Distributed optical storage local autonomous reactive voltage control system and method

Abstract

The invention discloses a distributed optical storage local autonomous reactive voltage control system and method, and the system achieves the efficient extraction of transformer area features through a lightweight edge sensing terminal, generates a dynamic reactive adjustment instruction through a double-layer reinforcement learning frame, and achieves the cooperative control of optical storage equipment through the combination of a distributed consistency protocol and a priority response strategy. The method is suitable for a weak communication scene of a medium and low voltage distribution network, and a reliable voltage control technology is provided for high-proportion new energy access.

Description

Technical Field

[0001] The present invention relates to the field of power system automation technology, and in particular to a distributed photovoltaic storage on-site autonomous reactive voltage control system and method based on lightweight substation voltage situation perception and self-learning technology. Background Art

[0002] With the large-scale integration of distributed photovoltaic (PV) and energy storage systems (ESS) into distribution networks, power fluctuations on both the source and load sides have significantly increased, leading to frequent problems such as voltage overshoot and three-phase imbalance in substations. Traditional solutions rely on high-density automated metering (AMI) for data collection, but this approach consumes significant amounts of communication bandwidth and edge computing resources and struggles to meet real-time control requirements.

[0003] Existing voltage control primarily relies on a centralized optimization strategy at a master station, requiring coordination of multiple nodes via a remote communication system. However, this model has two significant drawbacks: first, data transmission between the master station and terminal devices is delayed, resulting in delayed control response; second, the complexity of the centralized optimization algorithm grows exponentially with the number of nodes, making it impossible to run in real time within the limited computing resources at the edge of the substation.

[0004] Current reactive power compensation devices (such as SVGs and capacitor banks) often employ a periodic fixed switching strategy, making them unable to adapt to minute-by-minute source-load fluctuations. Although distributed photovoltaic storage systems offer potential for reactive power regulation, the lack of an effective multi-node coordination mechanism means their dynamic reactive power support capabilities are underutilized. These issues severely restrict the effectiveness of voltage quality control in high-penetration distributed energy scenarios. Summary of the Invention

[0005] In response to the above problems, the present invention provides a distributed photovoltaic storage on-site autonomous reactive voltage control system and method.

[0006] The technical solution of the present invention is: a distributed photovoltaic storage on-site autonomous reactive voltage control system, comprising a lightweight edge sensing terminal, a self-learning dynamic decision module and a distributed collaborative execution unit connected in sequence, wherein:

[0007] Lightweight edge sensing terminals are deployed at key nodes in the distribution area (including transformers, photovoltaic inverters, energy storage converters, and load terminals). They collect voltage, current, and power data in real time at a sampling frequency of 1kHz. They use sliding window filtering (window length 100ms) and principal component analysis (PCA) to reduce the data dimension and extract features such as voltage deviation and harmonic distortion rate (the compressed data dimension is ≤10).

[0008] The self-learning dynamic decision-making module is based on a two-layer reinforcement learning framework. The upper layer is a substation-level deep Q network (DQN), which inputs the overall substation voltage deviation, PV output forecast data, and load trend data, and outputs a global voltage regulation target. The lower layer is a device-level distributed actor-critic algorithm, which generates reactive output commands (Q_ref) for each PV storage device based on local voltage characteristics and the upper layer targets.

[0009] The distributed collaborative execution unit is embedded in the photovoltaic inverter and energy storage converter. It exchanges status information (voltage, reactive power) with adjacent nodes through a sparse communication protocol (LoRa / HPLC) and optimizes the objective function based on the consistency protocol:

[0010]

[0011] Where N is the total number of voltage nodes, V i is the voltage of the i-th voltage node, V ref is the reference voltage of the station area, I i is the current flowing through the i-th branch, R i is the impedance of the i-th branch;

[0012] α is the voltage deviation weight, and β is the line loss weight. α = 0.7, β = 0.3, and a priority response strategy is implemented (energy storage devices take precedence over photovoltaic inverters).

[0013] The lightweight edge sensing terminal supports multiple communication protocols such as LoRa, HPLC, RS485, with a communication interval of 5 seconds and a communication distance of 500 meters.

[0014] The sliding window filtering algorithm filters out 50 Hz power frequency interference and retains the dynamic characteristics of voltage fluctuations (time constant ≤ 100 ms).

[0015] The cumulative variance contribution rate of the principal component analysis (PCA) algorithm is ≥94%, achieving a transmission volume compression of more than 80% of the original data.

[0016] The self-learning dynamic decision-making module supports incremental online learning, embeds pre-trained models into edge devices through transfer learning technology, updates policy network weights based on real-time data every 15 minutes, and the computing resource occupancy rate of edge devices is ≤15%.

[0017] The decision cycle of the device-level distributed Actor-Critic algorithm is 1 second, the reactive output regulation accuracy is ±5kVar, the energy storage device response time is ≤1 second, and the photovoltaic inverter response time is ≤2 seconds.

[0018] The priority response strategy of the distributed collaborative execution unit includes: when the node voltage exceeds the limit

[0019] (|V i -V ref |>5% per unit value), the regulation weight of this node is increased by 50%; the reactive power regulation priority of the energy storage device is higher than that of the photovoltaic inverter, and it is preferred to provide capacitive or inductive reactive support (regulation range ±100kVar).

[0020] In the objective function of the consistency protocol, the weights α and β of the voltage deviation term and the line loss term support dynamic adjustment, with initial values ​​of 0.7 and 0.3, respectively.

[0021] A distributed photovoltaic storage on-site autonomous reactive power and voltage control method comprises the following steps:

[0022] Step 1: Lightweight Situational Awareness

[0023] The edge sensing terminal collects key node data at a frequency of 1kHz, removes noise through 100ms sliding window filtering, and uses the PCA algorithm to compress the data dimension to within 10 dimensions, extracting feature quantities such as voltage deviation and harmonic distortion rate.

[0024] Step 2: Self-learning decision generation

[0025] The upper layer DQN network generates the global voltage regulation target based on the overall characteristics of the substation area, and the lower layer Actor-Criti c algorithm combines the local voltage status with the global target to calculate the reactive output command (Q_ref) of each device;

[0026] Step 3: Distributed collaborative execution

[0027] Each photovoltaic storage device exchanges status information through sparse communication every 5 seconds, optimizes the objective function based on the consistency protocol, and responds preferentially to nodes with voltage exceeding the limit. The energy storage device and the photovoltaic inverter work together to perform reactive power regulation to achieve dynamic voltage balance in the substation area (response time ≤ 5 seconds).

[0028] In step 2, incremental online learning is supported, and the pre-trained model is embedded in the edge device through transfer learning technology. The pre-trained model is trained for more than 50,000 steps based on historical substation data (including at least three weather scenarios such as sunny / rainy / cloudy). The policy network weights are updated every 15 minutes during online learning.

[0029] In step 3, the communication protocol is compatible with LoRa and HPLC, the single-node single communication data volume is ≤100 bytes, and the communication bandwidth requirement is reduced by more than 70% compared with the traditional solution.

[0030] The present invention has the following advantages:

[0031] 1. Improved response speed: Breaking through the minute-level delay of traditional centralized control, if the photovoltaic output in the substation suddenly increases (change rate >10% / second), the voltage regulation response time can be shortened to ≤3 seconds, effectively dealing with temporary voltage rise / sag problems.

[0032] 2. Communication efficiency optimization: Through data compression and sparse communication, if the LoRa communication protocol is adopted, the communication bandwidth requirement can be reduced to <50kbps, which is more than 70% lower than the traditional solution, and the single-node single data transmission volume is reduced by 80%, solving the network congestion caused by high-density data transmission.

[0033] 3. Operational performance enhancement:

[0034] Voltage qualification rate: In a typical substation test, if the operating condition change frequency is ≤5 times / hour, the voltage qualification rate can be increased from 88% to 99%, meeting the distribution network voltage quality standard (GB / T12325).

[0035] Reduction of network losses: 12%-18% reduction compared to centralized model predictive control (MPC). If the load power factor of the substation is lower than 0.9, the network loss reduction effect can be further increased to 20%.

[0036] Device utilization: Edge device computing resource occupancy rate is ≤15%, and it can optionally adapt to low-cost embedded hardware such as ARMCortex-A53 and Raspberry Pi.

[0037] 4. Compatibility and scalability: Supports multiple types of communication protocols and photovoltaic storage equipment. If the renewable energy penetration rate in the substation area is ≤30%, this system can be directly deployed; optionally, through parameter adaptive adjustment, it can be compatible with higher penetration rate scenarios (such as above 50%). BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is the system architecture diagram of the present invention (edge ​​perception layer - self-learning decision layer - device execution layer). DETAILED DESCRIPTION

[0039] like Figure 1 As shown, the distributed photovoltaic storage local autonomous reactive voltage control system provided by the present invention includes the following core modules to form a hierarchical control architecture, specifically:

[0040] 1. Lightweight edge perception terminal

[0041] Deployment and data collection: Edge terminals are deployed at transformers, photovoltaic inverters, energy storage converters, and load terminals in the substation area. They integrate voltage / current sensors and optional multi-protocol communication modules (supporting LoRa, HPLC, RS485, or other wireless communication protocols compliant with the IEEE802.15.4 standard) to collect voltage, current, active power, and reactive power data in real time at a sampling frequency of 1kHz.

[0042] Data processing and feature extraction: A 100ms sliding window filtering algorithm or a wavelet transform algorithm can be optionally used to filter out 50Hz power frequency interference and retain the dynamic characteristics of voltage fluctuations (time constant ≤ 100ms). Principal component analysis (PCA) or other dimensionality reduction algorithms (such as LDA) can be used to optionally compress the original 100+-dimensional data to within 10 dimensions (cumulative variance contribution rate ≥ 94%), extract key features such as voltage deviation (unit value, accuracy ±0.5%) and harmonic distortion rate, and achieve more than 80% data compression.

[0043] Communication mechanism: supports sparse communication with an interval of 5 seconds, the single-node single data transmission volume is ≤100 bytes, and the communication distance covers 500 meters. If the communication environment is a high-density area, the communication interval can be optionally adjusted to 2 seconds to ensure real-time data.

[0044] 2. Self-learning dynamic decision-making module

[0045] Two-layer reinforcement learning framework: The upper-layer substation-level deep Q network (DQN) takes as input the overall substation voltage deviation exceeding a preset threshold (e.g., ±5% per unit), the photovoltaic output forecast for the next 15 minutes (with a resolution of 1kW), and the load trend for the past 24 hours (with a 1-minute sampling interval), and outputs the global voltage regulation target (reference voltage V_ref for each node).

[0046] Lower-layer device-level distributed Actor-Critic algorithm: Based on the local voltage status (real-time collected values) and upper-layer targets, it generates reactive power output commands (Q_ref) with a 1-second decision cycle. The PV inverter has an adjustment range of ±50kVar (response time ≤2 seconds), the energy storage converter has an adjustment range of ±100kVar (response time ≤1 second), and the adjustment accuracy is ±5kVar.

[0047] Incremental online learning: The pre-trained model is trained for more than 50,000 steps based on historical data that optionally includes at least three weather scenarios, including sunny, rainy, and cloudy. The edge device updates the policy network weights every 15 minutes using transfer learning technology. If the edge device uses a low-cost ARM Cortex-A53 processor, the computing resource utilization rate is ≤15%.

[0048] 3. Distributed collaborative execution unit

[0049] Consistency protocol optimization: Each PV storage device exchanges voltage and reactive power status through sparse communication, optimizing the following functions with the goal of minimizing voltage deviation and line loss:

[0050]

[0051] Where N is the total number of voltage nodes, V i is the voltage of the i-th voltage node, V ref is the reference voltage of the station area, I i is the current flowing through the i-th branch, R i is the impedance of the i-th branch;

[0052] The voltage deviation item weight α=0.7, and the line loss item weight β=0.3, and dynamic adjustment (such as automatic switching of α / β values ​​according to load peak and valley periods) is optionally supported.

[0053] Priority response strategy: If the node voltage exceeds the limit (|V i -V ref |>5% pu), the node regulation weight is increased by 50%; the energy storage device can optionally take precedence over the photovoltaic inverter to perform reactive compensation due to its faster response speed (≤1 second). If the voltage over-limit type is temporary rise (V i >1.05pu), inductive reactive power support is given priority; if it is a temporary sag (V i <0.95pu), capacitive reactive power support is provided first.

[0054] The present invention realizes efficient extraction of substation characteristics through lightweight edge perception terminals, generates dynamic reactive power regulation instructions using a two-layer reinforcement learning framework, and realizes collaborative control of photovoltaic storage devices by combining a distributed consistency protocol with a priority response strategy.

[0055] A distributed photovoltaic energy storage local autonomous reactive power and voltage control method, comprising:

[0056] Lightweight situational awareness: The edge terminal collects electrical data from key nodes at a frequency of 1kHz. After optionally removing noise through 100ms sliding window filtering or median filtering, the data dimension is compressed to within 10 dimensions using the PCA algorithm to extract feature quantities such as voltage deviation and harmonic distortion rate. The processing cycle is ≤2 seconds.

[0057] Self-learning decision generation: The upper DQN network generates a global voltage regulation target (such as V ref ), if the target voltage deviates from the real-time voltage by more than 3% per unit, the lower-layer Actor-Critical algorithm combines the local voltage status with the global target to calculate the reactive power output instruction Q of each device. refThe pre-trained model incrementally updates the strategy network based on real-time data every 15 minutes, and the update cycle can be shortened to 5 minutes when the output of renewable energy fluctuates violently.

[0058] Distributed collaborative execution: PV and energy storage devices exchange status information every 5 seconds via the LoRa / HPLC protocol. If the communication link is briefly interrupted, a local cache strategy can be optionally enabled to maintain the previous cycle's adjustment instructions. Based on the consistency protocol, the objective function is optimized, and out-of-limit nodes are responded to first. The energy storage and photovoltaic devices coordinately adjust the reactive power output to achieve dynamic voltage balance in the substation area, with an overall response time of ≤5 seconds.

[0059] The following are some examples:

[0060] 1. Example 1: Typical 10kV distribution substation deployment

[0061] Application scenario: Urban substation with a new energy penetration rate of 25%, including 10 photovoltaic inverters (50kW each), 2 energy storage converters (100kV each), and the edge terminal uses the ARM Cortex-A53 processor (main frequency 1.2GHz, memory 512MB).

[0062] (1) System hardware configuration

[0063] Edge perception terminal:

[0064] Sensors: voltage transformer (accuracy 0.2 level), current transformer (accuracy 0.5 level), sampling frequency 1kHz.

[0065] Communication module: Optional LoRa wireless module (frequency band 470MHz, communication distance 500 meters) and HPLC power line carrier module (supporting IEEE1901.1 protocol), the default communication interval is 5 seconds.

[0066] Hardware model: Advantech UNO-1122G, dimensions 100 × 100 × 50 mm, operating temperature -20°C to +60°C.

[0067] Optical storage equipment:

[0068] PV inverter: Huawei SUN2000-50KTL, reactive power regulation range ±50kVar, response time ≤2 seconds. Energy storage converter: Sungrow PCS-100K, reactive power regulation range ±100kVar, response time ≤1 second, battery capacity 200kWh.

[0069] (2) Data processing and decision parameters

[0070] Lightweight perception:

[0071] Sliding window filtering: The window length is 100ms (100 sampling points), which filters out 50Hz fundamental interference and retains the 2nd to 5th harmonic characteristics.

[0072] PCA dimensionality reduction: Input 100-dimensional raw data (instantaneous values ​​of voltage and current) and output 10-dimensional feature quantities (voltage deviation, phase imbalance, harmonic distortion rate, etc.), with a cumulative variance contribution rate of 95.2%.

[0073] Self-learning decision making:

[0074] DQN network structure: input layer 10 dimensions → fully connected layer (64 neurons, ReLU activation) → output layer 3 dimensions (each node V_ref), target update cycle 100 steps.

[0075] Actor-Critic algorithm: The policy network outputs Q_ref (continuous value), the critic network evaluates the state value, the learning rate is 1e-4, and the experience replay buffer capacity is 10,000.

[0076] Online learning: The latest one-hour data (including 300 working condition samples) is extracted every 15 minutes for transfer learning, and the CPU utilization rate of edge devices is 12% to 15%.

[0077] (3) Collaborative control process

[0078] Voltage over-limit response (if the detected A phase voltage is >1.05pu):

[0079] At time T0: the edge terminal detects the phase A voltage of 1.06 pu and triggers the over-limit flag.

[0080] T2 seconds: The over-limit information is broadcast via LoRa, and the energy storage converter responds first, outputting -50kVar inductive reactive power (suppressing voltage rise).

[0081] T5 seconds: The PV inverter outputs -30kVar reactive power based on the local ΔV = +0.06pu. After coordinated regulation, the phase A voltage drops to 1.02pu.

[0082] Consistency protocol execution:

[0083] The objective function weights α=0.7,β=0.3, and each node exchanges V every 5 seconds. i , Q i Data, solve the optimal Q by gradient descent method ref , calculation error ≤1%.

[0084] 2. Example 2: Optimization of High Penetration Low Voltage Station (380V)

[0085] Application scenario: Rural area, with a photovoltaic penetration rate of 35%, the load is mainly single-phase motors (power factor 0.8), there is a three-phase imbalance problem, and the communication method uses RS485 (wired, distance 200 meters).

[0086] (1) Differentiated configuration

[0087] Edge Terminal:

[0088] Add a three-phase current sensor to calculate the three-phase imbalance in real time (Formula: (I max -I min ) / I avg ), the over-limit threshold is 15%.

[0089] The harmonic compensation mode can be optionally enabled. When THD>5%, energy storage devices are preferentially dispatched to provide harmonic reactive power compensation.

[0090] Control strategy adjustment:

[0091] Priority response: If the voltage of a phase exceeds the limit and the three-phase imbalance is greater than 15%, the regulation weight of the equipment in that phase is increased by 100% (α=1.4).

[0092] Actor-Critic algorithm: The input adds the three-phase current amplitude and phase difference, and outputs phase-split reactive power commands (Qa_ref, Qb_ref, Qc_ref) to achieve precise phase regulation.

[0093] (2) Typical working condition processing

[0094] Three-phase unbalance adjustment (B phase voltage 0.92pu, A / C phase 1.0pu):

[0095] At T0: the voltage deviation of phase B is detected to be -8%, triggering phase regulation.

[0096] T1 second: The energy storage converter outputs +80kVar capacitive reactive power to phase B, increasing the voltage to 0.98pu.

[0097] T3 seconds: The adjacent node PV inverter adds +20kVar, and the final B phase voltage stabilizes at 0.99pu, and the three-phase imbalance is reduced from 20% to 5%.

[0098] Communication interruption emergency:

[0099] If the RS485 link failure lasts for more than 10 seconds, the edge terminal can optionally switch to local independent control mode: based on the last received global target (V_ref = 1.0pu), it adjusts according to the preset strategy (such as PV output +5kVar for every 1% voltage deviation) until communication is restored.

[0100] III. Example 3: Robustness Test under Extreme Working Conditions

[0101] Test conditions: Typhoon weather caused a sudden drop in photovoltaic output (from full power to 20% within 10 minutes), while the load suddenly increased (+50%), simulating a voltage sag scenario (V = 0.9pu).

[0102] (1) Model Adaptive Mechanism

[0103] Incremental learning triggers:

[0104] When a voltage deviation of >5% is detected for three consecutive cycles (3 seconds), the edge terminal automatically shortens the learning cycle to 5 minutes and prioritizes collecting current operating condition data (sample weight increased by 50%).

[0105] Strategic Network Updates:

[0106] New emergency regulation rule: If the voltage sags and the energy storage SOC is greater than 30%, the energy storage converter directly outputs +100kVar capacitive reactive power (beyond conventional gradient calculation).

[0107] The computing resource utilization of edge devices temporarily increased to 20% for a duration of ≤15 minutes, and overload protection was not triggered.

[0108] (2) Test results

[0109] Voltage recovery time: It takes 4.2 seconds from the occurrence of a sag to stabilization at 0.98 pu, which is better than traditional droop control (120 seconds).

[0110] Grid loss changes: The peak grid loss during the regulation process was 15kW, and dropped to 8kW after stabilization, a 40% reduction compared to centralized control.

[0111] IV. Example 4: Communication Protocol Compatibility Verification

[0112] Test configuration: Mixed deployment of LoRa (5 nodes) and HPLC (3 nodes) to verify cross-protocol collaboration capabilities.

[0113] Data interaction process:

[0114] The edge terminal uniformly encapsulates data through a protocol conversion module (supporting Modbus / TCP). The LoRa node transmission interval is 5 seconds, the HPLC node interval is 3 seconds, and the system automatically synchronizes timestamps.

[0115] Collaborative precision:

[0116] During cross-protocol communication, the voltage deviation calculation error is ≤0.3% per unit value, and the reactive power instruction consistency reaches 98%, meeting the GB / T 35727-2017 "Technical Provisions for Distributed Generation Access to Distribution Networks".

[0117] The working principle of the present invention is as follows:

[0118] 1. Lightweight edge-sensing technology: A combination of sliding window filtering and PCA dimensionality reduction is proposed to achieve efficient feature extraction on the edge. If the original data contains harmonic interference, the feature extraction accuracy can be improved by adding a wavelet transform preprocessing step, significantly reducing the data transmission volume and computational load.

[0119] 2. Two-layer reinforcement learning decision-making framework: A two-layer control framework is constructed that combines substation-level global optimization with device-level local autonomy. If the upper-layer DQN network detects a substation voltage trend deviation (duration > 10 minutes), it can optionally trigger the emergency adjustment mode of the lower-layer Actor-Critic algorithm, improving autonomous decision-making capabilities under complex working conditions.

[0120] 3. Distributed collaborative control mechanism: Design a sparse communication strategy and priority response rules based on the consistency protocol (such as "if the voltage exceeds the limit, energy storage takes priority"), optionally support adaptive reconstruction in the event of node failure (such as skipping offline nodes for collaborative computing), and break through the collaborative control bottleneck in weak communication scenarios.

[0121] The above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the same. Although specific embodiments are described in detail herein, those skilled in the art may modify them or replace some of the technical features with equivalents, and such modifications do not deviate from the core concept and scope of protection embodied in the embodiments of the present invention.

Claims

1. A distributed photovoltaic storage local autonomous reactive voltage control system, characterized in that: It includes lightweight edge perception terminals, self-learning dynamic decision modules and distributed collaborative execution units connected in sequence, among which, Lightweight edge sensing terminals are deployed at key nodes in the distribution area to collect voltage, current, and power data in real time, perform data dimensionality reduction, and extract feature quantities. The self-learning dynamic decision-making module is based on a two-layer reinforcement learning framework. The upper layer is a substation-level deep Q network, which inputs the overall substation voltage deviation, PV output forecast data, and load trend data, and outputs a global voltage regulation target. The lower layer is a device-level distributed Actor-Critic algorithm, which generates reactive power output instructions for each PV storage device based on local voltage characteristics and the upper layer targets. The distributed collaborative execution unit is embedded in the photovoltaic inverter and energy storage converter. It exchanges status information with adjacent nodes through a sparse communication protocol, optimizes the objective function based on the consistency protocol, and executes the priority response strategy.

2. A distributed photovoltaic storage local autonomous reactive power and voltage control system according to claim 1, characterized in that: The lightweight edge sensing terminal supports multiple communication protocols, including LoRa, HPLC, and RS485, with a communication interval of 5 seconds and a communication distance of 500 meters.

3. The distributed photovoltaic storage local autonomous reactive power and voltage control system according to claim 1, characterized in that: In the lightweight edge perception terminal, data dimension reduction is performed through sliding window filtering and principal component analysis algorithm.

4. The distributed photovoltaic storage local autonomous reactive power and voltage control system according to claim 1, characterized in that: The self-learning dynamic decision-making module supports incremental online learning, embeds pre-trained models into edge devices through transfer learning technology, and updates the policy network weights based on real-time data every 15 minutes.

5. The distributed photovoltaic storage local autonomous reactive power and voltage control system according to claim 4 is characterized in that: The pre-trained model is trained for more than 50,000 steps based on historical station data.

6. The distributed photovoltaic storage local autonomous reactive power and voltage control system according to claim 1, characterized in that: The consistency protocol optimization objective function is: Where N is the total number of voltage nodes, V i is the voltage of the i-th voltage node, V ref is the reference voltage of the station area, I i is the current flowing through the i-th branch, R i is the impedance of the i-th branch; α is the weight of the voltage deviation term, and β is the weight of the line loss term.

7. The distributed photovoltaic storage local autonomous reactive power and voltage control system according to claim 1, characterized in that: The priority response strategy of the distributed collaborative execution unit includes: when the node voltage exceeds the limit, the node adjustment weight is increased by 50%; The reactive power regulation priority of energy storage equipment is higher than that of photovoltaic inverters, and it gives priority to providing capacitive or inductive reactive power support.

8. A distributed solar-storage on-site autonomous reactive voltage control method, characterized in that: The distributed photovoltaic storage on-site autonomous reactive power and voltage control system according to any one of claims 1 to 7 comprises the following steps: Step 1: Lightweight Situational Awareness The edge sensing terminal collects key node data at a frequency of 1kHz, removes noise through 100ms sliding window filtering, and uses the principal component analysis algorithm to compress the data dimension to within 10 dimensions to extract feature quantities. Step 2: Self-learning decision generation The upper-level substation-level deep Q network generates a global voltage regulation target based on the overall characteristics of the substation. The lower-level Actor-Critic algorithm combines the local voltage status with the global target to calculate the reactive output command of each device. Step 3: Distributed collaborative execution Each photovoltaic storage device exchanges status information through sparse communication every 5 seconds, optimizes the objective function based on the consistency protocol, and responds preferentially to nodes with voltage exceeding the limit. The energy storage device and the photovoltaic inverter work together to perform reactive power regulation to achieve dynamic voltage balance in the substation area.

9. The distributed photovoltaic storage local autonomous reactive power and voltage control method according to claim 8, characterized in that: In step 2, incremental online learning is supported, and the pre-trained model is embedded in the edge device through transfer learning technology. The pre-trained model is trained for more than 50,000 steps based on historical station data, and the policy network weights are updated every 15 minutes during online learning.

10. The distributed photovoltaic storage local autonomous reactive power and voltage control method according to claim 8, characterized in that: In step 3, the communication protocol is compatible with LoRa and HPLC, and the amount of data transmitted by a single node in a single communication is ≤ 100 bytes.

Images