in the low-frequency range (50Hz-1kHz), while for nonlinear loads such as frequency converters, the range is extended to the mid-to-high frequency range (1kHz-100kHz) for harmonic impedance analysis. This dynamic frequency adjustment capability enables the system to accurately identify complex impedance characteristics under mixed load scenarios. To ensure that the test signal injection process does not affect the normal power supply quality, a real-time signal compensation mechanism is established between the wideband impedance detection unit and the current phase signal acquisition channel. An inverse superposition cancellation algorithm is used to eliminate the interference of the injected signal on the original current phase measurement, ensuring that the phase measurement accuracy is controlled at the milliradian level. When the load type identification unit detects a change in load characteristics (such as the starting of a large motor or the commissioning of a frequency converter), it immediately triggers the sensor array's adaptive sampling frequency adjustment mechanism. For sudden load changes, the sampling frequency is increased to more than three times the normal level, and a high-frequency noise filtering function is activated to avoid signal aliasing while ensuring data integrity. The obtained load impedance spectrum signal is not only used for real-time monitoring but is also synchronously transmitted to the virtual model parameter update interface of the modeling module. The equivalent impedance network at the load end in the virtual model is corrected through an impedance parameter iterative algorithm, improving the simulation accuracy of the digital twin's load characteristics by more than 40%. This dynamic parameter correction capability is particularly suitable for new energy microgrid scenarios and can effectively solve the problem of sudden impedance characteristic changes caused by the connection of power electronic equipment such as photovoltaic inverters and energy storage converters. In addition, the module is designed with redundant signal channels. When the main acquisition channel malfunctions, the backup channel can take over the signal acquisition task within 10 milliseconds and send equipment maintenance prompts to the early warning module through the self-diagnostic system, ensuring the continuity and reliability of data acquisition.

[0024] Furthermore, the key technology of the dynamic weighted evaluation mechanism in the analysis module lies in the construction of an adaptive weight allocation system for multi-dimensional evaluation indicators. This system dynamically adjusts the weight coefficients of each evaluation indicator based on the intelligent identification results of the system's operating scenario. For example, in a commercial building power supply scenario, it automatically activates the economy-first mode, increasing the weight of the power supply cost indicator to 1.5 times that of the conventional mode, while simultaneously introducing a time-based electricity price factor to real-time correct the cost calculation model. In contrast, in locations with stringent power quality requirements, such as hospitals and data centers, it switches to a safety-first mode, increasing the combined weight of voltage stability and harmonic distortion rate to over 70%, and implementing strict power quality standard thresholds. The weight adjustment algorithm is built upon a deep neural network. By analyzing characteristic patterns in historical operating data (such as load curve patterns, fault occurrence time distribution, and power output fluctuation characteristics), it establishes a scenario classification model and automatically generates the optimal weight combination scheme. To achieve continuous optimization of the evaluation strategy, a bidirectional data channel is established between this module and the modeling module, feeding back the actual effect data after each switching operation to the evaluation model. By comparing the simulated prediction results with the actual operating data, the characteristic parameters in the weight allocation strategy are dynamically corrected. For example, if statistics show that the impact of harmonic distortion rate on equipment lifespan is underestimated after multiple switching operations, the system will automatically adjust the baseline weight coefficient of this indicator and strengthen the monitoring of harmonic components in subsequent evaluations. This closed-loop optimization mechanism enables the evaluation model to adaptively evolve with long-term factors such as equipment aging and changes in power grid structure, improving evaluation accuracy by more than 35% compared to traditional fixed-weight models. Simultaneously, this module also features an abnormal weight combination identification function. When the weight value of a certain indicator is detected to continuously deviate from a reasonable range (e.g., the power supply cost weight exceeds 80% in 10 consecutive operations), a manual review mechanism will be automatically triggered, and a system parameter verification request will be sent to maintenance personnel to prevent the algorithm from falling into a local optimum. In scenarios with high penetration of new energy sources, this module also introduces a source-load matching evaluation sub-model. By analyzing the spatiotemporal correlation between the output characteristics of distributed power sources and the load demand curve, it optimizes the selection strategy for switching timing, maximizing the utilization rate of clean energy.

[0025] In one embodiment of the invention, the key innovation of the modeling module lies in achieving dynamic reconstruction of the power supply network topology and time-domain synchronization of simulation. The module's built-in virtual model employs a modular modeling approach, decomposing the power supply network into a standardized component library including power nodes, transmission lines, switching equipment, and load units. Each component contains multi-dimensional attributes such as electrical parameters, thermodynamic parameters, and aging coefficients. When distributed power source access or load node changes are detected (e.g., adding a photovoltaic grid-connected point or switching on important loads), the topology reconstruction engine automatically scans the changes in the physical system's connection relationships, generates a new network topology based on graph theory algorithms, and constructs the corresponding simulation model using equivalent circuit transformation technology. During the model verification phase, the system performs convergence tests and parameter sensitivity analyses. For high-impedance-ratio loops that may cause simulation divergence, virtual damping elements are automatically inserted to ensure computational stability. To achieve time synchronization between the virtual and physical systems, the module developed a high-precision time-domain alignment technology. By deploying hardware timestamp units on the physical system side, microsecond-level time tags are added to each data packet of the acquired signal. On the virtual model side, a variable-step-size simulation algorithm, combined with cubic spline interpolation, is used to compensate for timescale deviations caused by signal transmission delays, ensuring that the synchronization error between the simulation clock and the physical system clock is controlled within 100 microseconds. This time-domain synchronization capability is crucial for transient process simulation, especially when analyzing microsecond-level transient events such as short-circuit faults or lightning surges, ensuring that the phase error between the predicted waveform of the impact current and the actual recorded waveform is less than 1%. Addressing the high dynamic characteristics of the wind-solar-storage integrated system, the modeling module also introduced a real-time parameter identification algorithm, identifying key parameters such as the equivalent impedance and inertial time constant of the new energy power generation equipment online every 5 minutes and updating them to the virtual model database. When the system detects that the deviation between the transient impact peak during the actual switching process and the simulated predicted value exceeds the tolerance threshold, it initiates a model parameter self-calibration process. This process uses a particle swarm optimization algorithm to adjust parameters such as the equivalent inductance and capacitance in the virtual model until the error between the simulation results and the actual measured values ​​returns to a reasonable range. This dynamic calibration mechanism allows the model to maintain high-precision simulation capabilities even under long-term changes such as equipment aging and increased contact resistance. Furthermore, the modeling module also features multi-scenario parallel simulation capabilities, capable of simultaneously running simulations of various preset scenarios, including normal operating conditions, N-1 faults, and extreme weather, providing multi-dimensional solution simulation data support for the decision-making module.

[0026] like Figure 1As shown, the core innovation of the decision-making module lies in the design of its multi-objective optimization engine and comprehensive evaluation system. This engine constructs a mathematical model incorporating multi-dimensional constraints such as power supply reliability, equipment loss costs, and operational risk levels. It then uses Pareto optimality theory to screen candidate switching schemes. During the computation process, a multi-dimensional space is first established, including constraints such as voltage fluctuation tolerance, maximum permissible harmonic content, and minimum power supply continuity time. A genetic algorithm is then used to intelligently search the feasible solution space, generating a set of non-dominated solutions that satisfy all hard constraints. For each candidate scheme, the system quantitatively evaluates its comprehensive benefit indicators, including parameters such as expected equipment lifespan depreciation, power quality improvement, and operating cost change rate, forming a multi-dimensional feature vector. The fuzzy comprehensive evaluation model uses a trapezoidal membership function to handle the fuzzy boundary problems of each indicator. The relative importance weights of each evaluation dimension are determined through an expert system weighting method, and finally, a comprehensive score value for each scheme is calculated. To enhance decision-making credibility, the optimization engine's output is cross-validated with equipment health assessment data provided by the early warning module. When the remaining lifespan of switching devices involved in a recommended solution is detected to be below a safety threshold, the system automatically lowers the priority of that solution and triggers a device replacement prompt. This validation mechanism effectively avoids decision-making biases that may result from relying solely on simulation data, demonstrating significant advantages, particularly in addressing gradual faults such as contactor contact oxidation and line insulation aging. In renewable energy microgrid application scenarios, the optimization engine also integrates clean energy utilization indicators. By analyzing the matching degree between photovoltaic output prediction curves and load demand, it prioritizes switching paths that can maximize the absorption of renewable energy. When the system is in islanded operation mode, the engine automatically increases the weighting coefficients of frequency stability and voltage regulation capability to ensure the microgrid's autonomous operation capability after switching operations.

[0027] Specifically, the core technology of the hybrid switching architecture and action coordination mechanism of the execution module lies in the collaborative control strategy of semiconductor switches and mechanical contactors. The semiconductor switch group uses reverse-conducting IGBT devices to form parallel branches, and is equipped with an active current sharing circuit to ensure balanced current distribution in each parallel unit. Its gate drive circuit is designed with nanosecond-level overcurrent protection, which can achieve rapid blocking within 3 microseconds when a short-circuit current is detected. The mechanical contactor group adopts a double-break magnetic blowout arc extinguishing structure, and the surfaces of the moving and stationary contacts are coated with silver-based composite materials. The matching electromagnetic operating mechanism is equipped with a buffer device to reduce closing bounce. The action timing coordination unit controls the opening and closing phases of the two types of devices through a high-precision clock synchronization system. In the switching operation initiation phase, the semiconductor switch is first triggered to conduct at the current zero-crossing point, using its arc-free characteristics to complete the initial current transfer; after the load current stabilizes, the mechanical contactor closes at the voltage zero-crossing point to undertake the steady-state current carrying task. This step-by-step operation strategy reduces the arc energy during the switching process to less than 20% of the traditional method. The timing coordination unit is also designed with a dynamic adjustment function. When high-frequency harmonic components are detected in the load current, it automatically corrects the trigger phase of the semiconductor switch to avoid stress concentration in the device caused by conduction during the harmonic peak. In fault handling scenarios, if the early warning module issues an emergency trip command, the coordination unit will initiate a reverse operation sequence: first, the mechanical contactor is disconnected, and then the semiconductor switch completes the final isolation at a preset safe phase point. This operation sequence can effectively avoid the risk of contact welding during high-current interruption. To improve system reliability, the execution module is also equipped with a dual control loop. The main loop uses fiber optic transmission of PWM drive signals, and the backup loop transmits analog control signals through an isolation transformer coupling. When the main loop fails, it can seamlessly switch to the backup channel.

[0028] Furthermore, the core technology of the early warning module lies in the construction of an intelligent diagnostic system based on a high-frequency signal feature library. The feature library is established through long-term collection of various typical fault arc samples, including fingerprint features of 12 types of faults such as contactor contact wear, insulation degradation, and loose connections. Each type of fault corresponds to a specific combination of time-frequency domain feature vectors. The signal analysis unit uses a sliding time window technique to process the switch feedback signal in real time. The length of each time window can be adaptively adjusted. During steady-state operation, a 200-millisecond window is used for routine monitoring; when a transient process is detected, it automatically switches to a 10-millisecond short window for high-resolution analysis. The spectrum analysis algorithm combines Fast Fourier Transform and wavelet packet decomposition techniques to simultaneously capture steady-state features and transient change information over a wide frequency range. When an abnormal signal with a matching degree exceeding a set threshold is detected, the location algorithm accurately calculates the fault location segment by comparing the amplitude attenuation characteristics and time delay differences of multiple sensor signals, achieving a location accuracy at the branch level. To verify the fault type, the system triggers a dedicated simulation exercise in the modeling module. This reconstructs the physical scenario corresponding to the abnormal signal in a virtual environment, adjusting fault parameters to maximize the correlation between the simulated waveform and the actual measured data. The difference comparison unit uses a dynamic time warping algorithm to handle phase deviations in time-series data, and combines this with correlation coefficient matrix analysis to confirm the fault type. For novel, unknown faults, the system automatically generates feature templates and initiates an online learning mechanism to update the feature library. This module also includes a fault evolution prediction function. By establishing a Markov state transition model, it analyzes the changing trends of fault characteristics. When potential hazards such as a continuous decrease in insulation resistance or a linear increase in contact resistance are detected, preventative maintenance suggestions are generated in advance.

[0029] like Figure 1As shown, the edge computing node constructs a localized intelligent processing system. Its hardware platform adopts a multi-core heterogeneous architecture, including a dedicated signal processing core, an AI inference acceleration core, and a communication coprocessor. After the raw monitoring signal enters the edge node, it first passes through an adaptive filter to eliminate power frequency interference and random noise. The filter parameters are dynamically adjusted based on real-time spectrum analysis results, reducing the noise floor to less than 1 / 5 of the original level while ensuring the integrity of signal features. The feature extraction unit adopts a parallel processing pipeline design, simultaneously performing time-domain statistical calculations, frequency-domain feature extraction, and time-frequency joint analysis. Key feature vectors are transmitted to the analysis module after dimensionality reduction using a compressed sensing algorithm. The data compression algorithm uses an improved sparse autoencoder network, reducing the amount of transmitted data by 60% while maintaining a data reconstruction accuracy of over 98%. The secure communication link adopts an end-to-end encryption scheme based on national cryptographic standards, adding a data integrity check code at the application layer to prevent data tampering during transmission. The cloud management platform deploys a digital twin training cluster, which downloads the anonymized running data uploaded by the edge nodes weekly. It optimizes the control strategy model using deep reinforcement learning algorithms, and the updated model parameters are pushed to the edge nodes after digital signature. This collaborative optimization mechanism enables the system to quickly respond to local real-time demands while continuously absorbing global operational experience. Especially when dealing with changes in regional power grid parameters or new load integration scenarios, it can rapidly adapt to environmental changes through hot model updates. Edge nodes are also designed with an offline emergency mode, which automatically activates the latest policy model stored locally when a network interruption is detected, and records complete operational data to be retransmitted after network recovery, ensuring the system's continuous operation capability under extreme conditions.

[0030] Furthermore, the core of the closed-loop verification mechanism lies in constructing a dynamic parameter calibration channel between the virtual and real systems. This mechanism uses a high-precision measurement unit to capture key parameters such as the peak transient inrush current, voltage drop amplitude, and recovery time during the switching process of the execution module in real time, and compares them with the simulated prediction values ​​of the modeling module synchronously at the millisecond level. When the deviation between the actual measured value and the simulation result exceeds the tolerance threshold, the parameter identification engine automatically starts. First, it performs multi-dimensional feature extraction on the difference data, including waveform similarity analysis, spectral energy distribution comparison, and time-series phase offset calculation, generating a difference feature vector containing a deviation type label. The model correction instruction generation unit selects the corresponding calibration strategy according to the feature vector category: for steady-state parameter deviations (such as equivalent impedance mismatch), the least squares fitting algorithm is used to adjust the line resistance and inductance parameters in the virtual model; for dynamic characteristic deviations (such as transient response delay), the inertial element time constant in the simulation algorithm is updated through transfer function reconstruction technology. The generation of topology update markers relies on network structure change detection. If, during the verification process, the parameter adjustment range of a branch exceeds the set limit three times consecutively, the system determines that the branch may have a topology connection error and triggers a topology scan program to reconfirm the wiring relationships of the physical system. Parameter adjustment coefficients generated during calibration are fed back to the virtual model database of the modeling module via an encrypted data channel, while a version control log records detailed information for each correction. The synchronous transmission mechanism for calibration results is designed with a priority queue. For critical parameters affecting the calculation of evaluation weights (such as load impedance characteristics), the weight coefficient recalculation process of the analysis module is immediately triggered; while minor parameter updates are buffered and processed during system idle periods. This hierarchical processing strategy ensures the real-time performance of core functions while avoiding excessive consumption of system resources. In the application scenario of new energy power stations, the closed-loop verification mechanism has added a special calibration function for the equivalent model of distributed power sources. When the output of the photovoltaic array changes suddenly or the wind turbine switches in or out, the system automatically calls the preset energy type feature template, compares the actual output characteristics with the model prediction curve, and dynamically corrects the MPPT algorithm parameters of the photovoltaic unit or the transient reactance value of the doubly fed motor in the virtual model, so as to ensure that the simulation model always keeps up with the evolution of the operating status of the physical equipment.

[0031] like Figure 1As shown, the priority coverage logic focuses on solving the problem of rapid response under emergency faults, and its technical implementation relies on a multi-level condition monitoring network and intelligent decision-making architecture. The equipment health assessment system constructs a life prediction model for key components by integrating multi-dimensional sensor data, including mechanical life curves based on the number of switching operations, thermal aging models based on temperature rise history, and contact loss calculations based on arc energy analysis. When the assessment result indicates that the remaining life of a key component (such as the main contactor operating mechanism) is lower than the safety threshold, the health status encoder immediately generates a structured early warning message containing the equipment ID, fault level, and expected failure time. The priority determination unit initiates the corresponding response mechanism according to the fault level: for a level one emergency fault (such as a remaining life of less than 24 hours), a red warning code is immediately triggered and a protective switching command is activated; for a level two serious fault, an orange code is generated to initiate a backup plan. The protective switching command contains a preset safe path sequence, which is generated during the system initialization phase by traversing the safety assessment of all possible power supply paths and is updated periodically according to changes in the power grid topology. When the command is executed, the regular control signals of the decision module are forcibly interrupted, the system switches to fault emergency mode, and prioritizes the backup channel with the highest electrical isolation for load transfer. The fault isolation procedure employs a three-tiered, in-depth protection strategy: first, it disconnects the fast circuit breakers upstream and downstream of the faulty section; then, it implements reverse current blocking via a semiconductor switch; and finally, it switches the grounding switch to create a visible disconnection point. Throughout this process, the early warning module continuously monitors the effectiveness of the isolation operation. If the fault current is not completely cut off, a backup protection scheme is activated, such as forcibly disconnecting associated loads or triggering a local power outage contingency plan. When sending a maintenance request, the maintenance request generation unit simultaneously attaches a detailed diagnostic report of the faulty equipment, including historical operation records, key parameter trend charts, and a list of recommended replacement parts. To address complex fault scenarios, the system also features a cross-regional collaborative mechanism. When the local backup power capacity is insufficient, it requests support from adjacent power distribution areas via cross-site communication links, forming a multi-node collaborative power supply network. This priority coverage logic demonstrates significant advantages in critical load scenarios such as data centers, enabling seamless switching before complete equipment failure and reducing the impact of the fault to the level of a single functional module.

[0032] like Figure 2The diagram illustrates an adaptive switching method for power supply and distribution according to the present invention. It includes: S1: collecting voltage waveform signals, current phase signals, and load impedance spectrum signals, and generating a real-time monitoring signal; S2: dynamically weighting and evaluating voltage stability, harmonic distortion rate, frequency deviation, load matching degree, and power supply cost based on the real-time monitoring signal, and generating an evaluation signal; S3: constructing a virtual model of the power supply system and inputting the evaluation signal into the virtual model to generate a simulated signal including impact prediction and recovery time; S4: sorting the simulated signal using an algorithm and generating a control signal containing path selection instructions; S5: controlling semiconductor devices to switch the physical connection state of the power supply according to the control signal, and simultaneously generating a switching feedback signal; S6: forming a device health assessment result based on the switching feedback signal and the abnormal arc signal when the execution module switches the connection state, and transmitting it to the decision module.

[0033] like Figure 2As shown, the technological innovation of the edge computing architecture and cloud collaboration mechanism lies in the construction of a layered and progressive intelligent processing system. Edge computing nodes utilize a heterogeneous computing platform, integrating FPGA chips to accelerate signal preprocessing and running localized analysis algorithms with multi-core processors. At the data acquisition end, nodes have built-in adaptive noise reduction modules that eliminate electromagnetic interference through a hybrid algorithm of wavelet threshold denoising and Kalman filtering, while preserving abrupt changes in the signal waveform. The feature extraction pipeline adopts a parallel architecture design, simultaneously performing time-domain statistical calculations (such as peak factor and waveform factor), frequency-domain feature extraction (harmonic amplitude ratio, spectral centroid), and time-frequency joint analysis (wavelet packet energy entropy). Key feature vectors are uploaded after dimensionality compression through principal component analysis. The local decision engine deploys a lightweight neural network model, capable of completing over 80% of routine switching decisions without relying on the cloud, only forwarding decision requests for complex scenarios to the cloud. The cloud management platform constructs a distributed training cluster, using a federated learning framework to integrate the running data of multiple edge nodes, performing global model optimization while protecting data privacy. The model update mechanism incorporates a differential privacy protection layer, adding Gaussian noise during parameter upload to ensure that sensitive information from individual sites cannot be reverse-engineered. The strategy iteration and optimization process employs a multi-objective evolutionary algorithm to search for the optimal control strategy within tens of millions of historical data samples, and verifies the effectiveness of the new strategy through a digital twin simulation environment. Updated models are released incrementally, generating incremental packages by comparing parameter differences between old and new versions. Edge nodes only need to download megabyte-level update files to complete the model upgrade. To ensure communication reliability, the system adopts a dual-channel redundant transmission design. The primary channel uses a 5G network for low-latency communication, while the backup channel maintains basic data transmission via power line carrier. In extreme network outage scenarios, edge nodes initiate autonomous operation, maintaining basic switching functions based on emergency plans in their local knowledge base and recording complete operation logs to be synchronized to the cloud for post-event analysis after network recovery. This architecture is particularly suitable for widely distributed distribution networks, leveraging the powerful computing resources of the cloud for global optimization while ensuring real-time control requirements through edge intelligence, demonstrating strong collaborative recovery capabilities in response to regional power grid failures.

[0034] This invention discloses an adaptive switching system and method for power supply and distribution. It collects voltage, current, and load impedance signals in real time using multiple types of sensors, constructs an analysis module including a dynamic weighted evaluation model, and comprehensively calculates power quality indicators. It utilizes digital twin technology to create a virtual power supply system, simulating transient responses under different switching paths, and selects the optimal solution through reinforcement learning algorithms. It employs a hybrid actuator combining semiconductor switches and mechanical contactors in parallel, prioritizing rapid switching using solid-state devices to reduce arc impact. A closed-loop verification mechanism is established to compare actual switching parameters with simulation results and dynamically correct model accuracy. A high-frequency arc feature library is introduced to identify abnormal signals, and preventative switching is triggered by equipment lifespan prediction. Through edge computing and cloud-based collaborative optimization strategies, a dynamic balance between power supply cost and security is achieved.

[0035] Therefore, the adaptive switching system and method for power supply and distribution of the present invention solves the problem that traditional power supply and distribution switching systems rely on a single threshold to determine the power supply status and cannot comprehensively evaluate power quality.

[0036] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An adaptive switching system for power supply and distribution, characterized in that, include: The data acquisition module collects voltage waveform signals, current phase signals, and load impedance spectrum signals, and generates real-time monitoring signals. The analysis module dynamically weights and evaluates voltage stability, harmonic distortion rate, frequency deviation, load matching degree, and power supply cost based on the real-time monitoring signals, and generates an evaluation signal. The weight coefficients of the dynamic weighting evaluation in the analysis module are automatically adjusted according to the power supply scenario mode. When the system is in the economy priority mode, the weight ratio of the power supply cost index is increased. When the system is in the safety priority mode, the combined weight of voltage stability and harmonic distortion rate is strengthened. The weight adjustment process achieves pattern recognition through feature extraction of historical operating data and is linked with the virtual model parameter update of the modeling module to form a closed-loop optimization mechanism for the evaluation strategy. The modeling module constructs a virtual model of the power supply system and inputs the evaluation signal into the virtual model to generate a simulated signal that includes impact prediction and recovery time. The virtual model of the modeling module includes a dynamic reconstruction function of the power supply network topology. When a distributed power source is detected or a load node change is detected, the virtual model automatically generates a new equivalent circuit model and verifies its convergence. A time-domain synchronization channel is established between the virtual model and the physical system. The time scale deviation caused by signal transmission delay is compensated by an interpolation algorithm to ensure that the impact prediction result is aligned with the time-domain characteristics of the actual transient process. The decision module sorts and processes the analog signals using an algorithm to generate control signals containing path selection instructions. The algorithm of the decision module includes a multi-objective optimization engine, which quantifies power supply reliability indicators, equipment loss costs, and switching operation risks into multi-dimensional constraints. It uses Pareto front analysis to screen non-dominated solution sets and introduces a fuzzy comprehensive evaluation model to prioritize candidate solutions. The output results of the multi-objective optimization engine are cross-validated with the equipment health assessment results of the early warning module to form a quantitative scoring mechanism for decision credibility. An execution module controls the semiconductor device to switch the physical connection state of the power supply according to the control signal, and generates a switch feedback signal at the same time; The early warning module generates a device health assessment result based on the switch feedback signal and the abnormal arc signal when the execution module switches the connection state, and transmits it to the decision module.

2. The adaptive switching system for power supply and distribution according to claim 1, characterized in that, The data acquisition module includes a wideband impedance detection unit. This unit acquires the load impedance spectrum signal by injecting a test signal at a specific frequency and is linked with the load type identification unit. When a change in load type is detected, the sampling frequency range of the sensor array is dynamically adjusted. The analysis results of the load impedance spectrum signal are synchronously transmitted to the modeling module to correct the equivalent impedance parameters of the load end in the virtual model. A signal compensation mechanism is established between the wideband impedance detection unit and the current phase signal acquisition channel to eliminate the interference of the test signal injection on the original current phase measurement.

3. The adaptive switching system for power supply and distribution according to claim 1, characterized in that, The execution module includes a fast semiconductor switch group and a mechanical contactor group arranged in parallel. The semiconductor switch group is configured to prioritize responding to high-frequency switching commands and undertake transient current transfer. The mechanical contactor group automatically closes after the semiconductor switch completes conduction to maintain steady-state current carrying. An action timing coordination unit is set between the semiconductor switch group and the mechanical contactor group to ensure that the opening and closing operations of the two groups of devices meet the preset phase synchronization requirements.

4. The adaptive switching system for power supply and distribution according to claim 1, characterized in that, The early warning module includes a high-frequency signal feature library, which stores waveform pattern features of typical fault arcs. It performs continuous spectrum analysis on the switch feedback signal through a sliding time window. When an abnormal signal with a matching degree exceeding a set threshold is detected, an early warning command containing fault location information is automatically generated. The early warning command triggers the modeling module to start simulation and deduction of a specific fault scenario, and compares the simulation results with real-time monitoring data to confirm the fault type.

5. The adaptive switching system for power supply and distribution according to claim 1, characterized in that, It includes an edge computing node, which is deployed between the data acquisition module and the analysis module and configured to perform localized preprocessing on the real-time monitoring signal, including signal noise reduction, feature extraction and data compression. The edge computing node establishes a secure communication link with the cloud management platform, regularly uploads the de-identified running data and receives the updated algorithm model to realize remote iterative optimization of the control strategy.

6. The adaptive switching system for power supply and distribution according to claim 1, characterized in that, A closed-loop verification channel is established between the modeling module and the execution module. When the deviation between the measured transient impact peak value and the simulated predicted value exceeds the tolerance range during the actual switching process, a model correction instruction is automatically generated and fed back to the virtual model. The correction instruction includes parameter adjustment coefficients and topology update flags, triggering the virtual model to perform adaptive calibration of structural parameters. At the same time, the calibration results are synchronized to the evaluation weight calculation process of the analysis module.

7. An adaptive switching method for an adaptive switching system for power supply and distribution using any one of claims 1-6, characterized in that, include: S1: Collect voltage waveform signals, current phase signals, and load impedance spectrum signals, and generate real-time monitoring signals; S2: Based on the real-time monitoring signal, dynamically weighted evaluation of voltage stability, harmonic distortion rate, frequency deviation, load matching degree and power supply cost is performed, and an evaluation signal is generated; S3: Construct a virtual model of the power supply system and input the evaluation signal into the virtual model to generate a simulated signal that includes impact prediction and recovery time; S4: The analog signals are sorted using an algorithm to generate control signals containing path selection instructions; S5: Control the semiconductor device to switch the physical connection state of the power supply according to the control signal, and generate a switch feedback signal at the same time; S6: Based on the switch feedback signal and the abnormal arc signal when the execution module switches the connection state, a device health assessment result is generated and transmitted to the decision module.

Images