Voltage treatment device supporting remote automatic adjustment and treatment method

Abstract

The invention provides a voltage treatment device and method supporting remote automatic adjustment, and the device comprises a multi-parameter sensing module, an edge calculation unit, a dual-mode communication module, and an adjustment execution module. The method comprises the steps of multi-parameter synchronous acquisition, data preprocessing and noise suppression, dynamic weight fusion calculation, reinforcement learning decision generation and adjustment instruction execution and feedback. Dynamic fusion calculation is carried out on multiple parameters through the edge calculation unit, the algorithm reasoning time is shortened, a precise adjustment instruction is generated, the real-time response capability is remarkably improved, the response delay is reduced, edge calculation is supported, the edge decision reliability is improved, high-precision multi-parameter cooperative adjustment is achieved, and the communication reliability and the system robustness are enhanced. The communication packet loss rate is reduced, and seamless switching of power-on lines is realized.

Description

Technical Field

[0001] The present application belongs to the field of power grid remote control technology, and specifically relates to a voltage management device and management method that supports remote automatic regulation. Background Art

[0002] With the widespread integration of renewable energy sources (such as photovoltaic and wind power) and the increasing adoption of highly volatile loads (such as electric vehicle fast-charging stations and data centers), power systems are facing increased voltage fluctuations, severe harmonic pollution, and untimely regulation. Traditional local regulators are slow to adjust, lack remote control capabilities, and are unable to respond to sudden issues. Cloud-based remote regulation systems suffer from long communication links, high latency, and high signal loss rates, making them unable to meet the demand for rapid control. Furthermore, current regulation strategies lack learning capabilities and cannot be optimized based on historical data. Summary of the Invention

[0003] The present application provides a voltage management device and a management method that support remote automatic regulation to solve or partially solve the problems raised in the above background technology.

[0004] This application provides a voltage management device that supports remote automatic regulation, including: a multi-parameter sensing module, an edge computing unit, a dual-mode communication module, and a regulation execution module. The functions of each module are as follows:

[0005] Multi-parameter sensing module, synchronously collects multiple parameters of the three-phase power grid line, including at least voltage, current, ambient temperature and humidity;

[0006] The edge computing unit performs dynamic fusion calculations on multiple parameters and generates adjustment instructions based on the Q-learning model reinforcement learning strategy;

[0007] Adjustment execution module, to execute and provide feedback on adjustment instructions;

[0008] The dual-mode communication module is equipped with two wireless communication modules and an intelligent switching circuit.

[0009] Preferably, the multi-parameter sensing module includes:

[0010] High-precision voltage sensing unit, connected to phases A, B, and C of the three-phase power grid line, for real-time acquisition of voltage amplitude and phase angle;

[0011] The broadband current sensing unit is connected in series to the power grid to detect the effective value of the current and the total harmonic distortion rate THD;

[0012] Environmental Monitoring Unit, through I 2 The C bus is connected to the main control chip to collect the ambient temperature T and humidity H in real time.

[0013] Preferably, the edge computing unit includes:

[0014] The main control chip uses a collaborative architecture of an FPGA programmable logic array and an ARM Cortex-A53 processor. The FPGA is used to perform real-time data preprocessing, including Kalman filtering and FFT transformation. The ARM processor runs the Linux system and deploys a dynamic weight allocation algorithm.

[0015] An AI coprocessor embedded in the TensorFlow Lite framework loads a pre-trained reinforcement learning model Q-learning policy library to dynamically generate compensation strategies based on historical data.

[0016] The dynamic threshold generation circuit consists of a digital potentiometer and an operational amplifier. The SPI interface of the digital potentiometer is connected to the GPIO pin of the main control chip to receive adjustment instructions and change the resistance value. The output of the operational amplifier is connected to the IGBT drive circuit to generate the dynamic compensation current threshold I ref , and its calculation formula is:

[0017]

[0018] Among them, K p =0.8, K d =0.1.

[0019] Preferably, the adjustment execution module includes:

[0020] The three-level NPC inverter topology consists of IGBT modules, with a DC bus voltage of 1500V and a neutral point connected to the midpoint of the capacitor through a clamping diode;

[0021] The dynamic reactive power compensation circuit includes a parallel capacitor bank and a series inductor, which is used to ref Generate compensation current I comp , adjustment range 0-300A;

[0022] The fuse protection unit includes a fast-blow fuse and a voltage detection relay. When a sudden voltage change exceeding ±15% of the rated value is detected, the fuse is triggered to operate and cut off the main circuit. The response time is ≤1ms.

[0023] Preferably, the dual-mode communication module includes:

[0024] 5G communication submodule, used to transmit real-time voltage and current data and cloud optimization instructions;

[0025] The LoRa communication submodule is used to transmit key control instructions when the 5G signal strength is lower than a preset threshold, wherein the key control instructions include at least a fuse signal and a compensation current instruction;

[0026] The intelligent switching circuit includes MOS tubes Q1-Q3 and a voltage comparator U1. The input end of the comparator is connected to the 5G signal strength detection circuit, and the output end drives the MOS tube to switch the communication link.

[0027] Preferably, a power management module is further included, and the power management module includes:

[0028] Wide voltage input circuit, input voltage range AC 85V-265V, converted to DC 400V bus voltage through rectifier bridge BR1 and filter capacitor C4;

[0029] Isolated DC-DC converter, using flyback topology, outputs +12V, +5V, +3.3V multi-channel isolated power supply;

[0030] The energy storage unit, including supercapacitor groups SC1-SC3, is used to maintain power supply to the edge computing unit and communication module for ≥30 seconds when the grid is out of power.

[0031] This application also provides a voltage management method supporting remote automatic regulation, comprising the following steps:

[0032] S101: Multi-parameter synchronous acquisition, synchronous acquisition through SPI bus: three-phase voltage instantaneous value U a (t), U b (t), U c (t), effective current value I rms (t) and total harmonic distortion THD(t), ambient temperature T env (t) and humidity H env (t);

[0033] S102: Data preprocessing and noise suppression, performing improved Kalman filtering on raw data, state equation and observation

[0034] x k =Ax k-1 +Bu k-1 +w k

[0035] The measurement equation is defined as: k =Hx k +v k

[0036] Where: A is the state transfer matrix, which is set as A=diag(0.95,0.98,0.97) according to the inertia characteristics of the power grid, H is the observation matrix, H=diag(1,1,1), w k ~N(0,Q),v k ~N(0,R), Q=0.01I, R=0.1I;

[0037] S103: Dynamic weight fusion calculation, dynamically assigning parameter weights according to the grid operation status. The calculation formula is:

[0038]

[0039] Among them, α i is the parameter priority coefficient, i is the value of voltage U, current I and harmonic THD, voltage α U =0.6, current α I =0.3, harmonic α THD =0.1, S i (t) is the parameter sensitivity, J is the optimization objective function, β=0.05 is the time decay factor; j is the index of the sensor data source, n is the number of sensors, S j (t) is the signal value of the jth sensor at time t;

[0040] S104: Reinforcement learning decision generation, generating adjustment instructions based on the Q-learning model, specifically including:

[0041] State space definition:

[0042] Voltage deviation s1 = ΔU(t) = U ref -U avg (t), load change rate s2 = dP / dt, harmonic distortion rate s3 = THD(t), where U ref is the reference voltage, U avg (t) is the average voltage, P is the power;

[0043] Action space definition:

[0044] a1∈{1kHz,2kHz,5kHz}, is the IGBT switching frequency, a2∈[0,300A], is the compensation current amplitude;

[0045] Table update rules:

[0046]

[0047] Among them, the learning rate ɑ=0.2, the discount factor γ=0.9, and the reward function R=-(0.7ΔU 2 +0.2THD 2 +0.1f sw );

[0048] S105: Adjustment instruction execution and feedback:

[0049] Generate PWM control signal, carrier frequency f sw According to the setting of action a1, the duty cycle D is calculated by a2:

[0050] Among them I max =300A, I comp is the compensation current;

[0051] Real-time monitoring of output voltage U out (t), if any of the following conditions is met, the fuse protection is triggered:

[0052] |ΔU(t)|>15%U ref And lasts for more than 100ms;

[0053] THD(t)>10% and lasts for more than 500ms.

[0054] Preferably, the step S103 further includes a dynamic weight correction mechanism, which performs the following operations when a communication interruption is detected:

[0055] Call the local pre-stored strategy library and select the historical strategy that best matches the current power grid topology;

[0056] Loading strategy by priority: PV fluctuation scenario > load mutation scenario > equipment failure scenario;

[0057] The weight coefficient is temporarily adjusted to:

[0058]

[0059] Wherein, t0 is the network disconnection time, k=0.1, is the correction rate factor.

[0060] Preferably, the Q-learning model in step S104 adopts state space compression technology, specifically including:

[0061] Principal component analysis (PCA) was used to reduce the original state dimension from 10 to 3 dimensions, with a retained variance contribution rate of ≥ 95%;

[0062] The Q value function is fitted by the deep neural network DNN, and the network structure is:

[0063] Input layer: 3 nodes, compressed state;

[0064] Hidden layer: 2 layers, 64 nodes per layer, activation function ReLU;

[0065] Output layer: 6 nodes, number of action combinations.

[0066] Preferably, in step S105, the PWM signal generation adopts double closed-loop control, which specifically includes:

[0067] Outer loop voltage control:

[0068]

[0069] in,

[0070] Inner loop current control:

[0071]

[0072] in,

[0073] Compared with the prior art, this application has the following beneficial effects:

[0074] (1) This application uses an edge computing unit to dynamically fuse multiple parameters, reduce algorithm inference time, generate precise adjustment instructions, significantly improve real-time response capabilities, reduce response delays, support edge computing, improve edge decision reliability, achieve high-precision multi-parameter coordinated adjustment, enhance communication reliability and system robustness, reduce communication packet loss rate, and achieve seamless switching of power lines.

[0075] (2) This application adopts an edge computing and dual-mode communication fusion architecture, integrating the edge computing unit (FPGA+ARM+AI coprocessor) with the 5G+LoRa dual-mode communication module for the first time to achieve local real-time decision-making and highly reliable communication.

[0076] (3) This application uses a multi-parameter dynamic weight fusion algorithm to achieve dynamic balance of voltage, current and harmonics. It also uses a dynamic decision-making model based on reinforcement learning to improve the efficiency of strategy generation. It also implements an autonomous network disconnection strategy to support autonomous decision-making during communication interruptions. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] The present application is further described below with reference to the accompanying drawings and examples.

[0078] Figure 1 This is a schematic diagram of the device composition of this application.

[0079] Figure 2 This is a flowchart of the application method.

[0080] Figure 3 This is the topology circuit diagram of the three-level NPC inverter of the regulation execution module of this application.

[0081] Figure 4 This is a schematic diagram of the principle of the wide-voltage input circuit of the voltage management device of this application. DETAILED DESCRIPTION

[0082] For example, certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that hardware manufacturers may use different terms to refer to the same component. This specification and claims do not use differences in names as a way to distinguish components, but use differences in the functions of the components as the criteria for distinction. For example, "including" mentioned throughout the specification and claims is an open term, so it should be interpreted as "including but not limited to". "Approximately" means that within an acceptable error range, those skilled in the art can solve the technical problem within a certain error range and basically achieve the technical effect.

[0083] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "horizontal", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only used to facilitate the description of the present application and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction. Therefore, they should not be understood as limiting the present application.

[0084] In this application, unless otherwise specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they can refer to fixed connection, detachable connection, or integral connection; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; or internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on specific circumstances.

[0085] Example 1

[0086] like Figures 1 to 4 As shown, the present application provides a voltage management device that supports remote automatic regulation, including: a multi-parameter sensing module, an edge computing unit, a dual-mode communication module, and a regulation execution module. The functions of each module are as follows:

[0087] Multi-parameter sensing module, synchronously collects multiple parameters of the three-phase power grid line, including at least voltage, current, ambient temperature and humidity;

[0088] The edge computing unit performs dynamic fusion calculations on multiple parameters and generates adjustment instructions based on the Q-learning model reinforcement learning strategy;

[0089] Adjustment execution module, to execute and provide feedback on adjustment instructions;

[0090] The dual-mode communication module is equipped with two wireless communication modules and an intelligent switching circuit.

[0091] Specifically, the multi-parameter sensing module includes:

[0092] A high-precision voltage sensing unit, connected to phases A, B, and C of the three-phase power line, is used to collect voltage amplitude and phase angle in real time. Furthermore, the high-precision voltage sensing unit has a sampling frequency of ≥10kHz, an accuracy of ±0.1%, and an integrated PT100 temperature compensation circuit to eliminate ambient temperature drift errors;

[0093] The broadband current sensing unit is connected in series to the power grid to detect the effective value of the current and the total harmonic distortion (THD). Furthermore, the broadband current sensing unit has a bandwidth covering 0-10kHz and a harmonic analysis accuracy of ±0.5%.

[0094] Environmental Monitoring Unit, through I 2 The C bus is connected to the main control chip to collect the ambient temperature T and humidity H in real time. Preferably, the data update period is ≤100ms.

[0095] Specifically, the edge computing unit includes:

[0096] The main control chip uses a collaborative architecture of an FPGA programmable logic array and an ARM Cortex-A53 processor. The FPGA is used to perform real-time data preprocessing, including Kalman filtering and FFT transformation. The ARM processor runs the Linux system and deploys a dynamic weight allocation algorithm.

[0097] An AI coprocessor embedded in the TensorFlow Lite framework loads a pre-trained reinforcement learning model Q-learning policy library to dynamically generate compensation strategies based on historical data.

[0098] The dynamic threshold generation circuit consists of a digital potentiometer and an operational amplifier. The SPI interface of the digital potentiometer is connected to the GPIO pin of the main control chip to receive adjustment instructions and change the resistance value. The output of the operational amplifier is connected to the IGBT drive circuit to generate the dynamic compensation current threshold I ref , and its calculation formula is:

[0099]

[0100] Among them, K p =0.8, K d =0.1.

[0101] Specifically, the adjustment execution module includes:

[0102] The three-level NPC inverter topology consists of IGBT modules. The DC bus voltage is 1500V. The neutral point is connected to the midpoint of the capacitor through clamping diodes D1-D4. C1 and C2 are the voltage divider capacitors on the DC bus. The switching frequency of the IGBT module is 2kHz, and the dead time is 2μs. The circuit principle is as follows: Figure 2 As shown;

[0103] The dynamic reactive power compensation circuit includes a parallel capacitor bank and a series inductor, which is used to ref Generate compensation current I comp , adjustment range 0-300A;

[0104] The fuse protection unit includes a fast-blow fuse and a voltage detection relay. When a sudden voltage change exceeding ±15% of the rated value is detected, the fuse is triggered to operate and cut off the main circuit. The response time is ≤1ms.

[0105] Specifically, the dual-mode communication module includes:

[0106] 5G communication submodule, used to transmit real-time voltage and current data and cloud optimization instructions;

[0107] The LoRa communication submodule is used to transmit key control instructions when the 5G signal strength is lower than a preset threshold, wherein the key control instructions include at least a fuse signal and a compensation current instruction;

[0108] The intelligent switching circuit includes a MOS tube and a voltage comparator. The input end of the comparator is connected to the 5G signal strength detection circuit, and the output end drives the MOS tube to switch the communication link.

[0109] Furthermore, the 5G communication submodule adopts the 3GPP Release 16 protocol stack, supports NSA / SA dual-mode networking, has an uplink peak rate ≥500Mbps, and a downlink peak rate ≥1Gbps; the LoRa communication submodule operates in the 470MHz-510MHz frequency band, has a transmit power of +22dBm, and a receive sensitivity of -148dBm, and is used to transmit key control instructions when the 5G signal strength is lower than -95dBm; the intelligent switching circuit has a switching delay of ≤50ms.

[0110] Preferably, the voltage management device further includes a power management module, and the power management module includes:

[0111] Wide voltage input circuit, input voltage range AC 85V-265V, converted to DC400V bus voltage through rectifier bridge BR1 and filter capacitor C4. The circuit principle is as follows Figure 4 As shown;

[0112] The isolated DC-DC converter adopts flyback topology and outputs +12V, +5V, and +3.3V multi-channel isolated power supplies. Furthermore, the conversion efficiency is ≥90% and the ripple voltage is ≤50mV.

[0113] The energy storage unit, including the supercapacitor group, is used to maintain the power supply of the edge computing unit and the communication module for ≥30 seconds when the grid is out of power.

[0114] Example 2

[0115] Based on the voltage management device supporting remote automatic adjustment in Example 1, the present application further provides a voltage management method supporting remote automatic adjustment, comprising the following steps:

[0116] S101: Multi-parameter synchronous acquisition, synchronous acquisition through SPI bus: three-phase voltage instantaneous value U a (t), U b (t), U c (t), effective current value I rms (t) and total harmonic distortion THD(t), ambient temperature T env (t) and humidity H env (t);

[0117] S102: Data preprocessing and noise suppression, performing improved Kalman filtering on raw data, state equation and observation

[0118] x k =Ax k-1 +Bu k-1 +w k

[0119] The measurement equation is defined as: k =Hx k +v k

[0120] Where: A is the state transfer matrix, which is set as A=diag(0.95,0.98,0.97) according to the inertia characteristics of the power grid, H is the observation matrix, H=diag(1,1,1), w k ~N(0,Q),v k ~N(0,R), Q=0.01I, R=0.1I;

[0121] S103: Dynamic weight fusion calculation, dynamically assigning parameter weights according to the grid operation status. The calculation formula is:

[0122]

[0123] Among them, α i is the parameter priority coefficient, i is the value of voltage U, current I and harmonic THD, voltage α U =0.6, current α I =0.3, harmonic α THD =0.1, S i (t) is the parameter sensitivity, J is the optimization objective function, β=0.05 is the time decay factor; j is the index of the sensor data source, n is the number of sensors, S j(t) is the signal value (or characteristic value) of the jth sensor (or data source) at time t;

[0124] S104: Reinforcement learning decision generation, generating adjustment instructions based on the Q-learning model, specifically including:

[0125] State space definition:

[0126] Voltage deviation s1 = ΔU(t) = U ref -U avg (t), load change rate s2 = dP / dt, harmonic distortion rate s3 = THD(t), where U ref is the reference voltage, U avg (t) is the average voltage, P is the power;

[0127] Action space definition:

[0128] a1∈{1kHz,2kHz,5kHz}, is the IGBT switching frequency, a2∈[0,300A], is the compensation current amplitude;

[0129] Table update rules:

[0130]

[0131] Among them, the learning rate ɑ=0.2, the discount factor γ=0.9, and the reward function R=-(0.7ΔU 2 +0.2THD 2 +0.1f sw );

[0132] S105: Adjustment instruction execution and feedback:

[0133] Generate PWM control signal, carrier frequency f sw According to the setting of action a1, the duty cycle D is calculated by a2:

[0134] Among them I max =300A, I comp is the compensation current;

[0135] Real-time monitoring of output voltage U out (t), if any of the following conditions is met, the fuse protection is triggered:

[0136] |ΔU(t)|>15%U ref And lasts for more than 100ms;

[0137] THD(t)>10% and lasts for more than 500ms.

[0138] Preferably, in step S101, the sampling timestamp alignment error of the multi-parameter synchronous acquisition is ≤1ms.

[0139] Preferably, the step S103 further includes a dynamic weight correction mechanism, which performs the following operations when a communication interruption is detected:

[0140] Call the local pre-stored strategy library and select the historical strategy that best matches the current power grid topology;

[0141] Loading strategy by priority: PV fluctuation scenario > load mutation scenario > equipment failure scenario;

[0142] The weight coefficient is temporarily adjusted to:

[0143]

[0144] Among them, t0 is the off-grid time, k = 0.1 is the correction rate factor, and through dynamic correction, the automatic matching of the operation strategy and the grid topology is achieved, thereby improving the adaptability to the on-site environment.

[0145] Preferably, the Q-learning model in step S104 adopts state space compression technology, specifically including:

[0146] Principal component analysis (PCA) was used to reduce the original state dimension from 10 to 3 dimensions, with a retained variance contribution rate of ≥ 95%;

[0147] The Q value function is fitted by the deep neural network DNN, and the network structure is:

[0148] Input layer: 3 nodes, compressed state;

[0149] Hidden layer: 2 layers, 64 nodes per layer, activation function ReLU;

[0150] Output layer: 6 nodes, number of action combinations.

[0151] Preferably, in step S105, the PWM signal generation adopts double closed-loop control, which specifically includes:

[0152] Outer loop voltage control:

[0153]

[0154] in,

[0155] Inner loop current control:

[0156]

[0157] in,

[0158] The above describes the implementation methods of the present application in detail in conjunction with the accompanying drawings, but the present application is not limited to the above implementation methods. Various changes can be made within the scope of knowledge possessed by ordinary technicians in the relevant technical field without departing from the purpose of the present application.

Claims

1. A voltage management device supporting remote automatic regulation, characterized in that: include: The functions of the multi-parameter sensing module, edge computing unit, dual-mode communication module, and adjustment execution module are as follows: Multi-parameter sensing module, synchronously collects multiple parameters of the three-phase power grid line, including at least voltage, current, ambient temperature and humidity; The edge computing unit dynamically integrates multiple parameters and generates adjustment instructions based on the Q-learning model reinforcement learning strategy. Adjustment execution module, to execute and provide feedback on adjustment instructions; The dual-mode communication module is equipped with two wireless communication modules and an intelligent switching circuit.

2. The voltage management device supporting remote automatic regulation according to claim 1, characterized in that: The multi-parameter sensing module includes: High-precision voltage sensing unit, connected to phases A, B, and C of the three-phase power grid line, for real-time acquisition of voltage amplitude and phase angle; The broadband current sensing unit is connected in series to the power grid to detect the effective value of the current and the total harmonic distortion rate THD; Environmental Monitoring Unit, through I 2 The C bus is connected to the main control chip to collect the ambient temperature T and humidity H in real time.

3. The voltage management device supporting remote automatic regulation according to claim 1, characterized in that: The edge computing unit includes: The main control chip uses a collaborative architecture of an FPGA programmable logic array and an ARM Cortex-A53 processor. The FPGA is used to perform real-time data preprocessing, including Kalman filtering and FFT transformation. The ARM processor runs the Linux system and deploys a dynamic weight allocation algorithm. An AI coprocessor embedded in the TensorFlow Lite framework loads a pre-trained reinforcement learning model Q-learning policy library to dynamically generate compensation strategies based on historical data. The dynamic threshold generation circuit consists of a digital potentiometer and an operational amplifier. The SPI interface of the digital potentiometer is connected to the GPIO pin of the main control chip to receive adjustment instructions and change the resistance value. The output of the operational amplifier is connected to the IGBT drive circuit to generate the dynamic compensation current threshold I ref , and its calculation formula is: Among them, K p =0.8, K d =0.

1.

4. The voltage management device supporting remote automatic regulation according to claim 3, characterized in that: The adjustment execution module includes: The three-level NPC inverter topology consists of IGBT modules, with a DC bus voltage of 1500V and a neutral point connected to the midpoint of the capacitor through a clamping diode; The dynamic reactive power compensation circuit includes a parallel capacitor bank and a series inductor, which is used to ref Generate compensation current I comp , adjustment range 0-300A; The fuse protection unit includes a fast-blow fuse and a voltage detection relay. When a sudden voltage change exceeding ±15% of the rated value is detected, the fuse is triggered to operate and cut off the main circuit. The response time is ≤1ms.

5. The voltage management device supporting remote automatic regulation according to claim 1, characterized in that: The dual-mode communication module includes: 5G communication submodule, used to transmit real-time voltage and current data and cloud optimization instructions; The LoRa communication submodule is used to transmit key control instructions when the 5G signal strength is lower than a preset threshold, wherein the key control instructions include at least a fuse signal and a compensation current instruction; The intelligent switching circuit includes a MOS tube and a voltage comparator. The input end of the comparator is connected to the 5G signal strength detection circuit, and the output end drives the MOS tube to switch the communication link.

6. The voltage management device supporting remote automatic regulation according to claim 1, characterized in that: Also included is a power management module, the power management module including: Wide voltage input circuit, input voltage range AC 85V-265V, converted to DC 400V bus voltage through rectifier bridge and filter capacitor; Isolated DC-DC converter, using flyback topology, outputs +12V, +5V, +3.3V multi-channel isolated power supply; The energy storage unit, including the supercapacitor group, is used to maintain the power supply of the edge computing unit and the communication module for ≥30 seconds when the grid is out of power.

7. A voltage management method supporting remote automatic regulation, characterized in that: The steps include: S101: Multi-parameter synchronous acquisition, synchronous acquisition through SPI bus: three-phase voltage instantaneous value U a (t), U b (t), U c (t), effective current value I rms (t) and total harmonic distortion THD(t), ambient temperature T env (t) and humidity H env (t); S102: Data preprocessing and noise suppression, performing improved Kalman filtering on raw data, state equation and observation x k =Ax k-1 +Bu k-1 +w k The measurement equation is defined as: k =Hx k +v k Where: A is the state transfer matrix, which is set as A=diag(0.95,0.98,0.97) according to the inertia characteristics of the power grid, H is the observation matrix, H=diag(1,1,1), w k ~N(0,Q),v k ~N(0,R), Q=0.01I, R=0.1I; S103: Dynamic weight fusion calculation, dynamically assigning parameter weights according to the grid operation status. The calculation formula is: Among them, α i is the parameter priority coefficient, i is the value of voltage U, current I and harmonic THD, voltage α U =0.6, current α I =0.3, harmonic α THD =0.1, S i (t) is the parameter sensitivity, J is the optimization objective function, β=0.05 is the time decay factor; j is the index of the sensor data source, n is the number of sensors, S j (t) is the signal value (or characteristic value) of the jth sensor (or data source) at time t; S104: Reinforcement learning decision generation, generating adjustment instructions based on the Q-learning model, specifically including: State space definition: Voltage deviation s1 = ΔU(t) = U ref -U avg (t), load change rate s2 = dP / dt, harmonic distortion rate s3 = THD(t), where U ref is the reference voltage, U avg (t) is the average voltage, P is the power; Action space definition: a1∈{1kHz,2kHz,5kHz}, is the IGBT switching frequency, a2∈[0,300A], is the compensation current amplitude; Q-table update rules: Among them, the learning rate ɑ=0.2, the discount factor γ=0.9, and the reward function R=-(0.7ΔU 2 +0.2THD 2 +0.1f sw ); S105: Adjustment instruction execution and feedback: Generate PWM control signal, carrier frequency f sw According to the setting of action a1, the duty cycle D is calculated by a2: Among them I max =300A, I comp is the compensation current; Real-time monitoring of output voltage U out (t), if any of the following conditions is met, the fuse protection is triggered: |ΔU(t)|>15%U ref And lasts for more than 100ms; THD(t)>10% and lasts for more than 500ms.

8. The voltage management method supporting remote automatic regulation according to claim 7, characterized in that: The step S103 also includes a dynamic weight correction mechanism, which performs the following operations when a communication interruption is detected: Call the local pre-stored strategy library and select the historical strategy that best matches the current power grid topology; Loading strategy by priority: PV fluctuation scenario > load mutation scenario > equipment failure scenario; The weight coefficient is temporarily adjusted to: Wherein, t0 is the network disconnection time, k=0.1, is the correction rate factor.

9. The voltage management method supporting remote automatic regulation according to claim 7, characterized in that: In step S104, the Q-learning model adopts state space compression technology, which specifically includes: Principal component analysis (PCA) was used to reduce the original state dimension from 10 to 3 dimensions, with a retained variance contribution rate of ≥ 95%; The Q value function is fitted by the deep neural network DNN, and the network structure is: Input layer: 3 nodes, compressed state; Hidden layer: 2 layers, 64 nodes per layer, activation function ReLU; Output layer: 6 nodes, number of action combinations.

10. The voltage management method supporting remote automatic regulation according to claim 7, characterized in that: In step S105, the PWM signal generation adopts double closed-loop control, which specifically includes: Outer loop voltage control: in, Inner loop current control: in,

Images