A tunnel fire early warning method and system based on gas analysis

Abstract

The present invention relates to the technical field of tunnel fire monitoring and early warning, and specifically to a tunnel fire early warning method and system based on gas analysis. The tunnel fire early warning method based on gas analysis provided by the present invention comprises the following steps: obtaining multi-dimensional target data of tunnel gas; constructing an optimized prediction model for tunnel fire based on the multi-dimensional target data; obtaining an optimal prediction model through the optimized prediction model; and realizing the best early warning effect for tunnel fire using the optimal prediction model. The present invention constructs an optimal prediction model for tunnel fire using an optimized bidirectional long short-term memory neural network and multi-dimensional gas concentration data of the tunnel, effectively solving the problems of false alarm and missed alarm of fire early warning, realizing high-precision early warning for tunnel fire, and ensuring the safety of tunnel operation and the lives and property of personnel.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel fire monitoring and early warning, and in particular to a tunnel fire early warning method and system based on gas analysis. Background Art

[0002] Tunnels are important transportation infrastructures, and their internal environment is complex and changeable, with many potential safety hazards, especially fire accidents. Once the accumulation of flammable gases, toxic and harmful gases in tunnels reaches a critical concentration, it is very easy to cause fires or explosions, seriously threatening the safety of people's lives and property. Therefore, monitoring and early warning of tunnel fires are key technical means to achieve tunnel safety management and reduce disaster losses.

[0003] In the existing technology, temperature monitoring, smoke monitoring, image monitoring, and multi-parameter fusion monitoring are usually used to achieve early warning of tunnel fires. Although these methods have achieved early warning of tunnel fires to a certain extent, they still have shortcomings. For example, the temperature monitoring method is difficult to capture the tiny temperature changes in the early stage of the fire in time; the smoke monitoring method is difficult to work before the smoke spreads to the detector; the image monitoring method is expensive and easily affected by ambient temperature fluctuations; the multi-parameter fusion monitoring method is limited in its wide application due to its complex data processing requirements and high cost.

[0004] At present, there are not many studies on tunnel fire early warning by analyzing gas. The research work is low in cost and can detect fire signs earlier, but lacks scientificity and systematicity, resulting in unsatisfactory fire early warning results. Therefore, it is urgent to propose a more scientific, efficient and accurate tunnel fire early warning method based on gas analysis. Summary of the invention

[0005] In view of the defects in the prior art, the present invention provides a tunnel fire early warning method and system based on gas analysis.

[0006] On the one hand, the tunnel fire early warning method based on gas analysis includes the following steps: obtaining multi-dimensional target data of tunnel gas under normal conditions and fire conditions; based on the multi-dimensional target data, constructing an optimized prediction model for tunnel fire, wherein the optimized prediction model satisfies the following expression:

[0007] ,

[0008] in, represents the normalized weight of the similarity scoring function, The coding sequence is The normalized function weight of The coding sequence is The normalized function deviation of represents the normalization function, Indicates attention value ， Representing a time series; obtaining an optimal prediction model through the optimization prediction model; using the optimal prediction model, achieving the best early warning effect for tunnel fires. The present invention integrates multiple key steps such as data acquisition, model construction, optimization training and efficient early warning. By using the historical data of gas concentration under normal conditions and fire conditions to train the tunnel fire gas prediction model, the prediction model is continuously updated to achieve the best early warning effect for tunnel fires, thereby effectively avoiding or reducing the losses caused by fires and ensuring the safety of tunnel operations and the lives and property of personnel.

[0009] Optionally, the method of obtaining multidimensional target data of tunnel gas under normal conditions and fire conditions includes: obtaining target data of tunnel gas under normal conditions and fire conditions; preprocessing the target data to obtain multidimensional target data of tunnel gas. The present invention obtains and preprocesses multidimensional target data of gas from normal conditions to fire conditions; first, the target data of gas in the tunnel under normal operating conditions is obtained, which reflects the basic conditions of the tunnel environment; then, the target data of gas in the tunnel when a fire occurs is obtained, which is crucial for understanding the characteristics of the fire, the spread of the fire, and the changes in gas composition; finally, the raw data is preprocessed to obtain multidimensional target data of tunnel gas, which not only contains information in multiple dimensions such as time and space, but also covers key indicators such as gas concentration and component ratio, laying a solid foundation for the subsequent construction of a fire prediction model.

[0010] Optionally, the preprocessing of the target data to obtain the multi-dimensional target data of the tunnel gas includes: performing a first data processing on the target data, wherein the first data processing includes data processing using an outlier elimination algorithm, and the outlier elimination algorithm satisfies the following relationship:

[0011] ,

[0012] in, Indicates the normal value range of the data. Represents the average value of any gas characteristic parameter, represents the standard deviation; performing a second data processing on the target data, wherein the second data processing includes data processing using an interpolation missing value algorithm, and the interpolation missing value algorithm satisfies the following relationship:

[0013] ,

[0014] in, Indicates the pending The Gas parameter data, Indicates The Gas parameter data, Indicates The Gas parameter data, Indicates the dimension number of the gas parameter;

[0015] According to the result of the first data processing and the result of the second data processing, the multidimensional target data of the tunnel gas is obtained. The present invention significantly improves the quality of the multidimensional gas data by using the data processing of eliminating outliers and interpolating missing values, making it more accurate, complete and reliable, which is helpful for subsequent data analysis and model building, and provides a more scientific decision-making basis for tunnel fire warning and other work.

[0016] Optionally, the construction of an optimized prediction model for tunnel fires based on the multidimensional target data includes: establishing an algorithm-optimized model hidden layer structure based on the multidimensional target data; setting a self-attention mechanism based on the model hidden layer structure; and constructing an optimized prediction model for tunnel fires according to the self-attention mechanism. The present invention ensures that the model can deeply mine data features by establishing an algorithm-optimized model hidden layer structure; by introducing a self-attention mechanism, the model's ability to capture key information is enhanced, making the prediction results more accurate; by deeply analyzing the time characteristics of the sequence data, a tunnel fire optimization prediction model with a fully linked layer structure is formed, which fully guarantees the prediction accuracy.

[0017] Optionally, the model hidden layer structure includes a convolution hidden layer structure, an activation hidden layer structure and a pooling hidden layer structure; the convolution hidden layer structure satisfies the following expression:

[0018] The convolution hidden layer structure satisfies the following expression:

[0019] ,

[0020] in, represents the convolution hidden layer structure function, represents the weight, represents the convolution operation, The input of the representation layer, represents the bias, represents the activation function;

[0021] The activated hidden layer structure satisfies the following expression:

[0022] ,

[0023] in, represents the activation of the hidden layer structure function, represents the activation function, represents the characteristic function;

[0024] The pooled hidden layer structure satisfies the following expression:

[0025] ,

[0026] in, represents the pooling hidden layer structure function, represents the pooling function, represents the weight matrix, represents the bias term, represents the characteristic function, Represents the total number of elements. The present invention elaborates on the composition of the hidden layer structure of the model, including the convolution hidden layer, the activation hidden layer and the pooling hidden layer, and points out the specific expression of each layer structure, which has the effect of clearly showing the complexity and professionalism of the internal processing of the model, indicating that the model is finely designed and can extract and transform data features through various technical means such as convolution, activation and pooling, laying a solid foundation for subsequent predictive analysis.

[0027] Optionally, the self-attention mechanism satisfies the following expression:

[0028] ,

[0029] in, , Both represent learning matrices, represents the query vector, Indicates A coding sequence, represents a similarity scoring function; the present invention enhances the model's ability to capture key information by calculating the correlation between any two elements in the sequence; the self-attention mechanism involves matrix operations and weight allocation to ensure the flexibility and depth of information processing.

[0030] Optionally, obtaining the optimal prediction model through the optimization prediction model includes: obtaining optimal network parameters through the optimization prediction model; obtaining the optimal prediction model based on the optimal network parameters. The present invention performs multiple rounds of iterations and optimizations on the optimization prediction model, locks the optimal network parameters, and then constructs an optimal prediction model with excellent performance based on the optimal network parameters. The optimal prediction model not only has high prediction accuracy, but also has strong robustness, can more accurately grasp the inherent laws of the data, and provide a solid basis for decision-making, showing the strong potential of technology transformation into practical application value.

[0031] Optionally, the use of the optimal prediction model to achieve the best warning effect for tunnel fires includes: using the optimal prediction model to perform model training on multi-dimensional target data; obtaining the best prediction result by processing and analyzing the results of the model training; and achieving the best warning effect for tunnel fires based on the optimal prediction result. The present invention uses the constructed optimal prediction model to perform comprehensive model training on the multi-dimensional target data of tunnel gas, accurately captures fire precursors through deep learning and intelligent analysis, and outputs the best prediction result, which not only demonstrates the model's efficient processing capabilities and prediction accuracy, but also achieves the best warning effect for tunnel fires, effectively ensuring the safety of tunnel operations.

[0032] Optionally, obtaining the optimal prediction result by processing and analyzing the result of the model training includes: substituting the result of the model training into an absolute error expression to obtain an absolute error of the result, and the absolute error expression is as follows:

[0033] ,

[0034] in, represents the true value of concentration, represents the predicted concentration value;

[0035] The root mean square error of the result is obtained by substituting the result of the model training into the root mean square error expression. The root mean square error expression is as follows:

[0036] ,

[0037] in, Indicates the true value of concentration, represents the predicted concentration value;

[0038] The determination coefficient of the result is obtained by substituting the result of the model training into the determination coefficient expression, and the determination coefficient expression is as follows:

[0039] ,

[0040] in, represents the true value of concentration, represents the predicted concentration value, represents the average value of the true value of the concentration;

[0041] The optimal prediction result is obtained by analyzing the absolute error, the root mean square error and the determination coefficient. The present invention uses three precise evaluation tools, namely, absolute error, root mean square error and determination coefficient, to measure the degree of deviation between the predicted value and the actual value, quantify the dispersion of the prediction error, and reveal the advantages of the model prediction ability over random prediction, thereby comprehensively evaluating and optimizing the model performance, ensuring the accuracy and reliability of the prediction results, and finally locking in the optimal prediction results.

[0042] In the second aspect, the present invention provides a tunnel fire warning system based on gas analysis, the system includes an input device, a processor, an output device and a memory, the input device, the processor, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, the processor is configured to call the program instructions, and the system uses the tunnel fire warning method based on gas analysis. The tunnel fire warning system based on gas analysis provided by the present invention has a compact structure and stable performance, uses advanced algorithms to analyze data, and immediately triggers a warning once an abnormality is found, thereby realizing early detection, early reporting and early processing of fires, which can not only effectively prevent fires, but also respond quickly in the early stages of fires, ensure the safety of people and vehicles in tunnels, and improve the level of intelligent tunnel management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A logic flow chart of a tunnel fire early warning method based on gas analysis according to an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the structure of a tunnel fire warning system based on gas analysis according to an embodiment of the present invention;

[0045] Figure 3 The figure is an algorithm flow chart of a tunnel fire warning system based on gas analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described herein are only for illustration and are not intended to limit the present invention. In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it is obvious to those of ordinary skill in the art that these specific details do not need to be adopted to implement the present invention. In other examples, in order to avoid confusing the present invention, known circuits, software or methods are not specifically described.

[0047] Throughout the specification, references to "one embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment of the present invention. Therefore, the phrases "in one embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily all refer to the same embodiment or example. In addition, particular features, structures, or characteristics may be combined in one or more embodiments or examples in any suitable combination and / or subcombination. In addition, it should be understood by those of ordinary skill in the art that the figures provided herein are for illustrative purposes and that the figures are not necessarily drawn to scale.

[0048] See also Figure 1 , an embodiment of the present invention provides a tunnel fire early warning method based on gas analysis, the method comprising the following steps:

[0049] S1. Acquire multi-dimensional target data of tunnel gas;

[0050] Among them, step S1 includes the following steps:

[0051] S11. Obtain target data of tunnel gas under normal conditions and fire conditions;

[0052] Regarding step S11, in one embodiment, the fire simulation software FDS is used to simulate the fire, and obtain target data of tunnel gas under normal conditions and under fire conditions.

[0053] Specifically, under normal conditions and fire conditions, record the fire characteristic gases carbon monoxide (CO), carbon dioxide (CO 2 ), sulfur dioxide (SO 2 ), hydrogen sulfide (HS), hydrogen (H 2 ), and methane (CH 4 ) and other gas concentration information.

[0054] S12. Preprocessing the target data to obtain multi-dimensional target data of tunnel gas;

[0055] Wherein, step S12 further comprises the following steps:

[0056] S121. Performing a first data processing on the target data;

[0057] In one embodiment, the first data processing is performed using an outlier elimination algorithm, and the outlier elimination algorithm satisfies the following relationship:

[0058] ,

[0059] in, Indicates the normal value range of the data. Represents the average value of any gas characteristic parameter, Represents standard deviation; the normal value range of data refers to the normal fluctuation range of gas multidimensional data or its data characteristics determined by statistical methods. Using the normal value range of data as a reference standard for data is conducive to improving the quality and accuracy of tunnel gas multidimensional data. Different gas characteristic parameters of the multidimensional data set have unique variability, which can be quantified by calculating the standard deviation of the characteristic parameters. The standard deviation is an important indicator to measure the degree of data dispersion. Based on each column of data, it can reflect the degree of deviation between the value of any column of data and its average value.

[0060] In this embodiment, the expression satisfied by the algorithm for removing outliers is first defined using MATLAB software;

[0061] Furthermore, the multi-dimensional target data of tunnel gas obtained by using the fire simulation software FDS was imported into MATLAB;

[0062] Further, data elimination is judged according to the result of the expression;

[0063] Finally, the data outside the result range is eliminated to obtain the result of the first data processing.

[0064] S122. Performing a second data processing on the target data;

[0065] In one embodiment, the second data processing is performed using an interpolation missing value algorithm, and the interpolation missing value algorithm satisfies the following relationship:

[0066] ,

[0067] in, Indicates the pending The Gas parameter data, Indicates The Gas parameter data, Indicates The Gas parameter data, The dimension number representing the gas parameter; in an optional embodiment, the missing value is filled with any of the following data values:

[0068] 1. The average of other non-missing values ​​in the data;

[0069] 2. The median in the data;

[0070] 3. The mode in the data, that is, the value that occurs most often.

[0071] In this embodiment, the expression satisfied by the missing value interpolation algorithm is first defined using MATLAB software;

[0072] Further, the obtained result data after the first data processing is imported into MATLAB;

[0073] Further, the result data is substituted into the expression satisfied by the missing value interpolation algorithm to perform calculation to obtain the missing value interpolation;

[0074] Finally, the interpolation missing values ​​are interpolated into the result data to obtain the result of the second data processing.

[0075] In practical applications, interpolating missing values ​​can solve the problem of data integrity, but it may also introduce certain biases or uncertainties. Therefore, when choosing an interpolation method, the characteristics of the data and the analysis requirements should be considered to ensure the reliability of the data processing results.

[0076] S123. Obtain multi-dimensional target data of tunnel gas based on the result of the first data processing and the result of the second data processing.

[0077] In one embodiment, a data set is created using MATLAB software, and the data set includes results obtained after the first data processing and the second data processing.

[0078] S2. Based on the multi-dimensional target data, an optimized prediction model for tunnel fire is constructed.

[0079] Among them, step S2 includes the following steps:

[0080] S21. Based on the multidimensional target data, establishing an algorithm-optimized model hidden layer structure;

[0081] In one embodiment, an algorithm-optimized model hidden layer structure is established, wherein the model hidden layer structure includes a convolution hidden layer structure, an activation hidden layer structure, and a pooling hidden layer structure, and the convolution hidden layer structure satisfies the following expression:

[0082] ,

[0083] in, represents the convolution hidden layer structure function, represents the weight, represents the convolution operation, The input of the representation layer, represents the bias, Represents the activation function; the convolutional hidden layer structure is used to extract local features of the input data.

[0084] Furthermore, the activated hidden layer structure satisfies the following expression:

[0085] ,

[0086] in, represents the activation of the hidden layer structure function, represents the activation function, Represents the feature function; uses the activation hidden layer structure to increase nonlinearity, thereby enhancing the feature expression.

[0087] Furthermore, the pooled hidden layer structure satisfies the following expression:

[0088] ,

[0089] in, represents the pooling hidden layer structure function, represents the pooling function, represents the weight matrix, represents the bias term, represents the characteristic function, Represents the total number of elements. The pooling hidden layer structure is used to extract key information and reduce the dimension, thereby reducing parameters.

[0090] S22. Based on the hidden layer structure of the model, a self-attention mechanism is set;

[0091] In one embodiment, a bidirectional long short-term memory network (BiLSTM) is used to deeply analyze the temporal characteristics of sequence data, integrate temporal and spatial characteristics, form a comprehensive understanding of the data, improve the computational efficiency and generalization ability of the model, reduce the number of parameters to reduce the risk of overfitting, and enhance the model's ability to capture and predict complex data structures.

[0092] Furthermore, a self-attention mechanism is set, and the self-attention mechanism satisfies the following expression:

[0093] ,

[0094] in, , Both represent learning matrices, represents the query vector, Indicates A coding sequence, represents the similarity scoring function;

[0095] It should be understood that BiLSTM is a variant of the traditional long short-term memory network (LSTM), which takes into account past and future information while processing sequence data, which enables BiLSTM to capture long-term dependencies in time series data. In addition, BiLSTM consists of two LSTM layers, which traverse the input sequence in forward and reverse directions respectively. The forward LSTM layer processes the temporal information from the beginning to the end of the sequence, while the reverse LSTM layer processes from the end of the sequence to the beginning, capturing the reverse temporal dependency. The information in both directions is finally merged before the output layer, allowing the model to understand the data characteristics more comprehensively.

[0096] S23. Based on the self-attention mechanism, an optimized prediction model for tunnel fire is constructed.

[0097] In one embodiment, an optimized prediction model for tunnel fire is constructed, and the optimized prediction model satisfies the following expression:

[0098] ,

[0099] in, represents the normalized weight of the similarity scoring function, The coding sequence is y The normalized function weight of The coding sequence is y The normalized function deviation of represents the normalization function, Indicates attention value ， Represents a time series. In the process of building a model using the self-attention mechanism, it is necessary to capture the long-distance dependencies in the sequence data. In the self-attention mechanism, each data point calculates the relationship or similarity with all other points to generate an attention weight, so that the model can adaptively focus on the key information in the input data, thereby improving the accuracy of the prediction and the generalization ability of the model.

[0100] S3. Obtain the optimal prediction model through the optimization prediction model.

[0101] Among them, step S3 includes the following steps:

[0102] S31. Obtaining optimal network parameters through the optimization prediction model;

[0103] In one embodiment, the optimization prediction model is trained using the multi-dimensional target data of the tunnel gas, and the data training includes multiple iterative operations.

[0104] Further, the predicted result of each data training is compared with the actual result;

[0105] Furthermore, a set of data whose predicted results are most consistent with the actual results is continuously selected;

[0106] Furthermore, the set of data that best matches the actual results is converted into expression parameters in the optimized prediction model, and the expression parameters are used as optimal network parameters, and the optimal network parameters include but are not limited to normalized function deviation and training attention value.

[0107] S32. Obtain an optimal prediction model based on the optimal network parameters.

[0108] In one embodiment, an optimal prediction model is constructed using optimal network parameters obtained through data training.

[0109] Specifically, the optimal network parameters obtained through data training are substituted into the expression corresponding to the optimized prediction model. The expression is the expression corresponding to the optimal prediction model, that is, the optimal prediction model is successfully constructed.

[0110] S4. Utilize the optimal prediction model to achieve the best early warning effect for tunnel fire.

[0111] Wherein, step S4 includes the following steps:

[0112] S41. Using the optimal prediction model, perform model training of multi-dimensional target data;

[0113] In one embodiment, the processed data is divided into a training set and a test set, and the training set data is used to train the optimal prediction model.

[0114] During the training process, the model parameters are adjusted and cross-validation and other techniques are used to optimize the model performance and reduce the risk of overfitting or underfitting.

[0115] S42. Obtaining the optimal prediction result by processing and analyzing the results of the model training;

[0116] Wherein, step S42 further comprises the following steps:

[0117] S421. Obtain the absolute error of the result by substituting the result of the model training into the absolute error expression, and the absolute error expression is as follows:

[0118] ,

[0119] in, represents the true value of concentration, represents the predicted concentration value;

[0120] S422. Substituting the result of the model training into the root mean square error expression, the root mean square error of the result is obtained, and the root mean square error expression is as follows:

[0121] ,

[0122] in, represents the true value of concentration, represents the predicted concentration value;

[0123] S423. Substituting the result of the model training into the determination coefficient expression, the determination coefficient of the result is obtained, and the determination coefficient expression is as follows:

[0124] ,

[0125] in, represents the true value of concentration, represents the predicted concentration value, represents the average of the true values;

[0126] S424. Obtain the optimal prediction result based on the analysis of the absolute error, the root mean square error and the determination coefficient.

[0127] When the absolute error is closest to 0, the prediction result is a better prediction result;

[0128] When the root mean square error is closest to 0, the prediction result is a better prediction result;

[0129] When the determination coefficient is closest to 1, the prediction result is a better prediction result;

[0130] When the above three conditions are met at the same time, the prediction result is the optimal prediction result.

[0131] In an optional embodiment, two prediction models, namely, a traditional long short-term memory network LSTM and a bidirectional long short-term memory algorithm CNN-BiLSTM-Attention, are trained using simulated data;

[0132] Furthermore, the absolute error MAE , Root Mean Square Error RMSE and the coefficient of determination R 2 , as shown in Table 1.

[0133] Table 1 Prediction results of two prediction models

[0134]

[0135] According to Table 1, the prediction results of the LSTM prediction model and the CNN-BiLSTM-Attention prediction model are analyzed as follows:

[0136] (1) CNN-BiLSTM-Attention compared to LSTM MAE The value is smaller, so the prediction result of the CNN-BiLSTM-Attention prediction model is better.

[0137] (2) CNN-BiLSTM-Attention compared to LSTM RMSE The value is smaller, so the prediction result of the CNN-BiLSTM-Attention prediction model is better.

[0138] (3) CNN-BiLSTM-Attention compared to LSTM R 2 The value is larger, so the prediction result of the CNN-BiLSTM-Attention prediction model is better.

[0139] In summary, the CNN-BiLSTM-Attention prediction model used in the present invention is more accurate in prediction results than the traditional LSTM prediction model and is easier to train to obtain the optimal prediction results.

[0140] S43. Based on the optimal prediction result, the best early warning effect for tunnel fire is achieved.

[0141] In this embodiment, by analyzing MAE , RMSE and R 2 The optimal prediction result is determined, and the optimal prediction result is closest to the actual situation, which is easy to achieve the best early warning effect for tunnel fire.

[0142] See also Figure 2 , Figure 2 FIG. 1 is a schematic diagram of the structure of a tunnel fire warning system based on gas analysis in an embodiment of the present invention. The tunnel fire warning system includes an input device, a processor, an output device and a memory. The input device, the processor, the output device and the memory are interconnected. The memory is used to store a computer program. The computer program includes program instructions. The processor is configured to call the program instructions. The system uses the tunnel fire warning method based on gas analysis. The algorithm flow of the system is shown in FIG. Figure 3 .

[0143] In this embodiment, the input device uses a touch screen, which is mainly used to input multi-dimensional target data of tunnel gas, including concentration data of normal conditions and fire conditions;

[0144] The processor has built-in model algorithms, including an outlier removal algorithm, a missing value interpolation algorithm, a model hidden layer structure algorithm, a self-attention mechanism algorithm, and a normalization processing algorithm, for calculating model expressions for data input by an input device;

[0145] The output device uses a display screen, which is mainly used to display the prediction results obtained by the processor calculation;

[0146] The memory adopts a high-speed solid-state hard disk to store the original data input by the input device and the result data processed by the processor. The memory has the characteristics of fast reading and writing speed, large capacity and high reliability, and can meet the needs of large data storage.

[0147] In summary, in view of the defects of existing tunnel fire gas analysis and prediction methods and the shortcomings in practical applications, such as possible false alarms, missed alarms and insensitivity to early signs of fire, the present invention improves the tunnel fire prediction capability, solves the limitations and inefficiency problems caused by different combustion products, improves the efficiency of tunnel fire gas prediction schemes, and has high industrial utilization value.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A tunnel fire early warning method based on gas analysis, characterized in that: The method comprises the following steps: Acquire multidimensional target data of tunnel gas under normal conditions and under fire conditions; the acquiring multidimensional target data of tunnel gas under normal conditions and under fire conditions comprises: acquiring target data of tunnel gas under normal conditions and under fire conditions; preprocessing the target data to acquire multidimensional target data of tunnel gas; Based on the multi-dimensional target data, an optimized prediction model for tunnel fire is constructed. The optimized prediction model for tunnel fire based on the multi-dimensional target data comprises: based on the multi-dimensional target data, a model hidden layer structure optimized by an algorithm is established; the model hidden layer structure comprises a convolution hidden layer structure, an activation hidden layer structure and a pooling hidden layer structure; the convolution hidden layer structure satisfies the following expression: in, represents the convolution hidden layer structure function, represents the weight, represents the convolution operation, The input of the representation layer, Indicates bias, represents the activation function; the activation hidden layer structure satisfies the following expression: in, represents the activation of the hidden layer structure function, represents the activation function, represents the characteristic function; the pooled hidden layer structure satisfies the following expression: in, represents the pooling hidden layer structure function, represents the pooling function, represents the weight matrix, represents the bias term, represents the characteristic function, Indicates the total number of elements; Based on the hidden layer structure of the model, a self-attention mechanism is set; the self-attention mechanism satisfies the following expression: in, , Both represent learning matrices, represents the query vector, Indicates A coding sequence, represents the similarity scoring function; According to the self-attention mechanism, an optimized prediction model for tunnel fire is constructed; the optimized prediction model satisfies the following expression: in, represents the normalized weight of the similarity scoring function, The coding sequence is The normalized function weight of The coding sequence is The normalized function deviation of represents the normalization function, Indicates the attention value, represents a time series; Obtaining an optimal prediction model through the optimization prediction model; The optimal prediction model is used to achieve the best early warning effect for tunnel fire; the use of the optimal prediction model to achieve the best early warning effect for tunnel fire includes: using the optimal prediction model to perform model training of multi-dimensional target data; obtaining the optimal prediction result by processing and analyzing the results of the model training; the processing and analyzing the results of the model training to obtain the optimal prediction result includes: substituting the results of the model training into an absolute error expression to obtain the absolute error of the result, and the absolute error expression is as follows: in, Indicates the true value of concentration, represents the predicted concentration value; the root mean square error of the result is obtained by substituting the result of the model training into the root mean square error expression, and the root mean square error expression is as follows: in, Indicates the true value of concentration, represents the concentration prediction value; the determination coefficient of the result is obtained by substituting the result of the model training into the determination coefficient expression, and the determination coefficient expression is as follows: in, Indicates the true value of concentration, represents the predicted concentration value, represents the average value of the true value of the concentration; and obtaining the optimal prediction result by analyzing the absolute error, the root mean square error and the determination coefficient; According to the optimal prediction result, the best early warning effect for tunnel fire is achieved.

2. A tunnel fire early warning method based on gas analysis according to claim 1, characterized in that: The preprocessing of the target data to obtain multi-dimensional target data of the tunnel gas comprises: The target data is subjected to a first data processing, wherein the first data processing includes data processing using an outlier elimination algorithm, and the outlier elimination algorithm satisfies the following relationship: in, Indicates the normal value range of the data. Represents the average value of any gas characteristic parameter, represents standard deviation; The target data is subjected to a second data processing, wherein the second data processing includes data processing using an interpolation missing value algorithm, and the interpolation missing value algorithm satisfies the following relationship: in, Indicates the pending The Gas parameter data, Indicates The Gas parameter data, Indicates The Gas parameter data, The dimension number representing the gas parameter; According to the result of the first data processing and the result of the second data processing, multi-dimensional target data of the tunnel gas is obtained.

3. A tunnel fire early warning method based on gas analysis according to claim 1, characterized in that: The step of obtaining the optimal prediction model by optimizing the prediction model comprises: Obtaining optimal network parameters through the optimization prediction model; According to the optimal network parameters, an optimal prediction model is obtained.

4. A tunnel fire warning system based on gas analysis, the system using the tunnel fire warning method based on gas analysis according to any one of claims 1 to 3, characterized in that: The system includes an input device, a processor, an output device and a memory, wherein the input device, the processor, the output device and the memory are connected to each other, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions.

Images