A method and apparatus for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion

Abstract

This invention belongs to the field of new energy power system safety and discloses a method and device for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion. The method includes: acquiring multi-source feature data of the target photovoltaic power plant; the multi-source feature data includes electrical parameters, thermal imaging data, environmental data, and gas composition data; preprocessing the multi-source feature data to obtain preprocessed multi-source feature data; extracting and fusing key features from the preprocessed multi-source feature data, and using principal component analysis to reduce dimensionality, obtaining a feature vector after feature-level fusion; inputting the feature vector after feature-level fusion into a neural network-based multi-classifier fusion model for decision-level fusion to obtain a decision-level fusion result; and outputting the fire risk assessment result of the target photovoltaic power plant based on the decision-level fusion result. This invention effectively improves the accuracy, real-time performance, and reliability of fire risk assessment of distributed photovoltaic power plants through multi-source data fusion.

Description

Technical Field

[0001] This invention belongs to the field of safety technology for new energy power systems, and relates to risk management technology for distributed photovoltaic power stations, particularly to a method and device for fire risk assessment of distributed photovoltaic power stations based on multi-source data fusion. Background Technology

[0002] With the rapid development of renewable energy technologies, distributed photovoltaic (PV) power stations, with their advantages of flexible deployment and low construction costs, are widely used in commercial and industrial rooftops, public buildings, and residential buildings. However, PV power stations often operate outdoors for extended periods, making them highly susceptible to aging components, short circuits, and overloads, which are major causes of PV fires. Once a PV power station catches fire, it not only damages equipment and shuts down the power generation system but also easily ignites surrounding structures and vegetation. Therefore, real-time fire monitoring, hazard warning, and risk control for PV power stations are crucial to ensuring their safe and stable operation.

[0003] The causes of fires in photovoltaic power plants are complex and often concealed, encompassing both electrical fault risks such as abnormal voltage and current, and non-electrical hazards such as equipment aging due to high temperatures, abnormal environmental temperature and humidity, and spontaneous combustion of accumulated debris. These risks are diverse and overlapping. Currently, the mainstream prevention and control methods in the industry are mostly single-dimensional monitoring approaches, which cannot comprehensively cover all types of fire risk sources, resulting in significant shortcomings in monitoring coverage. Furthermore, the outdoor operation of photovoltaic power plants makes monitoring data susceptible to interference from weather changes and fluctuations in normal equipment start-up and shutdown, leading to frequent false alarms and missed alarms, and poor monitoring reliability. In addition, existing risk assessments largely rely on the historical experience of maintenance personnel and static data, lacking real-time, quantitative monitoring methods, and thus failing to meet the current intelligent and refined safety operation and maintenance needs of distributed photovoltaic power plants.

[0004] To address the issues of incomplete fire monitoring and lagging risk assessment in distributed photovoltaic (PV) power plants, the industry has gradually developed various fire monitoring and risk assessment technologies to support the safe operation and maintenance of power plants. Among these, single-sensor monitoring technology uses temperature and smoke sensors deployed at key locations within the power plant to collect fire characteristic data such as temperature and smoke. When the monitored data exceeds a preset threshold, an alarm is triggered, enabling basic fire anomaly monitoring. Electrical parameter monitoring technology primarily collects voltage and current data during the operation of PV equipment. By identifying abrupt changes or deviations in electrical parameters, it determines whether there are fire hazards caused by electrical faults within the power plant. Image recognition technology uses camera equipment to collect images of the power plant and relies on image recognition algorithms to identify visible fire characteristics such as flames and smoke, enabling the monitoring of sudden fires such as open flames and smoke. Traditional risk assessment methods rely on historical operating data and fault cases of the power plant, combined with the practical experience of operation and maintenance personnel, to manually determine and periodically investigate the fire risks of the power plant.

[0005] While the aforementioned existing technologies can achieve basic investigation and monitoring of fire risks in photovoltaic power plants, they still fall short of achieving comprehensive, precise, and proactive fire prevention and control. Specifically, single-sensor monitoring can only achieve single-dimensional data collection, with a limited monitoring dimension. It is easily affected by outdoor environment and normal equipment operation status, resulting in a high false alarm rate and poor monitoring stability. Electrical parameter monitoring only identifies electrical fault risks and cannot cover non-electrical fire causes such as equipment aging and environmental anomalies, thus failing to comprehensively characterize the overall fire risk status of the power plant. The recognition accuracy of image recognition technology is greatly affected by lighting, weather, and image resolution, resulting in low recognition accuracy under complex operating conditions. Furthermore, it can only identify visible fires that have already occurred and lacks the ability to provide early warning of potential hazards. Traditional manual assessment methods are highly subjective, lack quantitative standards, cannot achieve real-time dynamic monitoring, and the assessment results are lagging and inaccurate, making it difficult to meet the current development needs of intelligent safety operation and maintenance of large-scale distributed photovoltaic power plants. Summary of the Invention

[0006] To address the technical problems existing in the prior art, this invention provides a method and device for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion, in order to solve the technical problem that the existing technology still cannot achieve a comprehensive, accurate, and proactive fire prevention and control effect.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion, comprising: Acquire multi-source characteristic data of the target photovoltaic power station; the multi-source characteristic data includes electrical parameters, thermal imaging data, environmental data, and gas composition data; Preprocess the multi-source feature data to obtain preprocessed multi-source feature data; Key features are extracted and fused from the preprocessed multi-source feature data, and dimensionality is reduced using principal component analysis to obtain the feature vector after feature-level fusion. The feature vector after feature-level fusion is input into a neural network-based multi-classifier fusion model for decision-level fusion to obtain the decision-level fusion result. Based on the decision-level fusion results, the fire risk assessment results of the target photovoltaic power station are output.

[0008] Furthermore, the electrical parameters include the operating current and voltage data of key electrical equipment in the target photovoltaic power plant; Thermal imaging data includes thermal images of pre-defined monitoring locations at the target photovoltaic power plant; Environmental data includes temperature, humidity, light intensity, wind speed, and wind direction at the target photovoltaic power station's predetermined location; The gas composition data includes the concentrations of characteristic gases within the target photovoltaic power plant.

[0009] Furthermore, the process of preprocessing the multi-source feature data to obtain preprocessed multi-source feature data is as follows: The multi-source feature data is subjected to denoising filtering and normalization to obtain preprocessed multi-source feature data. The denoising filtering adopts an adaptive filtering algorithm, a filtering method based on image statistics, or a low-pass filtering method. The normalization processing adopts a linear normalization method or a Z-Score normalization method.

[0010] Furthermore, the feature vector after feature-level fusion includes feature data extracted from electrical parameters, feature data extracted from thermal imaging data, feature data extracted from environmental data, and feature data extracted from gas composition data.

[0011] Furthermore, feature data extracted from electrical parameters includes current change rate and voltage harmonic content; feature data extracted from thermal imaging data includes the area ratio of high-temperature regions and temperature gradient; feature data extracted from environmental data includes temperature and humidity change rate and the interaction influencing factors of light intensity and wind speed and direction; and feature data extracted from gas composition data includes the concentration growth rate of characteristic gases and the concentration ratio of different characteristic gases.

[0012] Furthermore, the neural network-based multi-classifier fusion model employs a pre-trained neural network model, including an input layer, an intermediate layer, and an output layer. The number of neurons in the input layer is determined based on the dimension of the feature vector after feature-level fusion. The intermediate layer includes several hidden neurons and uses ReLU as the activation function. The output layer includes low-risk neurons, medium-risk neurons, and high-risk neurons.

[0013] This invention also provides a distributed photovoltaic power station fire risk assessment system based on multi-source data fusion, used to implement the aforementioned distributed photovoltaic power station fire risk assessment method based on multi-source data fusion, comprising: The data acquisition module is used to acquire multi-source characteristic data of the target photovoltaic power station; the multi-source characteristic data includes electrical parameters, thermal imaging data, environmental data and gas composition data; The data preprocessing module is used to preprocess multi-source feature data to obtain preprocessed multi-source feature data; The feature-level fusion module is used to extract and fuse key features from the preprocessed multi-source feature data, and to reduce the dimensionality using principal component analysis to obtain the feature vector after feature-level fusion. The decision-level fusion module is used to input the feature vector after feature-level fusion into a neural network-based multi-classifier fusion model for decision-level fusion to obtain the decision-level fusion result. The risk assessment module is used to output fire risk assessment results for the target photovoltaic power plant based on the decision-level fusion results.

[0014] The present invention also provides an electronic device, comprising: A processor is used to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, performs the aforementioned method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion.

[0015] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion.

[0016] The present invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the aforementioned method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: The fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion provided by this invention effectively improves the accuracy, real-time performance, and reliability of fire risk assessment for distributed photovoltaic power plants by comprehensively considering multiple factors through multi-source data fusion. Specifically, it collects four types of multi-source monitoring data from the photovoltaic power plant: electrical parameters, thermal imaging data, environmental data, and gas composition data. This comprehensively covers fire risk factors such as electrical faults, high-temperature aging of equipment, environmental anomalies, and the natural and diverse intersection of potential hazards, effectively avoiding the shortcomings of incomplete coverage in existing risk monitoring technologies. Furthermore, by preprocessing the multi-source data, fusing key features, and performing principal component dimensionality reduction, it can effectively filter out outdoor weather-related factors. Environmental and operational noise, such as interference and fluctuations in normal equipment start-up and shutdown, significantly improves the accuracy of effective risk feature extraction, addressing the problems of high false alarm and false negative rates and poor stability in traditional monitoring methods. In addition, the adoption of a multi-classifier fusion model based on neural networks enables decision-level risk assessment, eliminating the need for reliance on human experience and static data-driven lagging evaluation modes. This allows for real-time, quantitative, and dynamic assessment of power plant fire risks, identifying both existing visible fires and predicting potential fire hazards, effectively achieving proactive risk prevention and control. This significantly improves the comprehensiveness, accuracy, and real-time nature of fire risk assessment for distributed photovoltaic power plants, adapting to the intelligent and refined safety operation and maintenance needs of large-scale distributed photovoltaic power plants.

[0018] Furthermore, the specific types of multi-source characteristic data are precisely defined, with electrical parameters, thermal imaging data, environmental data, and gas composition data being specifically limited to photovoltaic equipment operating current and voltage, thermal images of monitored locations, on-site temperature, humidity, light, wind speed, and characteristic gas concentrations. This corresponds to various core causes of fires in distributed photovoltaic power stations, thus comprehensively covering the electrical operating status of photovoltaic power stations, equipment heating status, on-site environmental conditions, and fault-generated gas status. This achieves all-round perception of fire risk sources and provides comprehensive and reliable raw data support for multi-source data fusion and accurate risk assessment.

[0019] Furthermore, by combining adaptive filtering, image statistical filtering, and low-pass filtering with linear normalization and Z-Score normalization, the original multi-source data is standardized and preprocessed. This effectively filters out various operating conditions and environmental noises and eliminates invalid interference data. At the same time, normalization processing unifies the dimensions and numerical ranges of monitoring data of different categories and magnitudes, solving the technical problem that electrical, image, environmental, and gas data have large scale differences and cannot be directly fused and calculated. This significantly improves the accuracy and stability of subsequent feature extraction and data fusion.

[0020] Furthermore, by limiting the feature vectors after feature-level fusion, an integrated fusion representation of multi-source data features can be achieved, thereby enabling a complete and intuitive mapping of the overall operational safety status of the photovoltaic power station. This provides fusion feature data with complete dimensions and strong correlation for subsequent decision-level fusion judgment, effectively improving the comprehensiveness and integrity of risk assessment.

[0021] Furthermore, by extracting current and voltage change rates, temperature gradients, environmental factor interaction coefficients, and gas concentration growth rates, the technology effectively overcomes the shortcomings of existing technologies, which can only identify steady-state abnormalities and cannot capture early-stage hidden faults. In particular, by extracting dynamic change features, coupled correlation features, and ratio features, subtle abnormal changes in the early stages of faults can be accurately captured, enabling early detection of potential fire hazards and significantly improving the anticipation and sensitivity of risk warnings.

[0022] Furthermore, a three-layer neural network structure adapted to the feature vector dimension is adopted, combined with the ReLU activation function and three-level risk classification output, to construct a decision-level fusion model. Leveraging the strong nonlinear fitting capability of the pre-trained neural network, it can uncover the complex correlation between multi-source fusion features and fire risk levels. Simultaneously, the structured design of the model's input layer, hidden layer, and output layer can adapt to multi-dimensional fusion feature data. The ReLU activation function avoids gradient saturation problems, improving model computational efficiency and fitting accuracy. Ultimately, it achieves quantitative classification output of low, medium, and high risks, enabling real-time, quantitative, and intelligent classification of fire risks, effectively improving the accuracy and objectivity of risk assessment results.

[0023] The fire risk assessment system, electronic equipment, computer-readable storage medium, and computer program product for distributed photovoltaic power plants based on multi-source data fusion provided by this invention possess all the advantages of the aforementioned fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart of the fire risk assessment method for distributed photovoltaic power plants provided in Example 1; Figure 2 This is a structural block diagram of the distributed photovoltaic power station fire risk assessment system provided in Example 2; Figure 3 This is a structural block diagram of the electronic device provided in Example 3. Detailed Implementation

[0026] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0027] This invention provides a method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion, comprising the following steps: Step 100: Obtain multi-source characteristic data of the target photovoltaic power station; wherein, the multi-source characteristic data includes electrical parameters, thermal imaging data, environmental data and gas composition data.

[0028] Step 200: Preprocess the multi-source feature data to obtain preprocessed multi-source feature data.

[0029] Step 300: Extract and fuse key features from the preprocessed multi-source feature data, and use principal component analysis to reduce dimensionality to obtain the feature vector after feature-level fusion.

[0030] Step 400: Input the feature vector after feature-level fusion into a neural network-based multi-classifier fusion model for decision-level fusion to obtain the decision-level fusion result.

[0031] Step 500: Based on the decision-level fusion results, output the fire risk assessment results of the target photovoltaic power station.

[0032] In the above embodiments, by simultaneously collecting four types of multi-source heterogeneous data—electrical parameters, thermal imaging data, environmental data, and gas composition—it is possible to simultaneously capture multiple types of fire causes, such as electrical faults, equipment overheating, environmental anomalies, and the accumulation of combustible gas. Secondly, in the feature-level fusion stage, principal component analysis is used for dimensionality reduction. The statistical correlation between data is used to eliminate redundant information and random noise introduced by external factors such as weather changes and equipment start-up and shutdown from multi-source sensors, thereby improving the effective signal-to-noise ratio from an information theory perspective and enhancing the robustness of feature representation. Thirdly, at the decision level, a multi-classifier fusion model based on neural networks is introduced. Through the parallel discrimination and weighted fusion of feature vectors by multiple classifiers, the statistical advantages of ensemble learning are used to reduce the probability of misjudgment by a single classifier under complex operating conditions, achieving higher generalization ability and decision reliability than a single model. This invention integrates multi-source heterogeneous data into a unified fire risk quantification index and performs dynamic risk assessment based on real-time data streams, replacing the traditional lagging assessment mode that relies on human experience and static data. This simultaneously solves the core problems of missing monitoring dimensions, false alarms and missed alarms caused by environmental interference, and insufficient proactive risk warning.

[0033] The following specific embodiments further explain the fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion provided by the present invention: Example 1 As attached Figure 1 As shown in Example 1, this method provides a fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion, including the following steps: Step 1: Obtain multi-source characteristic data of the target photovoltaic power station; the multi-source characteristic data includes electrical parameters, thermal imaging data, environmental data and gas composition data.

[0034] Electrical parameters include operating current and voltage data of key electrical equipment in the target photovoltaic power station; thermal imaging data includes thermal images of preset monitoring locations in the target photovoltaic power station; environmental data includes temperature, humidity, light intensity, wind speed and direction at preset locations in the target photovoltaic power station; gas composition data includes the concentration of characteristic gases in the target photovoltaic power station; optionally, the concentration of characteristic gases includes the concentration of carbon monoxide, sulfur dioxide and hydrogen.

[0035] The specific implementation process is as follows: By reading data from pre-configured electrical parameter acquisition units, thermal imaging data acquisition units, environmental data acquisition units, and gas composition monitoring units, multi-source characteristic data of the target photovoltaic power station is obtained. Among them, the electrical parameter acquisition unit is used to collect electrical parameters of the target photovoltaic power station; the thermal imaging data acquisition unit is used to collect thermal imaging data of the target photovoltaic power station; the environmental data acquisition unit is used to collect environmental data of the target photovoltaic power station; and the gas composition monitoring unit is used to collect gas composition data of the target photovoltaic power station.

[0036] The electrical parameter acquisition unit refers to the current sensor and voltage sensor pre-deployed on the key electrical equipment of the target photovoltaic power station, used to collect the operating current and voltage data of the key electrical equipment in real time. The key electrical equipment includes, for example, photovoltaic modules, inverters, and combiner boxes. By reading the real-time data collected by the current sensor and voltage sensor, the electrical parameters of the target photovoltaic power station can be obtained. Optionally, the current sensor has a range of 0-100A and an accuracy of ±0.1%FS, and the voltage sensor has a range of 0-1000V and an accuracy of ±0.2%FS. The data acquisition frequency of both the current sensor and the voltage sensor is set to be once every 1 second.

[0037] Interpretive methods involve deploying current and voltage sensors on photovoltaic modules, inverters, and combiner boxes to collect real-time operating current and voltage data with accuracy down to the microamp and millivolt level. This data is used to monitor the operating status of electrical equipment and detect early signs of potential electrical faults that could lead to fires. For example, abnormal fluctuations in current or sudden drops or rises in voltage may indicate internal short circuits, open circuits, or overloads, which are significant causes of fires.

[0038] The thermal imaging data acquisition unit refers to a thermal imager pre-deployed at a preset location in the target photovoltaic power station. The thermal imager performs an all-round scan of the entire power station and acquires thermal images of various parts of the power station at preset time intervals. Optionally, the thermal imager has a resolution of 640×480 pixels and a temperature measurement accuracy of ±2℃, and scans the power station at a time interval of 10 seconds.

[0039] Interpretive methods utilize thermal imagers to perform a comprehensive scan of the entire power plant, acquiring thermal images of various parts of the power plant at preset time intervals to form a thermal imaging sequence. Thermal imaging technology can penetrate smoke and dust, intuitively presenting the temperature distribution of object surfaces. It can not only detect high-temperature areas caused by local overheating and hot spot effects of photovoltaic modules, but also discover abnormal heat dissipation parts of electrical equipment, providing intuitive thermal information for fire risk assessment.

[0040] The environmental data acquisition unit refers to temperature and humidity sensors, light intensity sensors, and wind speed and direction sensors pre-installed at different locations within the power plant. These sensors continuously collect data on ambient temperature and humidity, light intensity, and wind speed and direction. Optionally, the temperature and humidity sensors have a temperature measurement range of -40℃ to 120℃ and an accuracy of ±0.5℃; the humidity measurement range is 0-100%RH with an accuracy of ±3%RH; and the light intensity sensor has a range of 0-2000W / m. 2 The accuracy is ±5%FS; the wind speed and direction sensor has a wind speed measurement range of 0-50m / s and an accuracy of ±0.3m / s; the wind direction measurement accuracy is ±5°; the temperature and humidity sensor, light intensity sensor, and wind speed and direction sensor all collect data once every 5 seconds.

[0041] Explanatoryly, temperature and humidity sensors, light intensity sensors, and wind speed and direction sensors are installed at different locations in the power plant to continuously collect environmental temperature, humidity, light intensity, wind speed, and wind direction data. Considering that dry, high-temperature, and strong-wind environments can accelerate the spread of fire, and that while humid environments may reduce flammability, they may affect the insulation performance of electrical equipment, environmental temperature, humidity, light intensity, and wind speed and direction are closely related to the occurrence and development of fires. Real-time monitoring can provide external conditions to support risk assessment.

[0042] A gas composition monitoring unit refers to a pre-deployed array of gas sensors used to sample and analyze the air within the power plant in real time, detecting the concentration of combustible or harmful gases such as carbon monoxide, sulfur dioxide, and hydrogen. Optionally, the gas sensor array includes a carbon monoxide sensor, a sulfur dioxide sensor, and a hydrogen sensor. The carbon monoxide sensor has a range of 0-100 ppm and an accuracy of ±1 ppm; the sulfur dioxide sensor has a range of 0-50 ppm and an accuracy of ±0.5 ppm; the hydrogen sensor has a range of 0-100 ppm and an accuracy of ±1 ppm; the sampling frequency of the carbon monoxide sensor, sulfur dioxide sensor, and hydrogen sensor is once every 30 seconds.

[0043] Interpretively, by setting up a gas sensor array to sample and analyze the air inside the power plant in real time, the concentration of combustible or harmful gases such as carbon monoxide, sulfur dioxide, and hydrogen can be detected. Specific gases are released during processes such as the combustion of photovoltaic module encapsulation materials and the overheating decomposition of electrical equipment insulation materials. By monitoring changes in gas composition, potential fire hazards can be detected in advance, achieving ultra-early warning.

[0044] Step 2: Preprocess the multi-source feature data to obtain preprocessed multi-source feature data. The preprocessing includes denoising filtering and normalization. Optionally, denoising filtering may employ an adaptive filtering algorithm, an image statistics-based filtering method, or a low-pass filtering method; normalization may employ linear normalization or Z-Score normalization.

[0045] In this embodiment 1, different filtering methods are used to denoise the collected multi-source feature data. For electrical parameters, an adaptive filtering algorithm is used to remove power frequency interference and high-frequency noise, ensuring smooth and stable current and voltage signals for accurate subsequent analysis. Adaptive filtering algorithms include, for example, the minimum mean algorithm or the recursive least squares algorithm. For thermal imaging data, a filtering method based on image statistics is used to eliminate noise caused by the thermal imager's own noise and environmental thermal radiation fluctuations, improving image clarity. For environmental data and gas composition data, a low-pass filtering method is used to filter out random interference during the acquisition process and restore the true data change trend.

[0046] To ensure comparability of data of different types and ranges, linear normalization or Z-score normalization methods are used to map electrical parameters, thermal imaging temperature values, environmental data, and gas concentrations to the [0,1] interval or conform to a standard normal distribution, thereby eliminating differences in data dimensions and laying the foundation for subsequent fusion of multi-source feature data.

[0047] Step 3: Extract and fuse key features from the preprocessed multi-source feature data, and perform dimensionality reduction using principal component analysis to obtain the feature vector after feature-level fusion. The feature vector after feature-level fusion includes feature data extracted from electrical parameters, thermal imaging data, environmental data, and gas composition data. Specifically, the feature data extracted from electrical parameters includes current change rate and voltage harmonic content; the feature data extracted from thermal imaging data includes the area ratio of high-temperature regions and temperature gradient; the feature data extracted from environmental data includes temperature and humidity change rate and the interaction factors between light intensity and wind speed and direction; and the feature data extracted from gas composition data includes the concentration growth rate of characteristic gases and the concentration ratio of different characteristic gases.

[0048] The specific implementation process is as follows: Step 31: Extract key features from the preprocessed multi-source feature data and perform feature fusion to obtain a fused feature vector; specifically, for electrical parameters, extract the current change rate and voltage harmonic content; for thermal imaging data, extract the area ratio of high-temperature regions and temperature gradient; for environmental data, extract the temperature and humidity change rate and the interaction factor between light intensity and wind speed; for gas composition data, extract the specific gas concentration growth rate and the ratio of different gas concentrations.

[0049] Step 32: Using principal component analysis, project the fused feature vector into a low-dimensional space to reduce data redundancy while retaining key information, improve fusion efficiency, and obtain the feature vector after feature-level fusion.

[0050] Step 4: Input the feature vector after feature-level fusion into a neural network-based multi-classifier fusion model for decision-level fusion to obtain the decision-level fusion result.

[0051] Optionally, the neural network-based multi-classifier fusion model employs a pre-trained neural network model, including an input layer, an intermediate layer, and an output layer; the number of neurons in the input layer is determined based on the dimension of the feature vector after feature-level fusion; the intermediate layer includes several hidden neurons and uses ReLU as the activation function; the output layer includes low-risk neurons, medium-risk neurons, and high-risk neurons.

[0052] Specifically, the feature vector after feature-level fusion is used as the input layer neuron, and multiple hidden neurons are set in the intermediate layer for feature learning and abstraction. The output layer corresponds to different fire risk level values, which serve as the decision-level fusion result. The fire risk level values ​​include low risk level, medium risk level, and high risk level. The low risk level value is the output value of the low risk neuron, the medium risk level value is the output value of the medium risk neuron, and the high risk level value is the output value of the high risk neuron.

[0053] The training process of the pre-trained neural network model involves training the model with a large amount of labeled historical data, enabling the model to learn the complex nonlinear relationship between various features and fire risk, and ultimately achieving accurate classification and assessment of fire risk of distributed photovoltaic power stations. It should be noted that the specific training process of the neural network model is not the focus of this application and will not be elaborated here.

[0054] Step 5: Based on the decision-level fusion results, output the fire risk assessment results for the target photovoltaic power station. Specifically, the fire risk level is assessed based on the decision-level fusion results. When the risk level exceeds a preset threshold, an early warning message is issued via an audible and visual alarm device and a remote communication module.

[0055] Specifically, the decision-level fusion result is compared with a pre-set risk level threshold to obtain the fire risk level. Optionally, the pre-set risk level thresholds include a low-risk level threshold, a medium-risk level threshold, and a high-risk level threshold. Among them, the low-risk level threshold is the output value of the low-risk neuron at its maximum and less than 0.3; the medium-risk threshold is the output value of the medium-risk neuron at its maximum and between 0.3 and 0.7; and the high-risk threshold is the output value of the high-risk neuron at its maximum and greater than 0.7.

[0056] Interpretive assessments, based on decision-level fusion results and pre-set risk level thresholds, quantify the current fire risk level of the power plant. When the assessment result is low risk, it indicates that the power plant is operating normally and the fire hazard is minimal. Medium risk means that there are certain potential problems and monitoring needs to be strengthened. High risk requires the immediate activation of the emergency plan and the adoption of emergency measures such as cutting off faulty circuits and activating fire-fighting equipment.

[0057] In this embodiment 1, it also includes: when the decision-level fusion result exceeds the preset risk level threshold, a warning message is issued through an audible and visual alarm device and a remote communication module; specifically, the warning message is sent to the mobile phone of the operation and maintenance personnel and the monitoring center through the audible and visual alarm device and the remote communication module in the target photovoltaic station; the warning message not only includes the current fire risk level, but also lists in detail the main factors that cause the risk to increase (such as electrical faults in a certain area, high temperature and dry environment, etc.), so that the operation and maintenance personnel can quickly locate the problem and take targeted measures to minimize the fire loss.

[0058] Engineering example illustration: To verify the effectiveness of the method described in Example 1, a distributed photovoltaic power station with an installed capacity of 5MW is selected as an example. The power station contains 20,000 photovoltaic modules, distributed in an area of ​​approximately 100 acres, and is equipped with 10 inverters and corresponding combiner boxes and other electrical equipment.

[0059] (1) In this example power plant, various sensors are pre-deployed according to the following layout, as detailed below: (11) Electrical parameter sensors: Current sensors (range 0-100A, accuracy ±0.1%FS) and voltage sensors (range 0-1000V, accuracy ±0.2%FS) are installed at the input and output terminals of each inverter and on the connection lines of 10% (i.e. 2000) of photovoltaic modules. The current and voltage sensors transmit the collected data in real time via wired transmission, and the data acquisition frequency is set to once every 1 second.

[0060] (12) Thermal imager: A thermal imager (resolution of 640×480 pixels, temperature measurement accuracy of ±2℃) is installed at each of the four corners and the center of the example power plant to cover the entire power plant area; the thermal imager scans the power plant every 10 seconds and transmits thermal image data through a wireless network.

[0061] (13) Environmental sensors: 20 temperature and humidity sensors (temperature measurement range -40℃-120℃, accuracy ±0.5℃; humidity measurement range 0-100%RH, accuracy ±3%RH), 10 light intensity sensors (range 0-2000W / m², accuracy ±5%FS) and 5 wind speed and direction sensors (wind speed measurement range 0-50m / s, accuracy ±0.3m / s; wind direction measurement accuracy ±5°) are evenly distributed in the power station. The data collected by the environmental sensors are transmitted in real time through a wired network, and the acquisition frequency is once every 5 seconds.

[0062] (14) Gas sensor array: Ten gas sensor arrays are installed in key areas such as the power distribution room, inverter room and photovoltaic module dense area of ​​the power station. Each array includes a carbon monoxide sensor (range 0-100ppm, accuracy ±1ppm), a sulfur dioxide sensor (range 0-50ppm, accuracy ±0.5ppm) and a hydrogen sensor (range 0-100ppm, accuracy ±1ppm). The gas sensors transport and analyze the collected gas samples in real time through pipeline sampling. The data acquisition frequency is once every 30 seconds.

[0063] (2) Data acquisition and preprocessing (21) Data Acquisition: Each sensor continuously acquires data according to a predetermined frequency and transmission method. At a certain moment, the current value acquired from the inverter input terminal is 45.3A and the voltage value is 380V. The thermal imager scan found a local high temperature point in a photovoltaic module area, with the highest temperature reaching 85℃ and the area of ​​the high temperature area accounting for about 0.5%. The temperature and humidity sensor recorded a temperature of 32℃, a humidity of 40%RH, and a light intensity of 1000W / m. 2 The wind speed was 3 m / s; the gas sensor array detected a carbon monoxide concentration of 2 ppm, a sulfur dioxide concentration of 0.5 ppm, and a hydrogen concentration of 1 ppm.

[0064] (22) Data preprocessing (221) Noise denoising filter: For electrical parameter data, an adaptive filtering algorithm is used to remove power frequency interference. Taking current data as an example, let the original current sequence be I(n), the power frequency interference frequency be f0 = 50Hz, and the sampling frequency be f s =1Hz, the current sequence is transformed to the frequency domain I(k) by Discrete Fourier Transform (DFT), the frequency domain values ​​corresponding to f0 and its harmonic components are identified and removed, and then the denoised current sequence I'(n) is obtained by Inverse Discrete Fourier Transform (IDFT). After processing, the current value becomes 45.2A, which is more stable and accurate.

[0065] The thermal imaging data is processed using a median filtering method based on image statistics, with a 3×3 filtering window to remove isolated noise caused by the thermal imager's own thermal noise, making the high-temperature area clearer and more realistic.

[0066] Environmental and gas composition data are filtered using a low-pass filter to remove random high-frequency interference during the acquisition process. For example, a cutoff frequency of f is set for temperature and humidity data. c A low-pass filter with a frequency of 0.1 Hz filters out high-frequency interference such as ambient thermal fluctuations, and corrects the temperature value to 31.8℃ and the humidity value to 39%RH.

[0067] (222) Data normalization: Using the Z-score normalization method, for all data collected at a certain time, the mean is first calculated. and standard deviation Taking voltage data as an example, suppose the set of voltage samples collected over a period of time is V={v1,v2,…,v...} n}, then the mean Standard deviation After normalizing the current voltage value v=380V, we get Similarly, other data are normalized to make them comparable and facilitate subsequent fusion.

[0068] (3) Feature-level fusion Extracting the rate of change of current from electrical parameters Assuming the current was 45.0A at the previous moment and is 45.2A at the current moment, the time interval... =1s, then the rate of change of current is (45.2-45.0) / 1=0.2A / s; the voltage harmonic content is extracted, and the voltage signal spectrum is analyzed by Fast Fourier Transform (FFT) to calculate the ratio of harmonic components to fundamental components. The analysis shows that the voltage harmonic content is 3%.

[0069] Thermal imaging data extracts the area percentage of high-temperature regions (S). high / S total (The area of ​​the high-temperature region is known to be approximately 0.5%) and temperature gradient. The temperature gradient is calculated by the temperature difference between adjacent pixels in the thermal image. Assuming that the temperature difference between adjacent pixels in a certain direction in a high-temperature region is 5℃ and the pixel spacing is 1mm, the temperature gradient is 5℃ / mm.

[0070] Environmental data is used to extract the rate of change of temperature and humidity. Assuming the previous time was 30℃ and the humidity was 38%RH, and the current time is 31.8℃ and the humidity is 39%RH, the time interval is... =5s, then the temperature change rate is (31.8-30) / 5=0.36℃ / s, and the humidity change rate is (39-38) / 5=0.2%RH / s; extract the interaction factor between light intensity and wind speed. ,in Light intensity (W / m²) Let be the wind speed (m / s); given that the light intensity is 1000 W / m² and the wind speed is 3 m / s, then the interaction factor is 1000. 3 = 3000 Wm / s.

[0071] Gas composition data selects the rate of increase of specific gas concentrations. For example, assuming the carbon monoxide concentration was 1 ppm at the previous moment and 2 ppm at the current moment, the time interval... =30s, then the carbon monoxide concentration growth rate is (2-1) / 30=0.033ppm / s; calculate the concentration ratio of different gases, such as the carbon monoxide to sulfur dioxide concentration ratio is 2 / 0.5=4.

[0072] Principal Component Analysis (PCA) is used for dimensionality reduction, projecting the extracted high-dimensional feature vectors into a low-dimensional space. Assuming the original feature vector dimension is n=10, the eigenvalues ​​and eigenvector matrix are calculated using the PCA algorithm. The eigenvectors corresponding to the first m=5 eigenvalues ​​are selected as new basis vectors. The original feature vectors are then projected into the new basis vector space to obtain the dimensionality-reduced feature vectors, which reduces data redundancy while retaining key information.

[0073] (4) Decision-level integration A multi-classifier fusion model based on a neural network was pre-constructed. The feature vectors after feature-level fusion were used as input layer neurons, with a total of 5 input layer neurons. Ten hidden neurons were set in the intermediate layer, using ReLU as the activation function for feature learning and abstraction. The output layer corresponded to different fire risk levels and was set as low-risk neurons, medium-risk neurons, and high-risk neurons, for a total of 3 neurons. During the model construction process, a large amount of labeled historical data (a total of 10,000 sets of data) from the power plant and other similar power plants over the past year was collected to train the neural network model. The backpropagation algorithm was used to adjust the weights and biases of the neural network, enabling the model to learn the complex nonlinear relationship between each feature and fire risk.

[0074] In the current engineering example, the feature vector after feature-level fusion is input into the trained neural network model. After forward propagation calculation, the output values ​​of the output layer neurons are 0.1 (low-risk neuron), 0.6 (medium-risk neuron), and 0.3 (high-risk neuron). Based on the principle of maximizing the output value, the current power plant fire risk is determined to be medium risk.

[0075] (5) Risk assessment and early warning implementation Risk Assessment: Based on the results output by the multi-source data fusion module, and combined with pre-set risk level thresholds (low risk threshold: maximum output value less than 0.3; medium risk threshold: maximum output value between 0.3 and 0.7; high risk threshold: maximum output value greater than 0.7), the current fire risk of the power plant is quantitatively assessed. In this embodiment, it is determined to be of medium risk, indicating that the power plant has certain potential problems and requires enhanced monitoring.

[0076] Early warning: Warning information is sent to the mobile phones of maintenance personnel and the monitoring center via the on-site audible and visual alarm devices and remote communication modules. The warning information not only indicates that the current fire risk level is medium, but also lists in detail the main factors that cause the risk to increase. For example, in terms of electrical parameters, the current change rate is 0.2A / s, and the voltage harmonic content is 3%; thermal imaging shows that the high temperature area accounts for about 0.5% of the area of ​​a certain photovoltaic module, and the temperature gradient is 5℃ / mm; in terms of environmental factors, the temperature change rate is 0.36℃ / s, the humidity change rate is 0.2%RH / s, and the interaction factor between light intensity and wind speed is 3000Wm / s; in terms of gas composition, the carbon monoxide concentration growth rate is 0.033ppm / s, and the carbon monoxide to sulfur dioxide concentration ratio is 4. This facilitates maintenance personnel to quickly locate the problem and take targeted measures to minimize fire losses.

[0077] The fire risk assessment method for distributed photovoltaic power plants described in Embodiment 1 collects four types of multi-source monitoring data: electrical parameters, thermal imaging data, environmental data, and gas composition data. The collected data undergoes adaptive filtering, image-based statistical filtering, and low-pass filtering for noise reduction. It is then preprocessed using Z-score normalization to eliminate interference and unify dimensions. Next, key features such as current change rate, high-temperature area ratio, and temperature and humidity change rate are extracted from each data source. After dimensionality reduction using principal component analysis, a multi-classifier fusion model based on a neural network is constructed for decision-level fusion. Finally, based on the decision-level fusion results and a preset threshold, the fire risk is quantitatively assessed. When the risk exceeds the threshold, an alarm is triggered via audible and visual signals and a remote communication module to warn maintenance personnel and the monitoring center. Detailed risk factors are also listed. Embodiment 1 overcomes the limitations of existing single-source assessments, significantly improving the accuracy, real-time performance, and reliability of risk assessment, providing strong technical support for fire prevention and control in distributed photovoltaic power plants.

[0078] Example 2 As attached Figure 2 As shown, this embodiment 2 provides a distributed photovoltaic power station fire risk assessment system based on multi-source data fusion, which is used to implement the distributed photovoltaic power station fire risk assessment method based on multi-source data fusion described in embodiment 1 above. It includes a data acquisition module, a data preprocessing module, a feature-level fusion module, a decision-level fusion module, and a risk assessment module.

[0079] The system comprises the following modules: a data acquisition module for acquiring multi-source feature data of the target photovoltaic power station, including electrical parameters, thermal imaging data, environmental data, and gas composition data; a data preprocessing module for preprocessing the multi-source feature data; a feature-level fusion module for extracting and fusing key features from the preprocessed multi-source feature data, and using principal component analysis for dimensionality reduction to obtain a feature vector after feature-level fusion; a decision-level fusion module for inputting the feature vector after feature-level fusion into a neural network-based multi-classifier fusion model for decision-level fusion to obtain a decision-level fusion result; and a risk assessment module for outputting a fire risk assessment result for the target photovoltaic power station based on the decision-level fusion result.

[0080] Example 3 As attached Figure 3 As shown, this embodiment 3 provides an electronic device, including: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the distributed photovoltaic power station fire risk assessment method based on multi-source data fusion; or, the processor executing the computer program to implement the functions of each module in the above-mentioned distributed photovoltaic power station fire risk assessment system based on multi-source data fusion.

[0081] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a preset function, the instruction segments describing the execution process of the computer program in the electronic device.

[0082] The electronic device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above are examples of electronic devices and do not constitute a limitation on the electronic device. It may include more components than described above, or combine certain components, or different components. For example, the electronic device may also include a communication interface, input / output devices, network access devices, and a bus.

[0083] The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor, or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the electronic device via various communication interfaces and lines.

[0084] The memory can be used to store the computer program and / or module. The processor implements various functions of the electronic device by running or executing the computer program and / or module stored in the memory and by calling the data stored in the memory.

[0085] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function (such as sound playback, image playback, etc.). The data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart memory cards, secure digital cards, flash memory cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0086] Example 4 This embodiment 4 also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for fire risk assessment of a distributed photovoltaic power station based on multi-source data fusion.

[0087] If the modules / units integrated in the distributed photovoltaic power station fire risk assessment system based on multi-source data fusion are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0088] Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion. This can also be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-mentioned method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or a preset intermediate form, etc.

[0089] The computer-readable storage medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0090] Example 5 This embodiment 5 provides a computer product, which includes a computer program stored in a computer-readable storage medium. The processor of the electronic device reads the computer program from the computer-readable storage medium and executes the computer program, so that the electronic device can execute the distributed photovoltaic power station fire risk assessment method based on multi-source data fusion described in embodiment 1, which will not be repeated here.

[0091] It should be noted that those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above methods.

[0092] The fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion described in this invention integrates heterogeneous data from multiple sources, covering photovoltaic module operating parameters, environmental monitoring data, and power plant operation and maintenance records. Through data fusion algorithms and model construction, it breaks through the limitations of assessment based on a single data source, accurately quantifies the fire risk level of distributed photovoltaic power plants, and provides key technical support for the safe operation and maintenance, early warning, and emergency decision-making of power plants, effectively ensuring the long-term stable, safe, and efficient operation of distributed photovoltaic power plants.

[0093] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

Claims

1. A method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion, characterized in that, include: Acquire multi-source characteristic data of the target photovoltaic power station; the multi-source characteristic data includes electrical parameters, thermal imaging data, environmental data, and gas composition data; Preprocess the multi-source feature data to obtain preprocessed multi-source feature data; Key features are extracted and fused from the preprocessed multi-source feature data, and dimensionality is reduced using principal component analysis to obtain the feature vector after feature-level fusion. The feature vector after feature-level fusion is input into a neural network-based multi-classifier fusion model for decision-level fusion to obtain the decision-level fusion result. Based on the decision-level fusion results, the fire risk assessment results of the target photovoltaic power station are output.

2. The method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion according to claim 1, characterized in that, Electrical parameters include the operating current and voltage data of key electrical equipment in the target photovoltaic power plant; Thermal imaging data includes thermal images of pre-defined monitoring locations at the target photovoltaic power plant; Environmental data includes temperature, humidity, light intensity, wind speed, and wind direction at the target photovoltaic power station's predetermined location; The gas composition data includes the concentrations of characteristic gases within the target photovoltaic power plant.

3. The method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion according to claim 1, characterized in that, The process of preprocessing multi-source feature data to obtain preprocessed multi-source feature data is as follows: The multi-source feature data is subjected to denoising filtering and normalization to obtain preprocessed multi-source feature data. The denoising filtering adopts an adaptive filtering algorithm, a filtering method based on image statistics, or a low-pass filtering method. The normalization processing adopts a linear normalization method or a Z-Score normalization method.

4. The method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion according to claim 1, characterized in that, The feature vector after feature-level fusion includes feature data extracted from electrical parameters, feature data extracted from thermal imaging data, feature data extracted from environmental data, and feature data extracted from gas composition data.

5. The method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion according to claim 4, characterized in that, Feature data extracted from electrical parameters includes current change rate and voltage harmonic content; feature data extracted from thermal imaging data includes the area ratio of high-temperature regions and temperature gradient; feature data extracted from environmental data includes temperature and humidity change rate and the interaction factors between light intensity and wind speed and direction; feature data extracted from gas composition data... This includes the concentration growth rate of the characteristic gas and the concentration ratio of different characteristic gases.

6. The method for fire risk assessment of distributed photovoltaic power plants based on multi-source data fusion according to claim 1, characterized in that, The neural network-based multi-classifier fusion model uses a pre-trained neural network model, including an input layer, an intermediate layer, and an output layer. The number of neurons in the input layer is determined based on the dimension of the feature vector after feature-level fusion. The intermediate layer includes several hidden neurons and uses ReLU as the activation function. The output layer includes low-risk neurons, medium-risk neurons, and high-risk neurons.

7. A fire risk assessment system for distributed photovoltaic power plants based on multi-source data fusion, characterized in that, The method for implementing the distributed photovoltaic power plant fire risk assessment method based on multi-source data fusion as described in any one of claims 1-6 includes: The data acquisition module is used to acquire multi-source characteristic data of the target photovoltaic power station; the multi-source characteristic data includes electrical parameters, thermal imaging data, environmental data and gas composition data; The data preprocessing module is used to preprocess multi-source feature data to obtain preprocessed multi-source feature data; The feature-level fusion module is used to extract and fuse key features from the preprocessed multi-source feature data, and to reduce the dimensionality using principal component analysis to obtain the feature vector after feature-level fusion. The decision-level fusion module is used to input the feature vector after feature-level fusion into a neural network-based multi-classifier fusion model for decision-level fusion to obtain the decision-level fusion result. The risk assessment module is used to output fire risk assessment results for the target photovoltaic power plant based on the decision-level fusion results.

8. An electronic device, characterized in that, include: A processor is used to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, performs the fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the fire risk assessment method for distributed photovoltaic power plants based on multi-source data fusion as described in any one of claims 1-6.

Images