Mine vehicle early fire early warning method and system based on multi-source information fusion

Abstract

The invention discloses a mining vehicle early fire early warning method and system based on multi-source information fusion, and belongs to the technical field of fire alarm, and the method comprises the steps: carrying out the multi-source monitoring of a mining vehicle, and obtaining the environment, working condition and visual monitoring data of the vehicle; fire sign sensitive feature detection is carried out on the vehicle environment monitoring data; fire sign sensitive feature detection is carried out on the vehicle operation condition data; fire sign sensitive feature detection is carried out on the vehicle visual monitoring data; performing single-factor fire deduction on the environment, working condition and visual fire sensitive feature vectors according to the vehicle fire historical event set; and performing multi-factor coupling fire behavior deduction according to the fire behavior deduction first map, and executing early fire behavior early warning according to the fire behavior deduction second map. The technical problems that in the prior art, mining vehicle fire early warning depends on single-dimensional monitoring data, information is one-sided, and early multi-factor collaborative risks are difficult to capture, so that fire missed judgment, misjudgment and untimely early warning are caused are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fire alarm technology, in particular to a mine vehicle early fire warning method and system based on multi-source information fusion. BACKGROUND

[0002] The mine vehicle is the core equipment for underground transportation in the mine, and the operation environment is closed and dusty, the equipment load is high, the fire is easy to spread rapidly to cause explosion and casualties, and early warning is a key link to ensure the safety production of the mine.

[0003] At present, the mine vehicle fire warning mainly relies on single-dimensional monitoring technology: only temperature / smoke sensors are used to monitor environmental parameters, which is difficult to capture local high temperature, burning marks and other visual features; only rely on working condition sensors to monitor current, oil pressure, which cannot associate the accelerating effect of high temperature environment on equipment aging; and lack of multi-dimensional data collaborative analysis, which is difficult to identify early coupling risks of environment, working condition and vision, resulting in high false alarm rate and often triggering alarm after the fire has spread, missing the early disposal opportunity. SUMMARY

[0004] The present application provides a mine vehicle early fire warning method and system based on multi-source information fusion, which aims to solve the technical problems that the existing mine vehicle fire warning relies on single-dimensional monitoring data, the information is one-sided, it is difficult to capture early multi-factor collaborative risks, resulting in fire misjudgment, false alarm and delayed warning.

[0005] In view of the above problems, the present application provides a mine vehicle early fire warning method and system based on multi-source information fusion.

[0006] The first aspect of the present application provides a mine vehicle early fire warning method based on multi-source information fusion, which comprises: multi-source monitoring of the mine vehicle to obtain vehicle environmental monitoring data, vehicle operating condition data and vehicle visual monitoring data; fire sign sensitive feature detection of the vehicle environmental monitoring data according to a vehicle fire history event set to obtain an environmental fire sensitive feature vector; fire sign sensitive feature detection of the vehicle operating condition data according to the vehicle fire history event set to obtain a working condition fire sensitive feature vector; fire sign sensitive feature detection of the vehicle visual monitoring data according to the vehicle fire history event set to obtain a visual fire sensitive feature vector; single-factor fire inference of the environmental fire sensitive feature vector, the working condition fire sensitive feature vector and the visual fire sensitive feature vector according to the vehicle fire history event set to obtain a fire inference first graph; multi-factor coupling fire inference according to the fire inference first graph to obtain a fire inference second graph, and early fire warning according to the fire inference second graph.

[0007] Further, the vehicle environment monitoring data is subjected to fire sign sensitive feature detection according to the vehicle fire history event set to obtain an environment fire sensitive feature vector, including: collecting sensor feature data corresponding to each environment monitoring parameter in the vehicle environment monitoring data to obtain a plurality of monitoring sensor feature sequences; performing anomaly detection according to the plurality of monitoring sensor feature sequences to obtain a plurality of sensor anomaly detection results; performing adaptive interference correction on the vehicle environment monitoring data according to the plurality of sensor anomaly detection results to obtain an environment monitoring feature sequence; performing fire indication degree evaluation on the environment monitoring feature sequence according to the vehicle fire history event set to obtain an environment monitoring fire indication degree sequence; and performing fire sign sensitive attention allocation on the environment monitoring feature sequence according to the environment monitoring fire indication degree sequence to generate the environment fire sensitive feature vector.

[0008] Further, the vehicle environment monitoring data is subjected to fire sign sensitive feature detection according to the vehicle fire history event set to obtain an environment fire sensitive feature vector, including: collecting sensor feature data corresponding to each environment monitoring parameter in the vehicle environment monitoring data to obtain a plurality of monitoring sensor feature sequences; performing anomaly detection according to the plurality of monitoring sensor feature sequences to obtain a plurality of sensor anomaly detection results; performing adaptive interference correction on the vehicle environment monitoring data according to the plurality of sensor anomaly detection results to obtain an environment monitoring feature sequence; performing fire indication degree evaluation on the environment monitoring feature sequence according to the vehicle fire history event set to obtain an environment monitoring fire indication degree sequence; and performing fire sign sensitive attention allocation on the environment monitoring feature sequence according to the environment monitoring fire indication degree sequence to generate the environment fire sensitive feature vector.

[0009] Further, according to the vehicle fire history event set, the environmental fire sensitive feature vector, the working condition fire sensitive feature vector and the visual fire sensitive feature vector are single-factor fire evolution, and a fire evolution first graph is obtained, including: according to the vehicle fire history event set, the environmental fire history event set, the working condition fire history event set and the visual fire history event set are obtained by single-factor classification; according to the environmental fire history event set, the environmental fire sensitive feature vector is subjected to environmental factor fire evolution, and a first fire evolution result is obtained; according to the working condition fire history event set, the working condition fire sensitive feature vector is subjected to working condition factor fire evolution, and a second fire evolution result is obtained; according to the visual fire history event set, the visual fire sensitive feature vector is subjected to visual factor fire evolution, and a third fire evolution result is obtained; the environmental fire sensitive feature vector, the working condition fire sensitive feature vector, the visual fire sensitive feature vector, the first fire evolution result, the second fire evolution result and the third fire evolution result are arranged, and the fire evolution first graph is generated.

[0010] Further, according to the environmental fire history event set, the environmental fire sensitive feature vector is subjected to environmental factor fire evolution, and a first fire evolution result is obtained, including: activating a fire path tracing architecture, the fire path tracing architecture including a top event node, an intermediate event node and a bottom event node, the top event node, the intermediate event node and the bottom event node being connected through a logical relationship node; according to the fire path tracing architecture, the environmental fire history event set is subjected to fire path tracing, and an environmental fire tracing path set is obtained; according to the environmental fire tracing path set, a disturbance injection is performed, and an environmental fire disturbance path set is obtained; according to the environmental fire tracing path set and the environmental fire disturbance path set, an adversarial training is performed, and an environmental factor fire evolution model is generated; the environmental fire sensitive feature vector is input into the environmental factor fire evolution model, and the first fire evolution result is obtained.

[0011] Further, according to the fire evolution first graph, a multi-factor coupled fire evolution is performed, and a fire evolution second graph is obtained, including: according to the fire evolution first graph, an environmental working condition coupled fire evolution is performed, and a fourth fire evolution result is obtained; according to the fire evolution first graph, an environmental visual coupled fire evolution is performed, and a fifth fire evolution result is obtained; according to the fire evolution first graph, a working condition visual coupled fire evolution is performed, and a sixth fire evolution result is obtained; according to the fire evolution first graph, a three-factor coupled fire evolution is performed, and a seventh fire evolution result is obtained; according to the fourth fire evolution result, the fifth fire evolution result, the sixth fire evolution result and the seventh fire evolution result, the fire evolution first graph is optimized, and the fire evolution second graph is generated.

[0012] Further, according to the fire evolution first graph, environment working condition coupling fire evolution is performed to obtain a fourth fire evolution result, including: environment working condition coupling factor event extraction is performed according to the vehicle fire history event set to obtain an environment working condition coupling fire event set; path tracing is performed on the environment working condition coupling fire event set according to a fire path tracing architecture to obtain an environment working condition coupling fire path set; an environment working condition coupling fire evolution model is trained according to the environment working condition coupling fire path set; and the fourth fire evolution result is output based on the environment fire sensitive feature vector and the working condition fire sensitive feature vector according to the environment working condition coupling fire evolution model.

[0013] Further, the vehicle visual monitoring data includes infrared thermal image data and visible light image data of the mining vehicle.

[0014] Further, early fire suppression management is performed on the mining vehicle according to the fire evolution second graph.

[0015] Another aspect of the present application provides a multi-source information fusion early fire warning system for a mining vehicle, which is used to execute the multi-source information fusion early fire warning method for a mining vehicle described in the present application. The system comprises: a multi-source monitoring obtaining module, which performs multi-source monitoring on a mining vehicle to obtain vehicle environment monitoring data, vehicle running working condition data and vehicle visual monitoring data; an environment fire sensitive feature vector obtaining module, which detects fire sign sensitive features from the vehicle environment monitoring data according to a vehicle fire history event set to obtain an environment fire sensitive feature vector; a working condition fire sensitive feature vector obtaining module, which detects fire sign sensitive features from the vehicle running working condition data according to the vehicle fire history event set to obtain a working condition fire sensitive feature vector; a visual fire sensitive feature vector obtaining module, which detects fire sign sensitive features from the vehicle visual monitoring data according to the vehicle fire history event set to obtain a visual fire sensitive feature vector; a fire evolution first graph obtaining module, which performs single-factor fire evolution on the environment fire sensitive feature vector, the working condition fire sensitive feature vector and the visual fire sensitive feature vector according to the vehicle fire history event set to obtain a fire evolution first graph; and an early fire warning execution module, which performs multi-factor coupling fire evolution according to the fire evolution first graph to obtain a fire evolution second graph, and executes early fire warning according to the fire evolution second graph.

[0016] One or more technical solutions provided in the present application have at least the following technical effects or advantages: The technical scheme provided by the application is as follows: the multi-source monitoring data of the mine vehicle environment, working condition and vision are integrated, the fire-sensitive feature vectors of each dimension are extracted, the first atlas is generated through single-factor fire deduction, the second atlas is generated through multi-factor coupling deduction, and finally the early fire warning and suppression management are executed based on the second atlas, so that the technical problems in the prior art, such as the dependence on single-dimensional data, one-sided information, the difficulty in capturing early multi-factor collaborative risks, the fire condition misjudgment and the delay of early warning, are solved, and the technical effects of comprehensively capturing early signs of fire, improving the accuracy and timeliness of early warning, effectively suppressing the spread of fire and enhancing the running safety of the mine vehicle are achieved.

[0017] The above description is only a summary of the technical scheme of the application. In order to more clearly understand the technical means of the application, the specific embodiments of the application can be implemented according to the content of the description, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of the early fire warning method of the mine vehicle based on multi-source information fusion is provided for the embodiments of the application.

[0019] Figure 2 A flowchart of obtaining the first atlas of fire deduction in the early fire warning method of the mine vehicle based on multi-source information fusion is provided for the embodiments of the application.

[0020] Figure 3 A structural diagram of the early fire warning system of the mine vehicle based on multi-source information fusion is provided for the embodiments of the application.

[0021] Explanation of reference signs: multi-source monitoring obtaining module 11, environment fire-sensitive feature vector obtaining module 12, working condition fire-sensitive feature vector obtaining module 13, vision fire-sensitive feature vector obtaining module 14, first atlas of fire deduction obtaining module 15, early fire warning execution module 16. DETAILED DESCRIPTION

[0022] The general idea of the technical scheme provided by the application is as follows: The embodiments of the application provide an early fire warning method and system of a mine vehicle based on multi-source information fusion. The multi-source monitoring data of the mine vehicle environment, working condition and vision are integrated, the fire-sensitive feature vectors of each dimension are extracted, the first atlas is generated through single-factor fire deduction, the second atlas is generated through multi-factor coupling deduction, and finally the early fire warning and suppression management are executed based on the second atlas, so that the technical effects of comprehensively capturing early signs of fire, improving the accuracy and timeliness of early warning, effectively suppressing the spread of fire and enhancing the running safety of the mine vehicle are achieved.

[0023] After introducing the basic principle of the application, the various non-limiting embodiments of the application will be specifically introduced in combination with the drawings of the specification.

[0024] Embodiment one As Figure 1 indicated, the embodiment of the present application provides a mine vehicle early fire warning method based on multi-source information fusion, which comprises: Step S100: Multi-source monitoring is performed on the mine vehicle to obtain vehicle environment monitoring data, vehicle operating condition data and vehicle visual monitoring data.

[0025] Specifically, the vehicle environment monitoring data refers to the environmental physical parameters of the space where the mine vehicle is located, reflecting the environmental state of the vehicle periphery and the interior. The vehicle operating condition data refers to the core state parameters of the vehicle itself when running, reflecting the working conditions of each system of the vehicle, including power, transmission, electrical and the like. The vehicle visual monitoring data refers to the key parts of the vehicle obtained by imaging equipment.

[0026] By deploying multiple types of monitoring equipment for all-around perception of the mine vehicle, specifically, temperature sensors, smoke sensors, flammable gas sensors are installed in key areas such as the vehicle engine compartment and the driver's cabin to collect environmental monitoring data in real time; the operating condition data is read through the OBD interface of the vehicle or the additional condition sensors; high-definition visible light cameras are installed in the cable slot, brake system and other parts prone to fire to periodically take pictures and obtain visual monitoring data. At the same time, the data collection module is used to aggregate and time-series store the sensor data, providing original data support for subsequent feature extraction.

[0027] This step realizes comprehensive coverage of the mine vehicle fire-related information through multi-source monitoring, including not only potential risk signals at the environmental level, but also abnormal states of equipment operation, and visual indicators, avoiding one-sidedness of a single data source, and laying a multi-dimensional and high-fidelity data foundation for subsequent fire-sensitive feature extraction and accurate early warning.

[0028] Step S200: Fire sign sensitive feature detection is performed on the vehicle environment monitoring data according to a vehicle fire history event set to obtain an environmental fire-sensitive feature vector.

[0029] Specifically, the vehicle fire history event set refers to a complete record set of past fire events of the mine vehicle, containing time series data of each event, such as changes in environmental parameters before and during the fire, fire causes such as circuit short circuit, oil leakage, fire development stages and final results, etc. The fire sign sensitive feature detection refers to identifying features highly related to fire and capable of early indicating fire risk from the vehicle environment monitoring data, such as temperature abnormal rising rate, process of sudden change of specific gas concentration, the core being to distinguish between "normal environmental fluctuations" and "fire precursor signals".

[0030] Specifically, the environmental data in the historical event set is subjected to feature mining, the importance weight of each environmental parameter such as temperature, smoke, and methane concentration before the occurrence of fire is calculated, and a fire-sensitive feature library is constructed. Then, the real-time vehicle environmental monitoring data is subjected to time sequence segmentation by using a sliding window method, and sensor noise is eliminated by using a Kalman filtering algorithm. Subsequently, whether the segmented data is abnormal is detected by using an isolation forest algorithm, and the abnormal data is compared with the fire-sensitive feature library by using cosine similarity. Finally, the features with similarity exceeding a threshold value are screened out, quantified into a vector according to the weight, and an environmental fire-sensitive feature vector is obtained.

[0031] This step guides the feature screening of real-time data by using the prior knowledge of the historical event set, can accurately capture the subtle signals in the early stage of fire, such as the abnormal rising rate of temperature instead of simply high temperature, can exclude environmental interference, and generates a sensitive feature vector focusing on the core parameters highly related to fire, thereby providing a high signal-to-noise ratio input for subsequent fire deduction, and significantly improving the pertinence and accuracy of early fire identification.

[0032] Step S300: fire sign-sensitive feature detection is performed on the vehicle operating condition data according to the vehicle fire history event set, and an operating condition fire-sensitive feature vector is obtained.

[0033] Specifically, the fire sign-sensitive feature detection identifies the abnormal state features with strong correlation with fire from the vehicle operating condition data, such as overload of the electrical system, sudden drop of the oil pressure, over-temperature of the mechanical components, etc., and the core is to distinguish “normal operating condition fluctuation” from “fire precursor anomaly” by using the historical fire law. The operating condition fire-sensitive feature vector refers to a multi-dimensional array formed after the operating condition-sensitive features are quantified, which collectively reflects the key information indicating fire risk in the vehicle operating state, such as the duration of over-limit motor current, the amplitude of battery voltage drop, the oil temperature rising rate of the transmission, the proportion of oil pressure deviating from the threshold, etc.

[0034] Firstly, the operating condition data in the historical event set is subjected to mining, the abnormal state features leading to fire due to operating conditions are extracted, the fire contribution weight of each operating condition parameter is calculated by using a random forest algorithm, and an operating condition fire-sensitive feature library is constructed. Then, the real-time collected operating condition data is subjected to filtering of high-frequency noise by using a wavelet transform algorithm, and time sequence segmentation is performed by using a sliding window. The abnormal fluctuation in the segment is identified by using a DBSCAN clustering algorithm. Subsequently, the abnormal fluctuation data is matched with the operating condition fire-sensitive feature library by using Jaccard similarity, the features with similarity exceeding a threshold value are screened out, quantified into a vector according to the weight, and an operating condition fire-sensitive feature vector is obtained.

[0035] This step refines the risk correlation law in the working condition data through historical fire events, can accurately capture the abnormal signals of the equipment before the fire, and at the same time, through noise filtering and cluster analysis, it can exclude normal working condition fluctuations. The generated sensitive feature vector focuses on the core risk point of the equipment operating state, provides high correlation and low redundancy input for subsequent fire deduction, effectively improves the early identification ability of the mine vehicle fire caused by equipment failure, and reduces the warning deviation caused by single parameter misjudgment.

[0036] Step S400: Fire sign sensitive feature detection is performed on the vehicle visual monitoring data according to the vehicle fire history event set, and a visual fire sensitive feature vector is obtained.

[0037] Specifically, the vehicle visual monitoring data refers to image information collected by imaging devices installed at key parts of the mine vehicle, such as the engine compartment, the cable slot, the braking system, etc., including visible light images and infrared thermal image data. The fire sign sensitive feature detection identifies the visual features directly related to the fire from the visual monitoring data, and the core is to distinguish between "normal visual features" and "fire precursor visual signals". The visual fire sensitive feature vector refers to a multidimensional array formed after quantizing the detected visual sensitive features, which concentrates the key information indicating fire risk in the visual data.

[0038] Specifically, the OpenCV library of Python is used to extract features from the visual data in the historical event set, and a CNN model is used to identify fire sign features, build a visual fire sensitive feature library, and label the correlation weight of each feature with fire; wherein, the CNN model is built by collecting mine vehicle fire history infrared thermal images and visible light images, labeling "circular high temperature area", "gray smoke texture" and "black burning spot" three types of features and corresponding parameters, and dividing the training and verification sets according to 7:3; a pre-trained ResNet-50 is used, the first 15 layers are frozen, the top layer is replaced with 3 types of feature output nodes, and a new parameter regression branch is added; cross entropy and MSE loss are used as the target, and Adam optimizer is used to train multiple rounds until the classification accuracy of the verification set meets the standard; the model convolution layer feature vector is extracted combined with the labeled parameters, thereby building the visual fire sensitive feature library. Then, the real-time collected visual monitoring data is preprocessed, noise is removed by Gaussian filtering, and the area with too high temperature is extracted from the infrared image by threshold segmentation method, and the object surface morphology change is identified from the visible light image by edge detection algorithm; then, the Faster R-CNN model is used to extract features from the preprocessed image, and real-time visual features are obtained, and the similarity comparison is performed between the features and the features in the visual fire sensitive feature library through the twin network; finally, the features with similarity exceeding the threshold are selected, quantified into vectors according to the correlation weight, and the visual fire sensitive feature vector is obtained.

[0039] This step realizes the accurate capture of early visual signals of fire through multi-modal analysis of visual data, provides intuitive and high-recognizability visual dimension input for subsequent multi-factor fire deduction, and significantly reduces the risk of early fire omission caused by the limitation of single data dimension.

[0040] Step S500: Single-factor fire deduction is performed on the environmental fire-sensitive feature vector, the working condition fire-sensitive feature vector and the visual fire-sensitive feature vector according to the vehicle fire history event set, and a first fire deduction graph is obtained.

[0041] Specifically, single-factor fire deduction refers to independent fire risk analysis of the environmental fire-sensitive feature vector, the working condition fire-sensitive feature vector and the visual fire-sensitive feature vector based on the vehicle fire history event set. The first fire deduction graph refers to the visualization and structured data set integrating the results of the three single-factor deductions, including the original values of the feature vectors, the corresponding single-factor fire probability and the deduction confidence.

[0042] Specifically, the vehicle fire history event set is classified by single factor, and is decomposed into an environmental fire history subset, a working condition fire history subset and a visual fire history subset according to the dominant factor causing the fire. Corresponding models are constructed, and corresponding real-time data is input to obtain the fire probability and confidence of each dimension. Finally, the original feature values, the corresponding single-factor fire probability and the deduction confidence of the three vectors are integrated into a structured graph using the Matplotlib library to generate the first fire deduction graph.

[0043] This step realizes the accurate decomposition of the fire risk of the three dimensions through single-factor independent deduction, eliminates the cross interference of different dimension features, and clearly locates the independent risk intensity of each dimension. The model is trained based on the classified historical subsets, so that the deduction result is more in line with the actual fire law of each dimension. The generated first fire deduction graph not only provides a single-dimensional risk benchmark for subsequent multi-factor coupling, but also quickly locates the high-risk dimension, improves the risk tracing ability of early fire warning, and reduces the deduction error caused by multi-factor mixing.

[0044] Step S600: Multi-factor coupled fire deduction is performed according to the first fire deduction graph, a second fire deduction graph is obtained, and early fire warning is performed according to the second fire deduction graph.

[0045] Specifically, multi-factor coupled fire deduction refers to the process of analyzing the synergistic effect between factors by integrating the single-factor deduction results of the environmental, working condition and visual dimensions, rather than simply superimposing them, and finally calculating the overall fire risk. The second fire deduction graph refers to a structured graph integrating the multi-factor coupling results, including the overall fire probability, the coupling weight of each factor and the risk level.

[0046] Specifically, the key data in the fire trend deduction first graph is extracted, an attention mechanism fusion model is constructed using the PyTorch library of Python, the probability value and the feature vector in three dimensions are input, the coupling weight between factors is calculated through the self-attention layer, and the coupling law in the historical data is introduced into the Bayesian network modeling; then, the single-factor probability and the coupling weight are weighted and fused to obtain the overall fire trend probability, and the fire trend deduction second graph is generated in combination with the preset threshold.

[0047] This step breaks through the limitation of single-factor deduction through multi-factor coupling analysis, captures the synergistic risk of "1+1>2", and improves the overall fire trend recognition accuracy; the risk level of the second graph and the coupling weight provide accurate basis for the early warning measures, and significantly improve the safety of mine transportation.

[0048] Further, according to the vehicle fire history event set, the fire sign sensitive feature vector of the vehicle environment monitoring data is detected, including: collecting the sensor feature data corresponding to each environment monitoring parameter in the vehicle environment monitoring data, obtaining a plurality of monitoring sensor feature sequences; performing anomaly detection according to the plurality of monitoring sensor feature sequences, obtaining a plurality of sensor anomaly detection results; performing adaptive interference correction on the vehicle environment monitoring data according to the plurality of sensor anomaly detection results, obtaining an environment monitoring feature sequence; performing fire indication degree evaluation on the environment monitoring feature sequence according to the vehicle fire history event set, obtaining an environment monitoring fire indication degree sequence; performing fire sign sensitive attention allocation on the environment monitoring feature sequence according to the environment monitoring fire indication degree sequence, and generating the environment fire sensitive feature vector.

[0049] Specifically, the fire indication degree evaluation refers to combining the vehicle fire history event set to calculate the correlation degree of each data point in the environment monitoring feature sequence with the fire trend. The environment monitoring fire indication degree sequence refers to a set formed by arranging the fire indication degree corresponding to each environment monitoring feature sequence in time. The fire sign sensitive attention allocation refers to assigning weights to the environment monitoring feature sequence according to the fire indication degree, and the high indication degree feature obtains higher attention weight and the low indication degree feature reduces the weight.

[0050] Specifically, using mine special sensors such as temperature sensors, smoke sensors, catalytic combustion methane sensors to collect environmental parameters of key areas of the vehicle, including the engine compartment, the driver's cabin, and the cargo compartment. The original data of individual parameters are sorted into monitoring sensor feature sequences by timestamp using the Pandas library of Python. Isolation forest algorithm is used to detect anomalies in each monitoring sensor feature sequence. For example, identify the abnormal point in the smoke concentration sequence that increases from 0.02 mg / m³ to 0.1 mg / m³. Output the sensor anomaly detection results containing the abnormal timestamp and deviation value. Based on the sensor anomaly detection results, use Kalman filter algorithm for adaptive interference correction. If the abnormal point is caused by temporary dust blocking the sensor, then correct 0.1 mg / m³ to 0.03 mg / m³ to get the noise-removed environmental monitoring feature sequence. Then, call the XGBoost algorithm to calculate the fire indication degree of each data point in the environmental monitoring feature sequence using the vehicle fire history event set as the training sample, forming the environmental monitoring fire indication degree sequence. Finally, assign weights to the environmental monitoring feature sequence through the Softmax attention mechanism, then extract the weighted values of each feature and integrate them into the environmental fire sensitive feature vector.

[0051] This step is to effectively eliminate sensor interference caused by dust and vibration in the mine environment through isolation forest anomaly detection and Kalman filter correction, ensuring the authenticity of environmental data. Second, based on the fire indication degree evaluation of the fire history event set, accurately identify features strongly associated with fire, and avoid irrelevant parameter interference analysis. Third, focus on high-risk features through sensitive attention allocation, and the generated environmental fire sensitive feature vector can condense core risk information, providing high signal-to-noise ratio input for subsequent single-factor fire deduction, significantly improving the identification accuracy of early fire signs, and reducing false positives and missed detections caused by data noise or irrelevant parameters.

[0052] Further, the fire indication degree of the environment monitoring feature sequence is evaluated according to the vehicle fire history event set, to obtain an environment monitoring fire indication degree sequence, including: identifying fire sign features according to the vehicle fire history event set to obtain a fire sign feature library; evaluating the confidence of each fire sign feature in the fire sign feature library according to the vehicle fire history event set to obtain a fire feature confidence sequence; based on the fire feature confidence sequence, cleaning the fire sign feature library according to a predetermined fire feature confidence, and establishing a confidence fire sign feature space; extracting an environment monitoring Rth feature from the environment monitoring feature sequence, where R is a positive integer; respectively evaluating the environment monitoring Rth feature according to each confidence fire sign feature in the confidence fire sign feature space to obtain an Rth twin comparison evaluation sequence; and performing maximum value optimization on the Rth twin comparison evaluation sequence to generate an Rth environment monitoring fire indication degree, and adding the Rth environment monitoring fire indication degree to the environment monitoring fire indication degree sequence.

[0053] Specifically, the fire sign feature library is a collection of identified fire sign features, each feature containing specific parameter thresholds and timing rules. The fire feature confidence sequence refers to a sequence formed by arranging the confidence of each fire sign feature in the feature library order. The predetermined fire feature confidence is a manually set screening threshold, used to retain high confidence features and exclude low correlation features. The confidence fire sign feature space is a feature collection composed of high confidence fire sign features after being cleaned by the predetermined confidence. The environment monitoring Rth feature refers to the Rth specific feature extracted from the environment monitoring feature sequence, where R is a positive integer, including a single parameter timing rule or a parameter combination feature.

[0054] Specifically, the vehicle fire history event set is mined for features using the Python Scikit-learn tool, and fire sign features are identified by association rule algorithms such as Apriori, which are organized into a fire sign feature library. The confidence of each feature in the library is calculated using a logistic regression algorithm: the ratio of the number of times the feature actually occurs after the feature appears to the total number of times the feature appears is calculated; then, a predetermined fire feature confidence is set to clean the feature library and establish a confidence fire sign feature space containing temperature, smoke and other features; then, the Rth feature is extracted from the environment monitoring feature sequence, and the feature is compared with the features in the confidence space using a twin neural network built with PyTorch to calculate the Rth twin comparison evaluation sequence; finally, the argmax function is used to perform maximum value optimization on the sequence to obtain the Rth environment monitoring fire indication degree, which is added to the environment monitoring fire indication degree sequence.

[0055] This step improves the reliability of fire indication degree through three layers of optimization: first, feature recognition and confidence evaluation based on historical data to ensure that the analysis object is a feature strongly associated with fire; second, construction of a confidence feature space to focus on high-reliability features and improve matching efficiency; third, twin comparison captures subtle temporal correlations between features through deep learning, and the final generated indication degree sequence can accurately quantify the fire risk of real-time environmental features, improving the accuracy of early fire sign detection, providing high-value input for subsequent sensitive feature extraction, and effectively reducing the deviation of early warning caused by feature misjudgment.

[0056] Further, as shown in Figure 2 According to the vehicle fire history event set, single-factor fire deduction is performed on the environmental fire sensitive feature vector, the working condition fire sensitive feature vector, and the visual fire sensitive feature vector to obtain a fire deduction first graph, including: performing single-factor classification on the vehicle fire history event set to obtain an environmental fire history event set, a working condition fire history event set, and a visual fire history event set; performing environmental factor fire deduction on the environmental fire sensitive feature vector based on the environmental fire history event set to obtain a first fire deduction result; performing working condition factor fire deduction on the working condition fire sensitive feature vector based on the working condition fire history event set to obtain a second fire deduction result; performing visual factor fire deduction on the visual fire sensitive feature vector based on the visual fire history event set to obtain a third fire deduction result; and arranging the environmental fire sensitive feature vector, the working condition fire sensitive feature vector, the visual fire sensitive feature vector, the first fire deduction result, the second fire deduction result, and the third fire deduction result to generate the fire deduction first graph.

[0057] Specifically, the vehicle fire history event set is classified by single factor using the Pandas library of Python, and the environmental fire history event set, the working condition fire history event set, and the visual fire history event set are split by filtering the parameter dimension that is only abnormal before the occurrence of fire; wherein, for the environmental fire sensitive feature vector, an environmental factor fire deduction model is constructed based on the environmental fire history subset as a training sample, the real-time environmental vector is input into the model to obtain the environmental dimension fire probability and confidence; for the working condition fire sensitive feature vector, an XGBoost model is trained based on the working condition fire history subset to obtain a working condition factor fire deduction model, and after the real-time working condition vector is input, the working condition dimension fire probability and confidence are obtained; for the visual fire sensitive feature vector, a CNN-LSTM hybrid model is trained based on the visual fire history subset to obtain a visual factor fire deduction model, and after the real-time visual vector is input, the visual dimension fire probability and confidence are obtained; finally, the original feature values, the corresponding single-factor fire probability, and the deduction confidence of the three vectors are integrated into a structured graph using the Matplotlib library to generate the fire deduction first graph.

[0058] This step excludes the cross interference of different dimensional characteristics by single-factor independent deduction, makes each dimensional fire risk more accurate, and at the same time, the classified historical subset makes the deduction model more in line with the fire law of a single dimension, thereby improving the deduction confidence. The generated first atlas clearly presents the risk difference of each dimension, provides a clear risk benchmark for subsequent multi-factor coupling deduction, and can quickly locate the high-risk dimension, thereby effectively reducing the deduction error caused by multi-factor confusion.

[0059] Further, the environment fire sensitive feature vector is subjected to environment factor fire deduction according to the environment fire historical event set, to obtain a first fire deduction result, including: activating a fire path tracing architecture, the fire path tracing architecture including a top event node, an intermediate event node and a bottom event node, the top event node, the intermediate event node and the bottom event node being connected through logical relationship nodes; tracing a fire path of the environment fire historical event set according to the fire path tracing architecture, to obtain an environment fire tracing path set; injecting perturbation according to the environment fire tracing path set, to obtain an environment fire perturbation path set; performing adversarial training according to the environment fire tracing path set and the environment fire perturbation path set, to generate an environment factor fire deduction model; inputting the environment fire sensitive feature vector into the environment factor fire deduction model, to obtain the first fire deduction result.

[0060] Specifically, the fire path tracing architecture is a hierarchical event tree model simulating the logic of fire development, including a top event node, an intermediate event node and a bottom event node, and the nodes are connected through logical relationship nodes such as AND and OR. The environment fire tracing path set is a specific fire development path set extracted from the environment fire historical event set based on the fire path tracing architecture. The perturbation injection refers to artificially adding interference factors in the environment fire tracing path, such as sensor measurement error, environmental sudden fluctuation, etc., to simulate the uncertainty in the actual scene. The environment fire perturbation path set refers to the fire path set containing interference factors generated after perturbation injection. The adversarial training refers to taking the environment fire tracing path set and the environment fire perturbation path set as positive and negative samples, training the model to identify the difference between the real fire path and the interference path, and improving the anti-interference ability of the model. The environment factor fire deduction model refers to a model that can predict the fire occurrence probability and confidence based on the environment fire sensitive feature vector after adversarial training.

[0061] Specifically, the fire path tracing architecture is constructed using the NetworkX library of Python, the top event node, the intermediate event node, and the bottom event node are defined, and they are connected through the logical relationship node; based on the architecture, the Pandas library is used to trace the path of the environmental fire history event set, extract the bottom event→intermediate event→top event chain of each record, form the environmental fire tracing path set, for example: methane leakage→concentration 1.8% LEL→temperature 15℃ / min→fire; then, the disturbance is injected to generate the environmental fire disturbance path set; then, the bidirectional LSTM model is built, the tracing path set is used as the positive sample, and the disturbance path set is used as the negative sample for adversarial training, and the iteration training is performed until the model has qualified identification accuracy for the real path and qualified anti-interference rate for the disturbance path, and the environmental factor fire deduction model is generated; finally, the environmental fire sensitive feature vector is input into the model to obtain the first fire deduction result. The bidirectional LSTM model includes that the input layer receives the environmental fire sensitive feature vector, the input size is set to 3; the middle layer is bidirectional LSTM, the forward and backward LSTM capture the forward and reverse dependencies of the time sequence respectively; the dropout layer is added to prevent overfitting; the full connection layer is connected; the output layer is activated by sigmoid, and the fire probability is output. The model uses cross-entropy as the loss function, and the Adam optimizer is trained to adapt to the environmental fire time sequence feature deduction.

[0062] Similarly, the working condition fire path tracing architecture is activated; based on the architecture, the working condition fire history event set is traced to obtain the working condition fire tracing path set; the parameter fluctuation disturbance is injected to generate the disturbance path set; the working condition factor deduction model is trained by the two; the working condition fire sensitive feature vector is input to obtain the second fire deduction result. In addition, the visual fire path tracing architecture is activated; based on the architecture, the visual fire history event set is traced to obtain the visual fire tracing path set; the image noise disturbance is injected to generate the disturbance path set; the visual factor deduction model is trained by the two; the visual fire sensitive feature vector is input to obtain the third fire deduction result.

[0063] This step restores the logical chain of fire caused by environmental and working condition and visual factors through the fire path tracing architecture, making the deduction interpretable; the disturbance injection and adversarial training enable the model to resist common sensor errors and parameter fluctuations in the mining environment, improving the anti-interference ability; the finally generated fire deduction result not only outputs the fire probability, but also traces the risk source, providing an environmental dimension basis with accuracy and interpretability for subsequent early warning, reducing the false alarm rate of single environmental factor fire.

[0064] Further, according to the fire evolution first graph, multi-factor coupled fire evolution is carried out to obtain a fire evolution second graph, including: according to the fire evolution first graph, environment working condition coupled fire evolution is carried out to obtain a fourth fire evolution result; according to the fire evolution first graph, environment vision coupled fire evolution is carried out to obtain a fifth fire evolution result; according to the fire evolution first graph, working condition vision coupled fire evolution is carried out to obtain a sixth fire evolution result; according to the fire evolution first graph, three-factor coupled fire evolution is carried out to obtain a seventh fire evolution result; and according to the fourth fire evolution result, the fifth fire evolution result, the sixth fire evolution result and the seventh fire evolution result, the fire evolution first graph is optimized to generate the fire evolution second graph.

[0065] Specifically, the environment probability, the working condition probability and the vision probability are extracted from the fire evolution first graph; a coupled model based on an attention mechanism is built to carry out two-by-two coupled evolution; when the environment and the working condition are coupled, the model calculates the synergistic correlation degree of the environment factor and the working condition factor “decrease of motor insulation performance under high temperature environment” through a self-attention layer, for example: decrease of motor insulation performance under high temperature environment. The first graph data is combined to output a fourth fire evolution result; when the environment and the vision are coupled, the correlation degree of the environment factor and the vision factor is calculated, for example: smoke diffusion range and high temperature zone spatial overlap degree, to output a fifth fire evolution result; when the working condition and the vision are coupled, the correlation degree of the working condition factor and the vision factor is calculated, for example: current abnormal duration and burning trace expansion speed, to output a sixth fire evolution result; then, a Bayesian network is used to model the three-factor interaction relationship to output a seventh fire evolution result; finally, a weighted fusion algorithm is used to optimize the single-factor risk assessment of the first graph, integrate the single-factor risk and the four types of coupled risks, and generate a fire evolution second graph containing “single factor, two-by-two coupled, three-factor risk and coupled influence degree”.

[0066] This step captures the synergistic risks ignored by single-factor evolution through multi-dimensional coupled analysis, making the fire risk assessment more comprehensive; the four types of coupled results and weights provide fine-grained basis for risk tracing; the optimized second graph improves the overall fire identification accuracy, and the coupled logic explanation reduces the uncertainty of early warning, providing more reliable decision support for subsequent hierarchical early warning.

[0067] Further, according to the first fire evolution graph, environment working condition coupling fire evolution is carried out to obtain a fourth fire evolution result, including: environment working condition coupling factor event extraction is carried out according to the vehicle fire history event set to obtain an environment working condition coupling fire event set; path tracing is carried out on the environment working condition coupling fire event set according to a fire path tracing architecture to obtain an environment working condition coupling fire path set; an environment working condition coupling fire evolution model is trained according to the environment working condition coupling fire path set; and the fourth fire evolution result is output based on the environment fire sensitive feature vector and the working condition fire sensitive feature vector according to the environment working condition coupling fire evolution model.

[0068] Specifically, the environment working condition coupling factor event extraction refers to screening and extracting fire events involving both environment parameter abnormalities and working condition parameter abnormalities from the vehicle fire history event set, focusing on cases where both factors jointly cause a fire. Similarly, it also includes fire events involving both environment parameter abnormalities and visual parameter abnormalities, and working condition parameter abnormalities and visual parameter abnormalities. The environment working condition coupling fire evolution model refers to a model that can learn the synergy of environment and working condition features, output the fire risk after coupling, and output the coupling fire probability and synergy weight, using the coupling path set as training data. Similarly, it also includes an environment visual coupling fire evolution model and a working condition visual coupling fire evolution model.

[0069] From the vehicle fire history event set, the Pandas library of Python is used to screen out events containing both environment and working condition abnormalities, extract environment features, working condition features and fire results, and form an environment working condition coupling fire event set; based on the previously constructed fire path tracing architecture, the NetworkX tool is used to trace the paths of the coupling event set, sort out the collaborative development chain of environment and working condition factors in each event, and generate an environment working condition coupling fire path set; after standardizing the feature data in the path set, a bidirectional LSTM model with attention mechanism is built with PyTorch, the input layer receives environment and working condition feature vectors, the attention layer calculates the synergy weight of the two, the LSTM layer captures the time sequence correlation, and the output layer outputs the coupling fire probability; the coupling path set is used as a training sample, and the model is trained until the accuracy of the verification set is qualified, to obtain an environment working condition coupling fire evolution model; finally, the real-time environment fire sensitive feature vector is input into the model, and the fourth fire evolution result is output.

[0070] From the fire history event set, use Pandas to filter the events with simultaneous abnormal environment and visual parameters, extract the environment features, visual features, and fire results to form the environment-visual coupling event set; based on the fire path tracing architecture, use NetworkX to generate the coupling path set; use PyTorch to build a CNN-LSTM fusion model, CNN extracts visual image features, LSTM processes environment time series data, and the attention layer calculates the spatial correlation weight between the two; with cross-entropy as the loss, the Adam optimizer is trained until the validation set accuracy is qualified, and the model construction is completed. Also, from the historical event set, use Pandas to extract fire events with abnormal working conditions and visual coordination, label the working condition features, visual features, and results to form the working condition-visual coupling event set; based on the path tracing architecture, use NetworkX to generate the coupling path set; use PyTorch to build a bidirectional LSTM attention model, input working condition time series and visual features, and the attention layer quantifies the time correlation between the two; with MSE and cross-entropy as the loss, the Adam optimizer is trained until the validation set accuracy is qualified, and the model construction is completed.

[0071] This step accurately captures the synergistic risks of the two by specifically extracting environment and working condition, environment and visual, and working condition and visual coupling events; the introduction of the attention mechanism enables the model to quantify the synergistic influence weight of the two, improving the result interpretability; the trained coupling model improves the recognition accuracy of the synergistic risks of environment and working condition, environment and visual, and working condition and visual, effectively addressing the underestimation of synergistic risks by single-factor analysis, significantly improving the comprehensiveness and accuracy of early fire warning.

[0072] Further, the vehicle visual monitoring data includes infrared thermal image data and visible light image data of the mining vehicle.

[0073] Specifically, the vehicle visual monitoring data includes infrared thermal image data and visible light image data, which are used to capture fire signs from different dimensions. Among them, the infrared thermal image data is the image data collected by the infrared thermal imager, which represents the temperature distribution of the object surface in grayscale or pseudo-color, and can intuitively present the temperature difference of the equipment, such as red for high-temperature area and blue for low-temperature area. Visible light image data is image data collected by a visible light camera, which reflects the visual features of objects such as color, shape, and texture, consistent with human vision.

[0074] Specifically, an infrared thermal imager and a visible light camera are deployed in key areas of the mine vehicle, such as the engine compartment, cable slot, braking system, etc., to respectively collect infrared thermal image data and visible light image data; the original data is preprocessed using the OpenCV library of Python, the infrared thermal image data is corrected for non-uniformity, sensor noise and temperature calibration are eliminated, the gray value is converted to actual temperature, and the visible light image data is dewarping processed and distortion corrected; then, a double-branch CNN model is built with PyTorch, the infrared branch extracts temperature gradient, area and other features of high temperature area through ResNet, and the visible light branch identifies smoke texture, burning marks and other morphological features through YOLOv5; finally, the features of the two branches are fused for visual fire sensitive feature detection.

[0075] The infrared thermal image data can identify abnormal high temperature of equipment in the early stage of fire, and make up for the blind area of local temperature sensing of environmental sensors; the visible light image data can capture morphological features such as smoke and burning marks, and complement the temperature features. The combination of the two makes the visual monitoring cover the whole process of "temperature anomaly and morphological change", compared with single visual data, the coverage rate of fire sign recognition is improved, and the missed judgment caused by the limitation of single data dimension is effectively reduced, such as high temperature hidden danger without smoke which is easily missed by visible light alone.

[0076] Further, the mine vehicle is subjected to early fire suppression management according to the second graph derived from the fire.

[0077] Specifically, early fire suppression management refers to taking targeted measures in the early stage of fire to prevent the development and expansion of fire through active intervention based on the risk level and coupling features of the second graph derived from the fire. The core is "early detection and early disposal". The suppression strategy library refers to a set of grading response schemes preset according to historical fire disposal experience and second graph features, for example, high risk corresponds to "automatic start of fire extinguishing device + power cut", medium risk corresponds to "local cooling + alarm prompt".

[0078] Specifically, the core information of the fire inference second graph is read in real time through the CAN bus, including the overall risk level, the dominant coupling factor and the key risk point; the decision tree algorithm based on Python analyzes the graph features and matches the corresponding scheme in the suppression strategy library. If it is high risk and the environment working condition coupling is dominant, the strategy of "engine compartment automatic fire extinguishing device start + motor load forced reduction + cut off unnecessary circuit" is called. If it is medium risk and the visual high temperature area is dominant, the "cooling system directional spraying + cockpit sound and light alarm" is executed. At the same time, the vehicle actuators such as fire extinguishing device electromagnetic valve, motor controller and cooling water pump are linked through PLC, and the edge computing module is used to collect the environment, working condition and visual data after suppression in real time, and feedback to the second graph for dynamic adjustment. For example, when the second graph shows that the three-factor coupling is high risk and focuses on the braking system, the system automatically starts the brake disc spray cooling and cuts off the brake motor power supply, and confirms the cooling effect through the camera.

[0079] This step directly links risk assessment and active suppression, avoiding the lag of only warning without disposal; the grading strategy based on the second graph and the coupling characteristics make the suppression measures more accurate; the real-time feedback adjustment mechanism ensures that the suppression effect can be verified, avoiding over-treatment. In practical application, it can shorten the response time of early fire disposal, reduce the spread rate of fire, and significantly reduce the damage and operation interruption of mine vehicles caused by fire.

[0080] In summary, the mine vehicle early fire warning method provided by the embodiment of the application has the following technical effects: 1. Through multi-source information fusion, multi-dimensional data such as environment, working condition and vision are integrated, breaking the limitation of single monitoring dimension. From feature extraction to multi-factor coupling inference, a complete early warning link is established to fully capture early signs of fire and avoid false positives or false negatives caused by incomplete information, providing comprehensive and three-dimensional data support for early fire identification of mine vehicles, and improving the integrity and reliability of the early warning system.

[0081] 2. Through single-factor classification and independent inference, historical events are divided into environment, working condition and vision subsets, excluding cross interference of different dimension features. Each dimension model focuses on learning the correlation between itself and fire, making single-factor risk assessment more accurate and clearly presenting independent risk intensity of each dimension, providing a clear benchmark for subsequent multi-factor coupling analysis, and enhancing risk tracing ability and interpretability of inference results.

[0082] 3. Through multi-factor coupling inference, the synergistic risk between environment, working condition and vision is captured, making up for the neglect of "1+1>2" additive effect in single-factor analysis. The second graph generated by integrating two-factor coupling and three-factor coupling results is optimized to fully reflect the overall risk situation and coupling weight, providing a more practical basis for early warning decision-making and improving the comprehensiveness and accuracy of early fire assessment.

[0083] Embodiment Two Based on the same inventive concept as the early fire warning method for mining vehicles in the foregoing embodiment, as shown in Figure 3 The embodiment of the present application provides a multi-source information fusion early fire warning system for mining vehicles, which comprises: A multi-source monitoring obtaining module 11 is configured to perform multi-source monitoring on the mining vehicle to obtain vehicle environment monitoring data, vehicle operating condition data and vehicle visual monitoring data; an environment fire sensitive feature vector obtaining module 12 is configured to detect fire sign sensitive features from the vehicle environment monitoring data according to a vehicle fire history event set to obtain an environment fire sensitive feature vector; an operating condition fire sensitive feature vector obtaining module 13 is configured to detect fire sign sensitive features from the vehicle operating condition data according to the vehicle fire history event set to obtain an operating condition fire sensitive feature vector; a visual fire sensitive feature vector obtaining module 14 is configured to detect fire sign sensitive features from the vehicle visual monitoring data according to the vehicle fire history event set to obtain a visual fire sensitive feature vector; a fire deduction first graph obtaining module 15 is configured to perform single-factor fire deduction on the environment fire sensitive feature vector, the operating condition fire sensitive feature vector and the visual fire sensitive feature vector according to the vehicle fire history event set to obtain a fire deduction first graph; and an early fire warning executing module 16 is configured to perform multi-factor coupled fire deduction according to the fire deduction first graph to obtain a fire deduction second graph, and perform early fire warning according to the fire deduction second graph.

[0084] Further, the environment fire sensitive feature vector obtaining module 12 is further configured to perform the following steps: collecting sensor feature data corresponding to each environment monitoring parameter in the vehicle environment monitoring data to obtain a plurality of monitoring sensor feature sequences; performing abnormality detection according to the plurality of monitoring sensor feature sequences to obtain a plurality of sensor abnormality detection results; performing adaptive interference correction on the vehicle environment monitoring data according to the plurality of sensor abnormality detection results to obtain an environment monitoring feature sequence; performing fire indication degree evaluation on the environment monitoring feature sequence according to the vehicle fire history event set to obtain an environment monitoring fire indication degree sequence; and performing fire sign sensitive attention allocation on the environment monitoring feature sequence according to the environment monitoring fire indication degree sequence to generate the environment fire sensitive feature vector.

[0085] Further, the environment fire sensitive feature vector obtaining module 12 is further configured to perform the following steps: identifying fire sign features according to the vehicle fire history event set, and obtaining a fire sign feature library; evaluating the confidence of each fire sign feature in the fire sign feature library according to the vehicle fire history event set, and obtaining a fire feature confidence sequence; cleaning the fire sign feature library according to a predetermined fire feature confidence based on the fire feature confidence sequence, and establishing a confidence fire sign feature space; extracting an environment monitoring Rth feature according to the environment monitoring feature sequence, where R is a positive integer; performing twin comparison evaluation of the environment monitoring Rth feature according to each confidence fire sign feature in the confidence fire sign feature space, and obtaining an Rth twin comparison evaluation sequence; performing maximum value optimization on the Rth twin comparison evaluation sequence, generating an Rth environment monitoring fire indication degree, and adding the Rth environment monitoring fire indication degree to the environment monitoring fire indication degree sequence.

[0086] Further, the fire evolution first graph obtaining module 15 is further configured to perform the following steps: performing single-factor classification according to the vehicle fire history event set, and obtaining an environment fire history event set, a working condition fire history event set, and a visual fire history event set; performing environment factor fire evolution on the environment fire sensitive feature vector according to the environment fire history event set, and obtaining a first fire evolution result; performing working condition factor fire evolution on the working condition fire sensitive feature vector according to the working condition fire history event set, and obtaining a second fire evolution result; performing visual factor fire evolution on the visual fire sensitive feature vector according to the visual fire history event set, and obtaining a third fire evolution result; collating the environment fire sensitive feature vector, the working condition fire sensitive feature vector, the visual fire sensitive feature vector, the first fire evolution result, the second fire evolution result, and the third fire evolution result, and generating the fire evolution first graph.

[0087] Further, the fire evolution first graph obtaining module 15 is further configured to perform the following steps: activating a fire path tracing architecture, the fire path tracing architecture including a top event node, an intermediate event node, and a bottom event node, the top event node, the intermediate event node, and the bottom event node being connected through logical relationship nodes; performing fire path tracing on the environment fire history event set according to the fire path tracing architecture, and obtaining an environment fire tracing path set; performing disturbance injection according to the environment fire tracing path set, and obtaining an environment fire disturbance path set; performing adversarial training according to the environment fire tracing path set and the environment fire disturbance path set, and generating an environment factor fire evolution model; inputting the environment fire sensitive feature vector into the environment factor fire evolution model, and obtaining the first fire evolution result.

[0088] Further, the early fire warning execution module 16 is further configured to perform the following steps: performing environment and working condition coupling fire deduction based on the fire deduction first graph to obtain a fourth fire deduction result; performing environment and visual coupling fire deduction based on the fire deduction first graph to obtain a fifth fire deduction result; performing working condition and visual coupling fire deduction based on the fire deduction first graph to obtain a sixth fire deduction result; performing three-factor coupling fire deduction based on the fire deduction first graph to obtain a seventh fire deduction result; and optimizing the fire deduction first graph based on the fourth fire deduction result, the fifth fire deduction result, the sixth fire deduction result and the seventh fire deduction result to generate the fire deduction second graph.

[0089] Further, the early fire warning execution module 16 is further configured to perform the following steps: performing environment and working condition coupling factor event extraction based on the vehicle fire event set to obtain an environment and working condition coupling fire event set; performing path tracing on the environment and working condition coupling fire event set based on a fire path tracing architecture to obtain an environment and working condition coupling fire path set; training an environment and working condition coupling fire deduction model based on the environment and working condition coupling fire path set; and outputting the fourth fire deduction result based on the environment and working condition coupling fire deduction model based on the environment fire sensitive feature vector and the working condition fire sensitive feature vector.

[0090] Further, the system is further configured to perform the following steps: the vehicle visual monitoring data includes infrared thermal image data and visible light image data of the mining vehicle.

[0091] Further, the system is further configured to perform the following steps: performing early fire suppression management on the mining vehicle based on the fire deduction second graph.

[0092] Any step of the above method can be stored as computer instructions or programs in an unlimited computer memory and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, and no redundant limitation is made herein.

[0093] Further, the above first or second possible order relationship may not represent a specific concept, and / or may refer to the selection of multiple elements individually or collectively. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method for early fire warning of mining vehicles based on multi-source information fusion, characterized in that, include: Multi-source monitoring of mining vehicles was conducted to obtain vehicle environmental monitoring data, vehicle operating condition data, and vehicle visual monitoring data. Based on the historical event set of vehicle fires, fire-sensitive feature detection is performed on the vehicle environmental monitoring data to obtain an environmental fire-sensitive feature vector. Based on the vehicle fire history event set, fire sign sensitive feature detection is performed on the vehicle operating condition data to obtain the operating condition fire sensitive feature vector. Based on the vehicle fire history event set, fire sign sensitive feature detection is performed on the vehicle visual monitoring data to obtain a visual fire sensitive feature vector. Based on the vehicle fire history event set, single-factor fire prediction is performed on the environmental fire sensitive feature vector, the working condition fire sensitive feature vector and the visual fire sensitive feature vector to obtain the first fire prediction map. Based on the first fire situation simulation map, a multi-factor coupled fire situation simulation is performed to obtain a second fire situation simulation map, and an early fire warning is executed based on the second fire situation simulation map.

2. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 1, characterized in that, Based on the vehicle fire history event set, fire-sensitive feature detection is performed on the vehicle environmental monitoring data to obtain an environmental fire-sensitive feature vector, including: Collect sensor feature data corresponding to each environmental monitoring parameter in the vehicle environmental monitoring data to obtain multiple monitoring sensor feature sequences; Anomaly detection is performed based on the multiple monitoring sensor feature sequences to obtain multiple sensor anomaly detection results; Based on the multiple sensor anomaly detection results, adaptive interference correction is performed on the vehicle environmental monitoring data to obtain an environmental monitoring feature sequence. The environmental monitoring feature sequence is evaluated for fire indicativeness based on the vehicle fire history event set to obtain the environmental monitoring fire indicative sequence. Based on the environmental monitoring fire indicator sequence, fire sign sensitive attention is assigned to the environmental monitoring feature sequence to generate the environmental fire sensitive feature vector.

3. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 2, characterized in that, Based on the historical event set of vehicle fires, the environmental monitoring feature sequence is evaluated for fire indicative intensity to obtain an environmental monitoring fire indicative intensity sequence, including: Fire sign feature identification is performed based on the vehicle fire history event set to obtain a fire sign feature library; Based on the historical event set of vehicle fires, the confidence level of each fire sign feature in the fire sign feature database is evaluated to obtain a fire feature confidence sequence. Based on the fire feature confidence sequence, the fire sign feature library is cleaned according to the predetermined fire feature confidence level to establish a confidence fire sign feature space; Based on the environmental monitoring feature sequence, extract the R-th feature of environmental monitoring, where R is a positive integer; Based on each confidence fire indication feature in the confidence fire indication feature space, the Rth feature of the environmental monitoring is evaluated by twin comparison to obtain the Rth twin comparison evaluation sequence; The maximum value of the Rth twin ratio evaluation sequence is optimized to generate the Rth environmental monitoring fire indicator, and the Rth environmental monitoring fire indicator is added to the environmental monitoring fire indicator sequence.

4. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 1, characterized in that, Based on the vehicle fire history event set, single-factor fire prediction is performed on the environmental fire sensitivity feature vector, the working condition fire sensitivity feature vector, and the visual fire sensitivity feature vector to obtain the first fire prediction map, including: Based on the vehicle fire history event set, single-factor classification was performed to obtain environmental fire history event set, working condition fire history event set and visual fire history event set; Based on the set of historical environmental fire events, environmental fire sensitive feature vectors are used to perform environmental factor fire inference to obtain the first fire inference result. Based on the historical event set of fires under the working conditions, the fire sensitive feature vector of the working conditions is used to extrapolate the fire situation of the working conditions and obtain the second fire situation extrapolation result. Based on the set of historical visual fire events, visual fire sensitivity feature vectors are used to perform visual factor fire inference to obtain the third fire inference result. The environmental fire-sensitive feature vector, the working condition fire-sensitive feature vector, the visual fire-sensitive feature vector, the first fire scenario prediction result, the second fire scenario prediction result, and the third fire scenario prediction result are organized to generate the first fire scenario prediction map.

5. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 4, characterized in that, Based on the historical environmental fire event set, environmental fire sensitivity feature vectors are used to perform environmental factor fire scenario estimation to obtain the first fire scenario estimation result, including: Activate the fire path tracing architecture, which includes a top event node, intermediate event nodes, and a bottom event node, and the top event node, intermediate event nodes, and bottom event nodes are connected by logical relationship nodes; Based on the fire path tracing architecture, the environmental fire historical event set is traced to obtain the environmental fire tracing path set. Based on the environmental fire tracing path set, a disturbance is injected to obtain the environmental fire disturbance path set; Based on the environmental fire tracing path set and the environmental fire disturbance path set, adversarial training is performed to generate an environmental factor fire inference model. The environmental fire-sensitive feature vector is input into the environmental factor fire prediction model to obtain the first fire prediction result.

6. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 1, characterized in that, Based on the first fire situation simulation map, a multi-factor coupled fire situation simulation is performed to obtain a second fire situation simulation map, including: Based on the first fire scenario simulation map, environmental conditions coupled with fire scenario simulation are performed to obtain the fourth fire scenario simulation result. Based on the first fire scenario simulation map, environmental visual coupling fire scenario simulation is performed to obtain the fifth fire scenario simulation result; Based on the first fire scenario simulation map, a visual coupling fire scenario simulation under working conditions is performed to obtain the sixth fire scenario simulation result. Based on the first fire scenario simulation map, a three-factor coupled fire scenario simulation was performed to obtain the seventh fire scenario simulation result. Based on the results of the fourth, fifth, sixth, and seventh fire scenario simulations, the first fire scenario simulation map is optimized to generate the second fire scenario simulation map.

7. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 6, characterized in that, Based on the first fire scenario simulation map, an environmental condition coupled fire scenario simulation is performed to obtain the fourth fire scenario simulation result, including: Based on the historical event set of vehicle fires, environmental and working condition coupling factor events are extracted to obtain an environmental and working condition coupled fire event set. Based on the fire path tracing architecture, the environmental condition coupled fire event set is traced to obtain the environmental condition coupled fire path set. Based on the set of fire path coupled with environmental conditions, train the fire simulation model coupled with environmental conditions. Based on the environmental fire-sensitive feature vector and the working condition fire-sensitive feature vector, and according to the environmental working condition coupled fire simulation model, the fourth fire simulation result is output.

8. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 1, characterized in that, The vehicle visual monitoring data includes infrared thermal image data and visible light image data of the mining vehicle.

9. The method for early fire warning of mining vehicles based on multi-source information fusion as described in claim 1, characterized in that, Early fire suppression management is carried out on the mining vehicles based on the second fire scenario projection map.

10. A multi-source information fusion early fire warning system for mining vehicles, characterized in that, The system is used to perform the multi-source information fusion method for early fire warning of mining vehicles according to any one of claims 1 to 9, the system comprising: The multi-source monitoring module performs multi-source monitoring on mining vehicles to obtain vehicle environmental monitoring data, vehicle operating condition data, and vehicle visual monitoring data. The environmental fire sensitive feature vector acquisition module performs fire sign sensitive feature detection on the vehicle environmental monitoring data based on the vehicle fire history event set to obtain the environmental fire sensitive feature vector. The fire-sensitive feature vector acquisition module for operating conditions detects fire signs based on the vehicle's historical fire event set and the vehicle's operating condition data to obtain the fire-sensitive feature vector for operating conditions. The visual fire sensitive feature vector acquisition module performs fire sign sensitive feature detection on the vehicle visual monitoring data based on the vehicle fire history event set to obtain the visual fire sensitive feature vector. The fire simulation first map acquisition module performs single-factor fire simulation based on the vehicle fire history event set, the environmental fire sensitive feature vector, the working condition fire sensitive feature vector, and the visual fire sensitive feature vector to obtain the fire simulation first map. The early fire warning execution module performs multi-factor coupled fire prediction based on the first fire prediction map to obtain the second fire prediction map, and executes early fire warning based on the second fire prediction map.

Images