A photovoltaic power station intelligent monitoring and data analysis system and method

Abstract

The application provides a photovoltaic power station intelligent monitoring and data analysis system and method, and belongs to the technical field of intelligent monitoring; the application comprises a sensor network module, a photovoltaic power station data analysis module, a power grid demand prediction module and a power supply adjustment module; the system combines the self data and environmental data of the photovoltaic power station through a data preprocessing unit, a digital twin modeling unit, a multi-modal data fusion unit, an adaptive fault diagnosis unit and a power generation prediction unit of the photovoltaic power station data analysis module to analyze, can accurately obtain the power information, efficiency information and equipment state information of the photovoltaic power station; at the same time, the method for determining the power grid electricity demand by combining the prediction results of the fitting function and the prediction results of the neural network algorithm realizes that the power of each point corresponding to the power curve of the fitting function is used for power supply of the power grid, so that the power supply for the power grid is more real-time, accurate and reliable.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and more specifically, to an intelligent monitoring and data analysis system for photovoltaic power plants. Background Technology

[0002] As the global energy structure transitions towards cleaner and lower-carbon energy, photovoltaic (PV) power generation, as an important form of renewable energy, is experiencing rapid growth in installed capacity. However, PV power plants are generally characterized by widely distributed equipment, complex operating environments, and power generation efficiency affected by multiple factors such as sunlight intensity, temperature, and dust obstruction. Traditional monitoring systems or methods often combine manual inspections with basic data collection, resulting in insufficient real-time monitoring, delayed fault response, and inadequate data value extraction. Furthermore, most existing monitoring solutions rely on single sensor networks or simple threshold alarm mechanisms, making it difficult to achieve refined monitoring at the component level and lacking the ability to conduct in-depth correlation analysis of massive amounts of operational data. This leads to the inability to provide early warnings of potential faults and a lack of data support for system energy efficiency optimization. Meanwhile, the demand for remote centralized management of distributed photovoltaic power plants is increasingly urgent, but traditional systems face technical bottlenecks in areas such as heterogeneous equipment compatibility, data transmission stability, and cloud-based collaborative analysis, limiting the improvement of operation and maintenance efficiency and the maximization of power generation revenue. Furthermore, with the development of artificial intelligence and the Internet of Things, how to deeply integrate machine learning algorithms with the characteristics of photovoltaic systems to build dynamic prediction models and intelligent decision-making systems has become a key technical challenge that the industry urgently needs to overcome. This invention aims to address the above-mentioned technical deficiencies and provide an innovative solution for the efficient operation and maintenance and intelligent management of photovoltaic power plants. Summary of the Invention

[0003] The purpose of this invention is to overcome the above-mentioned problems existing in the prior art and to greatly improve its technical effect on the basis of the original technology; this invention provides a photovoltaic power station intelligent monitoring and data analysis system, the system comprising:

[0004] Sensor network module; photovoltaic power plant data analysis module; power grid demand forecasting module; power supply adjustment module.

[0005] The sensor network module is deployed in each photovoltaic array, inverter, meteorological monitoring equipment, and power grid. Current and voltage sensors, AC voltage sensors, and infrared thermal imaging sensors are deployed at appropriate locations on each photovoltaic module in the photovoltaic array. The current and voltage sensors collect real-time data on the output current, voltage, and power of the photovoltaic modules. The AC voltage sensors detect the amplitude and waveform of the AC voltage generated by the inverter itself in real time, ensuring that the inverter's power devices operate within a safe range and preventing overvoltage breakdown. The infrared thermal imaging sensors periodically scan the surface temperature distribution of the photovoltaic modules and generate thermal patch maps. Dust accumulation detection sensors are deployed at appropriate locations on the edges of the photovoltaic solar panels to quantify the dust coverage rate of the solar panels based on the principle of light attenuation. Meteorological sensors are deployed near the photovoltaic array to simultaneously collect data on light intensity, ambient temperature and humidity, wind speed, and rainfall. The photovoltaic array consists of photovoltaic modules arranged in a specific order, which convert solar energy into direct current (DC), which is then converted into alternating current (AC) by an inverter. A portion of the AC is transmitted to the power grid for distribution to users. A portion of the AC is stored in the photovoltaic array's energy storage system.

[0006] The photovoltaic power station data analysis module is used to process the data collected by the sensor module. The module includes a data preprocessing unit, a digital twin modeling unit, a multimodal data fusion unit, an adaptive fault diagnosis unit, and a power generation prediction unit. The data preprocessing unit preprocesses the data collected by various sensors, including noise reduction, timestamp alignment, and outlier removal. The digital twin modeling unit constructs a three-dimensional virtual model of the power station using the collected data, dynamically mapping the operating status of the physical power station. The multimodal data fusion unit performs spatiotemporal correlation on the collected data and generates multidimensional feature vectors. The adaptive fault diagnosis unit identifies typical faults, including component aging, wiring faults, and inverter efficiency degradation, based on the multidimensional feature vectors and outputs a confidence score. The power generation prediction unit combines weather conditions and equipment health status, using artificial intelligence algorithms to generate a probability distribution curve for power generation over the next 48 hours.

[0007] Furthermore, the data preprocessing unit includes: preprocessing the collected photovoltaic power station data by denoising, timestamp alignment, and outlier removal using time-series data cleaning technology; the digital twin modeling unit includes: establishing a three-dimensional virtual model of the photovoltaic power station using digital twin modeling technology based on the collected data and the actual situation of the photovoltaic power station; simulating the impact of different dust coverage rates on the output characteristics of the components based on the established three-dimensional virtual model and a physical degradation model, and determining the relationship between dust coverage rate and power generation efficiency; the multimodal data fusion unit includes: using a spatiotemporal graph neural network to perform spatiotemporal correlation on various collected data and generate multidimensional feature vectors; the adaptive fault diagnosis unit includes: acquiring publicly available photovoltaic fault data and real-time false alarm information from operation and maintenance personnel, and... A basic model is trained using a publicly available photovoltaic fault dataset through a transfer learning strategy. Subsequently, an online active learning mechanism is used to dynamically adjust the basic model based on false alarm information reported by operation and maintenance personnel. The basic model identifies typical faults, including component aging, wiring faults, and inverter efficiency degradation, through multi-dimensional feature vectors and outputs a confidence score. The power generation prediction unit includes: using convolutional neural networks to extract spatial correlation features from meteorological data, using a time attention mechanism to capture the long-term trend of equipment performance degradation, and using a Bayesian neural network to quantify the uncertainty range of the prediction results. Therefore, artificial intelligence algorithms using convolutional neural networks, time attention mechanisms, and Bayesian neural networks are used to combine weather conditions and equipment health status to generate a probability distribution curve of power generation for the next 48 hours.

[0008] The power grid demand forecasting module is used to forecast the power grid demand connected to photovoltaic power plants. It obtains a fitting function and trains a neural network algorithm using historical power grid data, and combines the prediction results of the fitting function and the neural network algorithm to determine the power grid demand. The process of obtaining the fitting function and training the neural network algorithm using historical power grid data includes: first, obtaining a fitting function for power grid power and time using the least squares method based on historical data; and then training the historical power grid data using an LSTM algorithm to obtain a trained LSTM algorithm model. The training of the historical power grid data using the LSTM algorithm includes: first, extracting the date, week, and time information from the historical power grid data to form a dataset X = {(G...} i H i T i )}, where G i H represents the date information corresponding to the i-th data. i For the weekly information of the i-th data, H i ∈[1,7],H i T is an integer. i Corresponding to the time information for each day, T iThe time interval is 10 minutes; additionally, take T. i－1 To T i Average power P over time i The dataset Y = {P} is composed of i The LSTM algorithm is trained by taking dataset X as input and dataset Y as output, and the trained LSTM algorithm model is obtained.

[0009] Furthermore, the determination of grid electricity demand by combining the prediction results of the fitted function and the prediction results of the neural network algorithm includes: first, predicting the average power P of grid electricity consumption in the next 10 minutes using the trained LSTM algorithm model. L Secondly, the time periods predicted by the trained LSTM algorithm model are mapped to the time information of the fitted function to obtain the corresponding time period information of the fitted function. The power curve information corresponding to the time period information of the fitted function is extracted, and the average power information P within the corresponding time period information of the fitted function is calculated through the power curve information. N Finally, if P L ≤P N Then, for the next 10 minutes, power will be supplied to the grid according to the power curve corresponding to the time period information of the fitted function; if P L >P N Then, for the next 10 minutes, the power value at each point of the power curve corresponding to the time period information of the fitted function is multiplied by... A new power curve is obtained, and power is supplied to the grid for the corresponding time period using the new power curve.

[0010] The power supply adjustment module is used to distribute the power from multiple photovoltaic power plants to the power grid, ensuring sufficient power supply to users. Based on the power curve of the power supply to the grid during a given time period, it regulates the real-time power distribution from multiple photovoltaic power plants to the grid, ensuring the total power supplied by the multiple photovoltaic power plants at the corresponding time point equals the power at the corresponding point on the power curve. It also extracts the power curve for the next 48 hours from the fitted function to obtain the total power consumption for the next 48 hours. Substituting this total power consumption into the probability distribution curve of the power generation for the next 48 hours, it determines the probability of the total power generation in the probability distribution curve for the next 48 hours. It then checks if the probability is less than a probability threshold, which is set at 80%. If it is less than 80%, the module is prepared to utilize the stored power from the photovoltaic power plant's energy storage equipment during the next 48 hours of power supply.

[0011] In addition, this invention also provides a method for intelligent monitoring and data analysis of photovoltaic power plants. This method is based on an intelligent monitoring and data analysis system for photovoltaic power plants. The method includes: first, deploying various sensors in each photovoltaic array, inverter, meteorological monitoring equipment, and power grid; wherein current-voltage sensors, AC voltage sensors, and infrared thermal imaging sensors are deployed at appropriate locations in each photovoltaic module of the photovoltaic array; the current-voltage sensors collect real-time data on the output current, voltage, and power of the photovoltaic modules; the AC voltage sensors detect the amplitude and waveform of the AC voltage generated by the inverter itself in real-time to ensure that the power devices of the inverter operate within a safe range and prevent overvoltage breakdown; and the infrared thermal imaging sensors... Periodically scan the surface temperature distribution of photovoltaic modules and generate thermal patch maps; deploy dust accumulation detection sensors at appropriate locations on the edges of photovoltaic solar panels to quantify the dust coverage rate of solar panels based on the principle of light attenuation; deploy meteorological sensors near the photovoltaic array to simultaneously collect data on light intensity, ambient temperature and humidity, wind speed, and rainfall; process the data collected by various sensors; firstly, use time-series data cleaning technology to preprocess the collected photovoltaic power station data by denoising, timestamp alignment, and outlier removal; then, use digital twin modeling technology to establish a three-dimensional virtual model of the photovoltaic power station based on the collected photovoltaic data and the actual situation of the photovoltaic power station; based on the established three-dimensional virtual model and physical... The degradation model simulates the impact of different dust coverage rates on the output characteristics of the modules, and determines the relationship between dust coverage rate and power generation efficiency. A spatiotemporal graph neural network is used to perform spatiotemporal correlation on various collected data and generate multidimensional feature vectors. These data include photovoltaic data and meteorological data. Publicly available photovoltaic fault data and false alarm information from maintenance personnel are obtained. A basic model is trained using the publicly available photovoltaic fault dataset through a transfer learning strategy. Subsequently, an online active learning mechanism dynamically adjusts the basic model based on false alarm information from maintenance personnel. The basic model identifies typical faults, including module aging, wiring faults, and inverter efficiency degradation, through multidimensional feature vectors and outputs a confidence score. Finally, a... Convolutional neural networks extract spatial correlation features from meteorological data, a time attention mechanism captures the long-term trend of equipment performance degradation, and a Bayesian neural network quantifies the uncertainty range of prediction results. Artificial intelligence algorithms using convolutional neural networks, time attention mechanisms, and Bayesian neural networks combine weather conditions and equipment health status to generate a probability distribution curve for power generation in the next 48 hours. Historical grid electricity consumption data is used to obtain a fitting function and train the neural network algorithm. The prediction results of the fitting function and the neural network algorithm are combined to determine the grid's electricity demand. Based on the grid's electricity demand, the power from multiple photovoltaic power plants is allocated to the grid, ensuring sufficient power supply to users.

[0012] Furthermore, the method of obtaining the fitting function and training the neural network algorithm using historical power grid consumption data is as follows: First, based on the historical data, the least squares method is used to obtain the fitting function of power grid consumption and time; then, the LSTM algorithm is used to train the historical power grid consumption data to obtain a trained LSTM algorithm model; the training of the historical power grid consumption data using the LSTM algorithm includes: first, extracting the date information, week information, and time information of the historical power grid consumption data to form a dataset X = {(G i H i T i )}, where G i H represents the date information corresponding to the i-th data. i For the weekly information of the i-th data, H i ∈[1,7],H i T is an integer. i Corresponding to the time information for each day, T i The time interval is 10 minutes; additionally, take T. i－1 To T i Average power P over time i The dataset Y = {P} is composed of i The LSTM algorithm is trained using dataset X as input and dataset Y as output to obtain a trained LSTM algorithm model. The method for determining grid electricity demand by combining the prediction results of the fitted function and the neural network algorithm is as follows: First, the average power P of the grid electricity consumption within the next 10 minutes is predicted using the trained LSTM algorithm model. L Secondly, the time periods predicted by the trained LSTM algorithm model are mapped to the time information of the fitted function to obtain the corresponding time period information of the fitted function. The power curve information corresponding to the time period information of the fitted function is extracted, and the average power information P within the corresponding time period information of the fitted function is calculated through the power curve information. N Finally, if P L ≤P N Then, for the next 10 minutes, power will be supplied to the grid according to the power curve corresponding to the time period information of the fitted function; if P L >P N Then, for the next 10 minutes, the power value at each point of the power curve corresponding to the time period information of the fitted function is multiplied by... A new power curve is obtained, and power is supplied to the grid for the corresponding time period using the new power curve.

[0013] Furthermore, allocating the electricity from multiple photovoltaic power plants to the grid based on the grid's electricity demand includes: adjusting the power supply curve of the power supply to the grid from multiple photovoltaic power plants in real time to ensure sufficient power supply; ensuring that the total power provided by multiple photovoltaic power plants at the corresponding time within that time period equals the power at the corresponding point on the power curve; simultaneously, extracting the power curve of the fitted function for the next 48 hours to obtain the total electricity consumption for the next 48 hours, substituting the total electricity consumption into the probability distribution curve of the power generation for the next 48 hours, determining the probability of the total power generation in the probability distribution curve of the power generation for the next 48 hours, and judging whether the probability is less than a probability threshold, which is set to 80%; if it is less than 80%, then the stored electricity of the photovoltaic power plant's energy storage device must be ready to be called upon at any time during the power supply process for the next 48 hours.

[0014] The beneficial effects of this invention are:

[0015] This invention provides an intelligent monitoring and data analysis system and method for photovoltaic power plants; this invention has the following advantages:

[0016] 1. By combining the photovoltaic power station's own data and environmental data through the data preprocessing unit, digital twin modeling unit, multimodal data fusion unit, adaptive fault diagnosis unit, and power generation prediction unit, the power, efficiency, and equipment status information of the photovoltaic power station can be accurately obtained.

[0017] 2. A method is presented to determine the power demand of the power grid by combining the prediction results of the fitting function and the prediction results of the neural network algorithm, so that the predicted power demand of the power grid is more accurate and reliable. The power supply to the power grid is based on the power of each point corresponding to the curve of the fitting function, making the power supply to the power grid more real-time, accurate and reliable. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an intelligent monitoring and data analysis system for photovoltaic power plants according to the present invention.

[0019] Figure 2 This is a flowchart of a method for intelligent monitoring and data analysis of a photovoltaic power station according to the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings; it should be understood that the specific embodiments given herein are only for illustration and explanation of the present invention and cannot be used to limit the present invention.

[0021] It should be noted that many specific details are set forth in the following description in order to provide a full understanding of the present invention. However, the present invention may have other embodiments and variations thereof. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0022] like Figure 1 The diagram shown is a schematic of an intelligent monitoring and data analysis system for a photovoltaic power plant according to an embodiment of the present invention. The schematic diagram includes: a sensor network module S10; a photovoltaic power plant data analysis module S20; a power grid demand forecasting module S30; and a power supply adjustment module S40.

[0023] The sensor network module S10 is deployed in each photovoltaic array, inverter, meteorological monitoring equipment, and power grid. Specifically, current and voltage sensors, AC voltage sensors, and infrared thermal imaging sensors are deployed at appropriate locations on each photovoltaic module in the photovoltaic array. The current and voltage sensors collect real-time data on the output current, voltage, and power of the photovoltaic modules. The AC voltage sensors detect the amplitude and waveform of the AC voltage generated by the inverter itself in real time, ensuring that the inverter's power devices operate within a safe range and preventing overvoltage breakdown. The infrared thermal imaging sensors periodically scan the surface temperature distribution of the photovoltaic modules and generate thermal patch maps. Dust accumulation detection sensors are deployed at appropriate locations on the edges of the photovoltaic solar panels to quantify the dust coverage rate of the solar panels based on the principle of light attenuation. Meteorological sensors are deployed near the photovoltaic array to simultaneously collect data on light intensity, ambient temperature and humidity, wind speed, and rainfall.

[0024] In the above embodiments, specifically, the photovoltaic array is composed of photovoltaic modules arranged in a certain order, and the photovoltaic modules convert solar energy into direct current (DC), which is then converted into alternating current (AC) by an inverter. A portion of the AC is transmitted to the power grid and distributed to users; a portion of the AC is stored in the energy storage system of the photovoltaic array.

[0025] The photovoltaic power plant data analysis module S20 is used to process the data collected by the sensor network module S10. The photovoltaic power plant data analysis module S20 includes: a data preprocessing unit, a digital twin modeling unit, a multimodal data fusion unit, an adaptive fault diagnosis unit, and a power generation prediction unit. The data preprocessing unit preprocesses the data collected by various sensors, including noise reduction, timestamp alignment, and outlier removal. The digital twin modeling unit constructs a three-dimensional virtual model of the power plant using the collected data, dynamically mapping the operating status of the physical power plant. The multimodal data fusion unit performs spatiotemporal correlation on the collected data and generates multidimensional feature vectors. The adaptive fault diagnosis unit identifies typical faults, including component aging, wiring faults, and inverter efficiency degradation, based on multidimensional feature vectors and outputs confidence scores. The power generation prediction unit combines weather conditions and equipment health status, using artificial intelligence algorithms to generate a probability distribution curve of power generation for the next 48 hours.

[0026] In the above embodiments, specifically, the data preprocessing unit uses time-series data cleaning technology to preprocess the collected photovoltaic power station data by denoising, timestamp alignment, and outlier removal; the digital twin modeling unit uses digital twin modeling technology to establish a three-dimensional virtual model of the photovoltaic power station based on the collected data and the actual situation of the photovoltaic power station; based on the established three-dimensional virtual model, and using a physical degradation model, it simulates the impact of different dust coverage rates on the output characteristics of the components, and determines the relationship between dust coverage rate and power generation efficiency; the multimodal data fusion unit uses a spatiotemporal graph neural network to perform spatiotemporal correlation on the collected multiple data and generate multidimensional feature vectors; the adaptive fault diagnosis unit obtains publicly available photovoltaic fault data and false alarm information fed back in real time by operation and maintenance personnel, and uses transfer learning... The strategy utilizes publicly available photovoltaic fault datasets to train a basic model. Subsequently, an online active learning mechanism dynamically adjusts the basic model based on false alarm information reported by maintenance personnel. The basic model identifies typical faults, including component aging, wiring faults, and inverter efficiency degradation, through multi-dimensional feature vectors and outputs a confidence score. The power generation prediction unit includes: using convolutional neural networks to extract spatial correlation features from meteorological data, using a time attention mechanism to capture the long-term trend of equipment performance degradation, and using a Bayesian neural network to quantify the uncertainty range of the prediction results. Therefore, artificial intelligence algorithms using convolutional neural networks, time attention mechanisms, and Bayesian neural networks combine weather conditions and equipment health status to generate a probability distribution curve of power generation for the next 48 hours.

[0027] Among them, the power grid demand forecasting module S30 is used to forecast the power grid demand connected to the photovoltaic power station. It uses historical power grid electricity data to obtain a fitting function and train a neural network algorithm, and combines the prediction results of the fitting function and the prediction results of the neural network algorithm to determine the power grid demand.

[0028] Specifically, firstly, based on historical data, the least squares method is used to obtain the fitting function of power consumption and time in the power grid; then, the LSTM algorithm is used to train the historical power grid consumption data to obtain a trained LSTM algorithm model; the training of the historical power grid consumption data using the LSTM algorithm includes: firstly, extracting the date, week, and time information from the historical power grid consumption data to form a dataset X = {(G i H i T i )}, where G i H represents the date information corresponding to the i-th data. i For the weekly information of the i-th data, H i ∈[1,7],H i T is an integer. i Corresponding to the time information for each day, T i The time interval is 10 minutes; additionally, take T.i－1 To T i Average power P over time i The dataset Y = {P} is composed of i The LSTM algorithm is trained by taking dataset X as input and dataset Y as output, and the trained LSTM algorithm model is obtained.

[0029] In the above embodiment, specifically, the average power P of the power grid consumption within the next 10 minutes is predicted using the trained LSTM algorithm model. L Secondly, the time periods predicted by the trained LSTM algorithm model are mapped to the time information of the fitted function to obtain the corresponding time period information of the fitted function. The power curve information corresponding to the time period information of the fitted function is extracted, and the average power information P within the corresponding time period information of the fitted function is calculated through the power curve information. N Finally, if P L ≤P N Then, for the next 10 minutes, power will be supplied to the grid according to the power curve corresponding to the time period information of the fitted function; if P L >P N Then, for the next 10 minutes, the power value at each point of the power curve corresponding to the time period information of the fitted function is multiplied by... A new power curve is obtained, and power is supplied to the grid for the corresponding time period using the new power curve.

[0030] Among them, the power supply adjustment module S40 is used to distribute the power of multiple photovoltaic power plants to the power grid, so that the power grid has sufficient power to provide to users.

[0031] Specifically, based on the power curve of the power supply to the grid during the corresponding time period, multiple photovoltaic power stations are adjusted to allocate sufficient power to the grid in real time, so that the total power provided by the multiple photovoltaic power stations during the corresponding time period is equal to the power at the corresponding point on the power curve. The power curve of the fitting function for the next 48 hours is extracted to obtain the total power consumption for the next 48 hours. The total power consumption is substituted into the probability distribution curve of the power generation for the next 48 hours to determine the probability of the total power generation in the probability distribution curve of the power generation for the next 48 hours. It is then determined whether the probability is less than the probability threshold, which is set at 80%. If it is less than 80%, the stored power of the photovoltaic power station's energy storage device must be ready to be called at any time during the power supply process for the next 48 hours.

[0032] like Figure 2As shown, a flowchart of a method for intelligent monitoring and data analysis of a photovoltaic power station according to the present invention is presented. The flowchart includes: Step S100, firstly, various sensors are deployed in each photovoltaic array, inverter, meteorological monitoring equipment, and power grid; wherein, current and voltage sensors, AC voltage sensors, and infrared thermal imaging sensors are deployed at appropriate locations in each photovoltaic module of the photovoltaic array; the current and voltage sensors collect the output current, voltage, and power data of the photovoltaic modules in real time; the AC voltage sensors detect the amplitude and waveform of the AC voltage generated by the inverter itself in real time to ensure that the power devices of the inverter operate within a safe range and prevent overvoltage breakdown; the infrared thermal imaging sensors periodically scan the surface temperature distribution of the photovoltaic modules and generate... Thermal patch mapping; Dust accumulation detection sensors are deployed at appropriate locations on the edges of photovoltaic solar panels to quantify the dust coverage rate of the solar panels based on the principle of light attenuation; meteorological sensors are deployed near the photovoltaic array to simultaneously collect data on light intensity, ambient temperature and humidity, wind speed, and rainfall; Step S200: Processing the data collected by various sensors; First, time-series data cleaning technology is used to preprocess the collected photovoltaic power station data by denoising, timestamp alignment, and outlier removal; Subsequently, digital twin modeling technology is used to establish a three-dimensional virtual model of the photovoltaic power station based on the collected photovoltaic data and the actual situation of the photovoltaic power station; Based on the established three-dimensional virtual model, and using a physical degradation model, different dust types are simulated. The impact of dust coverage on component output characteristics was investigated, and the relationship between dust coverage and inverter efficiency was determined. A spatiotemporal graph neural network was used to perform spatiotemporal correlation on various collected data and generate multidimensional feature vectors. These data included photovoltaic (PV) data and meteorological data. Publicly available PV fault data and false alarm information from maintenance personnel were acquired. A basic model was trained using a transfer learning strategy on the publicly available PV fault dataset. Subsequently, an online active learning mechanism was used to dynamically adjust the basic model based on false alarm information from maintenance personnel. The basic model identified typical faults, including component aging, wiring faults, and inverter efficiency degradation, through multidimensional feature vectors and output confidence scores. Finally, a convolutional neural network was used to extract meteorological data. The system utilizes spatial correlation features to capture the long-term trend of equipment performance degradation using a time attention mechanism, and employs a Bayesian neural network to quantify the uncertainty range of the prediction results. It combines weather conditions and equipment health status using artificial intelligence algorithms based on convolutional neural networks, time attention mechanisms, and Bayesian neural networks to generate a probability distribution curve for power generation in the next 48 hours. In step S300, a fitting function and a trained neural network algorithm are obtained using historical grid electricity consumption data, and the grid electricity demand is determined by combining the prediction results of the fitting function and the neural network algorithm. In step S400, the electricity from multiple photovoltaic power plants is allocated to the grid based on the grid electricity demand, ensuring sufficient power supply to users.

[0033] In step S300, based on historical data, the least squares method is used to obtain the fitting function of power consumption and time in the power grid; the LSTM algorithm is used to train the historical power consumption data of the power grid to obtain a trained LSTM algorithm model; the training of the historical power consumption data of the power grid using the LSTM algorithm includes: firstly, extracting the date information, week information and time information of the historical power consumption data of the power grid to form a dataset X = {(G i H i T i )}, where G i H represents the date information corresponding to the i-th data. i For the weekly information of the i-th data, H i ∈[1,7],H i T is an integer. i Corresponding to the time information for each day, T i The time interval is 10 minutes; additionally, take T. i－1 To T i Average power P over time i The dataset Y = {P} is composed of i The LSTM algorithm is trained using dataset X as input and dataset Y as output to obtain a trained LSTM algorithm model. The determination of power grid demand, combining the prediction results of the fitting function and the neural network algorithm, includes: firstly, predicting the average power P of the power grid within the next 10 minutes using the trained LSTM algorithm model. L Secondly, the time periods predicted by the trained LSTM algorithm model are mapped to the time information of the fitted function to obtain the corresponding time period information of the fitted function. The power curve information corresponding to the time period information of the fitted function is extracted, and the average power information P within the corresponding time period information of the fitted function is calculated through the power curve information. N Finally, if P L ≤P N Then, for the next 10 minutes, power will be supplied to the grid according to the power curve corresponding to the time period information of the fitted function; if P L >P N Then, for the next 10 minutes, the power value at each point of the power curve corresponding to the time period information of the fitted function is multiplied by... A new power curve is obtained, and power is supplied to the grid for the corresponding time period through the new power curve. Power supply to the grid is controlled by controlling the power supply to prevent power waste.

[0034] In step S400, based on the power curve of the power supply to the grid during the corresponding time period, multiple photovoltaic power stations are adjusted to allocate sufficient power to the grid in real time, so that the total power provided by the multiple photovoltaic power stations during the corresponding time period is equal to the power at the corresponding point of the power curve. At the same time, the power curve of the fitting function for the next 48 hours is extracted to obtain the total power consumption for the next 48 hours. The total power consumption is substituted into the probability distribution curve of the power generation for the next 48 hours to determine the probability of the total power generation in the probability distribution curve of the power generation for the next 48 hours. It is then determined whether the probability is less than the probability threshold, which is set to 80%. If it is less than 80%, then the stored power of the photovoltaic power station's energy storage device must be ready to be called at any time during the power supply process for the next 48 hours.

Claims

1. A smart monitoring and data analysis system for photovoltaic power plants, characterized in that, The system includes: Sensor network module; photovoltaic power plant data analysis module; power grid demand forecasting module; power supply adjustment module; The sensor network module is deployed in various photovoltaic arrays, inverters, meteorological monitoring equipment, and power grids; the sensors include: current and voltage sensors, AC voltage sensors, infrared thermal imaging sensors, dust accumulation detection sensors, and meteorological sensors. The photovoltaic power station data analysis module is used to process the data collected by the sensor module. The module includes: a data preprocessing unit, a digital twin modeling unit, a multimodal data fusion unit, an adaptive fault diagnosis unit, and a power generation prediction unit. The data preprocessing unit preprocesses the data collected by various sensors, including noise reduction, timestamp alignment, and outlier removal. The digital twin modeling unit constructs a three-dimensional virtual model of the power station using the collected data, dynamically mapping the operating status of the physical power station. The multimodal data fusion unit performs spatiotemporal correlation on the collected data and generates multidimensional feature vectors. The adaptive fault diagnosis unit identifies typical faults, including component aging, wiring faults, and inverter efficiency degradation, based on the multidimensional feature vectors and outputs a confidence score. The power generation prediction unit combines weather conditions and equipment health status, using artificial intelligence algorithms to generate a probability distribution curve for power generation over the next 48 hours. The power grid demand forecasting module is used to predict the power grid demand connected to the photovoltaic power station. It obtains a fitting function and trains a neural network algorithm using historical power grid data, and combines the prediction results of the fitting function and the neural network algorithm to determine the power grid demand. First, the trained LSTM algorithm model predicts the average power P of the power grid within the next 10 minutes. L Secondly, the time periods predicted by the trained LSTM algorithm model are mapped to the time information of the fitted function to obtain the corresponding time period information of the fitted function. The power curve information corresponding to the time period information of the fitted function is extracted, and the average power information P within the corresponding time period information of the fitted function is calculated through the power curve information. N Finally, if P L ≤P N Then, for the next 10 minutes, power will be supplied to the grid according to the power curve corresponding to the time period information of the fitted function; if P L >P N Then, for the next 10 minutes, the power value at each point of the power curve corresponding to the time period information of the fitted function is multiplied by... A new power curve is obtained, and power is supplied to the grid for the corresponding time period using the new power curve; The power supply adjustment module is used to distribute the power from multiple photovoltaic power plants to the power grid, so that the power grid has sufficient power to provide to users.

2. The intelligent monitoring and data analysis system for photovoltaic power plants according to claim 1, characterized in that, The photovoltaic array comprises: photovoltaic modules arranged in a certain order, which convert solar energy into direct current (DC) through the photovoltaic modules, and then convert the DC into alternating current (AC) through an inverter. A portion of the AC is transmitted to the power grid and distributed to users by the power grid; a portion of the AC is stored in the energy storage system of the photovoltaic array.

3. The intelligent monitoring and data analysis system for photovoltaic power plants according to claim 1, characterized in that, The sensor network module includes: deploying current and voltage sensors, AC voltage sensors, and infrared thermal imaging sensors at appropriate locations on each photovoltaic module of the photovoltaic array; using current and voltage sensors to collect real-time output current, voltage, and power data of the photovoltaic modules; using AC voltage sensors to detect the amplitude and waveform of the AC voltage generated by the inverter itself in real time to ensure that the power devices of the inverter operate within a safe range and prevent overvoltage breakdown; using infrared thermal imaging sensors to periodically scan the surface temperature distribution of the photovoltaic modules and generate thermal patch maps; deploying dust accumulation detection sensors at appropriate locations on the edges of the photovoltaic solar panels to quantify the dust coverage rate of the solar panels based on the principle of light attenuation; and deploying meteorological sensors near the photovoltaic array to simultaneously collect data on light intensity, ambient temperature and humidity, wind speed, and rainfall.

4. The intelligent monitoring and data analysis system for photovoltaic power plants according to claim 1, characterized in that, The data preprocessing unit includes: preprocessing the collected photovoltaic power station data by denoising, timestamp alignment, and outlier removal using time-series data cleaning technology; the digital twin modeling unit includes: establishing a three-dimensional virtual model of the photovoltaic power station using digital twin modeling technology based on the collected data and the actual situation of the photovoltaic power station; simulating the impact of different dust coverage rates on the component output characteristics based on the established three-dimensional virtual model and a physical degradation model, and determining the relationship between dust coverage rate and power generation efficiency; the multimodal data fusion unit includes: performing spatiotemporal correlation of various collected data using a spatiotemporal graph neural network and generating multidimensional feature vectors; the adaptive fault diagnosis unit includes: acquiring publicly available photovoltaic fault data and real-time false alarm information from operation and maintenance personnel, and then... The mobile learning strategy uses publicly available photovoltaic fault datasets to train a basic model. Then, an online active learning mechanism dynamically adjusts the basic model based on false alarm information from maintenance personnel. The basic model identifies typical faults, including component aging, wiring faults, and inverter efficiency degradation, using multi-dimensional feature vectors and outputs a confidence score. The power generation prediction unit includes: using convolutional neural networks to extract spatial correlation features from meteorological data, using a time attention mechanism to capture the long-term trend of equipment performance degradation, and using a Bayesian neural network to quantify the uncertainty range of the prediction results. Therefore, artificial intelligence algorithms using convolutional neural networks, time attention mechanisms, and Bayesian neural networks combine weather conditions and equipment health status to generate a probability distribution curve for power generation in the next 48 hours.

5. The intelligent monitoring and data analysis system for photovoltaic power plants according to claim 1, characterized in that, The steps of obtaining the fitting function and training the neural network algorithm using historical power grid consumption data include: First, based on the historical data, using the least squares method to obtain the fitting function of power grid consumption and time; then, using the LSTM algorithm to train the historical power grid consumption data to obtain a trained LSTM algorithm model; the step of using the LSTM algorithm to train the historical power grid consumption data includes: First, extracting the date, week, and time information from the historical power grid consumption data to form a dataset X = {(G...} i H i T i )}, where G i H represents the date information corresponding to the i-th data. i For the weekly information of the i-th data, H i ∈[1,7],H i T is an integer. i Corresponding to the time information for each day, T i The time interval is 10 minutes; additionally, take T. i－1 To T i Average power P over time i The dataset Y = {P} is composed of i The LSTM algorithm is trained by taking dataset X as input and dataset Y as output, and the trained LSTM algorithm model is obtained.

6. The intelligent monitoring and data analysis system for photovoltaic power plants according to claim 1, characterized in that, The power supply adjustment module is used to distribute the power of multiple photovoltaic power plants to the power grid, including: adjusting the power supply curve of the power grid to the grid in a corresponding time period, adjusting the multiple photovoltaic power plants to distribute sufficient power to the grid in real time; so that the total power provided by the multiple photovoltaic power plants at the corresponding time in the time period is equal to the power at the corresponding point of the power curve; and extracting the power curve of the fitting function for the next 48 hours to obtain the total power consumption for the next 48 hours, substituting the total power consumption into the probability distribution curve of the power generation for the next 48 hours, determining the probability of the total power consumption in the probability distribution curve of the power generation for the next 48 hours, and judging whether the probability is less than the probability threshold, which is set to 80%; if it is less than 80%, then the power stored in the energy storage device of the photovoltaic power plant should be ready to be called at any time during the power supply process for the next 48 hours.

7. A method for intelligent monitoring and data analysis of photovoltaic power plants, characterized in that, The method is based on the intelligent monitoring and data analysis system for photovoltaic power plants described in claim 1. The method includes: first, deploying various sensors in each photovoltaic array, inverter, meteorological monitoring equipment, and power grid; wherein current and voltage sensors, AC voltage sensors, and infrared thermal imaging sensors are deployed at appropriate locations on each photovoltaic module of the photovoltaic array; the current and voltage sensors collect real-time data on the output current, voltage, and power of the photovoltaic modules; the AC voltage sensors detect the amplitude and waveform of the AC voltage generated by the inverter itself in real-time to ensure that the power devices of the inverter operate within a safe range and prevent overvoltage breakdown; the infrared thermal imaging sensors periodically scan the surface temperature distribution of the photovoltaic modules and generate a thermal patch map; a dust accumulation detection sensor is deployed at an appropriate location on the edge of the photovoltaic solar panel to quantify the dust coverage rate of the solar panel based on the principle of light attenuation; and a meteorological sensor is deployed near the photovoltaic array to simultaneously collect data on light intensity, ambient temperature and humidity, wind speed, and rainfall. Data collected from various sensors is processed. First, time-series data cleaning techniques are used to preprocess the collected photovoltaic power station data by denoising, timestamp alignment, and outlier removal. Then, digital twin modeling technology is used to establish a three-dimensional virtual model of the photovoltaic power station based on the collected photovoltaic data and the actual conditions of the photovoltaic power station. Based on the established three-dimensional virtual model, and using a physical degradation model, the impact of different dust coverage rates on the component output characteristics is simulated to determine the relationship between dust coverage rate and power generation efficiency. A spatiotemporal graph neural network is used to perform spatiotemporal correlation on various collected data and generate multi-dimensional feature vectors. These various data include photovoltaic data and meteorological data. Publicly available photovoltaic fault data and real-time false alarm information from operation and maintenance personnel are obtained. The transfer learning strategy trains a base model using publicly available photovoltaic fault datasets. Then, an online active learning mechanism dynamically adjusts the base model based on false alarms reported by maintenance personnel. The base model identifies typical faults, including component aging, wiring faults, and inverter efficiency degradation, using multi-dimensional feature vectors and outputs confidence scores. Finally, a convolutional neural network is used to extract spatial correlation features from meteorological data, a time attention mechanism is employed to capture the long-term trend of equipment performance degradation, and a Bayesian neural network is used to quantify the uncertainty range of the prediction results. Finally, artificial intelligence algorithms using convolutional neural networks, time attention mechanisms, and Bayesian neural networks are combined to integrate weather conditions and equipment health status, generating a probability distribution curve for power generation over the next 48 hours. The fitting function and the neural network algorithm were obtained by using historical data of power grid consumption, and the power grid demand was determined by combining the prediction results of the fitting function and the prediction results of the neural network algorithm. The electricity generated by multiple photovoltaic power plants is distributed to the power grid according to the grid's electricity demand, so that the grid has sufficient electricity to provide to users.

8. The intelligent monitoring and data analysis method for photovoltaic power plants according to claim 7, characterized in that, The process of obtaining the fitting function and training the neural network algorithm using historical power grid consumption data includes: obtaining the fitting function of power grid consumption and time using the least squares method based on historical data; training the historical power grid consumption data using the LSTM algorithm to obtain a trained LSTM algorithm model; the training of the historical power grid consumption data using the LSTM algorithm includes: firstly, extracting the date, week, and time information from the historical power grid consumption data to form a dataset X = {(G...} i H i T i )}, where G i H represents the date information corresponding to the i-th data. i For the weekly information of the i-th data, H i ∈[1,7],H i T is an integer. i Corresponding to the time information for each day, T i The time interval is 10 minutes; additionally, take T. i－1 To T i Average power P over time i The dataset Y = {P} is composed of i The LSTM algorithm is trained using dataset X as input and dataset Y as output to obtain a trained LSTM algorithm model. The determination of power grid demand, combining the prediction results of the fitting function and the neural network algorithm, includes: firstly, predicting the average power P of the power grid within the next 10 minutes using the trained LSTM algorithm model. L Secondly, the time periods predicted by the trained LSTM algorithm model are mapped to the time information of the fitted function to obtain the corresponding time period information of the fitted function. The power curve information corresponding to the time period information of the fitted function is extracted, and the average power information P within the corresponding time period information of the fitted function is calculated through the power curve information. N Finally, if P L ≤P N Then, for the next 10 minutes, power will be supplied to the grid according to the power curve corresponding to the time period information of the fitted function; if P L >P N Then, for the next 10 minutes, the power value at each point of the power curve corresponding to the time period information of the fitted function is multiplied by... A new power curve is obtained, and power is supplied to the grid for the corresponding time period using the new power curve.

9. The intelligent monitoring and data analysis method for a photovoltaic power station according to claim 7, characterized in that, Distributing electricity from multiple photovoltaic power plants to the grid based on grid demand includes: adjusting the power supply curves of multiple photovoltaic power plants to supply sufficient electricity to the grid in real time according to the determined power supply curves for the corresponding time periods; ensuring that the total power provided by multiple photovoltaic power plants at the corresponding time within the time period equals the power at the corresponding point on the power curve; simultaneously, extracting the power curves of the fitted function for the next 48 hours to obtain the total electricity consumption for the next 48 hours, substituting the total electricity consumption into the probability distribution curve of the electricity generation for the next 48 hours, determining the probability of the total electricity consumption in the probability distribution curve of the electricity generation for the next 48 hours, and judging whether the probability is less than a probability threshold, which is set at 80%; if it is less than 80%, then the electricity stored in the energy storage devices of the photovoltaic power plants must be ready to be used at any time during the power supply process for the next 48 hours.

Images