An interpretable machine learning assisted fire prediction evaluation method, system, device and medium based on multiple types of sensors

Abstract

A kind of fire prediction evaluation method, system, equipment and medium based on multi-type sensor explainable machine learning auxiliary, use multiple types of sensors for different scenarios, collect and integrate a large number of fire data sets, cover multiple different fire scenes, i.e. different time, place etc., and divide data set into training set and test set;Convolutional neural network submodel is constructed, the model is trained using data set, realize the universal modeling of fire prediction;Use principal component analysis method and combine entropy weight method to evaluate the disaster situation of the place where the fire occurs, realize the accurate universal evaluation of the disaster degree;The present application can accurately analyze and predict different types and sizes of fire situation by combining convolutional neural network submodel with principal component analysis method (PCA) and entropy weight method for disaster situation evaluation, improve the accuracy and operability of disaster situation evaluation, while enhancing the data processing capacity of the model and the explainability of the output.

Description

Technical Field

[0001] This invention belongs to the field of fire prediction and assessment technology, and specifically relates to a fire prediction and assessment method, system, device and medium based on interpretable machine learning assisted by multiple types of sensors. Background Technology

[0002] Effective fire prediction and assessment technologies are of paramount importance. With the advancement of science and technology, a range of advanced technologies have been applied to the early prediction and real-time assessment of fires, demonstrating immense potential.

[0003] Some existing fire detection systems utilize multi-sensor fusion technology to expand the spatiotemporal range of fire monitoring by monitoring parameters such as light, smoke, temperature, gas, and humidity. This technology combines signals from multiple sensors to improve the accuracy and timeliness of fire alarms. For example, multi-sensor fire alarm systems based on Dempster-Shafer evidence theory can effectively handle conflicting evidence between sensors and improve the reliability of fire warnings by processing and fusing data from multiple sensors.

[0004] In traditional methods, sensors are typically used to monitor environmental variables such as smoke, temperature, and gases, which are crucial indicators for early fire detection. However, these methods often rely on simple threshold judgments and struggle to adapt to increasing environmental complexity. To improve the accuracy of predictions and the granularity of assessments, researchers have begun exploring more sophisticated data analysis models. Specifically, machine learning techniques have been introduced into the field of fire monitoring due to their powerful data processing capabilities. Mert Nakip et al., in their paper "Recurrent Trend Predictive Neural Network for Multi-Sensor Fire Detection," investigated the application of recurrent trend predictive neural networks (rTPNN) in multi-sensor fire detection, validating the effectiveness of ML techniques in fire prediction and assessment modeling.

[0005] Existing traditional fire prediction technologies are subject to interference from various factors in complex environments, making it difficult to fully realize their advantages.

[0006] Current research has developed fire prediction models for various environments and has made significant progress with the support of machine learning techniques, playing a positive role in the development of fire prediction. However, due to limitations in experimental data, these models are only applicable to the specific scenarios presented, and may face some challenges when attempting to extend them to other environments.

[0007] In complex environments, fire prediction is significantly stochastic due to the combined influence of multiple environmental factors. For example, meteorological conditions (such as temperature, humidity, and wind speed), topographic features (such as slope and vegetation density), and human activities (such as agricultural burning and campfires) can all affect the occurrence and spread of fires. Furthermore, the rapid spread and uncertainty of fires further amplify the difficulty and randomness of fire prediction. It is worth noting that compared to traditional fixed monitoring systems, the occurrence and spread of fires typically exhibit strong temporal and spatial heterogeneity. In such cases, monitoring equipment may fail to capture key data on fire occurrence in real time, leading to insufficient accuracy and timeliness in prediction.

[0008] Meanwhile, in variable environments, computer vision faces challenges in fire prediction and dynamic assessment, including strong environmental dependence, high reliance on ambient lighting conditions, viewing angle and occlusion issues, high computational resource requirements, and high false alarm rates. Therefore, it is imperative to construct a universally applicable fire prediction and assessment model to analyze and improve the reliability of fire prediction and assessment.

[0009] Patent application CN202410324749 discloses a fire prediction system and method based on multi-source information fusion. This system includes collecting fire data through an internal building sensor system, using LSTM and ELM neural networks for data prediction and feature processing, and utilizing MATLAB's fuzzy control toolbox to establish fuzzy rules for final fire prediction. However, this method lacks spatial feature capture capabilities; feature selection is black-boxed, lacking transparency and interpretability; fuzzy rule setting is highly subjective; high-dimensional data processing is difficult; and fuzzy rules have limitations in weighting multiple indicators. Summary of the Invention

[0010] To overcome the shortcomings of the existing technologies, the present invention aims to provide a fire prediction and assessment method, system, device, and medium based on interpretable machine learning-assisted methods using multiple types of sensors. By combining convolutional neural network sub-models with principal component analysis (PCA) and entropy weighting methods for disaster assessment, this method can accurately analyze and predict fire situations of different types and scales, improving the accuracy and operability of disaster assessment while enhancing the model's data processing capabilities and the interpretability of its output. This method, through optimized data processing and assessment algorithms, effectively improves the response speed and processing accuracy to complex fire scenarios, thus providing more effective technical support for fire prevention and emergency response.

[0011] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0012] A fire prediction and assessment method based on interpretable machine learning using multiple types of sensors includes the following steps:

[0013] Step 1: Collect fire dataset T. Fire dataset T consists of time series data collected by multiple types of sensors and real labels of whether a fire occurred at the corresponding time. Preprocess the fire dataset T and divide it into training set and test set.

[0014] Step 2: Construct a convolutional neural network sub-model. Train the convolutional neural network sub-model using the training set obtained in Step 1. Input the test set obtained in Step 1 into the trained convolutional neural network sub-model. Calculate the accuracy and loss value to determine whether the output matches the corresponding true label. After multiple rounds of training and testing, complete the model training and obtain the optimal convolutional neural network sub-model. Obtain the sensor dataset A' currently deployed at multiple locations. Input the current sensor dataset A' into the optimal convolutional neural network sub-model after training to obtain the output. Use this as a basis to filter the current sensor dataset A', deleting the data corresponding to locations where fires will not occur, and obtain the sensor dataset A corresponding to locations where fires will occur.

[0015] Step 3: For the sensor dataset A corresponding to the location where a fire may occur in Step 2, use Principal Component Analysis (PCA) to determine the main factors affecting the occurrence of a fire at that location. Based on this, filter the data in the sensor dataset A corresponding to the location where a fire may occur obtained in Step 2 to obtain a dataset M containing only the main factors affecting the occurrence of a fire at that location.

[0016] Step 4: Use the entropy weight method to assess the disaster situation after a fire occurs at the location of the dataset M in Step 3, which contains only the main factors that affect the occurrence of the fire at that location.

[0017] In step 1, the collected fire dataset T is preprocessed using the following method:

[0018]

[0019] Where μ and σ represent the mean and standard deviation of each type of sensor data, respectively, the fire dataset T was then divided into training and testing sets.

[0020] The specific method for step 2 is as follows:

[0021] Step 2.1: The convolutional neural network sub-model is used to detect potential fire signals from environmental monitoring data; the architecture of the convolutional neural network sub-model is as follows:

[0022] f(x) = σ(w²ReLU(w₁*x + b₁) + b₂)

[0023] Where * denotes convolution operation, σ is the Sigmoid activation function, used to output the probability of predicting as positive (i.e., detecting fire); w1 and w2 are the model weights, and b1 and b2 are bias terms;

[0024] Step 2.2: Input the training set obtained in Step 1 into the convolutional neural network sub-model constructed in Step 2.1 to obtain the final prediction probability, i.e., p. If p ≥ 0.5, then the predicted label at the corresponding time step is... on the contrary,

[0025] Step 2.3: Input the final predicted probability p obtained in Step 2.2 and the true label at the corresponding time in the fire dataset T in Step 1 into the minimum binary cross-entropy loss function for training. During the training process, each batch of data undergoes forward propagation, loss calculation and backpropagation.

[0026] The formula for calculating the loss is as follows:

[0027]

[0028] Where, N train It is the number of samples in the training set, y i p is the true label of the i-th sample, with a value of 0 or 1. i It is the probability that the model predicts the i-th sample as positive.

[0029] Step 2.4: Evaluate the performance of the convolutional neural network sub-model trained in Step 2.3 on the test set obtained in Step 1. The evaluation metrics are accuracy and loss value, calculated using the following formulas:

[0030]

[0031] Where, N test y is the number of samples in the test set. i It is the true label of the i-th sample;

[0032] The formula for calculating the loss value is as follows:

[0033]

[0034] Where, N test y is the number of samples in the test set. i p is the true label of the i-th sample, with a value of 0 or 1. i It is the probability that the model predicts the i-th sample as positive.

[0035] Step 2.5: Repeat steps 2.3 and 2.4 to obtain multiple rounds of test data, plot the loss curve and accuracy curve during training and testing, verify the performance and stability of the convolutional neural network sub-model, and obtain the optimal convolutional neural network sub-model.

[0036] Step 2.6: Obtain the current sensor dataset A' deployed at multiple locations. The current sensor dataset A' consists of multiple sets of multi-type sensor data from different locations at the same time. Input the data from the current sensor dataset A' group by group into the optimal convolutional neural network sub-model in Step 2.5 to obtain the output value p, which is the probability of a fire occurring. If p ≥ 0.5, then the predicted label for the corresponding time moment is... Do not delete, otherwise, Delete the corresponding data group in the current sensor dataset A'. Repeat the above operation until all data groups in the current sensor dataset A' have been input and deleted. The final current sensor dataset A' is the sensor dataset A corresponding to the location where the fire will occur.

[0037] The specific method for step 3 is as follows:

[0038] Step 3.1: Standardize the sensor dataset A corresponding to the location where a fire might occur in Step 2 to obtain the standardized data matrix Z. The standardization formula is:

[0039]

[0040] Among them, X ij It is the value of the i-th sample on the j-th variable. s is the mean of the j-th variable. j It is the standard deviation of the j-th variable;

[0041] Step 3.2, Calculate the covariance matrix: Based on the standardized data matrix Z from Step 3.1, calculate the covariance matrix C, whose elements C... jk The covariance between variables j and k is:

[0042]

[0043] Step 3.3, Eigenvalue Decomposition: Perform eigenvalue decomposition on the covariance matrix C obtained in Step 3.2, specifically using the following formula:

[0044] CV = VD

[0045] We obtain eigenvalues ​​D and eigenvectors V, where V is a matrix containing all eigenvectors of C; D is a diagonal matrix whose diagonal elements are the corresponding eigenvalues, representing the variance contribution of each eigenvalue of C.

[0046] Step 3.4: Based on the magnitude of the eigenvalues ​​D obtained in Step 3.3, and using the Kaiser criterion, select the eigenvectors V corresponding to the first m eigenvalues ​​greater than 1 to form the principal component matrix P. Based on the principal component matrix P, obtain the dataset M containing only the main factors that affect the occurrence of the fire at that location.

[0047] The specific method for step 4 is as follows:

[0048] Step 4.1: For each type of data collected by multiple sensors in dataset M (which only contains the main factors affecting the occurrence of fire at this location) from Step 3, calculate its information entropy to assess the dispersion of the data. The higher the information entropy, the more irregular the data changes of that type of sensor; conversely, the lower the information entropy, the stronger the regularity of the changes. Information entropy E j Calculated using the following formula:

[0049]

[0050] Where, p ij It is the proportion of the value of the i-th sample on the j-th indicator to the sum of all sample values ​​for that indicator; k is a constant, taking a value of k. n is the number of samples;

[0051] Step 4.2, using the information entropy E obtained in step 4.1 j Calculate the weight w of each type of data collected by multiple sensors in a dataset M containing only the main factors influencing the occurrence of a fire at that location. j The formula for calculating using the complement of information entropy is as follows:

[0052]

[0053] Where m is the total number of indicators;

[0054] Step 4.3, use the weights w obtained in step 4.2 j We calculate the weighted sum of each type of data collected by multiple sensors in dataset M, which contains only the main factors influencing the occurrence of the fire at that location, to obtain a comprehensive score for disaster assessment.

[0055]

[0056] Where, x ij It is the value of the i-th sample on the j-th indicator;

[0057] Step 4.4: Determine the severity of the fire based on the overall score of the disaster assessment in Step 4.3. The higher the score, the greater the potential risk and severity of the fire.

[0058] This invention also provides an interpretable machine learning-assisted fire prediction and assessment system based on multiple types of sensors, comprising:

[0059] The fire dataset T collection and processing module is used to collect the fire dataset T, which consists of time series data collected by multiple types of sensors and real labels of whether a fire occurred at the corresponding time. The fire dataset T is preprocessed and divided into training set and test set.

[0060] The module for acquiring sensor dataset A corresponding to locations where fires are likely to occur is used to construct a convolutional neural network sub-model. It trains the sub-model using a training set and inputs a test set into it. By calculating accuracy and loss values, it determines whether the output matches the corresponding true labels. After multiple rounds of training and testing, the model training is completed, resulting in the optimal convolutional neural network sub-model. The module then acquires sensor dataset A' currently deployed at multiple locations. It inputs A' into the optimal convolutional neural network sub-model to obtain the output. This output is used to filter A', removing data corresponding to locations where fires are unlikely to occur, thus obtaining sensor dataset A corresponding to locations where fires are likely to occur.

[0061] The module for obtaining the dataset M containing only the main factors affecting the occurrence of fire at a specific location is used to determine the main factors affecting the occurrence of fire at a specific location using principal component analysis (PCA) on the sensor dataset A corresponding to the location where a fire may occur. Based on this determination, the data is filtered on the sensor dataset A corresponding to the location where a fire may occur to obtain the dataset M containing only the main factors affecting the occurrence of fire at that specific location.

[0062] The disaster assessment module is used to assess the disaster situation after a fire occurs at a given location using the entropy weight method on a dataset M containing only the main factors that influence the occurrence of a fire at that location.

[0063] This invention also provides a fire prediction and assessment device based on interpretable machine learning assisted by multiple types of sensors, comprising:

[0064] Memory: A computer program that stores the above-mentioned interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors, and is a computer-readable device;

[0065] Processor: Used to implement the aforementioned interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors when executing the computer program.

[0066] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the aforementioned interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors.

[0067] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0068] First, this invention employs a convolutional neural network sub-model, which has a strong ability to extract spatial features and is particularly suitable for processing disaster data with spatial distribution characteristics (such as fire images or geographic information). In contrast, LSTM and ELM are mainly good at processing time series data, and their ability to process spatial features is relatively weak.

[0069] Second, this invention employs Principal Component Analysis (PCA) for dimensionality reduction, which reduces data redundancy and improves the computational efficiency of the model. Meanwhile, the entropy weighting method can reasonably allocate the importance of each feature, thereby avoiding human bias and further enhancing the accuracy and scientific rigor of the model. In contrast, LSTM and ELM typically learn features adaptively through network training, but this may lead to overfitting or underfitting, and lacks transparency regarding feature importance.

[0070] Third, this invention employs the entropy weighting method, which combines the data dispersion of each feature, enabling the allocation of more reasonable weights to different factors in disaster assessment, thereby improving the rationality of the model's decision-making. While existing fuzzy control methods can use expert experience to set fuzzy rules, these rules are highly subjective and struggle to handle the relative importance of features when assigning weights to multiple indicators.

[0071] Fourth, the present invention employs principal component analysis (PCA) and entropy weight method to make the feature selection process of the assessment model more interpretable and to clarify the contribution of each factor to the disaster assessment results. In contrast, LSTM and ELM have strong black box characteristics and are difficult to explain the intermediate process of the model and the reasons for the formation of the prediction results.

[0072] Fifth, Principal Component Analysis (PCA) is used for dimensionality reduction, which can effectively address the multidimensional data issues that may be involved in disaster assessment, reducing data dimensionality and model complexity, and lowering computational costs. In contrast, LSTM and ELM suffer from increased model complexity and significantly longer training times when processing high-dimensional data, potentially leading to performance degradation.

[0073] In summary, this invention, by combining a convolutional neural network sub-model with principal component analysis (PCA) and entropy weighting, integrates advanced data fusion techniques and an innovative scoring mechanism to provide a comprehensive and accurate solution, significantly improving the efficiency and accuracy of fire prediction and assessment. The application of these technologies not only enhances the performance of the prediction model but also provides a powerful tool for disaster management and emergency response. Attached Figure Description

[0074] Figure 1This is a flowchart illustrating the implementation of the present invention.

[0075] Figure 2 This is a diagram showing the application results of the prediction model of this invention in a real-world scenario.

[0076] Figure 3 The training results of the model of this invention are visualized (loss value is less than 0.1).

[0077] Figure 4 The training results of the model of this invention are visualized (accuracy up to 96.1%).

[0078] Figure 5 The cumulative variance explained by PCA is used in this invention. Detailed Implementation

[0079] The technical solution adopted by the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0080] This invention aims to address the shortcomings of existing technologies by proposing an interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors. This model considers the complexity and variability of fire scenarios, the correlation between sensor data, and other influencing factors. The implementation scheme involves: using multiple types of sensors for different scenarios to collect and integrate a large amount of fire datasets, covering various fire scenarios including different times and locations, and dividing the datasets into training and testing sets; constructing a convolutional neural network sub-model and training it using the datasets to achieve universal modeling for fire prediction; and using principal component analysis combined with entropy weighting to assess the disaster situation in predicted fire locations, achieving an accurate and universal assessment of the degree of disaster.

[0081] The universal fire prediction and dynamic assessment model proposed in this invention, based on convolutional neural networks, principal component analysis (PCA), and entropy weighting, can accurately and efficiently predict and assess fire situations in various fire scenarios, providing strong support for fire protection and timely rescue.

[0082] This invention aims to address the challenges of fire risk monitoring and assessment in complex and ever-changing environments. The device integrates data from multiple types of sensors (such as temperature, humidity, smoke concentration, and carbon dioxide concentration) and utilizes advanced machine learning algorithms to achieve deep fusion and feature extraction of multi-dimensional data. Compared to traditional single-sensor systems, multi-sensor fusion can capture more comprehensive environmental information, effectively reducing false alarms and missed alarms, and improving the accuracy and response speed of fire detection.

[0083] The device can not only accurately monitor the likelihood of fire occurrence in different scenarios (such as indoor, outdoor, and industrial environments), but also dynamically assess the fire's spread and severity. Through real-time analysis and fusion of multiple environmental variables, the device can provide more reliable early warning information under complex conditions, providing a scientific basis for emergency decision-making and resource allocation. The application of multi-sensor fusion technology ensures the device's high adaptability and stability in various environments, effectively improving the overall performance of fire early warning and the efficiency of emergency response.

[0084] Reference Figure 1 This example presents an interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors. The implementation steps include the following:

[0085] Step 1: Collect fire dataset T. Fire dataset T consists of time series data collected by multiple types of sensors and real labels indicating whether a fire occurred at the corresponding time. Ensure that the dataset covers all fire scenarios. This process should fully consider various key factors, such as the diversity of fire sources and site environment. The fire dataset T is then divided into training set and test set.

[0086] Step 1.1: Collect fire dataset. In this embodiment, the fire sensor dataset obtained by Stefan Blattmann (Yu R, Kim KA Study of Novel Initial Fire Detection Algorithm Based on Deep Learning Method[J]. Journal of Electrical Engineering & Technology, 2024, (prepublish): 1-12.) is labeled as T.

[0087]

[0088] in, This represents the input sensor data. This indicates the output value corresponding to the input sensor data, i.e., whether a fire has occurred;

[0089] The types of sensors used are shown in Table 1:

[0090] Table 1 Sensor Types

[0091]

[0092]

[0093] Step 1.2, the fire dataset T from Step 1.1 is preprocessed using the following method:

[0094]

[0095] Where μ and σ represent the mean and standard deviation of each sensor data, respectively. The fire dataset T was then divided into a training set and a test set, with a ratio of 80% and 20%, respectively.

[0096] Step 2: Construct a convolutional neural network sub-model. Train the convolutional neural network sub-model using the training set obtained in Step 1. Input the test set obtained in Step 1 into the trained convolutional neural network sub-model. Calculate the accuracy and loss value to determine whether the output matches the corresponding true label. Figure 2 As shown; after multiple rounds of training and testing, the model training is completed, and the optimal convolutional neural network sub-model after training is obtained; the sensor dataset A' currently deployed in multiple locations is obtained, and the current sensor dataset A' is input into the optimal convolutional neural network sub-model after training to obtain the output result. Based on this, the current sensor dataset A' is filtered, and the data corresponding to the locations where fires will not occur is deleted, so as to obtain the sensor dataset A corresponding to the locations where fires will occur; thus realizing the universal modeling of fire prediction.

[0097] Step 2.1 employs a convolutional neural network sub-model to process time-series data collected from multiple sensors. This convolutional neural network sub-model aims to detect potential fire signals from environmental monitoring data. The architecture of this convolutional neural network sub-model is as follows:

[0098] f(x) = σ(w²ReLU(w₁*x + b₁) + b₂)

[0099] Where * denotes convolution operation, σ is the Sigmoid activation function, used to output the probability of predicting as positive (i.e., detecting fire); w1 and w2 are the model weights, and b1 and b2 are bias terms;

[0100] Step 2.2: Input the training set obtained in Step 1 into the convolutional neural network sub-model constructed in Step 2.1 to obtain the final prediction probability, i.e., p. If p ≥ 0.5, then the predicted label at the corresponding time step is... on the contrary,

[0101] Step 2.3: Input the final predicted probability p obtained in Step 2.2 and the true labels at the corresponding times in the fire dataset T from Step 1 into the minimum binary cross-entropy loss function for training to optimize the accuracy of fire detection. The Adam optimizer is used for parameter optimization, with an initial learning rate set to 0.00001. During training, each batch of data undergoes forward propagation, loss calculation, and backpropagation.

[0102] The formula for calculating the loss is as follows:

[0103]

[0104] Where, N train It is the number of samples in the training set, y i p is the true label of the i-th sample, with a value of 0 or 1. i It is the probability that the model predicts the i-th sample as positive.

[0105] Step 2.4: Evaluate the performance of the convolutional neural network sub-model trained in Step 2.3 on the test set obtained in Step 1. The main evaluation metrics are accuracy and loss value, calculated as follows:

[0106]

[0107] Where, N test y is the number of samples in the test set. i It is the true label of the i-th sample;

[0108] The formula for calculating the loss value is as follows:

[0109]

[0110] Where, N test y is the number of samples in the test set. i p is the true label of the i-th sample, with a value of 0 or 1. i It is the probability that the model predicts the i-th sample as positive.

[0111] Step 2.5: Repeat steps 2.3 and 2.4 to obtain multiple rounds of test data, plot the loss curve and accuracy curve during training and testing, verify the performance and stability of the convolutional neural network sub-model, and obtain the optimal convolutional neural network sub-model.

[0112] Figure 3 The diagram illustrates the changes in model loss during training, with the blue curve representing training loss and the orange curve representing test loss. As the number of iterations increases, both curves show a decreasing trend, indicating that the model is gradually learning, the loss is gradually decreasing, and the model performance is improving. It can be observed that the training loss is consistently slightly higher than the test loss, which may indicate that the model has good generalization ability and that there is no obvious overfitting.

[0113] Figure 4 The diagram illustrates the improvement in model accuracy during training. With each iteration, the accuracy gradually increased from 87.5% to approximately 96.1%, showing a steady upward trend. This indicates that the model is continuously optimizing and gradually becoming better at correctly classifying data, demonstrating good training results.

[0114] Overall, from Figure 3 and Figure 4 As can be seen, the convolutional neural network exhibits stable performance during training, with reduced loss and improved accuracy, demonstrating good learning ability and generalization performance.

[0115] Step 2.6: Obtain the sensor dataset A' currently deployed at multiple locations. This dataset consists of multiple sets of multi-type sensor data from different locations at the same time. Input the data from the current sensor dataset A' into the optimal convolutional neural network sub-model in Step 2.5 one by one to obtain the output value p, which is the probability of a fire occurring. If p ≥ 0.5, then the predicted label for the corresponding time moment is... Do not delete, otherwise, Delete the corresponding data group in the current sensor dataset A'. Repeat the above operation until all data groups in the current sensor dataset A' have been input and deleted. The final current sensor dataset A' is the sensor dataset A corresponding to the location where the fire will occur.

[0116] Step 3: For the sensor dataset A corresponding to the location where a fire may occur in Step 2, use Principal Component Analysis (PCA) to determine the main factors affecting the occurrence of a fire at that location. Based on this, filter the data in the sensor dataset A corresponding to the location where a fire may occur obtained in Step 2 to obtain a dataset M containing only the main factors affecting the occurrence of a fire at that location.

[0117] Principal component analysis is used to screen the main influencing factors, reducing the impact of bias and distortion in the weighting results caused by the fact that the main factors affecting fire vary in different environments and that there may be high multicollinearity among many factors (such as the concentration of various chemical gases).

[0118] The specific method for step 3 is as follows:

[0119] Step 3.1: To eliminate the influence of dimensions, the sensor dataset A corresponding to the location where the fire might occur in Step 2 is first standardized to obtain the standardized data matrix Z. The standardization formula is:

[0120]

[0121] Among them, X ij It is the value of the i-th sample on the j-th variable. s is the mean of the j-th variable. j It is the standard deviation of the j-th variable;

[0122] Step 3.2, Calculate the covariance matrix: Based on the standardized data matrix Z from Step 3.1, calculate the covariance matrix C, whose elements C...jk The covariance between variables j and k is represented as follows:

[0123]

[0124] Step 3.3, Eigenvalue Decomposition: Perform eigenvalue decomposition on the covariance matrix C obtained in Step 3.2, specifically using the following formula:

[0125] CV = VD

[0126] We obtain eigenvalues ​​D and eigenvectors V, where V is a matrix containing all eigenvectors of C; D is a diagonal matrix whose diagonal elements are the corresponding eigenvalues, representing the variance contribution of each eigenvalue of C.

[0127] Step 3.4: Based on the magnitude of the eigenvalues ​​D obtained in Step 3.3, and using the Kaiser criterion, select the eigenvectors V corresponding to the first 7 eigenvalues ​​greater than 1 to form the principal component matrix P. Based on the principal component matrix P, obtain the dataset M containing only the main factors influencing the occurrence of the fire at that location. These principal components, i.e., the matrix formed by the eigenvectors corresponding to the first m eigenvalues, explain the vast majority of the variance in the original data. Specifically, these eigenvectors represent the weights (i.e., loadings) of the original variables on each principal component, and each principal component represents a set of linear combinations designed to explain the largest portion of the data variance.

[0128] Step 4: Use the entropy weight method to assess the disaster situation after a fire occurs at the location of the dataset M in Step 3, which contains only the main factors that affect the occurrence of the fire at that location.

[0129] The specific method for step 4 is as follows:

[0130] Step 4.1: For each type of data collected by multiple sensors in dataset M (which only contains the main factors affecting the occurrence of fire at this location) from Step 3, calculate its information entropy to assess the dispersion of the data. The higher the information entropy, the more irregular the data changes of that type of sensor; conversely, the lower the information entropy, the stronger the regularity of the changes. Information entropy E j Calculated using the following formula:

[0131]

[0132] Where, p ij It is the proportion of the value of the i-th sample on the j-th indicator to the sum of all sample values ​​for that indicator; k is a constant, taking a value of k. n is the number of samples;

[0133] Step 4.2, using the information entropy E obtained in step 4.1 jCalculate the weight w of each type of data collected by multiple sensors in a dataset M containing only the main factors influencing the occurrence of a fire at that location. j The formula for calculating using the complement of information entropy is as follows:

[0134]

[0135] Where m is the total number of indicators;

[0136] like Figure 5 The table shows the weighted calculation results. It can be seen that Raw Ethanol has the highest weight at 42.92, followed by Raw H2 at 25.57. The weights of the other indicators are relatively smaller. Specifically, Pressure has a weight of 4.41, Temperature 11.85, Humidity 8.87, TVOC 2.94, and eCO2 3.44.

[0137] Step 4.3, use the weights w obtained in step 4.2 j We calculate the weighted sum of each type of data collected by multiple sensors in dataset M, which contains only the main factors influencing the occurrence of the fire at that location, to obtain a comprehensive score for disaster assessment.

[0138]

[0139] Where, x ij It is the value of the i-th sample on the j-th indicator;

[0140] Step 4.4: Determine the severity of the fire based on the overall score of the disaster assessment in Step 4.3. The higher the score, the greater the potential risk and severity of the fire.

[0141] The key points and protection points of this invention are as follows:

[0142] (1) Application of multi-sensor fusion technology: This invention utilizes various types of sensors (such as temperature sensors, humidity sensors, smoke sensors, and air pressure sensors) to monitor environmental data in real time. By fusing data from these different sensors, this invention can detect fires in various complex environments, improving the accuracy and reliability of fire early warning.

[0143] (2) Multi-environment adaptability: The present invention designs a fire detection model with strong universality, which can adapt to a variety of different fire scenarios (such as indoor, outdoor, industrial environments, etc.). Through extensive data training and testing, the model performs well in different environments and has high adaptability and generalization ability.

[0144] (3) Innovative Data Fusion and Processing Methods: To improve the accuracy of fire detection, this invention combines principal component analysis (PCA) and entropy weighting to process and analyze sensor data. These methods can reduce data redundancy, solve multicollinearity problems, and improve the interpretability and accuracy of the model.

[0145] (4) Efficient emergency management support: Through multi-sensor data fusion and intelligent analysis, this invention can provide more accurate fire information for emergency management, optimize resource allocation, and help formulate more effective emergency response plans.

[0146] (5) Deep Learning-Based Feature Extraction and Prediction Model: This invention introduces a convolutional neural network sub-model deep learning algorithm to automatically extract effective features from multi-dimensional sensor data. Compared with traditional rule and threshold judgment methods, these algorithms can more accurately predict the occurrence and development trend of fires, reducing false alarms and missed alarms.

[0147] This invention also provides an interpretable machine learning-assisted fire prediction and assessment system based on multiple types of sensors, comprising:

[0148] The fire dataset T collection and processing module is used to collect the fire dataset T in step 1. The fire dataset T consists of time series data collected by multiple types of sensors and real labels of whether a fire occurred at the corresponding time. The fire dataset T is preprocessed and divided into training set and test set.

[0149] The module for acquiring sensor dataset A corresponding to locations where fires may occur is used to implement the construction of a convolutional neural network sub-model in step 2. It trains the convolutional neural network sub-model using the training set obtained in step 1, inputs the test set obtained in step 1 into the trained convolutional neural network sub-model, calculates the accuracy and loss value, and determines whether the output matches the corresponding true label. After multiple rounds of training and testing, the model training is completed, and the optimal trained convolutional neural network sub-model is obtained. The module then acquires sensor dataset A' currently deployed at multiple locations, inputs the current sensor dataset A' into the optimal trained convolutional neural network sub-model, obtains the output, and uses this as a basis to filter the current sensor dataset A', deleting data corresponding to locations where fires will not occur, thus obtaining sensor dataset A corresponding to locations where fires may occur.

[0150] The module for obtaining the dataset M containing only the main factors affecting the occurrence of fire at this location is used in step 3 to determine the main factors affecting the occurrence of fire at this location using principal component analysis (PCA) on the sensor dataset A corresponding to the location where a fire will occur in step 2. Based on this, the data is filtered on the sensor dataset A corresponding to the location where a fire will occur obtained in step 2 to obtain the dataset M containing only the main factors affecting the occurrence of fire at this location.

[0151] The disaster assessment module is used in step 4 to perform a disaster assessment on the dataset M, which contains only the main factors that affect the occurrence of the fire at that location, based on the entropy weight method in step 3.

[0152] This invention also provides a fire prediction and assessment device based on interpretable machine learning assisted by multiple types of sensors, comprising:

[0153] Memory: A computer program that stores the above-mentioned interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors, and is a computer-readable device;

[0154] Processor: Used to implement the aforementioned interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors when executing the computer program.

[0155] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the aforementioned interpretable machine learning-assisted fire prediction and assessment method based on multiple types of sensors.

Claims

1. A multi-type sensor based explainable machine learning assisted fire prediction evaluation method, characterized in that, Comprising the following steps: Step 1, collect a fire data set T, the fire data set T is composed of time series data collected by multiple types of sensors and the true label of whether a fire occurs at the corresponding moment, preprocess the fire data set T, and divide it into a training set and a test set; The collected fire data set T is preprocessed by the following method: Wherein, μ and σ represent the mean and standard deviation of each sensor data respectively, and the fire data set T is then divided into a training set and a test set; Step 2, build a convolutional neural network submodel, train the convolutional neural network submodel using the training set obtained in step 1, input the test set obtained in step 1 into the trained convolutional neural network submodel, and determine whether the output result matches the corresponding true label by calculating the accuracy and loss value, complete the model training after multiple rounds of training and testing, and obtain the trained optimal convolutional neural network submodel; obtain the current sensor data set A' deployed in multiple locations, input the current sensor data set A' into the trained optimal convolutional neural network submodel, and obtain the output result as the basis for screening the current sensor data set A', deleting the data corresponding to the location where no fire will occur, and obtaining the sensor data set A corresponding to the location where the fire will occur; Step 3, using principal component analysis (PCA) to determine the main factors affecting the occurrence of fire at the location corresponding to the sensor data set A where the fire will occur in step 2, and based on the sensor data set A corresponding to the location where the fire will occur obtained in step 2, data screening is performed to obtain a data set M containing only the main factors affecting the occurrence of fire at the location; Step 4, using entropy weight method to evaluate the disaster situation after the fire occurs at the location in the data set M containing only the main factors affecting the occurrence of fire at the location in step 3.

2. The method of claim 1, wherein the method is based on a multi-type sensor-based explainable machine learning assisted fire prediction evaluation method. The specific method of step 2 is: Step 2.1, the convolutional neural network submodel is used to detect potential signals of fire from environmental monitoring data; the convolutional neural network submodel architecture is as follows: f(x) = σ(w2ReLU(w1*x+b1)+b2) Wherein, * represents convolution operation, σ is Sigmoid activation function, used to output the probability of predicting positive class (i.e. detecting fire); w1 and w2 are the weights of the model, and b1 and b2 are the bias terms; Step 2.2, input the training set obtained in step 1 into the convolutional neural network submodel constructed in step 2.1 to obtain the final prediction probability, i.e. p, if p≥0.5, the prediction label of the corresponding time is Conversely, Step 2.3, input the final prediction probability p obtained in step 2.2 and the true label of the corresponding moment in step 1 fire data set T into the minimum binary cross entropy loss function for training, and during the training process, each batch of data is forward propagated, loss calculated and back propagated; The loss calculation formula is as follows: where N train is the number of samples in the training set, y i is the true label of the ith sample, which is either 0 or 1, p i is the probability that the model predicts the ith sample as positive. Step 2.4, evaluate the performance of the convolutional neural network submodel trained in step 2.3 on the test set obtained in step 1, the evaluation index is accuracy and loss value, and the calculation formula is as follows: where N test is the number of samples in the test set, y i is the true label of the i-th sample. The calculation formula of the loss value is as follows: where N test is the number of samples in the test set, y i is the true label of the ith sample, which is either 0 or 1, p i is the probability that the model predicts the ith sample as positive. Step 2.5, repeat step 2.3 and step 2.4 to obtain multiple rounds of test data, draw loss curves and accuracy curves during training and testing, verify the performance and stability of the convolutional neural network submodel, and obtain the optimal convolutional neural network submodel; Step 2.6, obtain a current sensor data set A' deployed in multiple locations, the current sensor data set A' is composed of multiple groups of multi-type sensor data of different positions at the same time, input the data in the current sensor data set A' into the optimal convolutional neural network sub-model of step 2.5 group by group, and obtain an output value p, that is, the probability of fire occurrence, if p≥0.5, the prediction label of the corresponding time is No deletion operation, otherwise, Delete the corresponding data group in the current sensor data set A', repeat the above operation until all group data in the current sensor data set A' is input and deleted, and finally obtain the current sensor data set A', that is, the sensor data set A corresponding to the position where the fire will occur.

3. The method of claim 1, wherein the method is characterized by: The specific method of step 3 is: Step 3.1, standardize the sensor data set A corresponding to the location where the fire will occur in step 2 to obtain a standardized data matrix Z, and the standardization formula is: where X ij is the value of the ith sample on the jth variable, is the mean of the jth variable, s j is the standard deviation of the jth variable; Step 3.2, Compute the Covariance Matrix: Based on the standardized data matrix Z after Step 3.1, compute the covariance matrix C, whose elements Cjkrepresent the covariance between variable j and variable k: jk Step 3.2, Compute the Covariance Matrix: Based on the standardized data matrix Z after Step 3.1, compute the covariance matrix C, whose elements Cjkrepresent the covariance between variable j and variable k: Step 3.3, eigenvalue decomposition: perform eigenvalue decomposition on the covariance matrix C obtained in step 3.2, specifically through the following formula: CV=VD Obtain the eigenvalue D and the eigenvector V, wherein V is a matrix containing all the eigenvectors of C; D is a diagonal matrix whose diagonal elements are the corresponding eigenvalues, representing the variance contribution of each feature connected to C; Step 3.4, according to the size of the eigenvalue D obtained in step 3.3, based on Kaiser criterion, select the principal component matrix P composed of the eigenvectors corresponding to the first m eigenvalues greater than 1, and obtain the data set M of the main factors affecting the occurrence of fire at this location.

4. The method of claim 1, wherein the method is based on a multi-type sensor-based explainable machine learning assisted fire prediction evaluation method. The specific method of step 4 is: Step 4.1, for each type of data collected by the multi-type sensor in the data set M which only has an impact on the main factors of the fire in step 3, calculate the information entropy E of the data to evaluate the dispersion degree of the data, the higher the information entropy, the more irregular the change of the data of the type of sensor, and vice versa, the stronger the regularity of the change, the information entropy E of the data is calculated by the following formula: j E = -∑p (log2p) where p is the probability of the data of the type of sensor wherein p ij is the proportion of the value of the i-th sample on the j-th index to the sum of all sample values of the index; k is a constant, taken as n is the number of samples; Step 4.2, using the information entropy E obtained in step 4.1 j , calculate the weight of each data collected by the multi-type sensor in the data set M which only has the position affecting the main factors of fire occurrence, the weight w j The formula is as follows: Where m is the total number of indicators; Step 4.

3. Weighted sum of the data collected by the multi-type sensors in the dataset M that only have an impact on the main factors of fire occurrence using the weights w obtained in step 4.2 j A weighted sum of each data collected by the multi-type sensors in the dataset M that only have an impact on the main factors of fire occurrence is performed to obtain a comprehensive score of the disaster assessment: wherein x ij is the value of the ith sample on the jth indicator; Step 4.4, according to the high or low of the comprehensive score of disaster assessment in step 4.3, determine the severity of the fire, the higher the score, the greater the potential risk and severity of the fire.

5. A multi-type sensor based explainable machine learning assisted fire prediction assessment system based on the method of any one of claims 1 to 4, characterized in that, Comprise: A fire data set T collection and processing module for collecting a fire data set T, the fire data set T being composed of time series data collected by multiple types of sensors and real labels indicating whether a fire occurred at the corresponding time, preprocessing the fire data set T, and dividing the fire data set T into a training set and a test set; The collected fire data set T is preprocessed by the following method: Where μ and σ represent the mean and standard deviation of each sensor data respectively, and the fire data set T is then divided into a training set and a test set; A sensor data set A corresponding to the location where the fire will occur is obtained, a convolutional neural network submodel is constructed, the training set is used to train the convolutional neural network submodel, the test set is input into the trained convolutional neural network submodel, and whether the output result matches the corresponding real label is determined by calculating the accuracy and loss value. After multiple rounds of training and testing, the model training is completed, and the trained optimal convolutional neural network submodel is obtained; the current sensor data set A' deployed in multiple locations is obtained, the current sensor data set A' is input into the trained optimal convolutional neural network submodel, and the output result is obtained. The output result is used as a basis for screening the current sensor data set A', and the data corresponding to the location where the fire will not occur is deleted to obtain the sensor data set A corresponding to the location where the fire will occur. The position-influencing-fire-occurrence-main-factor-only-data-set M acquisition module is used to judge the position-influencing-fire-occurrence-main-factor of the sensor data set A corresponding to the position where the fire will occur by using the principal component analysis (PCA), and on this basis, data screening is performed on the sensor data set A corresponding to the position where the fire will occur to obtain the position-influencing-fire-occurrence-main-factor-only-data-set M; The disaster condition assessment module is used to assess the disaster condition after the fire occurs at the position by using the entropy weight method on the position-influencing-fire-occurrence-main-factor-only-data-set M.

6. A multi-type sensor based explainable machine learning assisted fire prediction evaluation device, characterized by, The method comprises: A memory for storing a computer program of the fire prediction and assessment method based on the multi-type sensor and the explainable machine learning assistance according to any one of claims 1-4, which is a computer-readable device; A processor for executing the computer program to realize the fire prediction and assessment method based on the multi-type sensor and the explainable machine learning assistance according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program, and the computer program can realize the fire prediction and assessment method based on the multi-type sensor and the explainable machine learning assistance according to any one of claims 1-4 when executed by a processor.

Images