New energy equipment fault early warning system for photovoltaic power station

Abstract

The invention belongs to the technical field of photovoltaic equipment fault early warning systems, and discloses a new energy equipment fault early warning system for a photovoltaic power station, which comprises the following steps: firstly, collecting photovoltaic system parameters according to a preset frequency to form a time sequence database; then preprocessing the parameters to generate a physical feature data set and an analysis feature data set; a dynamic Bayesian network model and a machine learning classification model are constructed, fault analysis is carried out based on the physical feature data set and the analysis feature data set, and two fault probability distributions are generated; calculating the consistency index of the two distributions, and adopting a corresponding decision strategy according to the consistency index to generate a comprehensive fault probability; and finally, determining a fault state according to a comparison result of the fault probability value and a preset threshold value, and generating an early warning signal. According to the invention, by fusing the physical model and the data driving method, intelligent fusion of results of the two models is realized, and the accuracy and reliability of fault early warning are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic device fault early warning systems, and more particularly to a new energy equipment fault early warning system for a photovoltaic power station. BACKGROUND

[0002] With the transformation of global energy structure and the rapid development of renewable energy, photovoltaic power generation, as an important part of clean energy, is being widely promoted and applied on a global scale. Photovoltaic power stations have become the main force in the development of new energy due to their advantages of no pollution, no noise, and renewable resources. However, with the rapid growth of photovoltaic installed capacity, the operation and maintenance management of photovoltaic power stations is facing increasing challenges, and fault diagnosis and early warning are key links to ensure the safe and efficient operation of photovoltaic power stations.

[0003] Currently, photovoltaic power station fault diagnosis technology has developed from traditional manual inspection to model-based intelligent diagnosis methods. Existing fault diagnosis systems usually use physical model-based or data-driven methods for fault judgment. However, these systems have a single problem in fault diagnosis models, making it difficult to cope with complex and variable fault scenarios in photovoltaic power stations. Most systems either use physical model-based methods or data-driven methods, lacking an organic integration of the two methods. Although physical model-based methods have good interpretability, they have limited recognition ability for complex nonlinear fault patterns; while data-driven methods can capture complex patterns, but lack physical mechanism support and have insufficient reliability in sparse or abnormal data situations. SUMMARY

[0004] In order to overcome the above problems of the prior art, the present application proposes a new energy equipment fault early warning system for a photovoltaic power station to solve the above problems.

[0005] The present application provides the following technical solutions: A new energy equipment fault early warning system for a photovoltaic power station, comprising: a data acquisition module for acquiring photovoltaic system parameters at a preset frequency to form a time series database; a data preprocessing module for preprocessing photovoltaic system parameters in the time series database to generate a physical feature dataset and an analysis feature dataset; a Bayesian inference module for constructing a dynamic Bayesian network model, performing fault inference based on the physical feature dataset, and generating a Bayesian fault probability distribution; a machine learning module for constructing a machine learning classification model, performing fault classification based on the analysis feature dataset, and generating a learning fault probability distribution; The cooperative decision module is configured to receive the Bayesian fault probability distribution and the learning fault probability distribution, calculate a consistency index of the two distributions, adopt a corresponding decision strategy according to the consistency index, and generate a comprehensive fault probability. The early warning generation module is configured to extract a probability value of each fault type from the comprehensive fault probability distribution, determine a fault state according to a comparison result of the fault probability value and a preset threshold, and generate a corresponding early warning signal.

[0006] Preferably, the photovoltaic system parameters include electrical parameters and environmental parameters; the electrical parameters include voltage, current and power; and the environmental parameters include component temperature, ambient temperature and irradiance.

[0007] Preferably, the physical feature data set includes a component temperature coefficient, a current-voltage curve feature point and a power-voltage curve feature point; and the analysis feature data set includes a time domain feature, a frequency domain feature and a statistical feature.

[0008] Preferably, the preprocessing of the photovoltaic system parameters in the time series database includes: standardizing the photovoltaic system parameters in the time series database, including abnormal value detection and elimination, missing data filling and noise filtering; extracting the physical feature data set according to the photovoltaic system parameters after the standardization, including calculating a temperature difference between the component temperature and the ambient temperature, recording the change of the temperature difference under different irradiance intensities, and obtaining a response coefficient of the component temperature to the irradiance intensity as the component temperature coefficient through linear regression analysis; constructing a current-voltage curve through the collected voltage and current data, extracting a current-voltage curve feature point, the current-voltage curve feature point including a short-circuit current, an open-circuit voltage, a current and a voltage corresponding to a maximum power point, and a slope change point of the current-voltage curve; constructing a power-voltage curve through the collected power and voltage data, extracting a power-voltage curve feature point, the power-voltage curve feature point including the maximum power point and a corresponding voltage, a rising rate and a falling rate of the power curve, and a turning point position of the power curve; extracting the analysis feature data set by analyzing the photovoltaic system parameters after the standardization, including performing sliding window analysis on the time series of the voltage, the current and the power, calculating the maximum value, the minimum value, the mean value, the variance and the trend slope in each window to form the time domain feature; performing fast Fourier transform on the time series of the voltage and the current to extract a power spectral density and a main frequency component to form the frequency domain feature; and calculating a skewness, a kurtosis and a quartile range of the voltage, the current, the power, the component temperature and the irradiance to form the statistical feature.

[0009] Preferably, the constructing of the dynamic Bayesian network model includes: The three-layer structure of the initial Bayesian network is constructed, including a fault node layer for representing possible fault types of the photovoltaic system, a state parameter layer for representing intermediate state variables of the photovoltaic system, and an observation parameter layer for representing parameters in the physical feature data set; A plurality of time slices are created, each slice containing a complete three-layer network structure; conditional probability relationships between nodes within the same time slice are defined; state transition probabilities of the state parameter nodes between adjacent time slices are defined; and observation probabilities between the state parameter nodes and the observation parameter nodes in each time slice are defined; Based on historical fault data and corresponding physical feature data, a dynamic Bayesian network is trained using an expectation maximization algorithm to obtain a trained dynamic Bayesian model; The fault inference based on the physical feature data set to generate the Bayesian fault probability distribution includes: The physical feature data set is input as observation evidence into the trained dynamic Bayesian network, and the posterior probability of each fault node is calculated by a joint tree algorithm to generate the Bayesian fault probability distribution, which is represented as a multi-dimensional vector of each fault type and corresponding probability value.

[0010] Preferably, the construction of the machine learning classification model includes: A multi-layer perception neural network is constructed, including an input layer for receiving the analysis feature data set, a hidden layer for feature conversion and abstract representation, and an output layer for representing the probability distribution of each fault type; a hierarchical feature fusion method is used to weight and combine the time domain features, frequency domain features and statistical features to form a comprehensive feature vector as the input of the neural network; Based on the historical labeled fault data, a neural network is trained using a back propagation algorithm, including initializing network weight parameters; calculating the prediction result of forward propagation; calculating the cross-entropy loss between the prediction result and the true label; updating the network parameters by gradient descent method; repeating the above steps until the loss function converges, to obtain a trained machine learning classification model; The fault classification based on the analysis feature data set to generate the learning fault probability distribution includes: The analysis feature data set is input into the trained machine learning classification model, and the probability values of each fault type are calculated by forward propagation to generate the learning fault probability distribution, which is represented as a multi-dimensional vector of each fault type and corresponding probability value.

[0011] Preferably, the calculation of the consistency index of the two distributions includes: The reciprocal of the bulldozer distance between the Bayesian fault probability distribution and the learning fault probability distribution is calculated as the consistency index of the two distributions; The method of adopting a corresponding decision strategy based on the consistency index to generate a comprehensive failure probability includes: When the consistency index is higher than the preset threshold, the Bayesian fault probability distribution and the learned fault probability distribution are directly averaged to generate a comprehensive fault probability distribution; When the consistency index is lower than the preset threshold, the stability indicators of the two models are quantified, and a weighted average is performed based on the stability indicators to generate a comprehensive failure probability distribution.

[0012] Preferably, the stability index of the two quantified models includes: For the dynamic Bayesian network model: Calculate the standard deviation matrix of the Bayesian fault probability distribution in at least two recent time windows, and take the square root of the sum of the squares of its elements as the volatility indicator ; Calculate the standard deviation of the rate of change of the physical characteristic data set relative to the historical average as an input stability indicator According to the formula , calculate the stability index of the dynamic Bayesian network model , where and They are the weight coefficients of the preset volatility index and input stability index respectively; For machine learning models: Calculate the standard deviation matrix of the learning failure probability distribution in at least two recent time windows, and take the square root of the sum of squares of its elements as the difference index ; Calculate the standard deviation of the rate of change of the feature data set relative to the historical average as an input adaptability indicator According to the formula , calculate the stability index of machine learning models , and are the weight coefficients of the preset difference index and input adaptability index respectively; The weighted average based on the stability index to generate a comprehensive failure probability distribution includes: According to the stability index of the two models and ,use As Bayesian model weights, use As machine learning model weights; Using the obtained weights, the Bayesian fault probability distribution and the learned fault probability distribution are weighted averaged to generate a comprehensive fault probability distribution.

[0013] Preferably, determining the fault state according to the comparison result of the fault probability value with the preset threshold value and generating the corresponding warning signal and diagnosis report includes: Extract the probability value of each fault type from the comprehensive fault probability distribution; compare the probability value of each fault type with the corresponding early warning threshold; for the fault type whose probability value exceeds the early warning threshold, generate the corresponding fault early warning signal.

[0014] The application provides a new energy equipment fault early warning system for a photovoltaic power station. The fault inference based on the physical feature dataset through the Bayesian inference module can fully utilize the physical mechanism and causal relationship of the photovoltaic system, has high interpretability and accuracy for faults with explicit physical manifestations; the fault classification based on the analysis feature dataset through the machine learning module can capture complex nonlinear relationships and hidden fault patterns, and has strong identification ability for data pattern sensitive faults. The complementary advantages of the two models greatly improve the identification ability of the system for various faults, making the fault early warning more comprehensive and reliable. The consistency index of the Bayesian fault probability distribution and the learning fault probability distribution is calculated, and a corresponding decision strategy is adopted based on the index, so that the intelligent fusion of the results of the two models is realized. The adaptive fusion mechanism can dynamically adjust the weights of the two models in the final decision according to the reliability of the models in different scenarios, effectively improving the overall prediction accuracy and robustness of the system. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A module schematic diagram of the new energy equipment fault early warning system for the photovoltaic power station. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0017] Embodiment 1 Please refer to Figure 1 In this embodiment, the new energy equipment fault early warning system for the photovoltaic power station comprises: A data acquisition module is configured to acquire photovoltaic system parameters at a preset frequency to form a time series database. The photovoltaic system parameters include electrical parameters and environmental parameters; the electrical parameters include voltage, current and power; and the environmental parameters include component temperature, environmental temperature and irradiance.

[0018] In this embodiment, multiple sensors can be installed in the photovoltaic power station to collect system parameters. Electrical parameter collection includes installing voltage sensors and current sensors at the photovoltaic strings and inverters. Power data is obtained through real-time calculation of voltage and current.

[0019] Environmental parameter collection includes installing temperature sensors on the back of photovoltaic modules to measure module temperature, installing temperature sensors around the array to measure ambient temperature, and installing irradiance meters on the array plane to measure solar radiation intensity.

[0020] The collection frequency can be set to once per minute, and all data is time-stamped. Data can be organized and stored by device ID, parameter type, and timestamp, forming a complete time series database that provides basic data support for subsequent analysis.

[0021] The data preprocessing module is used to preprocess the photovoltaic system parameters in the time series database to generate physical characteristic data sets and analysis characteristic data sets; The physical feature data set includes component temperature coefficients, current-voltage curve feature points, and power-voltage curve feature points; the analysis feature data set includes time domain features, frequency domain features, and statistical features.

[0022] The preprocessing of the photovoltaic system parameters in the time series database includes: Used to perform standardization processing on photovoltaic system parameters in a time series database, wherein the standardization processing includes outlier detection and elimination, missing data filling and noise filtering; Based on the standardized photovoltaic system parameters, a physical characteristic data set is extracted, including: calculating the temperature difference between the module temperature and the ambient temperature, recording the temperature difference changes under different irradiation intensities, and obtaining the module temperature response coefficient to irradiation intensity through linear regression analysis as the module temperature coefficient; A current-voltage curve is constructed using the collected voltage and current data, and characteristic points of the current-voltage curve are extracted. The characteristic points of the current-voltage curve include the short-circuit current, the open-circuit voltage, the current and voltage corresponding to the maximum power point, and the slope change point of the current-voltage curve. A power-voltage curve is constructed using the collected power and voltage data, and characteristic points of the power-voltage curve are extracted. The characteristic points of the power-voltage curve include the maximum power point and its corresponding voltage, the rate of rise and rate of fall of the power curve, and the position of the inflection point of the power curve. According to the analysis of the standardized photovoltaic system parameters, the analysis feature dataset is extracted, including: sliding window analysis on the time series of voltage, current and power, calculating the maximum, minimum, mean, variance and trend slope in each window to form the time domain features; fast Fourier transform on the time series of voltage and current, extracting the power spectral density and main frequency component to form the frequency domain features; calculating the skewness, kurtosis and interquartile range of voltage, current, power, component temperature and irradiance to form the statistical features.

[0023] In this embodiment, the data preprocessing module first standardizes the photovoltaic system parameters in the time series database. Standardization includes three steps: outlier detection and elimination, missing data filling and noise filtering.

[0024] After standardization, the data preprocessing module further extracts the physical feature dataset. First, calculate the temperature difference between the component temperature and the ambient temperature, and record the temperature difference change under different irradiance. Through linear regression analysis on these data, the response coefficient of component temperature to irradiance, i.e. the component temperature coefficient, is obtained. This coefficient reflects the thermal characteristics of the component and is an important indicator for judging the thermal spot, thermal decay and other faults of the component.

[0025] Next, the data preprocessing module constructs the current-voltage curve by collecting voltage and current data, and extracts its feature points. These feature points include short-circuit current, open-circuit voltage, current and voltage corresponding to the maximum power point, and the slope change point of the current-voltage curve. Short-circuit current reflects the photoelectric conversion capability of the component, and the maximum power point reflects the energy conversion efficiency of the system. The slope change point of the curve corresponds to the internal resistance change of the component, which can indicate potential local faults. Similarly, by constructing the power-voltage curve through power and voltage data, the maximum power point and its corresponding voltage, power curve rising rate and falling rate, and power curve inflection point position are extracted. The changes of these feature points can reflect the performance degradation and abnormal working state of the component or system.

[0026] In the extraction of the analysis feature dataset, the data preprocessing module first performs sliding window analysis on the time series of voltage, current and power, calculates the maximum, minimum, mean, variance and trend slope in each window to form the time domain features. Time domain features can reflect the dynamic change characteristics of system parameters, which is of great significance for detecting intermittent faults and performance degradation.

[0027] Secondly, the fast Fourier transform is performed on the time series of voltage and current, and the power spectral density and main frequency component are extracted to form the frequency domain features. Frequency domain analysis can reveal the periodic changes and hidden patterns in the time series, and is particularly suitable for detecting problems caused by oscillation, harmonic interference and grid interaction.

[0028] Finally, skewness, kurtosis and interquartile range of voltage, current, power, component temperature and irradiance are calculated to form statistical features. Statistical features reflect the shape of data distribution, which is suitable for detecting abnormal changes of system behavior.

[0029] a Bayesian inference module configured to construct a dynamic Bayesian network model, perform fault inference according to the physical feature dataset, and generate a Bayesian fault probability distribution; The constructing the dynamic Bayesian network model comprises: a three-layer structure of an initial Bayesian network is constructed, including a fault node layer for representing possible fault types of the photovoltaic system, a state parameter layer for representing intermediate state variables of the photovoltaic system, and an observation parameter layer for representing parameters in the physical feature dataset; a plurality of time slices are created, each containing a complete three-layer network structure; conditional probability relationships between nodes within the same time slice are defined; state transition probabilities of state parameter nodes between adjacent time slices are defined; and observation probabilities between state parameter nodes and observation parameter nodes in each time slice are defined; based on historical fault data and corresponding physical feature data, a dynamic Bayesian network is trained using an expectation maximization algorithm to obtain a trained dynamic Bayesian model; The performing fault inference according to the physical feature dataset and generating a Bayesian fault probability distribution comprises: the physical feature dataset is input as observation evidence into the trained dynamic Bayesian network, and a joint tree algorithm is used to calculate the posterior probability of each fault node to generate a Bayesian fault probability distribution, which is represented as a multi-dimensional vector of each fault type and corresponding probability value.

[0030] In this embodiment, the Bayesian inference module first constructs a three-layer structure of an initial Bayesian network. The fault node layer is located at the top layer of the network and contains main fault types that may occur in the photovoltaic system, such as component hot spot fault, bypass diode fault, component breakage, component PID effect, component string connection fault, and inverter fault. The state parameter layer is located at the middle layer of the network and represents intermediate state variables of the system, such as component temperature distribution state, component electrical property state, and inverter working state. These state variables cannot be directly observed but can be inferred through observation parameters. The observation parameter layer is located at the bottom layer of the network and corresponds to parameters in the physical feature dataset, including component temperature coefficient, current-voltage curve feature points, and power-voltage curve feature points.

[0031] To capture the dynamic changes of system states over time, the Bayesian inference module creates multiple time slices, each containing a complete three-layer network structure. The interval of time slices is set to 1 hour, enabling the model to capture short-term changes in system states. Within the same time slice, the Bayesian inference module defines the conditional probability relationship between nodes. For example, there is a conditional probability relationship between the component hot spot failure node and the component temperature distribution state node, indicating the probability distribution of the component temperature distribution state given the occurrence of hot spot failure.

[0032] Between adjacent time slices, the Bayesian inference module defines the state transition probability of state parameter nodes. The state transition probability describes the evolution of system states over time, for example, how the current time's component electrical property state affects the next time's component electrical property state. This time-dependent relationship enables the model to consider the development process of system failures and the influence of historical factors. In addition, the Bayesian inference module also defines the observation probability between state parameter nodes and observation parameter nodes in each time slice, describing the probability of observing a specific parameter value given the system state.

[0033] The training of the dynamic Bayesian network is a key step of the Bayesian inference module. Based on historical failure data and corresponding physical feature data, the Bayesian inference module uses the Expectation-Maximization (EM) algorithm to train the dynamic Bayesian network. The training process consists of two alternating steps: the expectation step and the maximization step. In the expectation step, the posterior distribution of hidden variables is estimated according to the current model parameters; in the maximization step, the model parameters are updated according to the estimated hidden variable distribution. Through multiple iterations, the EM algorithm gradually optimizes the model parameters, making the model better fit the historical data.

[0034] After training, the Bayesian inference module uses the trained dynamic Bayesian network for fault inference. When a new set of physical feature data arrives, the Bayesian inference module inputs it as observation evidence into the dynamic Bayesian network. Through the junction tree algorithm for exact inference, the posterior probability of each fault node is calculated. The junction tree algorithm first converts the Bayesian network into an acyclic undirected graph, then constructs a junction tree structure, and finally calculates the marginal posterior probability of the target node through the message passing mechanism on the tree. The result of Bayesian inference is the Bayesian fault probability distribution, represented as a multi-dimensional vector of each fault type and corresponding probability value.

[0035] The machine learning module is configured to construct a machine learning classification model, perform fault classification based on the analysis feature data set, and generate a learning fault probability distribution. The construction of the machine learning classification model includes: The multi-layer perceptron neural network is constructed, including an input layer for receiving the analysis feature dataset, a hidden layer for feature conversion and abstract representation, and an output layer for representing the probability distribution of each type of fault; a hierarchical feature fusion method is used to combine the time domain features, frequency domain features and statistical features in a weighted manner to form a comprehensive feature vector as the input of the neural network; Based on the historical labeled fault data, the neural network is trained using the back propagation algorithm, including initializing the network weight parameters, calculating the prediction results of forward propagation, calculating the cross-entropy loss between the prediction results and the true labels, updating the network parameters by gradient descent method, and repeating the above steps until the loss function converges to obtain the trained machine learning classification model. The fault classification according to the analysis feature dataset includes: The analysis feature dataset is input into the trained machine learning classification model, and the probability values of each type of fault are calculated by forward propagation to generate the learning fault probability distribution, which is represented as a multi-dimensional vector of each fault type and corresponding probability value.

[0036] In this embodiment, the machine learning module first constructs a multi-layer perceptron neural network. The neural network includes three main parts: input layer, hidden layer and output layer. The input layer is used to receive the analysis feature dataset, and the number of nodes is equal to the dimension of the feature vector. The hidden layer uses ReLU activation function, and the number of nodes of the output layer is equal to the number of fault types. The output layer uses Softmax activation function to convert the output of the neural network into probability distribution.

[0037] The neural network training process can be based on historical labeled fault data to train the network using the back propagation algorithm. First, the network weight parameters are initialized, then the training samples are input into the network, and the prediction results, i.e. the probability distribution of each type of fault, are calculated by forward propagation. Then, the cross-entropy loss between the prediction results and the true labels is calculated. After the loss calculation is completed, the machine learning module updates the network parameters by gradient descent method. To prevent overfitting, L2 regularization and Dropout technology can also be applied, with a Dropout rate of 0.3. The training process uses small batch training method, with a batch size of 64, and the above steps are repeated until the loss function converges or the maximum training period (200 periods) is reached.

[0038] After training, when a new analysis feature dataset arrives, the machine learning module inputs it into the trained machine learning classification model. Through forward propagation calculation, the model generates the probability values of each type of fault to form the learning fault probability distribution.

[0039] The collaborative decision-making module is used to receive the Bayesian fault probability distribution and the learned fault probability distribution, calculate the consistency index of the two distributions, adopt the corresponding decision strategy based on the consistency index, and generate a comprehensive fault probability; The calculation of the consistency index of the two distributions includes: Calculate the inverse of the bulldozer distance between the Bayesian fault probability distribution and the learned fault probability distribution as an indicator of the consistency of the two distributions; The method of adopting a corresponding decision strategy based on the consistency index to generate a comprehensive failure probability includes: When the consistency index is higher than the preset threshold, the Bayesian fault probability distribution and the learned fault probability distribution are directly averaged to generate a comprehensive fault probability distribution; When the consistency index is lower than the preset threshold, the stability indicators of the two models are quantified, and a weighted average is performed based on the stability indicators to generate a comprehensive failure probability distribution.

[0040] The stability indicators for quantifying the two models include: For the dynamic Bayesian network model: Calculate the standard deviation matrix of the Bayesian fault probability distribution in at least two recent time windows, and take the square root of the sum of the squares of its elements as the volatility indicator ; Calculate the standard deviation of the rate of change of the physical characteristic data set relative to the historical average as an input stability indicator According to the formula , calculate the stability index of the dynamic Bayesian network model , where and They are the weight coefficients of the preset volatility index and input stability index respectively; For machine learning models: Calculate the standard deviation matrix of the learning failure probability distribution in at least two recent time windows, and take the square root of the sum of squares of its elements as the difference index ; Calculate the standard deviation of the rate of change of the feature data set relative to the historical average as an input adaptability indicator According to the formula , calculate the stability index of machine learning models , and are the weight coefficients of the preset difference index and input adaptability index respectively; The weighted average based on the stability index to generate a comprehensive failure probability distribution includes: According to the stability index of the two models and ,use As Bayesian model weights, use As machine learning model weights; The obtained weights are used to perform a weighted average of the Bayesian failure probability distribution and the learned failure probability distribution to generate a comprehensive failure probability distribution.

[0041] In this embodiment, the collaborative decision module first calculates the reciprocal of the bulldozer distance between the Bayesian failure probability distribution and the learned failure probability distribution as a consistency indicator of the two distributions. The bulldozer distance is an effective measure of the difference between two probability distributions, and taking its reciprocal as a consistency indicator means that the smaller the distance, the higher the consistency. The larger the value of the consistency indicator, the closer the two distributions, i.e., the more consistent the judgments of the two models of different principles on the fault type.

[0042] When the prediction results of the two models are highly consistent, i.e., the consistency indicator is higher than a preset threshold such as 0.75, it indicates that the judgments of the two models on the current fault have high consistency, which usually means that the reliability of the prediction results is high. In this case, the collaborative decision module directly performs an arithmetic average of the Bayesian failure probability distribution and the learned failure probability distribution to generate a comprehensive failure probability distribution. The arithmetic average is a simple and effective method that can integrate the advantages of the two models and improve the stability of the prediction.

[0043] When the prediction results of the two models are low in consistency, i.e., the consistency indicator is lower than the preset threshold, it indicates that there is a large difference between the judgments of the two models on the current fault, which may mean that the prediction accuracy of one of the models is reduced under the current condition. At this time, a simple arithmetic average may introduce errors and reduce the accuracy of the prediction. Therefore, the collaborative decision module needs to further quantify the stability indicators of the two models and perform a weighted average based on the stability indicators to favor the more stable and reliable model.

[0044] For the stability quantification of the dynamic Bayesian network model, the collaborative decision module calculates the standard deviation matrix of the Bayesian failure probability distribution in the last at least two time windows, and takes the square root of the sum of squares of each element as the volatility indicator. The volatility indicator reflects the temporal stability of the prediction results of the Bayesian model, and the smaller the value, the more stable the model output. At the same time, the standard deviation of the change rate of the physical feature data set relative to the historical average value is calculated as the input stability indicator. The input stability indicator reflects the degree of change of the input data, and the smaller the value, the more stable the input data. The larger the stability indicator of the dynamic Bayesian network model calculated according to the formula, the more likely it is that the dynamic Bayesian network model is accurate in prediction; similarly, the larger the stability indicator of the machine learning model calculated according to the formula, the more likely it is that the machine learning model is accurate in prediction.

[0045] Based on the calculated stability indicators of the two models, the collaborative decision module further calculates the normalized weights.

[0046] Finally, the collaborative decision module uses the calculated weights to perform a weighted average of the Bayesian failure probability distribution and the learned failure probability distribution, generating a comprehensive failure probability distribution.

[0047] This stability-index-based weighted average strategy can adaptively adjust the influence of different models in the final decision, taking advantage of the complementary strengths of the Bayesian inference module and the machine learning module. The Bayesian inference module performs reasoning based on physical characteristics and causal relationships, and has high interpretability and accuracy for failures with clear physical mechanisms. The machine learning module performs data-driven classification and can capture complex nonlinear relationships and hidden failure patterns. Through the calculation of the consistency index and the stability index, the collaborative decision module can intelligently judge the reliability of the two models and generate a more accurate comprehensive failure probability distribution, thereby improving the overall prediction accuracy and reliability of the system.

[0048] The warning generation module is configured to extract the probability values of each fault type from the comprehensive failure probability distribution, determine the fault state according to the comparison result of the fault probability value and the preset threshold, and generate the corresponding warning signal.

[0049] The determination of the fault state according to the comparison result of the fault probability value and the preset threshold, and the generation of the corresponding warning signal and the diagnostic report include: Extracting the probability value of each fault type from the comprehensive failure probability distribution; comparing the probability value of each fault type with the corresponding warning threshold; for the fault type whose probability value exceeds the warning threshold, generating the corresponding fault warning signal.

[0050] In this embodiment, the warning generation module extracts the probability values of each fault type from the comprehensive failure probability distribution and compares them with the preset warning thresholds. Different fault types have different warning thresholds, such as 0.6 for component hot spot failure, 0.7 for bypass diode failure, 0.75 for component PID effect, etc.

[0051] When the probability value of a certain fault type exceeds its corresponding warning threshold, the warning generation module generates the corresponding warning signal. The relevant staff are notified through different means such as system prompts, SMS, phone calls, etc.

[0052] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one type. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0053] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

[0054] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A new energy equipment fault warning system for photovoltaic power stations, characterized in that: include: A data acquisition module is used to collect photovoltaic system parameters at a preset frequency to form a time series database; The data preprocessing module is used to preprocess the photovoltaic system parameters in the time series database to generate physical characteristic data sets and analysis characteristic data sets; The Bayesian reasoning module is used to build a dynamic Bayesian network model, perform fault reasoning based on the physical feature data set, and generate a Bayesian fault probability distribution; The machine learning module is used to build a machine learning classification model, classify faults based on the analysis feature data set, and generate a learning fault probability distribution; The collaborative decision-making module is used to receive the Bayesian fault probability distribution and the learned fault probability distribution, calculate the consistency index of the two distributions, adopt the corresponding decision strategy based on the consistency index, and generate a comprehensive fault probability; The early warning generation module is used to extract the probability value of each fault type from the comprehensive fault probability distribution, determine the fault state based on the comparison result of the fault probability value with the preset threshold, and generate the corresponding early warning signal.

2. A new energy equipment fault warning system for a photovoltaic power station according to claim 1, characterized in that: The photovoltaic system parameters include electrical parameters and environmental parameters; the electrical parameters include voltage, current and power; the environmental parameters include component temperature, ambient temperature and irradiance.

3. A new energy equipment fault warning system for a photovoltaic power station according to claim 2, characterized in that: The physical feature data set includes component temperature coefficients, current-voltage curve feature points, and power-voltage curve feature points; the analysis feature data set includes time domain features, frequency domain features, and statistical features.

4. A new energy equipment fault warning system for a photovoltaic power station according to claim 3, characterized in that: The preprocessing of the photovoltaic system parameters in the time series database includes: Used to perform standardization processing on photovoltaic system parameters in a time series database, wherein the standardization processing includes outlier detection and elimination, missing data filling and noise filtering; Based on the standardized photovoltaic system parameters, a physical characteristic data set is extracted, including: calculating the temperature difference between the module temperature and the ambient temperature, recording the temperature difference changes under different irradiation intensities, and obtaining the module temperature response coefficient to irradiation intensity through linear regression analysis as the module temperature coefficient; A current-voltage curve is constructed using the collected voltage and current data, and characteristic points of the current-voltage curve are extracted. The characteristic points of the current-voltage curve include the short-circuit current, the open-circuit voltage, the current and voltage corresponding to the maximum power point, and the slope change point of the current-voltage curve. A power-voltage curve is constructed using the collected power and voltage data, and characteristic points of the power-voltage curve are extracted. The characteristic points of the power-voltage curve include the maximum power point and its corresponding voltage, the rate of rise and rate of fall of the power curve, and the position of the inflection point of the power curve. The analysis is performed based on the standardized photovoltaic system parameters, and the feature data sets are extracted and analyzed. The analysis includes: performing sliding window analysis on the time series of voltage, current, and power, calculating the maximum value, minimum value, mean, variance, and trend slope within each window to form time domain features; performing fast Fourier transform on the time series of voltage and current to extract the power spectrum density and main frequency component to form frequency domain features; and calculating the skewness, kurtosis, and interquartile range of voltage, current, power, component temperature, and irradiance to form statistical features.

5. The new energy equipment failure warning system for photovoltaic power stations according to claim 1, characterized in that: The construction of the dynamic Bayesian network model includes: The three-layer structure of the initial Bayesian network is constructed, including a fault node layer, which is used to represent the possible fault types of the photovoltaic system; a state parameter layer, which is used to represent the intermediate state variables of the photovoltaic system; and an observation parameter layer, which is used to represent the parameters in the physical feature dataset. Create multiple time slices, each containing a complete three-layer network structure; define the conditional probability relationship between nodes within the same time slice; define the state transition probability of state parameter nodes between adjacent time slices; and define the observation probability between state parameter nodes and observation parameter nodes in each time slice; Based on historical fault data and corresponding physical feature data, the expectation maximization algorithm is used to train the dynamic Bayesian network to obtain a trained dynamic Bayesian model; The performing fault reasoning based on the physical feature data set to generate a Bayesian fault probability distribution includes: The physical feature dataset is input as observational evidence into the trained dynamic Bayesian network. The posterior probability of each fault node is calculated using the joint tree algorithm to generate a Bayesian fault probability distribution. The Bayesian fault probability distribution is represented as a multidimensional vector of each fault type and corresponding probability value.

6. A new energy equipment fault warning system for a photovoltaic power station according to claim 5, characterized in that: The construction of the machine learning classification model includes: Construct a multi-layer perceptron neural network, including an input layer for receiving and analyzing feature datasets; a hidden layer for feature conversion and abstract representation; and an output layer for representing the probability distribution of various fault types. A hierarchical feature fusion method is used to perform a weighted combination of time-domain features, frequency-domain features, and statistical features to form a comprehensive feature vector as the input to the neural network. Based on historically labeled fault data, a backpropagation algorithm is used to train a neural network. This includes initializing network weight parameters; calculating the forward propagation prediction results; calculating the cross-entropy loss between the prediction results and the true labels; updating the network parameters using the gradient descent method; and repeating these steps until the loss function converges to obtain a trained machine learning classification model. The fault classification based on the analysis feature data set and the generation of a learning fault probability distribution include: The analysis feature data set is input into the trained machine learning classification model, and the probability values ​​of various types of faults are calculated through forward propagation to generate a learned fault probability distribution. The learned fault probability distribution is represented as a multidimensional vector of each fault type and the corresponding probability value.

7. A new energy equipment fault warning system for a photovoltaic power station according to claim 6, characterized in that: The calculation of the consistency index of the two distributions includes: Calculate the inverse of the bulldozer distance between the Bayesian fault probability distribution and the learned fault probability distribution as an indicator of the consistency of the two distributions; The method of adopting a corresponding decision strategy based on the consistency index to generate a comprehensive failure probability includes: When the consistency index is higher than the preset threshold, the Bayesian fault probability distribution and the learned fault probability distribution are directly averaged to generate a comprehensive fault probability distribution; When the consistency index is lower than the preset threshold, the stability indicators of the two models are quantified, and a weighted average is performed based on the stability indicators to generate a comprehensive failure probability distribution.

8. A new energy equipment fault warning system for a photovoltaic power station according to claim 7, characterized in that: The stability indicators for quantifying the two models include: For the dynamic Bayesian network model: Calculate the standard deviation matrix of the Bayesian fault probability distribution in at least two recent time windows, and take the square root of the sum of the squares of its elements as the volatility indicator ; Calculate the standard deviation of the rate of change of the physical characteristic data set relative to the historical average as an input stability indicator According to the formula , calculate the stability index of the dynamic Bayesian network model , where and They are the weight coefficients of the preset volatility index and input stability index respectively; For machine learning models: Calculate the standard deviation matrix of the learning failure probability distribution in at least two recent time windows, and take the square root of the sum of squares of its elements as the difference index ; Calculate the standard deviation of the rate of change of the feature data set relative to the historical average as an input adaptability indicator According to the formula , calculate the stability index of machine learning models , and are the weight coefficients of the preset difference index and input adaptability index respectively; The weighted average based on the stability index to generate a comprehensive failure probability distribution includes: According to the stability index of the two models and ,use As Bayesian model weights, use As machine learning model weights; Using the obtained weights, the Bayesian fault probability distribution and the learned fault probability distribution are weighted averaged to generate a comprehensive fault probability distribution.

9. A new energy equipment fault warning system for a photovoltaic power station according to claim 8, characterized in that: Determining the fault state based on the comparison result between the fault probability value and the preset threshold value and generating the corresponding warning signal and diagnosis report includes: The probability value of each fault type is extracted from the comprehensive fault probability distribution; the probability value of each fault type is compared with the corresponding warning threshold; for the fault type whose probability value exceeds the warning threshold, the corresponding fault warning signal is generated.

Images