Wind power plant fire intelligent sensing and early warning method based on multi-sensor fusion

Abstract

The invention discloses a wind power plant fire intelligent sensing and early warning method based on multi-sensor fusion, and belongs to the technical field of fire sensing and early warning, and the method comprises the steps: obtaining data information collected by a plurality of sensors in real time through the plurality of sensors disposed in a wind power plant, and carrying out the data preprocessing, according to the method, the importance weights of the environmental parameters are dynamically adjusted, the importance degrees of the parameters are reflected in real time according to environmental changes, the accuracy of environmental condition monitoring and modeling is improved, the space-time correlation model is formed in combination with the multi-dimensional time sequence matrix and the space correlation matrix, the uncertainty of the environmental state is comprehensively evaluated, and the accuracy of environmental condition monitoring and modeling is improved. The dynamic environment modeling and prediction are carried out, the time and space factors are fully considered, the fire key features are extracted by adaptively adjusting the weight according to the change of the dynamic environment model and the environment parameters to the fire sensitivity, the information related to the fire can be captured in a more targeted manner, and the accuracy and effectiveness of fire feature extraction are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fire perception and early warning, and specifically refers to a wind farm fire intelligent perception and early warning method based on multi-sensor fusion. Background Art

[0002] With the development of wind power generation, the scale of wind farms has gradually expanded, and the risk of fire has also increased accordingly. Traditional fire detection systems mainly rely on a single type of sensor. These systems have problems such as slow response speed, high false alarm rate and inability to accurately locate the fire source.

[0003] However, the existing intelligent perception and early warning methods for wind farm fires still have certain defects. The existing intelligent perception and early warning methods for wind farm fires usually use fixed weights to evaluate the importance of various environmental parameters, and cannot adjust the weights in real time according to environmental changes, resulting in poor adaptability of the model in different scenarios. Especially when environmental conditions change rapidly, only one aspect of time series data or spatial information is considered, and the time and space factors are not fully combined for comprehensive evaluation, which limits the comprehensive understanding of the uncertainty of environmental status and makes it difficult to predict potential risks in advance. When extracting fire characteristics, traditional methods usually use fixed methods and fail to adaptively adjust according to changes in environmental parameters, resulting in the extracted features being not accurate enough, especially the ability to capture early signs of fire is limited. There is a lack of detailed analysis of local fire characteristics, and it is impossible to fully describe the subtle changes in fire characteristics. To this end, an intelligent perception and early warning method for wind farm fires based on multi-sensor fusion is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a wind farm fire intelligent perception and early warning method based on multi-sensor fusion to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a wind farm fire intelligent perception and early warning method based on multi-sensor fusion, comprising the following steps:

[0006] S1. Using multiple sensors deployed in the wind farm, real-time data information collected by multiple sensors is acquired and data preprocessing is performed.

[0007] S2, real-time monitoring and modeling of environmental conditions around the wind farm based on the pre-processed data;

[0008] S3, extracting key fire features from the preprocessed data based on environmental information, and fusing fire feature data through dimensionality reduction optimization;

[0009] S4. Perform real-time intelligent perception analysis based on the fused fire characteristic data to identify abnormal patterns;

[0010] S5. Obtain intelligent perception analysis results, automatically adjust system configuration according to the current environment and sensor status, and feed back to intelligent perception;

[0011] S6. Based on the abnormal detection situations identified by intelligent perception analysis, it immediately triggers the generation of automated reports to notify management personnel and integrates IoT devices for early warning.

[0012] S7. Optimize and adjust system operating parameters and multi-sensor parameter configuration information based on feedback, performance monitoring results, and environmental information.

[0013] Among them, the S1 obtains environmental data information collected by multiple sensors in real time through multiple sensors deployed in the wind farm, and performs data preprocessing; obtains wind farm data information in real time according to the multiple sensors in the wind farm, transmits the multi-sensor data to the central data processing unit, and the central processing unit performs preliminary filtering on the environmental data obtained by the sensors to remove abnormal values, checks and processes the abnormal values, converts the format of the preprocessed data and stores it in the database.

[0014] Wherein, the S2 is to monitor and model the environmental conditions around the wind farm in real time based on the pre-processed data; suppose the environmental parameter set of the pre-processed data is E(t)={E1(t), E2(t), ..., E i (t)}, let the importance weight of each environmental parameter w i (t) Dynamic adjustment is performed, and dynamic weight update is implemented as follows:

[0015]

[0016] In the formula, w i (t+1) represents the updated weight, w i (t) represents the weight of the i-th environment parameter at time t, η represents the learning rate, Represents the sum of the absolute values ​​of the rates of change of all parameters, R i (t) represents the rate of change of the i-th parameter at time t, the rate of change R i (t) is R i (t) = E i (t)-E i (t-Δt), measures the rate of change of the i-th parameter over time.

[0017] Among them, the said S2, assuming that the multidimensional time series matrix is ​​T(t), combined with the spatial correlation matrix S to form a spatiotemporal correlation model:

[0018]

[0019] In the formula, T sal(t) represents the spatiotemporal correlation model, T(t) represents the multidimensional time series matrix, represents the tensor product operator, which is used to combine time and space information, and S represents the spatial correlation matrix;

[0020] The uncertainty of the current environmental state is comprehensively evaluated based on the spatiotemporal correlation model data. The implementation formula is:

[0021]

[0022] In the formula, H(E(t)) represents the entropy value of the environment state, Indicates the accumulation of all n environmental parameters, w i (t) represents the weight of the i-th environmental parameter at time t, p(E i (t)) represents the probability distribution of the i-th environment parameter, and the probability distribution p(E i (t))=KDE(T sal (t) i ), KDE represents kernel density estimation, T sal (t) i represents the subset of spatiotemporal data related to the i-th parameter, and log represents the natural logarithm function;

[0023] According to T sal (t) and H(E(t)) for dynamic environment modeling and prediction, assuming the current environment state is s t , the strategic action taken is a t , execute the selected action a t , observe the immediate reward r t and the next state s t+1 , the dynamic environment modeling prediction implementation formula is:

[0024]

[0025] In the formula, Q(s t ,a t ) represents the dynamic environment model, s t Indicates the state is s t =(T sal (t),H(E(t))),r t represents the immediate reward, α represents the learning rate, γ represents the discount factor, Indicates the maximum Q value in the next state.

[0026] Among them, the said S3 extracts the key features of fire from the pre-processed data according to the environmental information, and fuses the fire feature data through dimensionality reduction optimization; according to the constructed dynamic environment model Q(s t ,a t), through adaptive weights, the selection is dynamically adjusted according to the changes in the sensitivity of each environmental parameter to fire, and the implementation formula is:

[0027]

[0028] In the formula, W(t) represents the environment parameter weight, η represents the learning rate, and Q(s t ,a t ) represents the output of the dynamic environment model, E(t) represents the environmental parameters at time point t, Represents the gradient of the impact of environmental parameters on the output of the environmental model;

[0029] Fire characteristics are extracted according to the weight of environmental parameters. Let the selected key parameter be F j (t), the fire feature extraction formula is:

[0030]

[0031] In the formula, represents the multi-scale eigenvalue at time point t, α k (t) represents the weighting coefficient of scale k, the value at time point t, W(F j ,s k ) represents the scale s k F j The result of wavelet transformation, K represents the number of scales.

[0032] Among them, S3 captures the rapidly changing local fire characteristics by dynamically adjusting the window size. Let the window width be w(t) and adjust and enhance it according to the local fluctuation intensity. The implementation formula is:

[0033]

[0034] In the formula, L j (t) represents the local fire characteristics of j features at time point t, T represents the time index variable, Represents the multi-scale feature value at time point t;

[0035] According to the enhanced local fire characteristics, the fire characteristics dimensionality reduction optimization is performed, and the implementation formula is:

[0036] q φ (z|x)=N(z;μ(x),σ(x)),

[0037] In the formula, q φ (z|x) represents the distribution of the latent variable z, φ represents the parameters of the encoder network, and x represents L j(t) Input local fire characteristics, μ(x) represents the mean of the potential z, σ(x) represents the standard deviation of the potential variable z, N represents the normal distribution function, and the output potential variable z is the fire characteristic F after dimensionality reduction rd (t);

[0038] According to the fire characteristics after dimensionality reduction optimization, the fusion is realized by the formula:

[0039] F fd (t) = GNN(F rd (t),A),

[0040] In the formula, F fd (t) represents the fused fire features, GNN represents the graph neural network, and feature fusion is performed. rd (t) represents the feature vector after dimensionality reduction, the value at time point t, and A represents the association between each sensor.

[0041] Wherein, said S4 performs real-time intelligent perception analysis based on the fused fire feature data to identify abnormal patterns; fd (t) Perform intelligent perception and set anomaly detection threshold based on historical data. The implementation formula is:

[0042]

[0043] In the formula, T(t) represents the anomaly detection threshold at time point t, α represents the learning efficiency, and F i (t) represents the fire characteristics after fusion F fd The i-th feature of (t), F i * (t) represents the mean value of fire characteristics under historical normal conditions, N represents the number of characteristics, and θ represents the baseline deviation;

[0044] Identify anomalies based on the anomaly detection threshold. The implementation formula is:

[0045]

[0046] In the formula, Er(t) represents the total anomaly score at time point t, M represents the number of different modalities, and β m represents the weight of the mth mode, S m (t) represents the local anomaly score of the mth modality, F m (t) and They represent the current fire characteristics of the mth mode and the feature mean under historical normal conditions, T(t) represents the anomaly detection threshold at time point t, and γ represents the attenuation coefficient. When the anomaly score exceeds the predetermined threshold, the abnormal mode type is identified.

[0047] Among them, the S5 obtains the results of intelligent perception analysis, and automatically adjusts the system configuration according to the current environment and sensor status, and feeds back to the intelligent perception; obtains the results of all anomaly detection, pattern recognition and classification in real-time intelligent perception analysis, and conducts a comprehensive evaluation, generates corresponding adjustment strategies based on the intelligent perception analysis results and the current environmental status, and prioritizes different adjustment strategies according to the severity and urgency of the anomaly, automatically adjusts the system configuration parameters according to the formulated strategy, and feeds back the adjusted system configuration information to the intelligent perception analysis in real time.

[0048] Among them, the S6, based on the abnormal detection situation identified by intelligent perception analysis, immediately triggers the generation of an automated report to notify the management personnel, and integrates the Internet of Things devices for early warning; obtains the latest abnormal detection results of the intelligent perception analysis, and if the abnormality score exceeds the preset alarm threshold, generates an automated report to notify the management personnel based on the abnormality type, and automatically controls the relevant Internet of Things devices for early warning based on the abnormality type, and monitors the status of the Internet of Things devices in real time.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. This invention dynamically adjusts the importance weights of environmental parameters to reflect the importance of each parameter in real time according to environmental changes, thereby improving the accuracy of environmental condition monitoring and modeling. It combines a multidimensional time series matrix and a spatial correlation matrix to form a spatiotemporal correlation model, comprehensively assesses the uncertainty of environmental conditions, and conducts dynamic environmental modeling and prediction. This fully considers temporal and spatial factors, enabling a more accurate description of the dynamic changes in the wind farm environment and the early prediction of potential risks.

[0051] 2. This invention extracts key fire features by adaptively adjusting weights based on the dynamic environmental model and changes in environmental parameter sensitivity to fire. This allows for more targeted capture of fire-related information, improving the accuracy and effectiveness of fire feature extraction. Dynamically adjusting the window size captures rapidly changing local fire features and performs enhanced processing, enabling detailed analysis of local variations in fire features. Combining multi-scale eigenvalues, this method provides a comprehensive and in-depth description of fire features. Dimensionality reduction and optimization of fire features reduces data redundancy and improves analysis efficiency. Graph neural networks are used for feature fusion, integrating information from various sensors to further enhance the characterization of fire features.

[0052] 3. This invention uses real-time intelligent perception analysis based on fused fire feature data, identifies abnormal patterns by setting anomaly detection thresholds and calculating anomaly scores, and can promptly detect fire hazards. Compared with traditional methods, this data-driven approach combined with real-time analysis can more quickly and accurately determine abnormal situations.

[0053] 4. The present invention acquires multi-source data in real time through multiple sensors deployed in the wind farm and performs preprocessing, which can effectively remove outliers and unify the data format, improve data quality, and provide a reliable basis for subsequent analysis. Compared with the use of a single sensor or unprocessed data, it can more comprehensively and accurately reflect the actual situation of the wind farm and reduce misjudgments and missed judgments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The operation process of the wind farm fire intelligent perception and early warning method based on multi-sensor fusion of the present invention Figure 1 ;

[0055] Figure 2 The operation process of the wind farm fire intelligent perception and early warning method based on multi-sensor fusion of the present invention Figure 2 ;

[0056] Figure 3 The operation process of the wind farm fire intelligent perception and early warning method based on multi-sensor fusion of the present invention Figure 3 ;

[0057] Figure 4 The operation process of the wind farm fire intelligent perception and early warning method based on multi-sensor fusion of the present invention Figure 4 . DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] Example

[0060] See also Figures 1-4 As shown, the present invention provides a technical solution: comprising the following steps:

[0061] S1. Using multiple sensors deployed in the wind farm, real-time data information collected by multiple sensors is acquired and data preprocessing is performed.

[0062] S2, real-time monitoring and modeling of environmental conditions around the wind farm based on the pre-processed data;

[0063] S3, extracting key fire features from the preprocessed data based on environmental information, and fusing fire feature data through dimensionality reduction optimization;

[0064] S4. Perform real-time intelligent perception analysis based on the fused fire characteristic data to identify abnormal patterns;

[0065] S5. Obtain intelligent perception analysis results, automatically adjust system configuration according to the current environment and sensor status, and feed back to intelligent perception;

[0066] S6. Based on the abnormal detection situations identified by intelligent perception analysis, it immediately triggers the generation of automated reports to notify management personnel and integrates IoT devices for early warning.

[0067] S7. Optimize and adjust system operating parameters and multi-sensor parameter configuration information based on feedback, performance monitoring results, and environmental information.

[0068] Among them, the S1 obtains environmental data information collected by multiple sensors in real time through multiple sensors deployed in the wind farm, and performs data preprocessing; obtains wind farm data information in real time according to the multiple sensors in the wind farm, transmits the multi-sensor data to the central data processing unit, and the central processing unit performs preliminary filtering on the environmental data obtained by the sensors to remove abnormal values, checks and processes the abnormal values, converts the format of the preprocessed data and stores it in the database.

[0069] Wherein, the S2 is to monitor and model the environmental conditions around the wind farm in real time based on the pre-processed data; suppose the environmental parameter set of the pre-processed data is E(t)={E1(t), E2(t), ..., E i (t)}, let the importance weight of each environmental parameter w i (t) Dynamic adjustment is performed, and dynamic weight update is implemented as follows:

[0070]

[0071] In the formula, w i (t+1) represents the updated weight, w i (t) represents the weight of the i-th environment parameter at time t, η represents the learning rate, Represents the sum of the absolute values ​​of the rates of change of all parameters, R i (t) represents the rate of change of the i-th parameter at time t, the rate of change R i (t) is R i (t) = E i (t)-E i (t-Δt), measures the rate of change of the i-th parameter over time.

[0072] Among them, the said S2, assuming that the multidimensional time series matrix is ​​T(t), combined with the spatial correlation matrix S to form a spatiotemporal correlation model:

[0073]

[0074] In the formula, T sal (t) represents the spatiotemporal correlation model, T(t) represents the multidimensional time series matrix, represents the tensor product operator, which is used to combine time and space information, and S represents the spatial correlation matrix;

[0075] The uncertainty of the current environmental state is comprehensively evaluated based on the spatiotemporal correlation model data. The implementation formula is:

[0076]

[0077] In the formula, H(E(t)) represents the entropy value of the environment state, Indicates the accumulation of all n environmental parameters, w i (t) represents the weight of the i-th environmental parameter at time t, p(E i (t)) represents the probability distribution of the i-th environment parameter, and the probability distribution p(E i (t))=KDE(T sal (t) i ), KDE represents kernel density estimation, T sal (t) i represents the subset of spatiotemporal data related to the i-th parameter, and log represents the natural logarithm function;

[0078] According to T sal (t) and H(E(t)) for dynamic environment modeling and prediction, assuming the current environment state is s t , the strategic action taken is a t , execute the selected action a t , observe the immediate reward r t and the next state s t+1 , the dynamic environment modeling prediction implementation formula is:

[0079]

[0080] In the formula, Q(s t ,a t ) represents the dynamic environment model, s t Indicates the state is s t =(T sal (t),H(E(t))),r t represents the immediate reward, α represents the learning rate, γ represents the discount factor, Indicates the maximum Q value in the next state.

[0081] Among them, the said S3 extracts the key features of fire from the pre-processed data according to the environmental information, and fuses the fire feature data through dimensionality reduction optimization; according to the constructed dynamic environment model Q(s t ,a t ), through adaptive weights, the selection is dynamically adjusted according to the changes in the sensitivity of each environmental parameter to fire, and the implementation formula is:

[0082]

[0083] In the formula, W(t) represents the environment parameter weight, η represents the learning rate, and Q(s t ,a t ) represents the output of the dynamic environment model, E(t) represents the environmental parameters at time point t, Represents the gradient of the impact of environmental parameters on the output of the environmental model;

[0084] Fire characteristics are extracted according to the weight of environmental parameters. Let the selected key parameter be F j (t), the fire feature extraction formula is:

[0085]

[0086] In the formula, represents the multi-scale eigenvalue at time point t, α k (t) represents the weighting coefficient of scale k, the value at time point t, W(F j ,s k ) represents the scale s k F j The result of wavelet transformation, K represents the number of scales.

[0087] Among them, S3 captures the rapidly changing local fire characteristics by dynamically adjusting the window size. Let the window width be w(t) and adjust and enhance it according to the local fluctuation intensity. The implementation formula is:

[0088]

[0089] In the formula, L j (t) represents the local fire characteristics of j features at time point t, T represents the time index variable, Represents the multi-scale feature value at time point t;

[0090] According to the enhanced local fire characteristics, the fire characteristics dimensionality reduction optimization is performed, and the implementation formula is:

[0091] q φ (z|x)=N(z;μ(x),σ(x)),

[0092] In the formula, qφ (z|x) represents the distribution of the latent variable z, φ represents the parameters of the encoder network, and x represents L j (t) Input local fire characteristics, μ(x) represents the mean of the potential z, σ(x) represents the standard deviation of the potential variable z, N represents the normal distribution function, and the output potential variable z is the fire characteristic F after dimensionality reduction rd (t);

[0093] According to the fire characteristics after dimensionality reduction optimization, the fusion is realized by the formula:

[0094] F fd (t) = GNN(F rd (t),A),

[0095] In the formula, F fd (t) represents the fused fire features, GNN represents the graph neural network, and feature fusion is performed. rd (t) represents the feature vector after dimensionality reduction, the value at time point t, and A represents the association between each sensor.

[0096] Wherein, said S4 performs real-time intelligent perception analysis based on the fused fire feature data to identify abnormal patterns; fd (t) Perform intelligent perception and set anomaly detection threshold based on historical data. The implementation formula is:

[0097]

[0098] In the formula, T(t) represents the anomaly detection threshold at time point t, α represents the learning efficiency, and F i (t) represents the fire characteristics after fusion F fd The i-th feature of (t), F i * (t) represents the mean value of fire characteristics under historical normal conditions, N represents the number of characteristics, and θ represents the baseline deviation;

[0099] Identify anomalies based on the anomaly detection threshold. The implementation formula is:

[0100]

[0101] In the formula, Er(t) represents the total anomaly score at time point t, M represents the number of different modalities, and β m represents the weight of the mth mode, S m (t) represents the local anomaly score of the mth modality, F m (t) and They represent the current fire characteristics of the mth mode and the feature mean under historical normal conditions, T(t) represents the anomaly detection threshold at time point t, and γ represents the attenuation coefficient. When the anomaly score exceeds the predetermined threshold, the abnormal mode type is identified.

[0102] Among them, the S5 obtains the results of intelligent perception analysis, and automatically adjusts the system configuration according to the current environment and sensor status, and feeds back to the intelligent perception; obtains the results of all anomaly detection, pattern recognition and classification in real-time intelligent perception analysis, and conducts a comprehensive evaluation, generates corresponding adjustment strategies based on the intelligent perception analysis results and the current environmental status, and prioritizes different adjustment strategies according to the severity and urgency of the anomaly, automatically adjusts the system configuration parameters according to the formulated strategy, and feeds back the adjusted system configuration information to the intelligent perception analysis in real time.

[0103] Among them, the S6, based on the abnormal detection situation identified by intelligent perception analysis, immediately triggers the generation of an automated report to notify the management personnel, and integrates the Internet of Things devices for early warning; obtains the latest abnormal detection results of the intelligent perception analysis, and if the abnormality score exceeds the preset alarm threshold, generates an automated report to notify the management personnel based on the abnormality type, and automatically controls the relevant Internet of Things devices for early warning based on the abnormality type, and monitors the status of the Internet of Things devices in real time.

[0104] Working principle: Multiple sensors deployed in wind farms acquire environmental data in real time. The multi-sensor data is transmitted to a central data processing unit for preliminary filtering, removing outliers, and format conversion before being stored in a database.

[0105] By monitoring the environmental conditions around the wind farm in real time based on the preprocessed data, a set of environmental parameters is set, and dynamically adjusted according to the importance weight of each environmental parameter, a spatiotemporal correlation model is formed in combination with the spatial correlation matrix, and the uncertainty of the current environmental state is comprehensively evaluated. Dynamic environmental modeling and prediction are realized based on the spatiotemporal correlation model data. According to the constructed dynamic environmental model, the environmental parameters are dynamically adjusted through adaptive weights to extract the key features of the fire. The rapidly changing local fire features are captured by dynamically adjusting the window size, and the fire feature dimensionality reduction optimization is performed. The feature fusion is performed through the graph neural network to obtain the reduced and fused fire features. The real-time intelligent perception analysis is performed based on the fused fire feature data, and the anomaly detection threshold is set for the fused fire features, and anomaly detection is performed. It can also identify and obtain the results of all anomaly detection, pattern recognition and classification in real-time intelligent perception analysis, generate corresponding adjustment strategies based on the results of intelligent perception analysis and the current environmental status, prioritize different adjustment strategies according to the severity and urgency of the anomaly, automatically adjust the system configuration parameters according to the adjustment strategy, and feed back the adjusted information to the intelligent perception module in real time, obtain the latest anomaly detection results of intelligent perception analysis, and determine the anomaly score. If the anomaly score exceeds the preset alarm threshold, it will generate an automated report to notify the management personnel based on the anomaly type, automatically control the relevant IoT devices for early warning, and monitor the status of IoT devices in real time. It will optimize and adjust the system operation parameters and multi-sensor parameter configuration information based on feedback, performance monitoring results and environmental information.

[0106] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

[0107] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. The intelligent perception and early warning method for wind farm fire based on multi-sensor fusion is characterized by: The following steps are involved: S1. Using multiple sensors deployed in the wind farm, real-time data information collected by multiple sensors is acquired and data preprocessing is performed. S2, real-time monitoring and modeling of environmental conditions around the wind farm based on the pre-processed data; S3, extracting key fire features from the preprocessed data based on environmental information, and fusing fire feature data through dimensionality reduction optimization; S4. Perform real-time intelligent perception analysis based on the fused fire characteristic data to identify abnormal patterns; S5. Obtain intelligent perception analysis results, automatically adjust system configuration according to the current environment and sensor status, and feed back to intelligent perception; S6. Based on the abnormal detection situations identified by intelligent perception analysis, it immediately triggers the generation of automated reports to notify management personnel and integrates IoT devices for early warning. S7. Optimize and adjust system operating parameters and multi-sensor parameter configuration information based on feedback, performance monitoring results, and environmental information.

2. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 1 is characterized by: Said S1, through multiple sensors deployed in the wind farm, obtains environmental data information collected by multiple sensors in real time and performs data preprocessing; obtains wind farm data information in real time according to the multiple sensors of the wind farm, transmits the multi-sensor data to the central data processing unit, the central processing unit performs preliminary filtering on the environmental data obtained by the sensors to remove abnormal values, checks and processes the abnormal values, converts the format of the preprocessed data and stores it in a database.

3. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 1 is characterized by: S2, real-time monitoring and modeling of environmental conditions around the wind farm based on the pre-processed data; Assume that the environmental parameter set of the preprocessed data is E(t)={E1(t),E2(t),...,E i (t)}, let the importance weight of each environmental parameter w i (t) Dynamic adjustment is performed, and dynamic weight update is implemented as follows: In the formula, w i (t+1) represents the updated weight, w i (t) represents the weight of the i-th environment parameter at time t, η represents the learning rate, Represents the sum of the absolute values ​​of the rates of change of all parameters, R i (t) represents the rate of change of the i-th parameter at time t, the rate of change R i (t) is R i (t) = E i (t)-E i (t-Δt), measures the rate of change of the i-th parameter over time.

4. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 1 is characterized by: S2, assuming that the multidimensional time series matrix is ​​T(t), combined with the spatial correlation matrix S to form a spatiotemporal correlation model: In the formula, T sal (t) represents the spatiotemporal correlation model, T(t) represents the multidimensional time series matrix, represents the tensor product operator, which is used to combine time and space information, and S represents the spatial correlation matrix; The uncertainty of the current environmental state is comprehensively evaluated based on the spatiotemporal correlation model data. The implementation formula is: In the formula, H(E(t)) represents the entropy value of the environment state, Indicates the accumulation of all n environmental parameters, w i (t) represents the weight of the i-th environmental parameter at time t, p(E i (t)) represents the probability distribution of the i-th environment parameter, and the probability distribution p(E i (t))=KDE(T sal (t) i ), KDE represents kernel density estimation, T sal (t) i represents the subset of spatiotemporal data related to the i-th parameter, and log represents the natural logarithm function; According to the spatiotemporal correlation model T sal (t) and the entropy value of the environmental state H(E(t)) are used to perform dynamic environmental modeling and prediction. Let the current environmental state be s t , the strategic action taken is a t , execute the selected action a t , observe the immediate reward r t and the next state s t+1 , the dynamic environment modeling prediction implementation formula is: In the formula, Q(s t ,a t ) represents the dynamic environment model, s t Indicates the state is s t =(T sal (t),H(E(t))),r t represents the immediate reward, α represents the learning rate, γ represents the discount factor, Indicates the maximum Q value in the next state.

5. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 1 is characterized by: S3 extracts the key fire features from the pre-processed data according to the environmental information, and fuses the fire feature data through dimensionality reduction optimization; according to the constructed dynamic environment model Q(s t ,a t ), through adaptive weights, the selection is dynamically adjusted according to the changes in the sensitivity of each environmental parameter to fire, and the implementation formula is: In the formula, W(t) represents the environment parameter weight, η represents the learning rate, and Q(s t ,a t ) represents the output of the dynamic environment model, E(t) represents the environmental parameters at time point t, Represents the gradient of the impact of environmental parameters on the output of the environmental model; Fire characteristics are extracted according to the weight of environmental parameters. Let the selected key parameter be F j (t), the fire feature extraction formula is: In the formula, represents the multi-scale eigenvalue at time point t, α k (t) represents the weighting coefficient of scale k, the value at time point t, W(F j ,s k ) represents the scale s k F j The result of wavelet transformation, K represents the number of scales.

6. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 5 is characterized by: The S3 captures the rapidly changing local fire characteristics by dynamically adjusting the window size. Let the window width be w(t) and adjust and enhance according to the local fluctuation intensity. The implementation formula is: In the formula, L j (t) represents the local fire characteristics of j features at time point t, T represents the time index variable, Represents the multi-scale feature value at time point t; According to the enhanced local fire characteristics, the fire characteristics dimensionality reduction optimization is performed, and the implementation formula is: q φ (z|x)=N(z;μ(x),σ(x)), In the formula, q φ (z|x) represents the distribution of the latent variable z, φ represents the parameters of the encoder network, and x represents L j (t) Input local fire characteristics, μ(x) represents the mean of the potential z, σ(x) represents the standard deviation of the potential variable z, N represents the normal distribution function, and the output potential variable z is the fire characteristic F after dimensionality reduction rd (t); According to the fire characteristics after dimensionality reduction optimization, the fusion is realized by the formula: F fd (t)=GNN(F rd (t),A), In the formula, F fd (t) represents the fused fire features, GNN represents the graph neural network, and feature fusion is performed. rd (t) represents the feature vector after dimensionality reduction, the value at time point t, and A represents the association between each sensor.

7. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 1 is characterized by: The S4 is to perform real-time intelligent perception analysis based on the fused fire feature data to identify abnormal patterns; fd (t) Perform intelligent perception and set anomaly detection thresholds based on historical data. The implementation formula is: In the formula, T(t) represents the anomaly detection threshold at time point t, α represents the learning efficiency, and F i (t) represents the fire characteristics after fusion F fd The i-th feature of (t), represents the mean value of fire characteristics under historical normal conditions, N represents the number of characteristics, and θ represents the baseline deviation; Identify anomalies based on the anomaly detection threshold. The implementation formula is: In the formula, Er(t) represents the total anomaly score at time point t, M represents the number of different modalities, and β m represents the weight of the mth mode, S m (t) represents the local anomaly score of the mth modality, F m (t) and They represent the current fire characteristics of the mth mode and the feature mean under historical normal conditions, T(t) represents the anomaly detection threshold at time point t, and γ represents the attenuation coefficient. When the anomaly score exceeds the predetermined threshold, the abnormal mode type is identified.

8. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 1 is characterized by: S5, obtaining intelligent perception analysis results, automatically adjusting system configuration according to the current environment and sensor status, and feeding back to the intelligent perception; Obtain all anomaly detection, pattern recognition, and classification results from real-time intelligent perception analysis, and conduct a comprehensive evaluation. Generate corresponding adjustment strategies based on the intelligent perception analysis results and the current environmental status, and prioritize different adjustment strategies based on the severity and urgency of the anomaly. Automatically adjust the system configuration parameters based on the formulated strategy, and feed the adjusted system configuration information back to the intelligent perception analysis in real time.

9. The wind farm fire intelligent perception and early warning method based on multi-sensor fusion according to claim 1 is characterized by: Said S6, based on the abnormal detection situation identified by intelligent perception analysis, immediately triggers the generation of an automated report to notify management personnel, and integrates IoT devices for early warning; Obtain the latest anomaly detection results of intelligent perception analysis. If the anomaly score exceeds the preset alarm threshold, an automated report is generated based on the anomaly type to notify management personnel. Based on the anomaly type, relevant IoT devices are automatically controlled to issue early warnings and the status of IoT devices is monitored in real time.

Images