Neural network fire trend prediction method based on improved GWO

Abstract

The invention relates to the technical field of fire prediction, in particular to an improved GWO-based neural network fire trend prediction method, which comprises the steps of collecting fire detection parameters; the detection parameters are preprocessed; establishing a fitness function according to a mean square error of a predicted value and an actual measurement value of the detection parameter; initializing a hyper-parameter and a fitness function of the deep learning network model, and optimizing the hyper-parameter of the deep learning network model by using an improved GWO algorithm; and inputting fire detection parameters collected in real time into the improved deep learning network model to predict the fire. The problems that an existing fire treatment method is low in treatment speed and poor in accuracy are solved.

Description

Technical Field

[0001] The invention relates to the technical field of fire prediction, and in particular to a neural network fire trend prediction method based on improved GWO. Background Art

[0002] Fire is one of the disasters that seriously threaten human life and property safety in society. With the acceleration of urbanization, the increase in building density and complexity, the probability and potential hazards of fire are also increasing; therefore, the research and development of fire prediction technology has important practical significance.

[0003] Traditional fire prevention mainly relies on fire prevention measures and manual monitoring, but these methods are often insufficient to cope with the suddenness and complexity of fires.

[0004] Li Jie et al. predicted the spread of cabin fire based on deep neural networks and trend feature vectors. They used the LSTM deep neural network model combined with the trend feature vector algorithm to predict the trend and temperature distribution of fire smoke flow. However, this method had the problems of slow data processing speed and insufficient accuracy.

[0005] The fire development trend prediction method with publication number CN111489006A uses a deep learning network to predict the fire development trend. This method also has the problem of slow processing speed when processing real-time fire data. Summary of the invention

[0006] Aiming at the deficiencies of the existing methods, the present invention solves the problems of slow processing speed and poor accuracy in the existing fire processing methods.

[0007] The technical solution adopted by the present invention is: the neural network fire trend prediction method based on improved GWO includes the following steps:

[0008] Step 1: Collect fire detection parameters;

[0009] As a preferred embodiment of the present invention, the detection parameters include: temperature distribution and CO concentration diffusion.

[0010] Step 2: preprocess the detection parameters;

[0011] As a preferred implementation of the present invention, the preprocessing includes: normalization processing.

[0012] Step 3: Establish a fitness function based on the mean square error between the predicted value and the actual measured value of the detection parameter;

[0013] Step 4: Initialize the hyperparameters and fitness function of the deep learning network model, and use the improved GWO algorithm to optimize the hyperparameters of the deep learning network model;

[0014] As a preferred implementation of the present invention, the hyperparameters include: the number of hidden layer nodes, the number of training times and the learning rate.

[0015] As a preferred implementation of the present invention, the deep learning network model is an LSTM network.

[0016] As a preferred embodiment of the present invention, the improved GWO algorithm includes:

[0017] Set the wolf population, maximum number of iterations, convergence factor, and position update weight;

[0018] Initialize the population;

[0019] As a preferred implementation of the present invention, logistic chaotic mapping is used to determine the initial position of the gray wolf population.

[0020] Construct an improved adaptive convergence factor, the formula is:

[0021]

[0022] Among them, a min and a max Respectively represent the minimum and maximum values ​​of the convergence factor; F i 、F min 、F max and F avg Respectively represent the current group fitness value, minimum fitness value, maximum fitness value and average fitness value;

[0023] Calculate the distance D between the current solution position and the three wolves α, β, δ α , D β , D δ , and then according to D α , D β , D δ Update wolfω position;

[0024] The positions of the three wolves are weighted by the fitness ratios among wolves α, β, and δ, and the positions are updated using the fitness values;

[0025] As a preferred embodiment of the present invention, the formula for assigning weights to the positions of the three wolves is:

[0026]

[0027] Among them, F α 、F β and F δ Represent the fitness values ​​of wolves α, β and δ respectively.

[0028] As a preferred embodiment of the present invention, the formula for updating the position using the fitness value is:

[0029] X1=X α -A1·D α

[0030] X2=X β -A2·D β

[0031] X3=X δ -A3·D δ

[0032]

[0033] Among them, w1, w2, w3 represent the weights of wolves α, β, and δ respectively; X1, X2, X3 represent the updated positions of wolves α, β, and δ respectively; X α , X β , X δ They represent the positions of wolves α, β, and δ before the update respectively; X(t+1) represents the position of the gray wolf after the update, and A1 to A3 represent the vector coefficients of wolves α, β, and δ.

[0034] As a preferred implementation manner of the present invention, it also includes: introducing Levy flight to update the position of the gray wolf.

[0035] Update the fitness value and the optimal solution, and stop iterative updating when the optimal solution is met or the maximum number of iterations is reached.

[0036] Step 5: Input the real-time collected fire detection parameters into the improved deep learning network model to predict the fire.

[0037] Beneficial effects of the present invention

[0038] 1. Improve the convergence factor of the GWO algorithm, improve the search accuracy and speed of the GWO algorithm, enhance the global search capability, and avoid falling into the local optimal solution;

[0039] 2. Fire prediction requires real-time data processing capabilities to quickly respond to fires. The traditional GWO algorithm processes fire data slowly. The adaptive convergence factor in the GWO algorithm is improved to improve the real-time performance of fire prediction.

[0040] 3. By setting different weights for wolves α, β, and δ, the position update of the gray wolf is adjusted, and the global search ability and local development ability of the algorithm are enhanced, which helps to accurately control the wolf pack to approach the optimal solution and improve the accuracy of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the neural network fire trend prediction method based on improved GWO of the present invention;

[0042] Figure 2 The improved GWO algorithm flow chart of the present invention;

[0043] Figure 3 This is a comparison chart of the prediction effects of the present invention and the existing method;

[0044] Figure 4 A comparison diagram of the prediction effects of the improved convergence factor of the present invention and the convergence factor of the traditional GWO;

[0045] Figure 5 It is a comparison diagram of convergence iteration curves of the improved convergence factor of the present invention and the convergence factor of the traditional GWO. DETAILED DESCRIPTION

[0046] The present invention is further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore it only shows the components related to the present invention.

[0047] like Figure 1 As shown, the neural network fire trend prediction method based on improved GWO includes the following steps:

[0048] Step 1: Construct a data set of temperature distribution and CO concentration diffusion when a fire occurs;

[0049] The detection parameters include: temperature distribution and CO concentration diffusion.

[0050] Dataset source: The fire smoke movement standard dataset provided by the National Institute of Standards and Technology (NIST) of the United States is used;

[0051] Collection method: Temperature distribution and CO concentration diffusion parameters are collected through temperature and smoke collectors.

[0052] Step 2: Preprocess the detection parameter data to reduce the error in the data processing process. The preprocessing includes: normalizing the data. The formula is:

[0053]

[0054] Among them, x new represents the normalized data, x represents the original data, and x min 、x max Respectively represent the minimum and maximum values ​​in the data set;

[0055] Step 3: Divide the preprocessed data into training sets and test sets, build a deep learning network model, set model parameters, and set the fitness function; use the mean square error (MSE) between the actual temperature / CO concentration and the predicted temperature / CO concentration to establish the fitness function. The formula for the mean square error is:

[0056]

[0057] in, Represents the predicted value, y i Indicates actual value.

[0058] Step 4: Initialize the deep learning network model parameters (hyperparameters and fitness function), and use the improved GWO algorithm to optimize the hyperparameters of the deep learning network; the optimized hyperparameters include the number of hidden layer nodes, the number of training times, and the learning rate;

[0059] The deep learning network model adopts the long short-term memory network LSTM. The present invention adopts a standard LSTM network, and the structure of the LSTM network is not described again.

[0060] Figure 2 Flowchart of the improved GWO algorithm of the present invention, the specific operation steps are as follows:

[0061] 1. Set the gray wolf population, maximum number of iterations, convergence factor and position update weight;

[0062] 2. The traditional GWO algorithm uses random initialization of the population, which cannot guarantee population diversity. The logistic chaotic mapping has the characteristics of randomness. Therefore, the logistic chaotic mapping is used to initialize the initial position of the gray wolf population. The logistic chaotic mapping formula is:

[0063] x(i)=4*x(i-1)*(1-x(i-1)) (1)

[0064] Among them, x(i-1) represents the position value of the gray wolf at the i-1th moment, and x(i) represents the position value of the gray wolf at the i-th moment.

[0065] 3. In order to improve the convergence speed and global search capability of GWO, the adaptive convergence factor a is improved i The formula is:

[0066]

[0067] Among them, a min and a max They represent the minimum and maximum values ​​of the convergence factor, respectively. In this embodiment, the minimum and maximum values ​​are 0 and 2; F i 、F min 、F max and F avg They respectively represent the current group fitness value, minimum fitness value, maximum fitness value and average fitness value.

[0068] The convergence factor in the traditional grey wolf algorithm converges slowly on complex problems and is prone to fall into local optimum. However, the present invention dynamically adjusts the convergence factor to prevent it from falling into local optimum, thereby improving the global search capability.

[0069] 4. Calculate the distance D between the current solution position and the three wolves α, β, and δ α , D β , D δ , and then according to D α , D β , D δ Update the position of wolf ω, the position adjustment formula is as follows:

[0070]

[0071] Among them, X α , X β , X δ Respectively represent the positions of α, β, and δ wolves; C1, C2, and C3 are random values; X represents the position of the current solution.

[0072] 5. Update the position of the fitness value. The fitness ratios of wolves α, β, and δ are used to assign weights w1, w2, and w3 to the positions of the three wolves, and then the position of the prey is solved. The weight formula is:

[0073]

[0074] Among them, w1, w2, w3 represent the wolf α, β and δ weights respectively; F α 、F β and F δ Respectively represent the fitness values ​​of wolves α, β, and δ; X1, X2, and X3 represent the updated positions of wolves α, β, and δ; X α , X β , X δ Respectively represent the positions of wolves α, β, and δ before the update; X(t+1) represents the position of the gray wolf after the update, and A1 to A3 represent the vector coefficients of wolves α, β, and δ; D α is the distance between wolf α and other gray wolves, D β is the distance between wolf β and other gray wolves, D δ is the distance between wolf δ and other gray wolf individuals.

[0075] By setting different weights for wolves α, β, and δ, the position update of the gray wolf is adjusted, which enhances the global search capability and local development capability of the algorithm, and helps to accurately control the wolf pack to approach the optimal solution, thereby improving the accuracy of the solution.

[0076] 6. Introduce Levy flight to update the position of the gray wolf and enhance the global search capability. The formula is:

[0077]

[0078] Among them, X * (t) represents the updated location information of the gray wolf, X(t) represents the current location information of the gray wolf; x is the stability index parameter of the Levy distribution, that is, the convergence factor a i .

[0079] 7. Update the fitness value and the optimal solution. Stop iterative updating when the optimal solution is met or the maximum number of iterations is reached.

[0080] Step 5: Construct an improved deep learning network, i.e., a prediction model for temperature and CO concentration, and determine the number of hidden layer nodes, number of training times, and learning rate; update the hyperparameters of the deep learning network by improving the GWO algorithm to obtain the optimal solution, and substitute the optimized hyperparameters (number of hidden layer nodes, number of training times, and learning rate) into the deep learning network model;

[0081] The improved GWO optimized deep learning network model is used to predict temperature and CO concentration; and the prediction results are combined with the path optimization algorithm to generate the optimal escape route.

[0082] Figure 3 The figure is a comparison chart of the effects of the method of the present invention and the existing methods. In order to uniformly evaluate the performance of the method of the present invention and the comparative methods, the method of the present invention (IGWO) is compared with ALGWO, LGWO and GWO using the following evaluation indicators: mean absolute error (MAE), mean absolute percentage error (MAPE) and R-squared value (R 2 ); The results show that IGWO has higher accuracy than other algorithms; the indicator formula is:

[0083]

[0084]

[0085] like Figure 4 The improved convergence factor of the algorithm of the present invention is compared with the evaluation effect of the traditional GWO algorithm. The evaluation indicators used by the method of the present invention (IGWO algorithm) and the GWO algorithm are the mean absolute error (MAE), mean absolute percentage error (MAPE) and R square value (R 2 ).

[0086] like Figure 5 This is a comparison diagram of the convergence iteration curves of the IGWO of the present invention and the traditional GWO. The traditional GWO algorithm processes fire data slowly. The present invention improves the convergence speed of the algorithm by improving the adaptive convergence factor in the Grey Wolf Algorithm.

[0087] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. The neural network fire trend prediction method based on improved GWO is characterized by: The following steps are involved: Step 1: Collect fire detection parameters; Step 2: preprocess the detection parameters; Step 3: Establish a fitness function based on the mean square error between the predicted value and the actual measured value of the detection parameter; Step 4: Initialize the hyperparameters and fitness function of the deep learning network model, and use the improved GWO algorithm to optimize the hyperparameters of the deep learning network model; Step 5: Input the real-time collected fire detection parameters into the improved deep learning network model to predict the fire.

2. The neural network fire trend prediction method based on improved GWO according to claim 1 is characterized in that: Improvements to the GWO algorithm include: Set the wolf population, maximum number of iterations, convergence factor, and position update weight; Initialize the population; Construct an improved adaptive convergence factor, the formula is: Among them, a min and a max Respectively represent the minimum and maximum values ​​of the convergence factor; F i 、F min 、F max and F avg Respectively represent the current group fitness value, minimum fitness value, maximum fitness value and average fitness value; Calculate the distance D between the current solution position and the three wolves α, β, δ α , D β , D δ , and then according to D α , D β , D δ Update wolfω position; The positions of the three wolves are weighted by the fitness ratios among wolves α, β, and δ, and the positions are updated using the fitness values; Update the fitness value and the optimal solution, and stop iterative updating when the optimal solution is met or the maximum number of iterations is reached.

3. The neural network fire trend prediction method based on improved GWO according to claim 2 is characterized in that: Hyperparameters include: The number of hidden layer nodes, the number of training times, and the learning rate.

4. The neural network fire trend prediction method based on improved GWO according to claim 2 is characterized in that: The initial position of the gray wolf population is determined using logistic chaos mapping.

5. The neural network fire trend prediction method based on improved GWO according to claim 2 is characterized in that: The formula for assigning weights to the positions of the three wolves is: Among them, F α 、F β and F δ Represent the fitness values ​​of wolves α, β and δ respectively.

6. The neural network fire trend prediction method based on improved GWO according to claim 2 is characterized in that: The formula for position update using fitness value is: Among them, w1, w2, w3 represent the weights of wolves α, β, and δ respectively; X1, X2, X3 represent the updated positions of wolves α, β, and δ respectively; X α , X β , X δ They represent the positions of wolves α, β, and δ before the update respectively; X(t+1) represents the position of the gray wolf after the update, and A1 to A3 represent the vector coefficients of wolves α, β, and δ.

7. The neural network fire trend prediction method based on improved GWO according to claim 2 is characterized in that: Also includes: Introduced Levi Flight to update the position of the gray wolf.

8. The neural network fire trend prediction method based on improved GWO according to claim 2 is characterized in that: The deep learning network model is an LSTM network.

9. The neural network fire trend prediction method based on improved GWO according to claim 1 is characterized in that: The detection parameters include: temperature distribution and CO concentration diffusion.

10. The neural network fire trend prediction method based on improved GWO according to claim 1 is characterized in that: Preprocessing includes: normalization processing.

Images