Photovoltaic power station monitoring system

Abstract

The invention discloses a photovoltaic power station monitoring system. Environmental data and equipment state data are collected in real time through a sensor; cleaning, de-noising and standardizing the collected original data; decomposing the generating capacity and the equipment state data through Fourier transform; performing convolution processing on the data through a multi-scale convolutional neural network, and extracting features of different time scales; through an LSTM model, a BP neural network and a GRU model, generating capacity prediction, equipment fault diagnosis and quick response of monitoring equipment are realized; integrating output results of the LSTM model, the BP neural network and the GRU model through weighted fusion to generate a unified decision support result; and through data visualization, generating capacity prediction, an equipment state and a historical trend are displayed, and a customized operation report is generated. According to the system, the intelligent level of a photovoltaic power generation system is effectively improved, and the fault prediction and operation monitoring capabilities of equipment are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station monitoring, and in particular to a photovoltaic power station monitoring system. Background Art

[0002] Photovoltaic power stations convert solar energy into electrical energy through solar panels. The power generation process is affected by environmental factors such as light, temperature, and humidity. With the continuous growth of global demand for renewable energy, photovoltaic power stations have occupied an increasingly important position in energy supply.

[0003] Existing photovoltaic power station monitoring systems often lack in-depth analysis of time series when processing data, especially in capturing long-term trends, seasonal changes and short-term fluctuations. Traditional monitoring systems often use a single model for power generation forecasting or equipment fault diagnosis, which cannot effectively handle complex nonlinear relationships and fluctuations on different time scales, resulting in limited prediction accuracy and fault diagnosis capabilities of the system. In addition, short-term abnormal fluctuations of equipment are often difficult to identify in advance, which can easily cause sudden failures, thereby affecting the power generation efficiency and equipment life of the power station.

[0004] In response to the above problems, the present invention provides a photovoltaic power station monitoring system, which combines time series decomposition, multi-scale feature extraction and multiple neural network models to process long-term trends, seasonal fluctuations and short-term equipment status fluctuations respectively, to achieve more accurate power generation prediction and earlier equipment failure warning. Summary of the invention

[0005] In view of the above problems, the present invention provides a photovoltaic power station monitoring system to solve the problems that the existing system model is single and cannot effectively handle complex nonlinear relationships and fluctuations on different time scales, and short-term abnormal fluctuations of equipment are often difficult to identify in advance.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides a photovoltaic power station monitoring system, characterized in that it includes:

[0008] Step S1, data collection, real-time collection of environmental data and equipment status data through sensors;

[0009] Wherein step S1 also includes the following sub-steps:

[0010] S1-1, the sensor uses an intelligent sensor based on the STM32F407 microcontroller, which collects environmental data and equipment status data in real time through the RS-485 bus;

[0011] S1-2, environmental data includes light intensity, temperature, humidity, wind speed, wind direction and rainfall; equipment status data includes current, voltage, power, equipment temperature and vibration information.

[0012] Step S2, data preprocessing, cleaning, denoising and standardizing the collected raw data;

[0013] Wherein step S2 also includes the following sub-steps:

[0014] S2-1, use the mean interpolation method to clean the original data, as shown in formula (1):

[0015]

[0016] Among them, X i is the data point of the current feature, and n is the total number of data points;

[0017] S2-2, use the moving average method to remove the noise in the data, as shown in formula (2):

[0018]

[0019] Among them, X t Represents data at time t;

[0020] S2-3, using standardization processing, converts features of different dimensions into standard normal distribution to ensure that the feature values ​​are in the same range, as shown in formula (3):

[0021]

[0022] Among them, X is the original eigenvalue, μ is the mean of the feature, and σ is the standard deviation of the feature.

[0023] Step S3, time series decomposition, decomposing the collected power generation and equipment status data into trend components, seasonal components and random components through Fourier transform;

[0024] Wherein step S3 also includes the following sub-steps:

[0025] S3-1, use Fourier transform to convert time series data into frequency domain, and extract the periodic characteristics of the data by identifying the frequency, as shown in formula (4):

[0026]

[0027] Among them, X(f) is the result in the frequency domain, x(t) is the time series signal, and f is the frequency.

[0028] S3-2, according to the results of the frequency domain, the data is decomposed into trend component, seasonal component and random component. The low-frequency component in the frequency domain results corresponds to the trend component; the medium and high-frequency components in the frequency domain results correspond to the seasonal component; and the high-frequency components in the frequency domain results correspond to the random component.

[0029] Step S4, multi-scale feature extraction, performing convolution processing on the data through a multi-scale convolutional neural network to extract features of different time scales;

[0030] Wherein step S4 also includes the following sub-steps:

[0031] S4-1, uses a multi-scale convolutional neural network to capture short-term changes at the hour level, that is, instantaneous changes in device status;

[0032] S4-2, using multi-scale convolutional neural networks to capture daily periodic fluctuations, that is, the fluctuation patterns of daily power generation and equipment status;

[0033] S4-3, using multi-scale convolutional neural networks to capture long-term trends at the weekly level, that is, to capture long-term seasonal changes or equipment aging trends;

[0034] S4-4, uses a multi-scale convolutional neural network to capture the above data changes and extract multi-level features of different time dimensions, as shown in formula (5):

[0035]

[0036] in, is the feature after the lth layer of convolution, is the convolution kernel weight of layer (1), X (1-1) is the input feature of the previous layer, f is the activation function, b (l) For bias.

[0037] Step S5, multi-model combination prediction, using LSTM model, BP neural network, and GRU model to achieve power generation prediction, equipment fault diagnosis, and rapid response of monitoring equipment;

[0038] Wherein, in step S5, the following sub-steps are also included:

[0039] S5-1, the processed data is input into the LSTM model for training. After the LSTM model is trained, it will accept new environmental data and predict future power generation changes, as shown in formulas (6)-(10)

[0040] i t =σ(W i ·[h t-1 , x t ]+b i ) Formula (6)

[0041] f t =σ(W f ·[h t-1 , x t ]+b f ) Formula (7)

[0042] o t =σ(W o ·[h t-1 , x t ]+b o ) Formula (8)

[0043] c t =f t ·c t-1 +i t tanh(W c ·[h t-1 , x t ]+b c ) Formula (9)

[0044] h t =o t ·tanh(c t ) Formula (10)

[0045] Among them, i t is the input gate, f t For the forget gate, t is the output gate, c t is the cell state update, h t To update the hidden state; LSTM processes long sequence data through the control mechanism of input gate, forget gate and output gate;

[0046] S5-2, analyze the health status of the equipment through the BP neural network to diagnose whether the equipment has a risk of failure. The equipment operation status data is input into the BP neural network. The network calculates the activation value of each layer through forward propagation until the health status, activation value and failure probability of the output equipment are output, as shown in formulas (11)-(12):

[0047]

[0048] y=σ(W (L) ·h (L-1) +b (L) ) Formula (12)

[0049] Where f(x) is the σ function; y is the output value, i.e., the failure probability of the equipment, W (L) is the weight matrix of the output layer, h (L -1) is the output value of the hidden layer, b (L)is the bias of the output layer, σ is the activation function, i.e., f(x), and the σ function is used to convert the output value into a probability (ranging between 0 and 1). If the failure probability is close to 1, there is a problem with the device; if it is close to 0, the device is in a relatively healthy state.

[0050] S5-3: The GRU model is used to quickly process short-term fluctuations of the equipment and identify whether the equipment has short-term abnormalities. The GRU model processes short-term state changes of the equipment and can respond to abnormal changes in the equipment state, capture short-term fluctuations and abnormal conditions of the equipment, and perform short-term fault prediction, as shown in formula (13):

[0051]

[0052] Among them, (X) is the short-term equipment fluctuation data, and fGRU(X) is the fault prediction result output by the GRU model.

[0053] Step S6, data fusion and decision support, integrating the output results of the LSTM model, BP neural network and GRU model through weighted fusion to generate a unified decision support result;

[0054] Wherein, in step S6, the following sub-steps are also included:

[0055] S6-1, use the weighted fusion algorithm to integrate the three model data, as shown in formula (14):

[0056]

[0057] Among them, w1, W2, and W3 are the weights of LSTM, BP, and GRU models respectively. For their respective prediction results;

[0058] S6-2, based on the fused prediction results, the system generates operational suggestions and decision support information, including:

[0059] The forecast results are evaluated, and if anomalies or significant deviations are detected, the system will issue a warning signal; based on historical data and forecast results, decision support reports for different operating scenarios are generated to help users choose the optimal operating strategy.

[0060] Step S7: data visualization, displaying power generation forecast, equipment status, historical trends through the interface, and generating customized operation reports;

[0061] Wherein, in step S7, the following sub-steps are also included:

[0062] S7-1, data visualization displays key data of the photovoltaic power station in real time through a graphical interface, including environmental data, equipment status, power generation forecast, real-time data and historical data trend charts;

[0063] S7-2, customized operation report Based on historical data and real-time monitoring information, customized operation reports are automatically generated. The report content can adjust system parameters and set warning thresholds as needed.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. By introducing time series decomposition and LSTM model, the present invention can decompose the power generation data of photovoltaic power stations into trend, seasonal and random components, and predict power generation based on long-term trends and cyclical changes, which can better capture long-term trends and seasonal fluctuations and significantly improve the accuracy of power generation prediction.

[0066] 2. The present invention uses a BP neural network combined with multi-scale feature extraction to perform a refined diagnosis of the health status of the equipment. By utilizing the multi-time scale characteristics of the equipment operation data, it can detect the potential failure risks of the equipment at an earlier stage, especially for the long-term aging and short-term fluctuations of the equipment, and effectively reduce the occurrence of sudden failures of the equipment.

[0067] 3. The present invention processes short-term equipment status data through the GRU model, which can capture instantaneous fluctuations in equipment operation, quickly respond to short-term anomalies, have higher sensitivity, and can issue fault warnings in time to avoid major equipment damage and downtime caused by untimely discovery of short-term faults.

[0068] 4. The present invention adopts a multi-scale convolutional neural network to extract data features at different time scales (hourly, daily, and weekly), which can comprehensively capture the changes of photovoltaic power stations at different time levels in the short, medium, and long term, effectively improving the system's ability to understand complex change patterns and making predictions and diagnosis more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It is understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0070] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but is only for selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0072] Please refer to Figure 1 The present invention provides a flow chart of a photovoltaic power station monitoring system, which includes the following steps:

[0073] Step S1, data collection, real-time collection of environmental data and equipment status data through sensors;

[0074] Wherein step S1 also includes the following sub-steps:

[0075] S1-1, the sensor uses an intelligent sensor based on the STM32F407 microcontroller, which collects environmental data and equipment status data in real time through the RS-485 bus;

[0076] S1-2, environmental data includes light intensity, temperature, humidity, wind speed, wind direction and rainfall; equipment status data includes current, voltage, power, equipment temperature and vibration information.

[0077] It should be noted that the STM32F407 microcontroller is often used in low-power devices with high real-time requirements. It is responsible for preliminary processing of sensor data, such as converting analog signals into digital signals or unifying the formats of different sensor data. The RS-485 bus protocol uses the RS-485 bus for transmission to ensure stable and long-distance data transmission, and can effectively cope with data transmission needs in various harsh environments on site.

[0078] Step S2, data preprocessing, cleaning, denoising and standardizing the collected raw data;

[0079] Wherein step S2 also includes the following sub-steps:

[0080] S2-1, use the mean interpolation method to clean the original data, fill the missing values ​​with the mean of the feature, and interpolate the missing values, as shown in formula (1):

[0081]

[0082] Among them, X i is the data point of the current feature, and n is the total number of data points;

[0083] S2-2, use the moving average method to remove noise in the data, and use the sliding window as the average value of 5 or 10 days, as shown in formula (2):

[0084]

[0085] Among them, X t Represents data at time t;

[0086] S2-3, using standardization processing, converts features of different dimensions into standard normal distribution to ensure that the feature values ​​are in the same range, as shown in formula (3):

[0087]

[0088] Among them, X is the original eigenvalue, μ is the mean of the feature, and σ is the standard deviation of the feature.

[0089] It should be noted that the collected raw data is cleaned, standardized and denoised to ensure the accuracy and availability of the data in preparation for input into the neural network for training.

[0090] Step S3, time series decomposition, decomposing the collected power generation and equipment status data into trend components, seasonal components and random components through Fourier transform;

[0091] Wherein step S3 also includes the following sub-steps:

[0092] S3-1, use Fourier transform to convert time series data into frequency domain, and extract the periodic characteristics of the data by identifying the frequency, as shown in formula (4):

[0093]

[0094] Among them, X(f) is the result in the frequency domain, x(t) is the time series signal, and f is the frequency;

[0095] S3-2, according to the results of the frequency domain, the data is decomposed into trend component, seasonal component and random component. The low-frequency component in the frequency domain results corresponds to the trend component; the medium and high-frequency components in the frequency domain results correspond to the seasonal component; and the high-frequency components in the frequency domain results correspond to the random component.

[0096] It should be noted that long-term trend analysis: by extracting trend components, the system can better understand the changing trends of power generation and equipment status over time, and support long-term performance evaluation and equipment maintenance strategies; seasonal fluctuation capture: seasonal components help the system identify periodic fluctuations, making power generation forecasts more accurate, especially when there are significant changes in different seasons or time periods; noise removal: by decomposing and removing random noise, the system's prediction and fault diagnosis models can focus on useful information and improve prediction accuracy.

[0097] Step S4, multi-scale feature extraction, performing convolution processing on the data through a multi-scale convolutional neural network to extract features of different time scales;

[0098] Wherein step S4 also includes the following sub-steps:

[0099] S4-1, uses a multi-scale convolutional neural network to capture short-term changes at the hour level, that is, instantaneous changes in device status;

[0100] S4-2, using multi-scale convolutional neural networks to capture daily periodic fluctuations, that is, the fluctuation patterns of daily power generation and equipment status;

[0101] S4-3, using multi-scale convolutional neural networks to capture long-term trends at the weekly level, that is, to capture long-term seasonal changes or equipment aging trends;

[0102] S4-4, uses a multi-scale convolutional neural network to capture the above data changes and extract multi-level features of different time dimensions, as shown in formula (5):

[0103]

[0104] in, is the feature after the lth layer of convolution, is the convolution kernel weight of the (l)th layer, X (l-1) is the input feature of the previous layer, f is the activation function, b (l) For bias.

[0105] It should be noted that short-term change capture: the hourly convolution layer captures the instantaneous changes of the equipment, which helps the system respond quickly to short-term anomalies; periodic fluctuation analysis: the daily convolution layer identifies the periodic fluctuations of daily power generation and equipment status, and adapts to the changes in day and night cycles; long-term trend identification: the weekly convolution layer identifies long-term changes in equipment status, such as equipment aging or seasonal effects, to help better predict future power generation and equipment status.

[0106] Step S5, multi-model combination prediction, using LSTM model, BP neural network, and GRU model to achieve power generation prediction, equipment fault diagnosis, and rapid response of monitoring equipment;

[0107] Wherein, in step S5, the following sub-steps are also included:

[0108] S5-1, the processed data is input into the LSTM model for training. After the LSTM model is trained, it will accept new environmental data and predict future power generation changes, as shown in formulas (6)-(10)

[0109] i t =σ(W i ·[h t-1 , x t ]+b i ) Formula (6)

[0110] f t =σ(W f ·[h t-1 , x t ]+b f ) Formula (7)

[0111] o t =σ(W o ·[h t-1 , x t ]+b o ) Formula (8)

[0112] c t =f t ·c t-1 +i t tanh(W c ·[h t-1 , x t ]+b c ) Formula (9)

[0113] h t =o t ·tanh(c t ) Formula (10)

[0114] Among them, i t is the input gate, f t For the forget gate, t is the output gate, c t is the cell state update, h t To update the hidden state; LSTM processes long sequence data through the control mechanism of input gate, forget gate and output gate;

[0115] S5-2, analyze the health status of the equipment through the BP neural network to diagnose whether the equipment has a risk of failure. The equipment operation status data is input into the BP neural network. The network calculates the activation value of each layer through forward propagation until the health status, activation value and failure probability of the output equipment are output, as shown in formulas (11)-(12):

[0116]

[0117] y=σ(W (L) ·h (L-1) +b (L) ) Formula (12)

[0118] Where f(x) is the σ function; y is the output value, i.e., the failure probability of the equipment, W (L) is the weight matrix of the output layer, h (L -1) is the output value of the hidden layer, b (L) is the bias of the output layer, σ is the activation function, i.e., f(x), and the σ function is used to convert the output value into a probability (ranging between 0 and 1). If the failure probability is close to 1, there is a problem with the device; if it is close to 0, the device is in a relatively healthy state.

[0119] S5-3: The GRU model is used to quickly process short-term fluctuations of the equipment and identify whether the equipment has short-term abnormalities. The GRU model processes short-term state changes of the equipment and can respond to abnormal changes in the equipment state, capture short-term fluctuations and abnormal conditions of the equipment, and perform short-term fault prediction, as shown in formula (13):

[0120]

[0121] Among them, (X) is the short-term equipment fluctuation data, and fGRU(X) is the fault prediction result output by the GRU model.

[0122] It should be noted that LSTM is good at processing long-term dependent data. By inputting historical environmental data and power generation data, it can predict the power generation trend in the future. Since the power generation of photovoltaic power stations is highly correlated with environmental factors such as time, light intensity, and temperature, and these data are typical time series data, LSTM can be used to capture the historical change trend of power generation and then predict future power generation.

[0123] Step S6, data fusion and decision support, integrating the output results of the LSTM model, BP neural network and GRU model through weighted fusion to generate a unified decision support result;

[0124] Wherein, in step S6, the following sub-steps are also included:

[0125] S6-1, use the weighted fusion algorithm to integrate the three model data, as shown in formula (14):

[0126]

[0127] Among them, w1, w2, and w3 are the weights of LSTM, BP, and GRU models respectively. For their respective prediction results;

[0128] S6-2, based on the fused prediction results, the system generates operational suggestions and decision support information, including:

[0129] The forecast results are evaluated, and if anomalies or significant deviations are detected, the system will issue a warning signal; based on historical data and forecast results, decision support reports for different operating scenarios are generated to help users choose the optimal operating strategy.

[0130] It should be noted that the weights can be adjusted adaptively. During the model training process, the weights are dynamically adjusted according to the prediction results of the current model and the weights are allocated according to the accuracy.

[0131] Step S7: data visualization, displaying power generation forecast, equipment status, historical trends through the interface, and generating customized operation reports;

[0132] Wherein, in step S7, the following sub-steps are also included:

[0133] S7-1, data visualization displays key data of the photovoltaic power station in real time through a graphical interface, including environmental data, equipment status, power generation forecast, real-time data and historical data trend charts;

[0134] S7-2, customized operation report Based on historical data and real-time monitoring information, customized operation reports are automatically generated. The report content can adjust system parameters and set warning thresholds as needed.

[0135] It should be noted that environmental data includes temperature and light intensity; equipment status data includes current, voltage, and inverter efficiency; and the system's prediction results include power generation prediction and equipment health status.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention has various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A photovoltaic power station monitoring system, characterized in that: The following steps are involved: Step S1, data collection, real-time collection of environmental data and equipment status data through sensors; Step S2, data preprocessing, cleaning, denoising and standardizing the collected raw data; Step S3, time series decomposition, decomposing the collected power generation and equipment status data into trend components, seasonal components and random components through Fourier transform; Step S4, multi-scale feature extraction, performing convolution processing on the data through a multi-scale convolutional neural network to extract features of different time scales; Step S5, multi-model combination prediction, using LSTM model, BP neural network, and GRU model to achieve power generation prediction, equipment fault diagnosis, and rapid response of monitoring equipment; Step S6, data fusion and decision support, integrating the output results of the LSTM model, BP neural network and GRU model through weighted fusion to generate a unified decision support result; Step S7: Data visualization, displaying power generation forecast, equipment status, historical trends through the interface, and generating customized operation reports.

2. A photovoltaic power station monitoring system according to claim 1, characterized in that: Wherein step S1 also includes the following sub-steps: S1-1, the sensor uses an intelligent sensor based on the STM32F407 microcontroller, which collects environmental data and equipment status data in real time through the RS-485 bus; S1-2, environmental data includes light intensity, temperature, humidity, wind speed, wind direction and rainfall; equipment status data includes current, voltage, power, equipment temperature and vibration information.

3. A photovoltaic power station monitoring system according to claim 1, characterized in that: Wherein step S2 also includes the following sub-steps: S2-1, use the mean interpolation method to clean the original data, as shown in formula (1): Among them, X i is the data point of the current feature, and n is the total number of data points; S2-2, use the moving average method to remove the noise in the data, as shown in formula (2): Among them, X t Represents data at time t; S2-3, using standardization processing, converts features of different dimensions into standard normal distribution to ensure that the feature values ​​are in the same range, as shown in formula (3): Among them, X is the original eigenvalue, μ is the mean of the feature, and σ is the standard deviation of the feature.

4. A photovoltaic power station monitoring system according to claim 1, characterized in that: Wherein step S3 also includes the following sub-steps: S3-1, use Fourier transform to convert time series data into frequency domain, and extract the periodic characteristics of the data by identifying the frequency, as shown in formula (4): Among them, X(f) is the result in the frequency domain, x(t) is the time series signal, and f is the frequency; S3-2, according to the results of the frequency domain, the data is decomposed into trend component, seasonal component and random component. The low-frequency component in the frequency domain results corresponds to the trend component; the medium and high-frequency components in the frequency domain results correspond to the seasonal component; and the high-frequency components in the frequency domain results correspond to the random component.

5. A photovoltaic power station monitoring system according to claim 1, characterized in that: Wherein step S4 also includes the following sub-steps: S4-1, uses a multi-scale convolutional neural network to capture short-term changes at the hour level, that is, instantaneous changes in device status; S4-2, using multi-scale convolutional neural networks to capture daily periodic fluctuations, that is, the fluctuation patterns of daily power generation and equipment status; S4-3, using multi-scale convolutional neural networks to capture long-term trends at the weekly level, that is, to capture long-term seasonal changes or equipment aging trends; S4-4, uses a multi-scale convolutional neural network to capture the above data changes and extract multi-level features of different time dimensions, as shown in formula (5): in, is the feature after the lth layer of convolution, is the convolution kernel weight of the (l)th layer, X (l-1) is the input feature of the previous layer, f is the activation function, b (l) For bias.

6. A photovoltaic power station monitoring system according to claim 1, characterized in that: Wherein, in step S5, the following sub-steps are also included: S5-1, the processed data is input into the LSTM model for training. After the LSTM model is trained, it will accept new environmental data and predict future power generation changes, as shown in formulas (6)-(10) i t = σ(W i · [h t-1 , x t + b i ) Equation (6) f t = σ(W f · [h t-1 , x t + b f ) Equation (7) the t =σ(W o ·[h t-1 ,x t ]+b o ) expression(8) c t = f t · c t-1 + i t · tanh(W c · [h t-1 , x t + b c ) Equation (9) h t =o t ·tanh(c t ) Formula (10) Among them, i t is the input gate, f t For the forget gate, t is the output gate, c t is the cell state update, h t To update the hidden state; LSTM processes long sequence data through the control mechanism of input gate, forget gate and output gate; S5-2, analyze the health status of the equipment through the BP neural network to diagnose whether the equipment has a risk of failure. The equipment operation status data is input into the BP neural network. The network calculates the activation value of each layer through forward propagation until the health status, activation value and failure probability of the output equipment are output, as shown in formulas (11)-(12): y=σ(W (L) ·h (L-1) +b (L) ) expression(12) Where f(x) is the σ function; y is the output value, i.e., the failure probability of the equipment, W (L) is the weight matrix of the output layer, h (L-1) is the output value of the hidden layer, b (L) is the bias of the output layer, σ is the activation function, i.e., f(x), and the σ function is used to convert the output value into a probability (ranging between 0 and 1). If the failure probability is close to 1, there is a problem with the device; if it is close to 0, the device is in a relatively healthy state. S5-3: The GRU model is used to quickly process short-term fluctuations of the equipment and identify whether the equipment has short-term abnormalities. The GRU model processes short-term state changes of the equipment and can respond to abnormal changes in the equipment state, capture short-term fluctuations and abnormal conditions of the equipment, and perform short-term fault prediction, as shown in formula (13): Among them, (X) is the short-term equipment fluctuation data, and fGRU(X) is the fault prediction result output by the GRU model.

7. A photovoltaic power station monitoring system according to claim 1, characterized in that: Wherein, in step S6, the following sub-steps are also included: S6-1, use the weighted fusion algorithm to integrate the three model data, as shown in formula (14): Among them, w1, w2, and w3 are the weights of LSTM, BP, and GRU models respectively. For their respective prediction results; S6-2, based on the fused prediction results, the system generates operational suggestions and decision support information, including: The forecast results are evaluated, and if anomalies or significant deviations are detected, the system will issue a warning signal; based on historical data and forecast results, decision support reports for different operating scenarios are generated to help users choose the optimal operating strategy.

8. A photovoltaic power station monitoring system according to claim 1, characterized in that: Wherein, in step S7, the following sub-steps are also included: S7-1, data visualization displays key data of the photovoltaic power station in real time through a graphical interface, including environmental data, equipment status, power generation forecast, real-time data and historical data trend charts; S7-2, customized operation report Based on historical data and real-time monitoring information, customized operation reports are automatically generated. The report content can adjust system parameters and set warning thresholds as needed.

Images