A photovoltaic access substation leakage current dynamic risk assessment and suppression method based on neural network and CVaR prediction algorithm

Abstract

The application discloses a photovoltaic access substation leakage current dynamic risk assessment and suppression method based on a neural network and a CVaR prediction algorithm, relates to the technical field of fault monitoring and safety protection of an intelligent power distribution system, and comprises the following steps: S1: collecting operation data and meteorological data of a photovoltaic access substation; S2: predicting a leakage current value in a future predetermined time interval in real time based on a trained LSTM model; S3: performing risk assessment on the predicted leakage current value based on a conditional value risk CVaR algorithm; and S4: comparing the risk assessment value with a safe leakage current threshold value.The method has the time sequence prediction advantage of an LSTM neural network, can find potential leakage current increase signs in advance, changes from passive response after the event to active early warning before the event, gains valuable time for taking control measures, and avoids safety accidents caused by response lag due to sudden leakage current increase.

Description

Technical Field

[0001] This invention relates to the field of fault monitoring and safety protection technology for intelligent power distribution systems, specifically a method for dynamic risk assessment and suppression of leakage current in photovoltaic access areas based on neural networks and CVaR prediction algorithms. Background Technology

[0002] With the large-scale integration of new energy sources such as photovoltaic power generation into intelligent power distribution systems, the operating characteristics of low-voltage distribution areas have changed significantly. Among these changes, photovoltaic grid-connected converter devices, due to their topology and operating mode, may generate leakage current to ground. This is especially true in photovoltaic systems without power frequency isolation transformers, where parasitic capacitance exists between the DC and AC sides of the converter device. Under the high-frequency switching action of the converter device, high-frequency leakage current is generated. Excessive leakage current can directly cause power distribution system faults in photovoltaic-connected distribution areas, including: accelerated aging of line insulation leading to single-phase grounding faults, causing the distribution area switch to trip and users to lose power; abnormal grounding of the converter device causing ground arcing, resulting in equipment burnout; and the superposition of leakage current causing residual current devices (RCDs) to malfunction, reducing power supply reliability. Therefore, relevant standards stipulate limits on the leakage current to ground of grid-connected equipment. Leakage risks are usually prevented by installing residual current devices (RCDs). However, in distribution areas with centralized photovoltaic access, the leakage currents of multiple converter devices may be superimposed, causing the total leakage current of the distribution area to exhibit dynamic fluctuation characteristics. Its peak value may momentarily exceed the safety threshold under sudden changes in weather conditions or fault conditions.

[0003] In traditional intelligent power distribution systems, leakage protection is mostly based on a fixed-value tripping method, which directly disconnects the photovoltaic grid when the leakage current exceeds a fixed threshold. This passive protection method cannot provide early warning and may disconnect photovoltaic power generation equipment when it is not necessary, resulting in a decrease in power supply reliability and power generation revenue. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms, comprising the following steps:

[0006] S1: Collect operational and meteorological data for the photovoltaic grid connection area;

[0007] S2: The Long Short-Term Memory (LSTM) neural network model is used to perform time-series learning and training on the operational data and meteorological data, and the leakage current value within a predetermined time interval is predicted in real time based on the trained LSTM model, including the input layer, hidden layer (i.e., LSTM unit), and output layer.

[0008] S3: The VaR threshold of the leakage current under the pre-set confidence level is determined by the conditional value risk (CVaR) algorithm, and the expected value of the leakage current exceeding the threshold is calculated as the risk assessment value.

[0009] S4: Compare the risk assessment value with the safe leakage current threshold. When the risk assessment value exceeds the safe leakage current threshold, generate a risk warning signal. Trigger the leakage current suppression control strategy according to the risk warning signal, adjust the output power of the photovoltaic converter and enable the grounding compensation device to limit the actual leakage current to within the safe leakage current threshold.

[0010] S5: After the control measures are implemented, new leakage current data is collected and fed back to the LSTM model. The model weights are retrained every 24 hours. Steps S1-S4 are executed in a loop to achieve dynamic control.

[0011] Furthermore, in step S1, the operating data includes the output current and output voltage of the photovoltaic converter, the equivalent capacitance of the distribution line to ground, and the total leakage current of the transformer area.

[0012] Furthermore, in step S1, the meteorological data includes light intensity and ambient temperature, with a sampling interval of 1 minute, and the collected data is preprocessed by normalization and missing value imputation.

[0013] Furthermore, in step S2, the state update formula for the LSTM unit includes:

[0014] Forgotten Gate Input gate ;

[0015] Candidate Memory Cell state ;

[0016] Output gate Hidden layer output ;

[0017] in, It is the sigmoid function; It is the hyperbolic tangent function; This is the weight matrix; For bias terms; This is the hidden layer state; The input vector; In cellular state; This is element-wise multiplication.

[0018] Furthermore, in step S3, to quantitatively assess the risk level of leakage current exceeding the limit, a risk assessment model based on the Conditional Value Risk (CVaR) algorithm is introduced, and an LSTM prediction model is used to predict future moments. The leakage current was used to obtain the predicted value. and its probability distribution, let random variable The future leakage current is represented, and its distribution is given by the prediction model, with a defined confidence level. ,like If it is 0.95, then the value risk is... Defined as the quantile of leakage current at this confidence level, i.e.:

[0019]

[0020] in, Let be a probability function, describing the probabilistic characteristics of the leakage current value; The infimum is satisfied. The smallest value.

[0021] Furthermore, conditional value risk Defined as exceeding The expected value of the partial leakage current is used to characterize the average risk level under extreme conditions:

[0022]

[0023] in, For the conditional expectation operator, characterize conditions, The expected value.

[0024] Furthermore, based on the above definition, the confidence level is calculated. Future leakage current The value is used as a risk indicator for leakage current. Exceeding the safe leakage current threshold If the leakage current risk is considered to be at an unacceptable level, control measures need to be triggered.

[0025] Furthermore, in step S4, the load reduction ratio of the control strategy is linearly determined according to the CVaR exceeding the limit ratio, i.e., the load reduction ratio. .

[0026] Furthermore, in step S4, the capacitance adjustment range of the grounding compensation device is 0~5μF.

[0027] This invention provides a method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms, which has the following beneficial effects:

[0028] 1. This method for dynamic risk assessment and suppression of leakage current in photovoltaic access areas based on neural networks and CVaR prediction algorithms integrates the time-series prediction advantages of LSTM neural networks. It can accurately predict future leakage current trends based on real-time operating data of photovoltaic areas in intelligent power distribution systems. Compared with traditional methods that rely solely on fixed threshold monitoring, this method can detect potential signs of increased leakage current in advance, transforming passive post-event response into proactive early warning. When the prediction results show that the future leakage current may approach or exceed the safety threshold, the system can issue an early warning signal in advance, buying valuable time for control measures and avoiding safety accidents caused by sudden increases in leakage current and delayed response.

[0029] 2. This method for dynamic risk assessment and suppression of leakage current in photovoltaic grid-connected areas, based on neural networks and CVaR prediction algorithms, applies the Conditional Value Risk (CVaR) index to the safety assessment of the distribution network. Compared with traditional indicators that only focus on whether limits are exceeded, CVaR can quantify the severity that leakage current may reach under extreme conditions. By calculating the expected value of the tail of the leakage current distribution, this method is more sensitive to risks with low probability but high consequences and can identify dangerous situations that may be overlooked by traditional methods. This quantitative approach to extreme risks improves the accuracy and completeness of risk assessment, helps the power grid to take targeted prevention and control measures, and thus improves the robustness and safety margin of the system. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall process of a method for dynamic risk assessment and suppression of leakage current in photovoltaic access areas based on neural networks and CVaR prediction algorithms according to the present invention.

[0031] Figure 2 This is a schematic diagram showing the comparison curves between the measured leakage current (orange) and the LSTM predicted leakage current (yellow) (from 12:00 PM to 12:00 AM).

[0032] Figure 3 This is a schematic diagram showing the change of CVaR risk value over time.

[0033] Figure 4 This is a schematic diagram comparing the leakage current levels under conditions of no control measures (orange) and with the control strategy of this invention (blue);

[0034] Figure 5 for Figure 1 Upper middle section diagram;

[0035] Figure 6 for Figure 1A schematic diagram of the middle section;

[0036] Figure 7 for Figure 1 Diagram of the lower middle section. Detailed Implementation

[0037] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0038] like Figures 1-7 As shown, the present invention provides a technical solution: a method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms, comprising the following steps:

[0039] S1: Collect operational and meteorological data for the photovoltaic grid connection area;

[0040] The operational data includes the output current and output voltage of the photovoltaic converter, the equivalent capacitance of the distribution line to ground, and the total leakage current of the transformer area. The meteorological data includes the light intensity and ambient temperature. The sampling interval is 1 minute. The collected data is normalized and missing value imputation is performed. Z-score standardization (mean is 0, standard deviation is 1) is used to normalize the data, and linear interpolation is used to imput missing values.

[0041] S2: The Long Short-Term Memory (LSTM) neural network model is used to perform time-series learning and training on the operational data and meteorological data, and the leakage current value within a predetermined time interval is predicted in real time based on the trained LSTM model. The LSTM prediction model adopts a long short-term memory network structure, including an input layer, a hidden layer (LSTM unit), and an output layer.

[0042] The training data consisted of 12 consecutive months of operational data from a photovoltaic power station, covering all four seasons and including extreme weather scenarios such as sunny, cloudy, and rainy days, totaling approximately 525,600 data sets. The LSTM model had 32 hidden layer units, used the Adam optimization algorithm, and had an initial learning rate of 0.001, which decayed by 10% every 50 training epochs. The loss function was the mean squared error (MSE), and the model was considered converged when the MSE ≤ 0.001 for 10 consecutive training epochs. To prevent overfitting, a dropout layer was added after the hidden layers, with a dropout ratio of 0.2.

[0043] The state update formulas for LSTM cells include:

[0044] Forgotten Gate Input gate ;

[0045] Candidate Memory Cell state ;

[0046] Output gate Hidden layer output ;

[0047] in, It is the sigmoid function; It is the hyperbolic tangent function; This is the weight matrix; For bias terms; This is the hidden layer state; The input vector; In cellular state; Element-wise multiplication;

[0048] By repeatedly training on historical data, the network parameters can be determined. and This enables the LSTM model to accurately learn the timing pattern of leakage current and obtain the predicted leakage current value for the next time step through the output layer. ;

[0049] S3: The VaR threshold of the leakage current under the pre-set confidence level is determined by the conditional value risk (CVaR) algorithm, and the expected value of the leakage current exceeding the threshold is calculated as the risk assessment value.

[0050] Confidence level With a confidence level of 0.95 = 95%, a 95% confidence level can cover more than 99% of non-extreme operating conditions, while avoiding excessive warnings;

[0051] To quantitatively assess the risk level of leakage current exceeding limits, a risk assessment model based on the Conditional Value Risk (CVaR) algorithm is introduced, and an LSTM prediction model is used to predict future time points. The leakage current was used to obtain the predicted value. and its probability distribution, let random variable The future leakage current is represented, and its distribution is given by the prediction model, with a defined confidence level. ,like If it is 0.95, then the value risk is... Defined as the quantile of leakage current at this confidence level, i.e.:

[0052]

[0053] in, Let be a probability function, describing the probabilistic characteristics of the leakage current value; The infimum is satisfied. The smallest value;

[0054] Based on this, conditional value risk Defined as exceeding The expected value of the partial leakage current is used to characterize the average risk level under extreme conditions:

[0055]

[0056] in, For the conditional expectation operator, characterize conditions, Expected value;

[0057] Based on the above definition, the confidence level is calculated. Future leakage current The value is used as a risk indicator for leakage current. Exceeding the safe leakage current threshold If this occurs, the risk of future leakage current is considered to be at an unacceptable level, and control measures need to be triggered.

[0058] S4: Compare the risk assessment value with the safe leakage current threshold. When the risk assessment value exceeds the safe leakage current threshold, generate a risk warning signal. Trigger the leakage current suppression control strategy according to the risk warning signal, adjust the output power of the photovoltaic converter and enable the grounding compensation device to limit the actual leakage current to within the safe leakage current threshold.

[0059] The load reduction ratio of the control strategy is determined linearly based on the CVaR exceedance ratio;

[0060] That is, the descent ratio

[0061] The capacitance adjustment range of the grounding compensation device is 0~5μF;

[0062] The safe leakage current threshold is set at 0.5A. Based on the limit requirements for leakage current protection of low-voltage distribution areas in "Installation and Operation of Residual Current Operated Protective Devices" (GB13955-2017), it can be dynamically adjusted for different distribution area types (such as residential distribution areas and industrial distribution areas): 0.5A is maintained for residential distribution areas, while it can be increased to 1.0A for industrial distribution areas (with larger load current). The adjustment trigger condition is automatic switching when the monthly average load current of the distribution area is >200A.

[0063] When adjusting the output power of the photovoltaic converter, the real-time load current of the distribution area (obtained through the distribution transformer monitoring terminal) must be referenced simultaneously. The load reduction step size is set to 5% / minute to avoid voltage fluctuations in the distribution area due to rapid load reduction. The grounding compensation device adopts a stepped adjustment (step interval 0.5μF, 10 steps in total). After adjustment, the leakage current is monitored every 10 seconds. If the leakage current still exceeds the threshold by 10%, the device is switched to the next step. The control strategy switching threshold is set as follows: when reactive power regulation can reduce the leakage current to within 90% of the threshold, reactive power regulation is used first; if the leakage current still exceeds the threshold by 5% after reactive power regulation, the grounding compensation device is activated; if the leakage current still exceeds the threshold after compensation, the load is reduced linearly according to the CVaR excess ratio. ;

[0064] S5: After the control measures are implemented, new leakage current data is collected and fed back to the LSTM model. The model weights are retrained every 24 hours. Steps S1-S4 are executed in a loop to achieve dynamic control. During retraining, only the newly added data in the last 24 hours is updated. Incremental training is adopted to avoid the inefficiency caused by repeated training of the full amount of data.

[0065] This method comprises four main parts: a data acquisition module, a leakage current prediction module, a risk assessment module, and a control execution module, and is implemented according to the following... Figure 1 The process shown enables real-time assessment and control of leakage current risk in photovoltaic power distribution areas. The functions of each module are as follows:

[0066] Data acquisition module: Real-time acquisition of operational status data and environmental data of the photovoltaic grid connection area, which will serve as the input basis for subsequent prediction models;

[0067] Leakage current prediction module: It uses an LSTM neural network model trained on historical data to process the collected real-time data and predict the leakage current value of the transformer area in the next short time (e.g., the next minute or a few minutes). LSTM can learn the nonlinear dynamic relationship between leakage current and factors such as light and temperature, thus giving a more accurate short-term prediction result.

[0068] Risk assessment module: Analyzes the leakage current results predicted by LSTM, and uses the CVaR algorithm to calculate the risk value of the leakage current at a predetermined confidence level. Specifically, based on the predicted value and its possible distribution, it calculates the expected level when the leakage current exceeds the safety threshold, which is used as a risk assessment index to quantify the severity of extreme leakage current events.

[0069] Control execution module: Makes control decisions based on risk assessment results. When the risk assessment value is lower than the safety threshold, the system continues to operate normally. If the risk value exceeds the threshold, an early warning is issued immediately and leakage current suppression control strategy is triggered, such as reducing the output power of the photovoltaic converter or putting on an additional grounding compensation device to limit the actual leakage current within a safe range in a timely manner. After the control action is executed, the system continues to enter the next cycle of monitoring. This cycle is repeated to dynamically control the leakage current risk.

[0070] Example:

[0071] Taking the photovoltaic system operation data and corresponding meteorological data of a certain day as an example, the data sampling interval is 1 minute. The meteorological data includes irradiance (W / m²) and ambient temperature (°C). The maximum irradiance of the day is about 900 W / m², with the peak occurring around noon. The temperature fluctuates between 15°C and 30°C throughout the day. The photovoltaic system is a grid-connected converter with a rated power of 30kW, connected to a 400V low-voltage network and connected to a 10kV medium-voltage network through a 100kVA distribution transformer. The power distribution line is a mixture of overhead lines and cables, with an equivalent capacitance to ground of about 5 microfarads. The converter adopts a transformerless topology, and its DC bus parasitic capacitance to ground is about 30 nanofarads. Therefore, during normal operation, the high-frequency leakage current generated by the converter is about 300 mA.

[0072] To monitor the overall leakage current level of the distribution area, a high-precision leakage current sensor was installed at the neutral point of the transformer in the distribution area to record the waveform of the leakage current to ground in the area over time. On the selected example day, the leakage current gradually increased during the day as the sunlight intensified; however, at noon, due to occasional cloud cover and changes in the operating status of the converter, there were drastic fluctuations in the leakage current and a sudden increase in the peak value. Based on this real data, the prediction and control effects of the method of the present invention will be simulated and the results analyzed.

[0073] First, a portion of the collected historical data is used to train the LSTM model, and the remaining data is used to test the prediction performance. The model input includes the sequence of leakage current, illumination, and temperature over the past few minutes, and the output is the predicted leakage current value for the next minute. Then, for each minute's prediction result, the CVaR risk value at a 95% confidence level is calculated based on the error statistical distribution of the prediction value, and compared with a preset safety threshold of 0.5A to determine whether control intervention is needed. When the predicted risk value exceeds the threshold, control measures are simulated in the simulation to limit the leakage current within the threshold, such as reducing photovoltaic output to reduce the peak leakage current.

[0074] like Figure 2 As shown, Figure 2The curves show a comparison between the measured leakage current (orange) and the LSTM predicted leakage current (yellow) (from 12:00 PM to 12:00 AM). This shows that the LSTM model can accurately predict the trend of leakage current changes during the day, including the process of leakage current increasing with the increase of photovoltaic power output. For most of the time, the predicted curve matches the measured curve well. There is only a certain deviation when there are a few sudden changes (for example, some steep peak values ​​are underestimated). This is mainly due to the difficulty in fully predicting the abnormal behavior and rapid changes of the converter device.

[0075] like Figure 3 As shown, Figure 3 The curve shows the change of the 95% CVaR risk value (blue) over time. The red dashed line represents the safety threshold of 0.5A. For most of the time, the leakage current risk value is below the threshold; however, at certain times around noon, the risk value suddenly rises and exceeds 0.5A, indicating a potential risk event of leakage current exceeding the limit. These high-risk periods correspond to the peak values ​​of the measured leakage current (e.g., Figure 1 The spikes that appeared around 13:00 and 15:00 indicate that the CVaR risk index successfully captured the dangerous situation of abnormally high leakage current, even if these dangers may only have a low probability of occurrence.

[0076] like Figure 4 As shown, Figure 4 The comparison shows the leakage current levels under no control measures (orange) and with the control strategy of this invention (blue). Around 13:00 and 15:00, the leakage current in the transformer area under no control showed a peak value that significantly exceeded the 0.5A threshold. However, after adopting the control strategy of this invention, the leakage current was effectively limited and did not exceed the threshold. In the simulation, once the predicted risk value exceeded the threshold, the system immediately reduced the output power of the photovoltaic converter, successfully reducing the leakage current peak value at the corresponding time to a safe range.

[0077] High-risk triggering conditions: , For safe leakage current threshold ;

[0078] Priority control measures (with minimal impact on power generation):

[0079] Adjust the reactive power output of the photovoltaic converter (by using the SVG function of the converter, increase inductive reactive power, offset part of the ground capacitance current, and reduce leakage current).

[0080] Put the grounding compensation device into operation (connect an adjustable capacitor in parallel with the neutral point of the transformer in the distribution area to divert leakage current).

[0081] Alternative control measures (when priority measures are ineffective):

[0082] Reduce the active power output of the photovoltaic converter (linearly reduce the load according to the "leakage current exceeding the standard ratio", for example, if CVaR exceeds the threshold by 10%, then reduce the load by 10% to avoid excessive load reduction).

[0083] Emergency control measures (leakage current exceeding the threshold for more than 5 seconds):

[0084] Disconnect photovoltaic power from the grid;

[0085] Simulation Results Analysis: The simulation results demonstrate that the proposed method exhibits significant advantages in leakage current risk prediction and control. Firstly, regarding the accuracy of leakage current prediction, the LSTM model successfully captures the dynamic relationship between leakage current and factors such as illumination and temperature, enabling the system to obtain relatively accurate leakage current predictions in most situations, providing a reliable basis for subsequent risk assessment. Even for some unpredictable sudden spikes, the risk level can be promptly identified through CVaR assessment. Secondly, the application of CVaR as a risk assessment indicator makes the measurement of leakage current risk more sensitive and comprehensive. Compared to traditional methods that only determine "whether it exceeds the limit," CVaR considers the degree of exceeding the limit, enabling the system to accurately predict leakage current even when it is near or slightly above the threshold. The system issues a high-risk alarm when the value is exceeded, as demonstrated in the simulation scenario around 13:00: although the predicted average leakage current did not exceed the limit significantly at that time, CVaR captured the possible extreme situation, thus triggering the control strategy in advance. Finally, the proposed control strategy proved its rapid effectiveness. In the simulation, once the CVaR risk value exceeded the limit, the system could take timely measures such as reducing photovoltaic output to prevent the actual leakage current from exceeding the limit. While ensuring safety, the control strategy has little impact on normal photovoltaic power generation: it only slightly reduces photovoltaic output during a very few high-risk periods, and the photovoltaic system still operates at maximum power for most of the day. Therefore, the method of this invention can minimize the impact on photovoltaic power generation revenue while ensuring safety and compliance, achieving a win-win situation for both safety and efficiency.

[0086] Real-world application case: In a residential photovoltaic area, there are 20 30kW converters with a total installed capacity of 600kW. A three-month field test was conducted on them, covering extreme weather such as high temperature (38℃) and heavy rain. The results showed that the number of leakage current exceeding the limit was reduced from 12 times in the traditional method to 2 times, and the photovoltaic power generation utilization rate was increased by 3.8% (due to the reduction of unnecessary tripping).

[0087] Cost-benefit analysis: The cost of new hardware (high-precision sensors, edge computing terminals) is about 20,000 yuan, and the annual power generation revenue due to reduced tripping losses is about 15,000 yuan. The cost can be recovered within 2 years, which has industrial promotion value.

[0088] Based on the above description, the method of the present invention integrates the time-series prediction advantages of LSTM neural networks, and can accurately predict the future leakage current change trend based on the real-time operation data of the photovoltaic power station. Compared with the traditional method that only relies on fixed threshold monitoring, this method can detect potential signs of leakage current increase in advance, realizing the transformation from passive response after the fact to proactive early warning. When the prediction result shows that the future leakage current may approach or exceed the safety threshold, the system can issue an early warning signal in advance, buy valuable time for taking control measures, and avoid safety accidents caused by sudden increase in leakage current and delayed response.

[0089] Furthermore, the Conditional Value Risk (CVaR) index is innovatively applied to the safety assessment of distribution networks. Compared with traditional indicators that only focus on whether limits are exceeded, CVaR can quantify the severity that leakage current may reach under extreme conditions. By calculating the expected value of the tail of the leakage current distribution, this method is more sensitive to risks with low probability but high consequences and can identify dangerous situations that may be overlooked by traditional methods. This quantitative means of extreme risks improves the accuracy and completeness of risk assessment, helps the power grid to take targeted prevention and control measures, and thus improves the robustness and safety margin of the system.

[0090] Moreover, the active suppression control strategy proposed in this invention can quickly control the leakage current within a safe range when a high risk is detected. For example, by reducing the output power of the photovoltaic converter or activating the compensation device, the peak leakage current can be instantly reduced so that it no longer exceeds the threshold. Based on simulation results, this control measure can effectively eliminate the leakage current over-limit phenomenon, and the triggering frequency is very low, with minimal impact on the photovoltaic power generation utilization rate. Compared with the existing practice of directly tripping to cut off the fault source, the method of this invention significantly improves the continuity of system operation and power quality, ensuring reliable power supply for users and equipment safety.

[0091] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms, characterized in that: Includes the following steps: S1: Collect operational and meteorological data for the photovoltaic grid connection area; S2: The Long Short-Term Memory (LSTM) neural network model is used to perform time-series learning and training on the operational data and meteorological data, and the leakage current value within a predetermined time interval is predicted in real time based on the trained LSTM model, including the input layer, hidden layer (i.e., LSTM unit), and output layer. S3: The VaR threshold of the leakage current under the pre-set confidence level is determined by the conditional value risk (CVaR) algorithm, and the expected value of the leakage current exceeding the threshold is calculated as the risk assessment value. S4: Compare the risk assessment value with the safe leakage current threshold. When the risk assessment value exceeds the safe leakage current threshold, generate a risk warning signal. Trigger the leakage current suppression control strategy according to the risk warning signal, adjust the output power of the photovoltaic converter and enable the grounding compensation device to limit the actual leakage current to within the safe leakage current threshold. S5: After the control measures are implemented, new leakage current data is collected and fed back to the LSTM model. The model weights are retrained every 24 hours. Steps S1-S4 are executed in a loop to achieve dynamic control. In step S3, to quantitatively assess the risk level of leakage current exceeding the limit, a risk assessment model based on the Conditional Value Risk (CVaR) algorithm is introduced, and an LSTM prediction model is used to predict future moments. The leakage current was used to obtain the predicted value. and its probability distribution, let random variable The future leakage current is represented, and its distribution is given by the prediction model, with a defined confidence level. ,like If it is 0.95, then the value risk is... Defined as the quantile of leakage current at this confidence level, i.e.: ； in, Let be a probability function, describing the probabilistic characteristics of the leakage current value; The infimum is satisfied. The smallest value; Conditional Value Risk Defined as exceeding The expected value of the partial leakage current is used to characterize the average risk level under extreme conditions: ； in, For the conditional expectation operator, characterize conditions, Expected value; Based on the above definition, the confidence level is calculated. Future leakage current The value is used as a risk indicator for leakage current. Exceeding the safe leakage current threshold If this occurs, the risk of future leakage current is considered to be at an unacceptable level, and control measures need to be triggered. In step S4, the load reduction ratio of the control strategy is determined linearly according to the CVaR exceedance ratio, i.e., the load reduction ratio. .

2. The method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms as described in claim 1, characterized in that: In step S1, the operating data includes the output current and output voltage of the photovoltaic converter, the equivalent capacitance of the distribution line to ground, and the total leakage current of the transformer area.

3. The method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms as described in claim 1, characterized in that: In step S1, the meteorological data includes light intensity and ambient temperature, with a sampling interval of 1 minute. The collected data is then normalized and preprocessed by filling in missing values.

4. The method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms as described in claim 1, characterized in that: In step S2, the state update formula for the LSTM unit includes: forget gate Input gate ; Candidate Memory Cell state ; Output gate Hidden layer output ; in, It is the sigmoid function; It is the hyperbolic tangent function; This is the weight matrix; For bias terms; This is the hidden layer state; The input vector; In cellular state; This is element-wise multiplication.

5. The method for dynamic risk assessment and suppression of leakage current in photovoltaic grid connection areas based on neural networks and CVaR prediction algorithms as described in claim 1, characterized in that: In step S4, the capacitance adjustment range of the grounding compensation device is 0~5μF.

Images