Absolute gas emission quantity prediction model based on quantum evolution particle swarm

Abstract

The invention provides an absolute gas emission quantity prediction model based on a quantum evolution particle swarm, and belongs to the technical field of gas emission quantity prediction.The absolute gas emission quantity prediction model comprises the steps that a time sequence of gas emission quantity related influence factors is reconstructed through a multi-phase space reconstruction method and serves as input of the prediction model; using a quantum particle swarm optimization algorithm to optimize hyper-parameters of the prediction model; establishing an absolute gas emission quantity prediction model; performing quantum hyper-parameter optimization on the absolute gas emission quantity prediction model; checking the established absolute gas emission quantity prediction model; multiple time series reconstruction is performed through multiple absolute gas emission quantity related influence factors, hyper-parameter optimization is performed on an SVM model based on a radial basis function by using a quantum particle swarm algorithm to establish an absolute gas emission quantity prediction model, and the absolute gas emission quantity prediction model has better convergence and higher prediction precision. And the accuracy of predicting the absolute gas emission quantity under the mine is effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of gas emission prediction, and in particular relates to an absolute gas emission prediction model based on quantum evolution particle swarm. Background Art

[0002] Gas enrichment in mines is one of the important disasters in coal mine production. Therefore, establishing an absolute gas emission prediction model using multiple relevant influencing factors of absolute gas quantity can effectively prevent such gas disasters.

[0003] At present, relevant scholars at home and abroad have proposed many prediction methods combining neural networks and support vector machines for the problem of absolute gas outburst. For example, an improved gravitational search algorithm (IGSA) is used to train and optimize the initial weights and thresholds of the BP neural network, which improves the prediction accuracy and convergence speed of the model; a method for predicting the danger of coal and gas outburst in tunnels using a BP neural network; the historical data of gas emission is decomposed using the Local Mean Decomposition (LMD) method to obtain PF components, and each component is extrapolated using the Support Vector Machines (SVM) function fitting method; the gas emission time series data is decomposed using the Empirical Mode Decomposition (EMD) method to obtain multiple intrinsic mode functions (IMFs), and an SVM prediction model optimized by the Particle Swarm Optimization (PSO) is established for each IMF; the absolute gas emission is decomposed into several intrinsic mode components by the Variational Modal Decomposition (VMD) method and their local characteristics are analyzed, and a Relevance Vector Machine (Relevance Vector Machine) is established according to each intrinsic mode component. Machine, RVM), and optimize the model parameters through differential evolution (DE) to improve the prediction accuracy; a weighted improved Least Squares Support Vector Machines (LSSVM) prediction model, which solves the robustness and sparsity problems that LSSVM itself lacks.

[0004] Since the factors affecting absolute gas emission are complex and changeable and there are nonlinear problems, the problems of difficult adjustment and slow convergence of the above-mentioned support vector machine and deep learning parameters have not been well solved, and the prediction accuracy of the model has not been significantly improved. In recent years, a large number of advanced optimization algorithms and various methods for processing time series have emerged, but they have not yet achieved good results in combination with gas emission prediction. Summary of the invention

[0005] Based on the above technical problems, the present invention provides an absolute gas emission prediction model based on quantum evolved particle swarm, which reconstructs multiple time series through multiple absolute gas emission related influencing factors, and uses quantum particle swarm algorithm to perform hyperparameter optimization of the SVM model based on radial basis function to establish an absolute gas emission prediction model. The absolute gas emission prediction model of the present invention has better convergence and higher prediction accuracy, which effectively improves the accuracy of absolute gas emission prediction in mines.

[0006] The specific technical solution is:

[0007] A prediction model for absolute gas emission based on quantum evolution particle swarm includes the following steps:

[0008] S1: The time series of gas emission related influencing factors are reconstructed using the multiple phase space reconstruction method and used as the input of the prediction model;

[0009] S2: Optimize prediction model hyperparameters using quantum particle swarm algorithm;

[0010] S3: Establish an absolute gas emission prediction model, as follows:

[0011]

[0012] In the formula, the kernel function K(x i ,x)Select REF function;

[0013]

[0014] Where σ 2 is the kernel function width of the absolute gas emission prediction model;

[0015] S4: Quantum hyperparameter optimization of the absolute gas emission prediction model;

[0016] S5: Test the established absolute gas emission prediction model.

[0017] In addition, the absolute gas emission prediction model based on quantum evolution particle swarm in the above technical solution provided by the present invention may also have the following additional technical features:

[0018] In the above technical solution, in step S2, the specific steps are as follows:

[0019] The gas outflow algorithm of quantum evolution particle swarm is as follows:

[0020] v ij (t+1)=ωv ij (t)+

[0021] c1r 1j (t)(p ij (t)-X ij (t))-

[0022] c2r 2j (t)(p gj (t)-X ij (t))

[0023] Where v ij (t) represents the spatial velocity of the i-th particle in the j-dimensional space at time t; c1 and c2 are particle acceleration factors; r 1j (t) and r 2j (t) is a random function;

[0024] Update the next generation of aggregation points based on the aggregation speed and aggregation point information:

[0025] X ij (t+1)=X ij (t)+v ij (t+1)

[0026] Assume that the evolving particles have M particle groups representing potential solutions to the gas outburst problem in the N-dimensional quantum optimization space:

[0027] X i (t) = [X i,1 (t),X i,2 (t),…,X i,N (t)],

[0028] i=1,2,…,M

[0029] The optimal aggregation point pbest of the evolutionary particle individuals is calculated based on the fitness value f[X i (t)]Continuously updated:

[0030]

[0031] Then calculate the average optimal aggregation point Cbest of the particle swarm and the global optimal aggregation point gbest:

[0032]

[0033] A random gas gathering point is calculated for the j dimensions of the evolving particle i:

[0034] p i,j (t) = φ j (t)·P i,j (t)+[1-φ j (t)]·G j (t),φ j (t)∈U(0,1)

[0035] Finally, the new gas gathering point is obtained according to the evolution particle equation:

[0036]

[0037] Where β is the control parameter.

[0038] In the above technical solution, in step S3, the absolute gas emission prediction model is obtained by the following steps:

[0039] Create a model function:

[0040] f(x)=W T X+b, i=1,2,...,n

[0041] Where W T and b are the normal vector and intercept of the model respectively, X is the time series of the reconstruction of the characteristics of the factors affecting the absolute gas emission;

[0042] To minimize the risk, the inequality constraint solution of the regression problem is transformed into an equality constraint solution:

[0043]

[0044] y i [W T φ(x i )+b]=1-e i

[0045] Where, γ is the regularization parameter of LSSVM gas emission model, φ(x i ) is x i The result of mapping in high-dimensional space;

[0046]

[0047] In the formula, the Lagrange multiplier α plays the role of model support;

[0048] Finally, the LSSVM gas emission model function is established:

[0049]

[0050] In the formula, the kernel function K(x i ,x)Select REF function:

[0051]

[0052] Where σ 2 is the kernel function width of the LSSVM gas emission model.

[0053] In the above technical solution, in step S4, the following steps are specifically included:

[0054] S1: Select the average relative error as the objective function:

[0055]

[0056] Where y i is the true value of the test sample of gas-related influencing factors, is the prediction result output after each iteration, and n is the number of training samples;

[0057] S2: Randomly generate N combinations of two hyperparameters and assign initial gas gathering points to the evolving particles:

[0058]

[0059] By calculating the respective fitness values ​​f[X i (0)] and determine the optimal aggregation point of the individual evolving particles and the global optimal clustering point

[0060] S3: Calculate the next generation of hyperparameter combinations

[0061] Repeat the above iterative process to obtain the t generation hyperparameter combination Until the number of iterations reaches the maximum value I max Or the fitness value of each generation does not change, and the optimal combination of parameters is obtained.

[0062] S4: According to the optimal model hyperparameter γ s and A QPSO-LSSVM gas emission prediction model was established, and the model was trained using the reconstructed absolute gas emission data set training samples.

[0063] Compared with the prior art, the absolute gas emission prediction model based on quantum evolution particle swarm of the present invention has the following beneficial effects:

[0064] The present invention reconstructs multiple time series through multiple absolute gas emission related influencing factors, and uses quantum particle swarm algorithm to perform hyperparameter optimization on the SVM model based on radial basis function to establish an absolute gas emission prediction model. The absolute gas emission prediction model of the present invention has better convergence and higher prediction accuracy, which effectively improves the accuracy of absolute gas emission prediction in mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is the waiting effect between particles;

[0066] Figure 2 is a delay time curve diagram;

[0067] Figure 3 is the embedding dimension graph;

[0068] Figure 4 It is the time series diagram of absolute gas emission;

[0069] Figure 5 It is the fitness change curve;

[0070] Figure 6 This is a comparison chart of the prediction results between QPSO-LSSVM and PSO-LSSVM;

[0071] Figure 7 This is a comparison chart of the prediction results between QPSO-LSSVM and IGSA-BP;

[0072] Figure 8 This is a comparison chart of the prediction results of QPSO-LSSVM and ACC-ENN. DETAILED DESCRIPTION

[0073] The following is a combination of specific implementation cases and attached Figure 1-8 The present invention is further described below, but the present invention is not limited to these embodiments.

[0074] A prediction model for absolute gas emission based on quantum evolution particle swarm includes the following steps:

[0075] S1: The time series of gas emission related influencing factors are reconstructed using the multiple phase space reconstruction method and used as the input of the prediction model;

[0076] S2: Optimize prediction model hyperparameters using quantum particle swarm algorithm;

[0077] S3: Establish an absolute gas emission prediction model, as follows:

[0078]

[0079] In the formula, the kernel function K( xi,x)Select the REF function;

[0080]

[0081] Where σ 2 is the kernel function width of the absolute gas emission prediction model;

[0082] S4: Quantum hyperparameter optimization of the absolute gas emission prediction model;

[0083] S5: Test the established absolute gas emission prediction model.

[0084] In an embodiment of the present invention, in step S2, the specific steps are as follows:

[0085] The gas outflow algorithm of quantum evolution particle swarm is as follows:

[0086] v ij (t+1)=ωv ij (t)+

[0087] c1r 1j (t)(p ij (t)-X ij (t))-

[0088] c2r 2j (t)(p gj (t)-X ij (t))

[0089] Where v ij (t) represents the spatial velocity of the i-th particle in the j-dimensional space at time t; c1 and c2 are particle acceleration factors; r 1j (t) and r 2j (t) is a random function;

[0090] Update the next generation of aggregation points based on the aggregation speed and aggregation point information:

[0091] X ij (t+1)=X ij (t)+v ij (t+1)

[0092] Assume that the evolving particles have M particle groups representing potential solutions to the gas outburst problem in the N-dimensional quantum optimization space:

[0093] X i (t) = [X i,1 (t),X i,2 (t),…,X i,N (t)],

[0094] i=1,2,…,M

[0095] The optimal aggregation point pbest of the evolutionary particle individuals is calculated based on the fitness value f[X i (t)]Continuously updated:

[0096]

[0097] Then calculate the average optimal aggregation point Cbest of the particle swarm and the global optimal aggregation point gbest:

[0098]

[0099] A random gas gathering point is calculated for the j dimensions of the evolving particle i:

[0100] p i,j (t) = φ j (t)·P i,j (t)+[1-φ j (t)]·G j (t),φ j (t)∈U(0,1)

[0101] Finally, the new gas gathering point is obtained according to the evolution particle equation:

[0102]

[0103] Where β is the control parameter.

[0104] In the embodiment of the present invention, in step S3, the absolute gas emission prediction model is obtained by the following steps:

[0105] Create a model function:

[0106] f(x)=W T X+b, i=1,2,…,n

[0107] Where W T and b are the normal vector and intercept of the model respectively, X is the time series of the reconstruction of the characteristics of the factors affecting the absolute gas emission;

[0108] To minimize the risk, the inequality constraint solution of the regression problem is transformed into an equality constraint solution:

[0109]

[0110] y i [W T φ(x i )+b]=1-e i

[0111] Where, γ is the regularization parameter of LSSVM gas emission model, φ(xi ) is x i The result of mapping in high-dimensional space;

[0112]

[0113] In the formula, the Lagrange multiplier α plays the role of model support;

[0114] Finally, the LSSVM gas emission model function is established:

[0115]

[0116] In the formula, the kernel function K(x i ,x)Select REF function:

[0117]

[0118] Where σ 2 is the kernel function width of the LSSVM gas emission model.

[0119] In an embodiment of the present invention, step S4 specifically includes the following steps:

[0120] S1: Select the average relative error as the objective function:

[0121]

[0122] Where y i is the true value of the test sample of gas-related influencing factors, is the prediction result output after each iteration, and n is the number of training samples;

[0123] S2: Randomly generate N combinations of two hyperparameters and assign initial gas gathering points to the evolving particles:

[0124]

[0125] By calculating the respective fitness values ​​f[X i (0)] and determine the optimal aggregation point of the individual evolving particles and the global optimal clustering point

[0126] S3: Calculate the next generation of hyperparameter combinations

[0127] Repeat the above iterative process to obtain the t generation hyperparameter combination Until the number of iterations reaches the maximum value I max Or the fitness value of each generation does not change, and the optimal combination of parameters is obtained.

[0128] S4: According to the optimal model hyperparameter γ s and A QPSO-LSSVM gas emission prediction model was established, and the model was trained using the reconstructed absolute gas emission data set training samples.

[0129] Example analysis:

[0130] 1. Time series variable feature reconstruction

[0131] The general influencing factors of absolute gas emission include absolute gas concentration, coal seam thickness, coal seam spacing, coal production volume, advancement speed, etc. There is often a complex nonlinear relationship between these factors. The multiple phase space reconstruction method is used to reconstruct the characteristic time series of various gas emission-related influencing factors.

[0132] Characteristic time series of factors affecting the single absolute gas emission volume [s1, s2, ..., s n ], the reconstructed phase space is:

[0133]

[0134] Where τ is the delay time, d is the embedding dimension, and M n is a phase point in the phase space.

[0135] 2. Gas parameter setting:

[0136] The mutual information method is used to determine the parameter τ, and the method is as follows:

[0137] Assume that there are two sets of nonlinear absolute gas emission related influencing factor characteristic time series [q1,q2,...,q m ]、[s1,s2,...,s n ].

[0138]

[0139] In the formula, H(Q) and H(S) are the entropies of the two groups of sequences, P Q (q i ), P S (s i ) is the absolute gas emission probability of each sequence.

[0140]

[0141] I(Q,S)=H(Q)+H(S)-H(Q,S) (4)

[0142] P Q,S (q i ,s j ) is P Q (qi ) and P S (s i ) is the joint distribution probability. Define (Q,S) = (s i ,s i+τ ), 1<i≤n-τ, I(Q,S) is the correlation function about the delay time τ. The first minimum point of I(Q,S) is the optimal value of the delay time τ.

[0143] The Cao method improved by the false nearest neighbor method is used to determine the parameter d, as follows:

[0144]

[0145] Where E(d) is the criterion for evaluating the time series reconstructed based on known τ and unknown d, and E * (d) Evaluate the correlation between the two sets of sequences. Let E1(d) = E(d+1) / E(d). If d is greater than a certain value d0, E1(d) will no longer change, and d0+1 is the minimum embedding dimension d. Let E2(d) = E * (d+1) / E * (d), the factors affecting the absolute gas emission are random time series, E2(d)≡1.

[0146] 3. Quantum evolution particle swarm gas outflow algorithm:

[0147] The gas outflow algorithm of quantum evolution particle swarm is as follows:

[0148]

[0149] Where v ij (t) represents the spatial velocity of the i-th particle in the j-dimensional space at time t; c1 and c2 are particle acceleration factors; r 1j (t) and r 2j (t) is a random function;

[0150] Update the next generation of aggregation points based on the aggregation speed and aggregation point information:

[0151] X ij (t+1)=X ij (t)+v ij (t+1) (8)

[0152] Since the gas outflow speed and gas gathering point of the particle swarm algorithm are updated slowly and it is easy to fall into the local optimal solution, the quantum behavior strategy evolutionary particle swarm algorithm is adopted. In quantum space, the motion state of the evolving particles is irregular and random, and the quantum system can better describe the nonlinear state of the particles under the action of the state superposition principle. In addition, the quantum behavior strategy introduces the average optimal gathering point C of the particles, and there is an obvious mutual waiting effect between the evolving particles, such as Figure 1 shown.

[0153] Assume that the evolving particles have M particle groups representing potential solutions to the gas outburst problem in the N-dimensional quantum optimization space:

[0154]

[0155] The optimal aggregation point pbest of the evolutionary particle individuals is calculated based on the fitness value f[X i (t)]Continuously updated:

[0156]

[0157] Then calculate the average optimal aggregation point Cbest of the particle swarm and the global optimal aggregation point gbest:

[0158]

[0159] A random gas gathering point is calculated for the j dimensions of the evolving particle i:

[0160] p i,j (t) = φ j (t)·P i,j (t)+[1-φ j (t)]·G j (t),φ j (t)∈U(0,1) (13)

[0161] Finally, the new gas gathering point is obtained according to the evolution particle equation:

[0162]

[0163] Where β is the control parameter, which can generally achieve good results by linearly reducing it from 1.0 to 0.5.

[0164] 4. Establish an absolute gas emission prediction model

[0165] The kernel function of traditional SVM has nonlinear problems, and its operation is slow and complicated. On this basis, RBF is added to transform the quadratic programming problem of SVM into a problem of solving a linear equation system, and the LSSVM gas emission model is established.

[0166] Create a model function:

[0167] f(x)=W T ·X+b,i=1,2,...,n (15)

[0168] Where W T and b are the normal vector and intercept of the model respectively, X is the time series of the reconstruction of the characteristics of the factors affecting the absolute gas emission;

[0169] To minimize the risk, the inequality constraint solution of the regression problem is transformed into an equality constraint solution:

[0170]

[0171] Where, γ is the regularization parameter of LSSVM gas emission model, φ(x i ) is x i The result of mapping in high-dimensional space;

[0172]

[0173] In the formula, the Lagrange multiplier α plays the role of model support;

[0174] Finally, the LSSVM gas emission model function is established:

[0175]

[0176] In the formula, the kernel function K(x i ,x)Select REF function:

[0177]

[0178] Where σ 2 is the kernel function width of the LSSVM gas emission model.

[0179] 5. Quantum hyperparameter optimization of gas emission prediction model

[0180] Two hyperparameter kernel function widths of the established gas emission prediction model and the regularization parameter γ s It plays a vital role in the prediction accuracy.

[0181] S1: Select the average relative error as the objective function:

[0182]

[0183] In formula (21), y i is the true value of the test sample of gas-related influencing factors, is the prediction result output after each iteration, and n is the number of training samples;

[0184] S2: Randomly generate N combinations of two hyperparameters and assign initial gas gathering points to the evolving particles:

[0185]

[0186] Calculate the fitness value f[X i (0)], and determine the optimal aggregation point of the individual evolving particles according to equations (10) and (12): and the global optimal clustering point

[0187] S3: Calculate the next generation of hyperparameter combinations according to equations (13) and (14)

[0188] Repeat the above iterative process to obtain the t generation hyperparameter combination Until the number of iterations reaches the maximum value I max Or the fitness value of each generation does not change, and the optimal combination of parameters is obtained.

[0189] S4: According to the optimal model hyperparameter γ s and A QPSO-LSSVM gas emission prediction model was established, and the model was trained using the reconstructed absolute gas emission data set training samples.

[0190] 6. Experimental studies

[0191] Four models, QPSO-LSSVM, PSO-LSSVM, IGSA-BP and ACC-ENN, were selected for experimental comparative analysis. The obtained absolute gas emission prediction results, average relative error and root mean square error results reflect the effect of the absolute gas emission prediction model established by combining the time series phase space reconstruction method with the QPSO algorithm to optimize the hyperparameters of the LSSVM model. The experimental data selected the historical monitoring data of the influencing factors related to the absolute gas emission in Licun Coal Mine.

[0192] 7. Experimental data selection

[0193] Since there are many factors affecting the absolute gas emission in the mine, the mining depth of coal mine X1 (m), coal seam thickness X2 (m), coal seam inclination X3 (deg), original gas content of mining layer X4 (m3 / t), coal seam spacing X5 (m), advancement speed X6 (m / d), recovery rate X7, daily output X8 (m3), gas content of adjacent layer X9 (m 3 / t), absolute gas outflow rate X10 (m 3 / min) of 1697 sets of data of ten related influencing factors, from which 1500 sets of measured data were randomly selected as research objects, 1200 sets of data were randomly selected as training samples of the absolute gas emission prediction model, and the remaining 300 sets of data were used as test samples of the prediction model. Ten sets of data of related influencing factors are listed, as shown in Table 1.

[0194] Table 1 Partial data of factors affecting absolute gas emission

[0195]

[0196] 8. Experimental Data Processing

[0197] Figure 2 , Figure 3 To solve the curve diagram of delay time τ and embedding dimension d, it can be concluded from the figure that τ is 3 and d is 15.

[0198] The above determined parameters are used to reconstruct the characteristic time series of absolute gas emission variable. The reconstructed time series is as follows: Figure 4 shown.

[0199] 9. Model performance testing

[0200] Determine the optimization interval of the two hyperparameters of the prediction model γ∈[0.1,1000], σ 2 ∈[1,1000], then the quantum particle swarm initialization parameters are configured, and the particle swarm population size P n =50, maximum number of iterations I max =50, control parameter β=0.8. Figure 5 It is the fitness value change curve of QPSO-LSSVM model, PSO-LSSVM model, IGSA-BP model and ACC-ENN model.

[0201] from Figure 5 It can be seen that the QPSO-LSSVM model has achieved global convergence after 17 iterations, while the other three models have slower convergence speeds and their global optimization capabilities are not as good as the QPSO algorithm.

[0202] Comparison of the absolute gas emission prediction results of the four models, QPSO-LSSVM, PSO-LSSVM, IGSA-BP and ACC-ENN. Figure 6 , Figure 7 , Figure 8 .

[0203] The comparison of the above prediction results shows that the absolute gas emission prediction accuracy of the QPSO-LSSVM model is better than that of the other three models, and the accuracy of the peak point is much higher than that of the other three models.

[0204] The average relative error ARE and root mean square error RMSE of each model are calculated according to formula (22) and formula (23).

[0205]

[0206] Table 2 Comparison of errors of three prediction models

[0207]

[0208] The average relative error and root mean square error of the four models are shown in Table 2. From the error comparison results, it can be seen that the average relative error and root mean square error of the QPSO-LSSVM prediction model are 8.76% and 10.26% respectively. Compared with the other three models, the error has been significantly reduced.

[0209] 10. Conclusion

[0210] (1) There are many factors affecting the absolute gas emission in mines. The time series variables of the relevant influencing factors are reconstructed using the multiple phase space reconstruction method. The reconstructed time series is obtained by adjusting the gas parameter delay time τ and the embedding dimension d as the input of the prediction model.

[0211] (2) The quantum strategy-improved PSO algorithm is used to optimize the hyperparameters of the prediction model, which overcomes the problem that hyperparameter optimization is prone to falling into local optimal solutions and improves the convergence speed of the algorithm during the hyperparameter optimization process.

[0212] (3) Combined with the QPSO algorithm, the hyperparameters of the LSSVM gas emission prediction model are optimized, and the QPSO-LSSVM absolute gas emission prediction model is established. Through comparative experiments on the absolute gas emission historical data of Licun Coal Mine, the results show that the QPSO-LSSVM prediction model has faster convergence speed and higher prediction accuracy than the other three models, and the prediction error is significantly lower than that of other models, which effectively realizes the accurate prediction of absolute gas emission in the mine.

[0213] In the description of the present invention, the term "plurality" refers to two or more than two. Unless otherwise clearly defined, the orientation or positional relationship indicated by the terms "upper" and "lower" is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention. The terms "connection", "installation", "fixation", etc. should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0214] In the description of the present invention, the description of the terms "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In the present invention, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0215] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A prediction model for absolute gas emission based on quantum evolution particle swarm, characterized in that: The steps include: S1: The time series of gas emission related influencing factors are reconstructed using the multiple phase space reconstruction method and used as the input of the prediction model; S2: Optimize prediction model hyperparameters using quantum particle swarm algorithm; S3: Establish an absolute gas emission prediction model, as follows: In the formula, the kernel function K(x i ,x)Select REF function; Where σ 2 is the kernel function width of the absolute gas emission prediction model; S4: Quantum hyperparameter optimization of the absolute gas emission prediction model; S5: Test the established absolute gas emission prediction model.

2. The absolute gas emission prediction model based on quantum evolution particle swarm according to claim 1 is characterized in that: In step S2, the specific steps are as follows: The gas outflow algorithm of quantum evolution particle swarm is as follows: v ij (t+1)=ωv ij (t)+ c1r 1j (t)(p ij (t)-X ij (t))- c2r 2j (t)(p gj (t)-X ij (t)) Where v ij (t) represents the spatial velocity of the i-th particle in the j-dimensional space at time t; c1 and c2 are particle acceleration factors; r 1j (t) and r 2j (t) is a random function; Update the next generation of aggregation points based on the aggregation speed and aggregation point information: X ij (t+1)=X ij (t)+v ij (t+1) Assume that the evolving particles have M particle groups representing potential solutions to the gas outburst problem in the N-dimensional quantum optimization space: X i (t)=[X i,1 (t),X i,2 (t),…,X i,N (t)], i=1,2,…,M The optimal aggregation point pbest of the evolutionary particle individuals is calculated based on the fitness value f[X i (t)]Continuously updated: Then calculate the average optimal aggregation point Cbest of the particle swarm and the global optimal aggregation point gbest: A random gas gathering point is calculated for the j dimensions of the evolving particle i: p i,j (t)=φ j (t)·P i,j (t)+[1-φ j (t)]·G j (t),φ j (t)∈U(0,1) Finally, the new gas gathering point is obtained according to the evolution particle equation: Where β is the control parameter.

3. The absolute gas emission prediction model based on quantum evolved particle swarm according to claim 2 is characterized in that: In step S3, the absolute gas emission prediction model is obtained by the following steps: Create a model function: f(x)=W T ·X+b,i=1,2,...,n Where W T and b are the normal vector and intercept of the model respectively, X is the time series of the reconstruction of the characteristics of the factors affecting the absolute gas emission; To minimize the risk, the inequality constraint solution of the regression problem is transformed into an equality constraint solution: y i [W T φ(x i )+b]=1-e i Where, γ is the regularization parameter of LSSVM gas emission model, φ(x i ) is x i The result of mapping in high-dimensional space; In the formula, the Lagrange multiplier α plays the role of model support; Finally, the LSSVM gas emission model function is established: In the formula, the kernel function K(x i ,x)Select REF function: Where σ 2 is the kernel function width of the LSSVM gas emission model.

4. The absolute gas emission prediction model based on quantum evolved particle swarm according to claim 3 is characterized in that: In the step S4, the following steps are specifically included: S1: Select the average relative error as the objective function: Where y i is the true value of the test sample of gas-related influencing factors, is the prediction result output after each iteration, and n is the number of training samples; S2: Randomly generate N combinations of two hyperparameters and assign initial gas gathering points to the evolving particles: By calculating the respective fitness values ​​f[X i (0)] and determine the optimal aggregation point of the individual evolving particles and the global optimal clustering point S3: Calculate the next generation of hyperparameter combinations Repeat the above iterative process to obtain the t generation hyperparameter combination Until the number of iterations reaches the maximum value I max Or the fitness value of each generation does not change, and the optimal combination of parameters is obtained. S4: According to the optimal model hyperparameter γ s and A QPSO-LSSVM gas emission prediction model was established, and the model was trained using the reconstructed absolute gas emission data set training samples.

Images