Ventilation parameter optimization method and dust concentration control method for tunneling

Abstract

The present invention discloses a ventilation parameter optimization method and dust concentration control method for a tunneling tunnel, including obtaining tunnel wind flow-dust data information of a target tunneling tunnel under different ventilation parameters; fitting the acquired data information based on a polynomial regression algorithm; performing data cleaning based on the difference between the fitting result and the actual data; taking ventilation parameters as the optimization object and dust concentration as the optimization target, adopting a multi-objective particle swarm optimization algorithm, a multi-objective genetic algorithm and a polynomial regression algorithm to optimize the ventilation parameters of the target tunneling tunnel; according to the obtained optimization solution set, the optimal ventilation parameters of the target tunneling tunnel are obtained based on the Pareto frontier solution set, and the ventilation parameter optimization of the target tunneling tunnel is completed. The present invention not only realizes the ventilation parameter optimization and dust concentration control of the target tunneling tunnel, but also has higher reliability and better accuracy.

Description

Technical Field

[0001] The invention belongs to the technical field of mines, and in particular relates to a ventilation parameter optimization method and a dust concentration control method for an excavation tunnel. Background Art

[0002] With the rapid development of economy and technology, the demand and mining volume of coal are also increasing. In the process of coal mining, dust pollution has always been a top priority for people. High concentrations of dust will affect the health of coal miners, damage underground equipment, and seriously affect the safety and reliability of the coal mining process. Therefore, it is particularly important to optimize the ventilation parameters and control dust in the tunneling process of coal mining.

[0003] At present, the traditional ventilation parameter optimization scheme and the corresponding dust control scheme for tunneling tunnels use traditional software to build a regression model for dust concentration data under the influence of ventilation parameters; however, the fitting effect of such schemes is poor and it is impossible to obtain an accurate regression model. At the same time, in terms of ventilation parameter optimization, the existing schemes use a combined optimization method to obtain the optimization results, but such schemes cannot obtain the global optimal solution under multiple optimization objectives. Summary of the invention

[0004] One of the purposes of the present invention is to provide a method for optimizing ventilation parameters of a tunneling tunnel with high reliability and good accuracy.

[0005] A second object of the present invention is to provide a dust concentration control method including the ventilation parameter optimization method of the tunneling tunnel.

[0006] The ventilation parameter optimization method for the tunneling tunnel provided by the present invention comprises the following steps:

[0007] S1. Obtaining the tunnel airflow-dust data information of the target tunnel under different ventilation parameters;

[0008] S2. Fitting the data information obtained in step S1 based on a polynomial regression algorithm;

[0009] S3. Based on the difference between the fitting result obtained in step S2 and the actual data, the data information obtained in step S1 is cleaned;

[0010] S4. According to the data information obtained in step S3, the ventilation parameters are optimized and the dust concentration is optimized, and the ventilation parameters of the target tunnel are optimized by using a multi-objective particle swarm optimization algorithm, a multi-objective genetic algorithm and a polynomial regression algorithm;

[0011] S5. According to the optimization solution set obtained in step S4, based on the Pareto front solution set, the optimal ventilation parameters of the target tunneling tunnel are obtained, and the ventilation parameter optimization of the target tunneling tunnel is completed.

[0012] The step S2 comprises the following steps:

[0013] Convert the data information obtained in step S1 into a design matrix;

[0014] A polynomial regression algorithm is used to fit the data information obtained in step S1, with the mean square error between the predicted value and the actual value as the loss function.

[0015] The step S2 specifically includes the following steps:

[0016] The sample data of the single ventilation parameters obtained in step S1 is converted into the following design matrix X: In the formula is the sample data of the mth ventilation parameter; n is the highest degree of the polynomial;

[0017] The polynomial regression algorithm is used to fit the data information obtained in step S1, with the mean square error between the predicted value and the actual value as the loss function; specifically, the following steps are included:

[0018] A. Initialize the first-order moment vector and the second-order moment vector 0; initialize the weight vector is a random value that conforms to the standard normal distribution Column vector of ; initialize stale weight values for ; Initialize the number of iterations i to 1;

[0019] B. For the i-th round of iteration:

[0020] The loss function value is calculated using the following formula: : In the formula is the actual value; is the predicted value, and , is the j-th row vector of the design matrix X;

[0021] Calculate the loss function value The gradient value ;

[0022] Update the first moment estimate for ,in is the first-order moment attenuation coefficient;

[0023] Update the second moment estimate for ,in is the second-order moment attenuation coefficient;

[0024] Update first-order moment bias correction for ;

[0025] Update second-order moment bias correction for ;

[0026] Update the weight vector for ,in is the learning rate, is the first positive number less than the set value;

[0027] C. Determine the result obtained in step B:

[0028] like , then stop the loop and use the obtained parameters as the optimal parameters of the polynomial regression algorithm; where, is the set threshold;

[0029] like , then increase the value of i by 1, The value of is updated to , and return to step B for the next round of iteration;

[0030] D. Use the following formula to calculate the evaluation index corresponding to the optimal parameters: Where RMSE is the first evaluation indicator; SSE is the sum of squared errors, and ; is the second evaluation indicator; SSR is the regression sum of squares, and , is the mean of the actual values; SST is the total sum of squares, and .

[0031] The step S3 specifically includes the following steps:

[0032] The difference between the fitting result obtained in step S2 and the actual data is used as the detected data set XX. ;

[0033] The z-score method is used to detect the data in the detected data set XX:

[0034] Calculate the mean of the detected data set XX for , calculate the standard deviation of the detected data set XX for ; is the value of the jjth element in the detected data set XX;

[0035] Calculate the z-score value of the jjth element in the detected data set XX for ;

[0036] Setting Thresholds , and make a judgment:

[0037] like , then the data corresponding to the jjth element in the detected data set XX is considered as abnormal data and is directly removed;

[0038] like , then the data corresponding to the jjth element in the detected data set XX is considered normal data and is retained;

[0039] Finally, all normal data is retained and data cleaning is completed.

[0040] The step S4 comprises the following steps:

[0041] According to the data information obtained in step S3, the air duct hanging height, the distance between the air duct and the side wall, and the distance between the air duct and the working face are used as ventilation parameters and optimization objects, the arrangement range of the air duct hanging height, the arrangement range of the distance between the air duct and the side wall, and the arrangement range of the distance between the air duct and the working face are used as constraint conditions, and the dust concentration at the driver's position in the dead-end tunnel, the dust concentration at the breathing height of the pedestrian side, and the dust concentration at the breathing height of the tunnel are used as optimization targets, and a multi-objective particle swarm optimization algorithm is used to optimize the ventilation parameters of the target excavation tunnel;

[0042] During the optimization process, a method combining multi-objective genetic algorithm and polynomial regression algorithm was used to calculate the dust concentration at the driver's side, the breathing height of the pedestrian side, and the breathing height of the lane in the dead-end lane.

[0043] The step S4 specifically comprises the following steps:

[0044] Build optimization goals:

[0045] Optimization problem A: The optimization variable is , the optimization goal is , the fixed parameters are and ;

[0046] Optimization problem B: The optimization variables are , the optimization goal is , the fixed parameters are and ;

[0047] Optimization Problem C: The optimization variables are , the optimization goal is , the fixed parameters are and ;

[0048] in, is the hanging height of the wind tube, is the distance between the wind tube and the side wall, is the distance between the wind tube and the working surface, is the dust concentration at the driver's side in the dead-end tunnel, is the dust concentration at the breathing height of pedestrians, is the dust concentration at the breathing height of the tunnel;

[0049] Build constraints: The multi-objective particle swarm optimization algorithm is used to optimize the ventilation parameters of the target tunneling tunnel, which specifically includes the following steps:

[0050] a. Initialize the model parameters and set the number of iterations to 0;

[0051] b. Initialize the population: Use a random generation strategy to generate the initial particle swarm position and initial particle velocity, and use them as the current particle swarm position and current particle velocity;

[0052] c. According to the current particle group position and current particle speed, a method combining a multi-objective genetic algorithm and a polynomial regression algorithm is used to calculate the dust concentration at the driver's side, the pedestrian's breathing height, and the breathing height of the lane in the dead-end lane, and the calculated dust concentration is used as the fitness value of each particle;

[0053] d. Perform non-dominated sorting on the fitness values ​​of each particle in the current population, and store the examples in the first layer in the REP set;

[0054] e. Adopt the adaptive grid algorithm to divide the particles in the REP set into different grids and calculate the probability of each grid being selected;

[0055] f. According to the probability obtained in step e, a grid is selected by the roulette method, and the individuals in the selected grid are used as the gBest candidate set; one of the individuals in the gBest candidate set is selected by random selection as the global optimal gBest;

[0056] g. Update the speed and position of all particles, and set the local optimal pBest of the particles in the current population to the particle itself;

[0057] h. Perform non-dominated sorting on the fitness values ​​of each particle in the current population, and store the examples in the first layer in the REP set; use an adaptive grid algorithm to divide the particles in the REP set into different grids, and calculate the probability of each grid being selected;

[0058] To achieve the update and maintenance of the REP set;

[0059] i. Update the local optimal pBest according to the dominance relationship:

[0060] If the newly generated particle dominates the parent particle, the local optimum pBest is updated;

[0061] If the newly generated particle is dominated by the parent particle, it will not be updated;

[0062] If the newly generated example is complementary to the parent particle, the local optimum pBest is updated according to the set probability;

[0063] j. The number of iterations increases by 1;

[0064] k. Determine the number of iterations:

[0065] If the number of iterations reaches the set maximum value, the iteration ends, and the Pareto frontier solution set and the corresponding fitness value are output according to the current REP set;

[0066] Otherwise, return to step g for the next iteration.

[0067] The method of combining the multi-objective genetic algorithm and the polynomial regression algorithm specifically comprises the following steps:

[0068] (1) Population initialization: According to the set range of the initial learning rate, a random strategy is used to generate the initial population, and the number of iterations is set to 0;

[0069] (2) Based on the initial population, a polynomial regression algorithm is used to calculate the fitness value of each individual in the initial population; the fitness value is the first evaluation index RMSE and the second evaluation index ;

[0070] (3) Perform fast non-dominated sorting on the initial population to divide it into several levels; in each level, calculate the crowding distance of each individual;

[0071] (4) Selecting parent individuals from the initial population, performing genetic recombination on the parent individuals to generate new offspring individuals; performing genetic mutation on the new offspring individuals to enhance the diversity of the population;

[0072] (5) The number of iterations increases by 2;

[0073] (6) Merge the parent and child populations, and use the polynomial regression algorithm to calculate the fitness value of each individual in the merged population; the fitness value is the first evaluation index RMSE and the second evaluation index ;

[0074] (7) Perform fast non-dominated sorting on the merged population to divide the merged population into several levels; in each level, calculate the crowding distance of each individual;

[0075] (8) Generate new parents, perform genetic recombination on the parents to generate new offspring; perform genetic mutation on the new offspring to enhance the diversity of the population;

[0076] (9) Determine the number of iterations:

[0077] If the number of iterations reaches the set value, the iteration is judged to be over and the optimal parameters are output;

[0078] Otherwise, the number of iterations increases by 1 and the process returns to step (6) for the next round of iterations.

[0079] The fast non-dominated sorting in step (3) and step (7) specifically includes the following steps:

[0080] 1) Setup and , For individuals The set of all solutions dominated by To dominate individuals The number of solutions, and Initialized to an empty set, Initialized to 0;

[0081] 2) Traverse every pair of individuals in the population :like Dominate Then join in ,like Dominate but The value of is increased by 1;

[0082] 3) Get all The individuals are 0, and the acquired individuals are classified into the first layer , the first layer As the current layer;

[0083] 4) For each individual in the current layer, check Solution , and each The corresponding The value of is reduced by 1;

[0084] 5) Yes Make a judgment: If is 0, then the corresponding Into a new layer, and the new layer as the current layer;

[0085] 6) Return to step 4) for the next round of iteration until all individuals in the population are stratified.

[0086] The calculation of the crowding distance of each individual in step (3) and step (7) specifically includes the following steps:

[0087] For each objective function , sort the objective function values ​​in ascending order, and set the crowding distances of the first and last solutions after sorting to infinity; the objective function includes RMSE and ;

[0088] For the remaining solutions after sorting, the corresponding crowding distance is calculated using the following formula: In the formula is the crowding distance of the kkth solution; is the objective function value of the kk+1th solution; is the objective function value of the kk-1th solution; is the objective function The maximum value of is the objective function The minimum value of .

[0089] The step (4) of selecting parent individuals from the initial population specifically includes the following steps:

[0090] Randomly select two individuals A and B;

[0091] Compare the non-dominated levels of individuals A and B:

[0092] If A and B are of different levels, the individual with the lower level is selected;

[0093] If A and B are at the same level, the individual with a larger crowding distance is selected;

[0094] Repeat the above steps until a set number of individuals are selected as parent individuals.

[0095] The genetic recombination of the parent individuals in step (4) and step (8) specifically includes the following steps:

[0096] Randomly select two parents and ;

[0097] Randomly generate crossover probabilities , and make a judgment:

[0098] If the crossover probability If it is greater than the set value, proceed to the next step;

[0099] If the crossover probability If it is less than or equal to the set value, there is no crossover and the algorithm ends;

[0100] The crossover ratio is calculated using the following formula: : In the formula For interval Random numbers in ; is the distribution index;

[0101] The following formula is used to generate offspring individuals and : For the generated offspring individuals and Make a judgment:

[0102] like and If both are within the set range, the generation of offspring individuals is completed;

[0103] like and If it exceeds the set range, the excess value will be deleted to within the set range to complete the generation of offspring individuals.

[0104] The genetic mutation of the new offspring individuals described in step (4) and step (8) specifically includes the following steps:

[0105] Generate mutation probability , and make a judgment:

[0106] like If it is greater than the set value, proceed to the next step;

[0107] like If it is less than or equal to the set value, no mutation is performed and the algorithm ends;

[0108] The disturbance is calculated using the following formula: : In the formula is the current gene value; for The lower limit of for The upper limit value of is a first parameter for controlling the range of variation, and , r is the interval The random numbers in is the variation distribution index; is a second parameter used to control the range of variation, and ;

[0109] Generate new gene values for , is the original gene value;

[0110] Generate new gene values Make a judgment:

[0111] like Within the set range, the generation of new genes is completed;

[0112] like If it is not within the set range, the exceeding value will be deleted to be within the set range to complete the generation of new genes.

[0113] The generation of a new parent described in step (8) specifically includes the following steps:

[0114] In order of levels, from the first level At the beginning, go down layer by layer and select individuals in turn until the number of selected individuals meets the set requirement, completing the generation of the new parent generation;

[0115] In the same level, individuals with larger crowding distance are preferred.

[0116] The adaptive grid algorithm specifically comprises the following steps:

[0117] Determine the maximum and minimum bounds for each dimension in the target space;

[0118] The modulus of the grid is calculated using the following formula: In the formula is the size of a grid; is the maximum boundary of the i-th dimension; is the minimum boundary of the i-th dimension; M is the number of grids to be divided in the i-th dimension;

[0119] Count the number of particles in the grid;

[0120] Traverse the particles in the external archive and calculate the grid number where the particle is located;

[0121] The grid position of the nth particle is calculated as ,in is the objective function value of the nth particle, To take integer value;

[0122] Determine the grid position of the particle by the particle number and calculate the probability of the nth grid being selected for , is the number of particles in the nth grid.

[0123] The non-dominated sorting specifically includes the following steps:

[0124] For every pair of individuals in the population , make a dominance relationship judgment;

[0125] If individual mm is not dominated by any other individual, then individual mm is classified into the first layer ;

[0126] Finally, The individuals in are the Pareto frontier solution set;

[0127] The dominance relationship is defined as:

[0128] Individual mm dominates individual nn if and only if individual mm is not inferior to individual nn in all objectives and individual mm is superior to individual nn in at least one objective.

[0129] The step S5 specifically includes the following steps:

[0130] According to the optimization solution set obtained in step S4, the corresponding Pareto front solution set is obtained, and the optimal solution is obtained by using the ideal unreachable point method, so as to obtain the optimal ventilation parameters of the target tunneling tunnel and complete the ventilation parameter optimization of the target tunneling tunnel.

[0131] The present invention also provides a dust concentration control method including the ventilation parameter optimization method of the tunneling tunnel, which also includes the following steps:

[0132] S6. Using the optimal ventilation parameters obtained in step S5, the dust concentration in the tunneling tunnel is controlled to complete the dust concentration control in the tunneling tunnel.

[0133] The ventilation parameter optimization method and dust concentration control method for the tunneling tunnel provided by the present invention perform data information fitting based on a polynomial regression algorithm, and combine a multi-objective particle swarm optimization algorithm, a multi-objective genetic algorithm, a polynomial regression algorithm and a Pareto frontier solution set scheme, which not only realizes the ventilation parameter optimization and dust concentration control of the target tunneling tunnel, but also has higher reliability and better accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0134] Figure 1 The figure is a schematic diagram of the method flow of the optimization method of the present invention.

[0135] Figure 2It is a schematic diagram of a full-scale physical model of a fully-mechanized tunnel according to an optimization method embodiment of the present invention.

[0136] Figure 3 Schematic diagram of the fitting curve of the driver regression model in the optimization method embodiment of the present invention, wherein Figure 3 (a) is a schematic diagram of the curve of the wind tube hanging height and the dust concentration at the driver's position. Figure 3 (b) is a diagram showing the curve of the distance between the wind tube and the side wall and the dust concentration at the driver's position. Figure 3 (c) is a schematic diagram of the curve of the distance between the wind tube and the head-on surface and the dust concentration at the driver's position.

[0137] Figure 4 It is a fitting curve diagram of the regression model at the breathing height of the pedestrian side in the optimization method embodiment of the present invention, wherein Figure 4 (a) is a schematic diagram of the curve of the wind tube hanging height and the dust concentration at the breathing height of the pedestrian side. Figure 4 (b) is a schematic diagram of the curve of the distance between the air duct and the side wall and the dust concentration at the breathing height of the pedestrian side. Figure 4 (c) is a schematic diagram of the curve of the distance between the wind tube and the head-on surface and the dust concentration at the breathing height of the pedestrian side.

[0138] Figure 5 It is a regression model fitting curve diagram at the tunnel breathing height of the optimization method embodiment of the present invention, wherein Figure 5 (a) is a schematic diagram of the dust concentration curve at the wind tube hanging height and the roadway breathing height. Figure 5 (b) is a schematic diagram of the curve of the distance between the wind tube and the side wall and the dust concentration at the breathing height of the tunnel. Figure 5 (c) is a schematic diagram of the curve of the distance between the wind tube and the head surface and the dust concentration at the breathing height of the tunnel.

[0139] Figure 6 Schematic diagram of the Pareto frontier solution set and the corresponding optimal result of the optimization method embodiment of the present invention, wherein Figure 6 (a) is a schematic diagram of the optimal results of the wind tube hanging height. Figure 6 (b) is a schematic diagram of the optimal result of the distance between the wind tube and the side wall. Figure 6 (c) is a schematic diagram of the optimal result of the distance between the wind tube and the head surface.

[0140] Figure 7 It is a comparative schematic diagram of an embodiment of the optimization method of the present invention, wherein Figure 7 (a) is a schematic diagram of dust concentration values ​​in each area under different schemes. Figure 7 (b) Schematic diagram of the optimization rate of different optimization schemes.

[0141] Figure 8 The figure is a flow chart of the control method of the present invention. DETAILED DESCRIPTION

[0142] like Figure 1 The figure shows a schematic diagram of the method flow of the optimization method of the present invention: the ventilation parameter optimization method of the tunneling tunnel provided by the present invention comprises the following steps:

[0143] S1. Obtain the tunnel airflow-dust data information of the target tunnel under different ventilation parameters.

[0144] S2. Fitting the data information obtained in step S1 based on a polynomial regression algorithm; comprising the following steps:

[0145] Convert the data information obtained in step S1 into a design matrix;

[0146] A polynomial regression algorithm is used to fit the data information obtained in step S1, with the mean square error between the predicted value and the actual value as the loss function;

[0147] The polynomial regression algorithm has the function of adaptively adjusting the learning rate and the ability to fine-tune and update parameters through the first-order moment and second-order moment of the gradient.

[0148] The specific implementation includes the following steps:

[0149] The sample data of the single ventilation parameters obtained in step S1 is converted into the following design matrix X: In the formula is the sample data of the mth ventilation parameter; n is the highest degree of the polynomial.

[0150] The polynomial regression algorithm is used to fit the data information obtained in step S1, with the mean square error between the predicted value and the actual value as the loss function; specifically, the following steps are included:

[0151] A. Initialize the first-order moment vector and the second-order moment vector 0; initialize the weight vector is a random value that conforms to the standard normal distribution (mean is 0 and standard deviation is 1). Column vector of ; initialize stale weight values for ; Initialize the number of iterations i to 1;

[0152] B. For the i-th round of iteration:

[0153] The loss function value is calculated using the following formula: : In the formula is the actual value; is the predicted value, and , is the j-th row vector of the design matrix X;

[0154] Calculate the loss function value The gradient value ;

[0155] Update first moment estimate for ,in is the first-order moment attenuation coefficient;

[0156] Update the second moment estimate for ,in is the second-order moment attenuation coefficient;

[0157] Update first-order moment bias correction for ;

[0158] Update second-order moment bias correction for ;

[0159] Update the weight vector for ,in is the learning rate, is the first positive number less than the set value;

[0160] C. Determine the result obtained in step B:

[0161] like , then stop the loop and use the obtained parameters as the optimal parameters of the polynomial regression algorithm; where, is the set threshold;

[0162] like , then increase the value of i by 1, The value of is updated to , and return to step B for the next round of iteration;

[0163] D. Use the following formula to calculate the evaluation index corresponding to the optimal parameters: Where RMSE is the first evaluation indicator; SSE is the sum of squared errors, and ; is the second evaluation indicator; SSR is the regression sum of squares, and , is the mean of the actual values; SST is the total sum of squares, and ;

[0164] The smaller the RMSE value, the smaller the deviation and the more accurate the regression model; The closer it is to 1, the greater the goodness of fit and the better the prediction performance of the regression model;

[0165] During specific implementation, the fitting effect can also be evaluated based on the calculated evaluation indicators; the higher the fitting effect, the higher the accuracy of the subsequent solution; when the fitting effect is lower than the set requirement, re-fitting can be considered;

[0166] S3. Based on the difference between the fitting result obtained in step S2 and the actual data, data cleaning is performed on the data information obtained in step S1; specifically comprising the following steps:

[0167] Since there are some abnormal points in the dust concentration data set, the polynomial regression algorithm is used for fitting, which leads to the deviation of the regression coefficient and the increase of the sum of square errors, thereby reducing the fitting accuracy of the data and easily causing overfitting. Therefore, the difference between the fitting result obtained in step S2 and the actual data is taken as the detected data set XX. ;

[0168] The z-score method is used to detect the data in the detected data set XX:

[0169] Calculate the mean of the detected data set XX for , calculate the standard deviation of the detected data set XX for ; is the value of the jjth element in the detected data set XX;

[0170] Calculate the z-score value of the jjth element in the detected data set XX for ;

[0171] Setting Thresholds , and make a judgment:

[0172] like , then the data corresponding to the jjth element in the detected data set XX is considered as abnormal data and is directly removed;

[0173] like , then the data corresponding to the jjth element in the detected data set XX is considered normal data and is retained;

[0174] Finally, all normal data is retained and data cleaning is completed.

[0175] S4. Based on the data information obtained in step S3, taking ventilation parameters as optimization objects and dust concentration as optimization targets, a multi-objective particle swarm optimization algorithm, a multi-objective genetic algorithm and a polynomial regression algorithm are used to optimize the ventilation parameters of the target tunneling tunnel; comprising the following steps:

[0176] Through analysis, it can be seen that ventilation parameters such as the hanging height of the wind duct, the distance between the wind duct and the side wall, the distance between the wind duct and the front face, and the wind speed have an impact on the dust concentration in the dead-end tunnel; when the ventilation parameters are the same, there are differences in the change trend of dust concentration in different areas of the dead-end tunnel, and the ventilation parameter values ​​corresponding to the highest and lowest dust concentration values ​​in different areas are different; therefore, in order to minimize the dust concentration in high dust areas such as the dust concentration at the driver's side, the dust concentration at the breathing height of the pedestrian side, and the dust concentration at the breathing height of the tunnel in the dead-end tunnel, and improve the working environment, it is of great significance to select the best ventilation parameters;

[0177] According to the data information obtained in step S3, the air duct hanging height, the distance between the air duct and the side wall, and the distance between the air duct and the working face are used as ventilation parameters and optimization objects, the arrangement range of the air duct hanging height, the arrangement range of the distance between the air duct and the side wall, and the arrangement range of the distance between the air duct and the working face are used as constraint conditions, and the dust concentration at the driver's position in the dead-end tunnel, the dust concentration at the breathing height of the pedestrian side, and the dust concentration at the breathing height of the tunnel are used as optimization targets, and a multi-objective particle swarm optimization algorithm is used to optimize the ventilation parameters of the target excavation tunnel;

[0178] During the optimization process, a method combining multi-objective genetic algorithm and polynomial regression algorithm was used to calculate the dust concentration at the driver's side, the breathing height of the pedestrian side, and the breathing height of the lane in the dead-end lane.

[0179] The specific implementation includes the following steps:

[0180] Build optimization goals:

[0181] Optimization problem A: The optimization variable is , the optimization goal is , the fixed parameters are and ;

[0182] Optimization problem B: The optimization variables are , the optimization goal is , the fixed parameters are and ;

[0183] Optimization Problem C: The optimization variables are , the optimization goal is , the fixed parameters are and ;

[0184] in, is the hanging height of the wind tube, is the distance between the wind tube and the side wall, is the distance between the wind tube and the working surface, is the dust concentration at the driver's side in the dead-end tunnel, is the dust concentration at the breathing height of pedestrians, It is the dust concentration at the breathing height of the tunnel.

[0185] Build constraints: The multi-objective particle swarm optimization algorithm is used to optimize the ventilation parameters of the target tunneling tunnel, which specifically includes the following steps:

[0186] The multi-objective particle swarm optimization algorithm is based on swarm intelligence. Each particle in the swarm represents a solution, that is, a ventilation parameter solution. By collectively sharing information among the birds, the swarm can find the optimal destination, that is, the optimal ventilation parameters. First, each particle in the initial population of the multi-objective particle swarm optimization algorithm is randomly distributed in various positions in the search space, and then the particle swarm moves in a disorderly and chaotic manner in the search space. On this basis, the particles at random positions learn and share position information within the particle swarm, and select the local optimal position and the global optimal position at this time. Then each particle combines its own experience with the experience transmitted by other particles to update its own position. Then, through continuous iterative updates, the entire particle swarm approaches the global optimal position and finally obtains the optimal ventilation parameters.

[0187] a. Initialize the model parameters and set the number of iterations to 0;

[0188] b. Initialize the population: Use a random generation strategy to generate the initial particle swarm position and initial particle velocity, and use them as the current particle swarm position and current particle velocity;

[0189] c. According to the current particle group position and current particle speed, a method combining a multi-objective genetic algorithm and a polynomial regression algorithm is used to calculate the dust concentration at the driver's side, the pedestrian's breathing height, and the breathing height of the lane in the dead-end lane, and the calculated dust concentration is used as the fitness value of each particle.

[0190] The method of combining the multi-objective genetic algorithm and the polynomial regression algorithm specifically comprises the following steps:

[0191] (1) Population initialization: According to the set range of the initial learning rate, a random strategy is used to generate the initial population, and the number of iterations is set to 0;

[0192] (2) Based on the initial population, a polynomial regression algorithm is used to calculate the fitness value of each individual in the initial population; the fitness value is the first evaluation index RMSE and the second evaluation index ;

[0193] (3) For the initial population, a fast non-dominated sort is performed to divide the initial population into several levels; in each level, the crowding distance of each individual is calculated; in each non-dominated layer, the crowding distance of each individual is calculated to measure the sparseness of the distribution of solutions around the individual; the larger the crowding distance, the sparser the area where the individual is located, and the individual is retained first to ensure the diversity of population distribution;

[0194] (4) Selecting parent individuals from the initial population, performing genetic recombination on the parent individuals to generate new offspring individuals; performing genetic mutation on the new offspring individuals to enhance the diversity of the population;

[0195] (5) The number of iterations increases by 2;

[0196] (6) Merge the parent and child populations, and use the polynomial regression algorithm to calculate the fitness value of each individual in the merged population; the fitness value is the first evaluation index RMSE and the second evaluation index ;

[0197] (7) Perform fast non-dominated sorting on the merged population to divide the merged population into several levels; in each level, calculate the crowding distance of each individual;

[0198] (8) Generate new parents, perform genetic recombination on the parents to generate new offspring; perform genetic mutation on the new offspring to enhance the diversity of the population;

[0199] (9) Determine the number of iterations:

[0200] If the number of iterations reaches the set value, the iteration is judged to be over and the optimal parameters are output;

[0201] Otherwise, the number of iterations increases by 1 and the process returns to step (6) for the next round of iterations.

[0202] The fast non-dominated sorting in step (3) and step (7) specifically includes the following steps:

[0203] 1) Setup and , For individuals The set of all solutions dominated by To dominate individuals The number of solutions, and Initialized to an empty set, Initialized to 0;

[0204] 2) Traverse every pair of individuals in the population :like Dominate Then join in ,like Dominate but The value of is increased by 1;

[0205] 3) Get all The individuals are 0, and the acquired individuals are classified into the first layer , the first layer As the current layer;

[0206] 4) For each individual in the current layer, check Solution , and each The corresponding The value of is reduced by 1;

[0207] 5) Yes Make a judgment: If is 0, then the corresponding Into a new layer, and the new layer as the current layer;

[0208] 6) Return to step 4) for the next round of iteration until all individuals in the population are stratified.

[0209] The calculation of the crowding distance of each individual in step (3) and step (7) specifically includes the following steps:

[0210] For each objective function , sort the objective function values ​​in ascending order, and set the crowding distances of the first and last solutions after sorting to infinity; the objective function includes RMSE and ;

[0211] For the remaining solutions after sorting, the corresponding crowding distance is calculated using the following formula: In the formula is the crowding distance of the kkth solution; is the objective function value of the kk+1th solution; is the objective function value of the kk-1th solution; is the objective function The maximum value of is the objective function The minimum value of .

[0212] The step (4) of selecting parent individuals from the initial population specifically includes the following steps:

[0213] Randomly select two individuals A and B;

[0214] Compare the non-dominated levels of individuals A and B:

[0215] If A and B are of different levels, the individual with the lower level is selected;

[0216] If A and B are at the same level, the individual with a larger crowding distance is selected;

[0217] Repeat the above steps until a set number of individuals are selected as parent individuals.

[0218] The genetic recombination of the parent individuals in step (4) and step (8) specifically includes the following steps:

[0219] Randomly select two parents and ;

[0220] Randomly generate crossover probabilities , and make a judgment:

[0221] If the crossover probability If it is greater than the set value, proceed to the next step;

[0222] If the crossover probability If it is less than or equal to the set value, there is no crossover and the algorithm ends;

[0223] The crossover ratio is calculated using the following formula: : In the formula For interval Random numbers in ; is the distribution index, usually (2-20) which controls the solution distribution after crossover;

[0224] The following formula is used to generate offspring individuals and : For the generated offspring individuals and Make a judgment:

[0225] like and If both are within the set range, the generation of offspring individuals is completed;

[0226] like and If it exceeds the set range, the excess value will be deleted to within the set range to complete the generation of offspring individuals.

[0227] The genetic mutation of the new offspring individuals described in step (4) and step (8) specifically includes the following steps:

[0228] Generate mutation probability , and make a judgment:

[0229] like If it is greater than the set value, proceed to the next step;

[0230] like If it is less than or equal to the set value, no mutation is performed and the algorithm ends;

[0231] The disturbance is calculated using the following formula: : In the formula is the current gene value; for The lower limit of for The upper limit value of is a first parameter for controlling the range of variation, and , r is the interval The random numbers in is the variation distribution index; is a second parameter used to control the range of variation, and ;

[0232] Generate new gene values for , is the original gene value;

[0233] Generate new gene values Make a judgment:

[0234] like Within the set range, the generation of new genes is completed;

[0235] like If it is not within the set range, the exceeding value will be deleted to be within the set range to complete the generation of new genes.

[0236] The generation of a new parent described in step (8) specifically includes the following steps:

[0237] In order of levels, from the first level At the beginning, go down layer by layer and select individuals in turn until the number of selected individuals meets the set requirement, completing the generation of the new parent generation;

[0238] In the same level, individuals with larger crowding distance are preferred.

[0239] The adaptive grid algorithm specifically comprises the following steps:

[0240] Determine the maximum and minimum bounds for each dimension in the target space;

[0241] The modulus of the grid is calculated using the following formula: In the formula is the size of a grid; is the maximum boundary of the i-th dimension; is the minimum boundary of the i-th dimension; M is the number of grids to be divided in the i-th dimension;

[0242] Count the number of particles in the grid;

[0243] Traverse the particles in the external archive and calculate the grid number where the particle is located;

[0244] The grid position of the nth particle is calculated as ,in is the objective function value of the nth particle, To take integer value;

[0245] Determine the grid position of the particle by the particle number and calculate the probability of the nth grid being selected for , is the number of particles in the nth grid.

[0246] d. Perform non-dominated sorting on the fitness values ​​of each particle in the current population, and store the examples in the first layer in the REP set;

[0247] In specific implementation, the non-dominated sorting comprises the following steps:

[0248] For every pair of individuals in the population , make a dominance relationship judgment;

[0249] If individual mm is not dominated by any other individual, then individual mm is classified into the first layer ;

[0250] Finally, The individuals in are the Pareto frontier solution set;

[0251] The dominance relationship is defined as:

[0252] Individual mm dominates individual nn if and only if individual mm is not inferior to individual nn in all objectives and individual mm is superior to individual nn in at least one objective.

[0253] e. Adopt the adaptive grid algorithm to divide the particles in the REP set into different grids and calculate the probability of each grid being selected;

[0254] f. According to the probability obtained in step e, a grid is selected by the roulette method, and the individuals in the selected grid are used as the gBest candidate set; one of the individuals in the gBest candidate set is selected by random selection as the global optimal gBest;

[0255] g. Update the speed and position of all particles, and set the local optimal pBest of the particles in the current population to the particle itself;

[0256] h. Perform non-dominated sorting on the fitness values ​​of each particle in the current population, and store the examples in the first layer in the REP set; use an adaptive grid algorithm to divide the particles in the REP set into different grids, and calculate the probability of each grid being selected;

[0257] To achieve the update and maintenance of the REP set;

[0258] i. Update the local optimal pBest according to the dominance relationship:

[0259] If the newly generated particle dominates the parent particle, the local optimum pBest is updated;

[0260] If the newly generated particle is dominated by the parent particle, it will not be updated;

[0261] If the newly generated example is complementary to the parent particle, the local optimum pBest is updated according to the set probability;

[0262] j. The number of iterations increases by 1;

[0263] k. Determine the number of iterations:

[0264] If the number of iterations reaches the set maximum value, the iteration ends, and the Pareto frontier solution set and the corresponding fitness value are output according to the current REP set;

[0265] Otherwise, return to step g for the next iteration.

[0266] S5. According to the optimization solution set obtained in step S4, based on the Pareto front solution set, the optimal ventilation parameters of the target tunneling tunnel are obtained, and the ventilation parameter optimization of the target tunneling tunnel is completed; specifically, the steps include:

[0267] In the process of multi-objective optimization of ventilation parameters using multi-objective particle swarm optimization algorithm, multiple objectives conflict and influence each other, and it is impossible to achieve the optimal value of these objectives at the same time; therefore, when optimizing ventilation parameters, the multi-objective particle swarm optimization algorithm will balance these objective values ​​to make them as optimal as possible, thereby obtaining a series of mutually non-dominant solution sets, namely, the Pareto front solution set; and there are multiple solutions that meet the requirements in the Pareto front solution set, so it is necessary to further select the best solution from them;

[0268] In terms of the Pareto frontier solution set optimization, the closer the optimal point is to the ideal unreachable point, the better; therefore, the ideal unreachable point of the present invention is the dust concentration at the driver , Dust concentration at the breathing height of pedestrians and dust concentration at breathing height in the tunnel All are 0;

[0269] Therefore, according to the optimization solution set obtained in step S4, the corresponding Pareto front solution set is obtained, and the optimal solution is obtained by using the ideal unreachable point method, so as to obtain the optimal ventilation parameters of the target tunneling tunnel and complete the optimization of the ventilation parameters of the target tunneling tunnel.

[0270] The method of the present invention is further described below in conjunction with an embodiment:

[0271] Taking the 14158 fully mechanized tunnel of a certain mine as the engineering background, a 1:1 equal-size physical model was established using SolidWorks 3D modeling software. The physical model mainly includes the EBZ200 cantilever tunneling machine, bridge conveyor, crawler and pressure-type air duct. The required air volume of the tunnel is 840 , using forced ventilation, the diameter of the wind tube is 1.2m, the hanging height (the distance from the center of the wind tube to the top of the tunnel) is 0.9m, the distance from the side (the center of the wind tube along the negative direction of the x-axis) is 0.8m, and the distance from the working face is 10m. The tunnel physical model and main size parameters are as follows Figure 2 shown.

[0272] This implementation case generates a total of 311 ventilation parameter data sets with single ventilation parameter changes, including 112 sample data of changes in the hanging height of the wind duct when the distance from the wind duct to the side wall and the distance from the front face are fixed values; 79 sample data of changes in the distance from the wind duct to the side wall when the hanging height of the wind duct and the distance from the front face are fixed values; 120 sample data of changes in the distance from the wind duct to the front face when the hanging height of the wind duct and the distance from the side wall are fixed values. On this basis, the FLUENT software is used to perform grid verification, numerical simulation and field verification on the constructed mathematical model and physical model to ensure that the numerical model can accurately reflect the wind flow and dust migration characteristics of the comprehensive excavation tunnel. Accordingly, the numerical simulation of the ventilation parameter data of 311 single ventilation parameter changes is performed based on the FLUENT software, and the dust concentration data set of the comprehensive excavation tunnel under the influence of different ventilation parameters is obtained, including the dust concentration data set at the driver in the comprehensive excavation tunnel, the dust concentration data set at the breathing height of the pedestrian side of the comprehensive excavation tunnel, and the dust concentration data set at the overall breathing height of the comprehensive excavation tunnel.

[0273] In order to analyze the influence of different ventilation parameters on the dust migration characteristics in the comprehensive excavation tunnel, the scheme of the present invention constructs a regression model for the dust concentration data set at the driver's side in the comprehensive excavation tunnel under the influence of different ventilation parameters, the dust concentration data set at the breathing height of the pedestrian side of the comprehensive excavation tunnel, and the dust concentration data set at the overall breathing height of the comprehensive excavation tunnel. That is, firstly, a polynomial regression algorithm is used to preliminarily construct a regression model for the dust sample data set. On this basis, the adaptive moment estimation polynomial regression algorithm is improved by using the outlier detection method, the optimal learning rate is selected, and the outliers in the dust sample data set are removed, so as to accurately construct a dust concentration regression model under the influence of different ventilation parameters.

[0274] Since the present invention adopts the dust concentrations in three areas in the comprehensive excavation tunnel, namely the driver's position, the pedestrian side breathing height and the overall breathing height of the tunnel, to evaluate the dust prevention and control effect in the comprehensive excavation tunnel, it is necessary to use the method of the present invention to respectively construct dust concentration regression models under the influence of different ventilation parameters in the three high-concentration areas, namely, the dust concentration regression model at the driver's position, the dust concentration regression model at the pedestrian side breathing height and the dust concentration regression model at the tunnel breathing height, so as to ensure accurate analysis of the dust migration characteristics in the comprehensive excavation tunnel under the influence of different ventilation parameters.

[0275] Since this implementation case is based on the wind tube hanging height , Distance between the air duct and the side wall Distance from the wind tube to the front face Taking the influence of three ventilation parameter changes on dust migration characteristics in fully mechanized tunnel as an example, the dust concentration regression model at the driver's position includes the dust concentration at the driver's position at different hanging heights of the wind tube, different distances between the wind tube and the side wall, and different distances between the wind tube and the front face. Regression model. The fitting curve of the regression model of dust concentration at the driver under the influence of different ventilation parameters is shown in the figure below. Figure 3 shown.

[0276] Depend on Figure 3 The analysis shows that the dust concentration regression model at the driver's position under the influence of three different ventilation parameters is All of them are greater than 0.95, and RMSE is less than 11, which well reflects the characteristics of the dust concentration data at the driver's location and has a good fitting effect; Figure 3 The bright area is the 95% confidence interval of the regression equation. The confidence interval range is small, indicating that the error between the predicted value and the actual value is small and the prediction effect is good. Figure 3 As can be seen from (a), as the height of the wind tube increases, the dust concentration at the driver first decreases and then increases. When the height of the wind tube is around 1.85m, the dust concentration at the driver reaches the lowest value, which is about 102 The dust concentration at the driver's side first decreases and then increases with the increase of the distance between the wind tube and the side wall. When the distance between the wind tube and the side wall is about 0.6m and 1.5m, the dust concentration at the driver's side reaches the highest and lowest concentration values, respectively, which are about 113 With 103 ,like Figure 3 (b) as shown; Figure 3 As shown in (c), as the distance between the wind tube and the front face increases, the dust concentration value at the driver's side changes in the trend of decrease-increase-decrease-sharp increase. When the distance between the wind tube and the front face is about 2.6m and 16m, the dust concentration value at the driver's side reaches the lowest and highest values, respectively, which are about 100 With 490 ; In summary, the three ventilation parameters all have an impact on the dust concentration at the driver's position, among which the distance between the wind duct and the head-on surface has the greatest impact on the dust concentration at the driver's position.

[0277] Similarly, the dust concentration regression model at the pedestrian side breathing height includes the different hanging heights of the air duct , Different distances between the air duct and the side Different distances between the wind tube and the front face Dust concentration at breathing height of the descending person Regression model. The fitting curve of the regression model of dust concentration at the breathing height of the descending person under the influence of different ventilation parameters is shown in the figure below. Figure 4 shown.

[0278] Depend on Figure 4 The analysis shows that the three different ventilation parameters affect the dust concentration regression model at the breathing height of the downhill pedestrian. The values ​​are all greater than 0.9, and the RMSE is less than 10, which shows a good fitting effect. The dust concentration data points at the breathing height on the pedestrian side have a low degree of dispersion, and the data points are distributed in the area near the fitting curve. Figure 4 The bright area is the 95% confidence interval of the regression curve, and its confidence interval range is within the acceptable range, indicating a good prediction effect. Figure 4 As can be seen in (a), with the increase of the hanging height of the wind tube, the dust concentration value at the breathing height on the pedestrian side first decreases and then increases, and the increase and decrease amplitudes are basically the same; when the hanging height of the wind tube is around 1.8m, the dust concentration value at the breathing height on the pedestrian side reaches the lowest value, which is about 25 The dust concentration at the breathing height on the pedestrian side increases first, then decreases, and then increases again as the distance between the air duct and the side wall increases. When the distance between the air duct and the side wall is about 0.75m and 1.55m, the dust concentration at the breathing height on the pedestrian side reaches the highest and lowest values, respectively, which are about 135 With 32 ,like Figure 4 (b) as shown; Figure 4As shown in (c), as the distance between the wind tube and the front face increases, the dust concentration value at the breathing height on the pedestrian side changes from decrease to increase to decrease to a sharp increase. When the distance between the wind tube and the front face is about 0.8m and 16m, the dust concentration value at the breathing height on the pedestrian side reaches the lowest and highest values, respectively, which are about 80 With 375 ; In summary, the three ventilation parameters have a great influence on the dust concentration value at the breathing height on the pedestrian side, among which the distance between the air duct and the head-on surface has the greatest influence on the dust concentration value at the breathing height on the pedestrian side, followed by the influence of the air duct hanging height.

[0279] Similarly, the dust concentration regression model at the tunnel breathing height includes the different hanging heights of the wind tube , Different distances between the air duct and the side Different distances between the wind tube and the front face Dust concentration at breathing height in lower tunnel Regression model; The fitting curve of the regression model of dust concentration at the breathing height of the tunnel under the influence of different ventilation parameters is shown in Figure 5 shown.

[0280] Depend on Figure 5 The analysis shows that the dust concentration regression model at the breathing height of the tunnel under the influence of three different ventilation parameters is The dust concentration regression model at the breathing height of the tunnel under the influence of the hanging height of the wind tube is greater than 0.9 and the RMSE is less than 7. It reached 0.9978 and RMSE was 6.401, with the best fitting effect. Figure 5 The bright-colored area is the 95% confidence interval of the regression curve. The confidence interval range is small, and the prediction effect is good. Figure 5 As can be seen from (a), with the increase of the hanging height of the wind tube, the dust concentration value at the breathing height of the tunnel first decreases slowly and then increases sharply, and the overall distribution shows a " " type trend; when the air duct hanging height is around 1.2m, the dust concentration value at the breathing height of the tunnel is the lowest, about 123 ;like Figure 5 As shown in (b), the dust concentration at the breathing height of the tunnel first decreases and then increases with the increase of the distance between the wind tube and the side wall. When the distance between the wind tube and the side wall is about 1.81m and 1.4m, the dust concentration at the breathing height of the tunnel reaches the highest and lowest values, respectively, which are about 178 With 117 ;like Figure 5 As shown in (c), as the distance between the wind tube and the front face increases, the dust concentration value at the breathing height of the tunnel changes from decline to rise to sharp rise. When the distance between the wind tube and the front face is about 4m and 16m, the dust concentration value at the breathing height on the pedestrian side reaches the lowest and highest values, respectively, which are about 130 With 240 ; In summary, the three ventilation parameters have a significant impact on the dust concentration value at the breathing height of the tunnel, among which the distance between the wind tube and the head surface has the greatest impact on the dust concentration value at the breathing height of the tunnel.

[0281] In the process of multi-objective optimization of ventilation parameters using the multi-objective particle swarm optimization algorithm, multiple objectives conflict and influence each other, and it is impossible to achieve the optimal value of these objectives at the same time; therefore, when optimizing ventilation parameters, the multi-objective particle swarm optimization algorithm will balance these objective values ​​to make them as optimal as possible, thereby obtaining a series of non-dominant solution sets, namely the Pareto frontier solution set; and there are multiple solutions that meet the requirements in the Pareto frontier solution set, so it is necessary to further select the best solution from them.

[0282] In terms of the optimization of the Pareto frontier solution set, the closer the optimal point is to the ideal unreachable point, the better; therefore, the ideal unreachable point of the present invention is that the dust concentration at the driver, the dust concentration at the breathing height of the pedestrian side, and the dust concentration at the breathing height of the alley are all 0.

[0283] The present invention uses a multi-objective particle swarm optimization algorithm to optimize and solve the three multi-objective optimization problems A, B, and C respectively, and obtains the Pareto frontier solution sets of A, B, and C respectively. On this basis, the optimal solution, that is, the optimal ventilation parameters, is selected using the ideal unreachable point method. The Pareto frontier solution sets of A, B, and C and the corresponding optimization results are shown in Figure 1. Figure 6 shown.

[0284] from Figure 6 (a) It can be seen that the Pareto frontier solution set of optimization problem A is mainly concentrated in the range of (1.0m, 2.0m), that is, the hanging height of the wind duct is (1.0m, 2.0m), and each optimization objective reaches the relative optimum; the Pareto frontier solution set of optimization problem B is mainly concentrated in the range of (1.3m, 1.6m), that is, the distance between the wind duct and the side wall is (1.3m, 1.6m), and each optimization objective reaches the relative optimum, such as Figure 6 (b) The Pareto frontier solution set of optimization problem C is mainly concentrated in the range of (1m, 4m), that is, the distance between the wind tube and the head surface is (1m, 4m), and each optimization objective reaches a relatively optimal value, as shown in Figure 6 The specific optimization results after multi-objective optimization are shown in Table 1, and the comparison between the optimal solution and the original solution is shown in Figure 7 As shown in the figure, the optimal results corresponding to the optimization problems A, B, and C are solutions A, B, and C.

[0285] Table 1 Schematic diagram of ventilation parameter optimization results

[0286]

[0287] from Figure 7 As can be seen from (a) and (b), Schemes A, B, and C have a significant effect on reducing the dust concentration at the driver's side, all of which are reduced by about 22%. Figure 6 The minimum dust concentration values ​​in (a), (b), and (c) are basically consistent, which proves the accuracy of the multi-objective optimization results of ventilation parameters; Schemes A and B significantly reduce the dust concentration at the breathing height on the pedestrian side, by 58.5% and 68.8% respectively, indicating that the optimized air duct hanging height and distance from the side wall have a significant effect on reducing the dust concentration at the breathing height on the pedestrian side; and compared with the original scheme, the dust concentration at the breathing height on the pedestrian side of Scheme C does not change much; Figure 6 (c) Analysis shows that when the wind tube is 1.7m to 10m away from the front face, the dust concentration value at the breathing height of the pedestrian side does not fluctuate much and is close to the minimum dust concentration value; Schemes A, B, and C have obvious effects on reducing the dust concentration at the breathing height of the tunnel, among which Scheme B has the best optimization effect, with an optimization rate of 19.9%; In summary, the multi-objective optimization of ventilation parameters of the comprehensive excavation tunnel of the present invention, the three optimization schemes A, B, and C obtained, compared with the original scheme, have reduced the dust concentration values ​​of each area in the comprehensive excavation tunnel to a certain extent, and the dust concentration value reduction effect in some areas is very significant. By comparing and analyzing Schemes A, B, and C, Scheme B has the best optimization effect, and the optimization rates of the corresponding three target dust areas are 22.5%, 68.8%, and 19.9%, respectively.

[0288] like Figure 8 The figure shows a method flow diagram of the control method of the present invention: the dust concentration control method including the ventilation parameter optimization method of the tunneling tunnel disclosed in the present invention also includes the following steps:

[0289] S1. Obtaining the tunnel airflow-dust data information of the target tunnel under different ventilation parameters;

[0290] S2. Fitting the data information obtained in step S1 based on a polynomial regression algorithm;

[0291] S3. Based on the difference between the fitting result obtained in step S2 and the actual data, the data information obtained in step S1 is cleaned;

[0292] S4. According to the data information obtained in step S3, the ventilation parameters are optimized and the dust concentration is optimized, and the ventilation parameters of the target tunnel are optimized by using a multi-objective particle swarm optimization algorithm, a multi-objective genetic algorithm and a polynomial regression algorithm;

[0293] S5. According to the optimization solution set obtained in step S4, based on the Pareto front solution set, the optimal ventilation parameters of the target tunneling tunnel are obtained, and the ventilation parameter optimization of the target tunneling tunnel is completed;

[0294] S6. Using the optimal ventilation parameters obtained in step S5, the dust concentration in the tunneling tunnel is controlled to complete the dust concentration control in the tunneling tunnel.

Claims

1. A method for optimizing ventilation parameters of a tunneling tunnel, characterized in that The steps include: S1. Obtaining the tunnel airflow-dust data information of the target tunnel under different ventilation parameters; S2. Fitting the data information obtained in step S1 based on a polynomial regression algorithm; S3. Based on the difference between the fitting result obtained in step S2 and the actual data, the data information obtained in step S1 is cleaned; S4. According to the data information obtained in step S3, the ventilation parameters are optimized and the dust concentration is optimized, and the ventilation parameters of the target tunnel are optimized by using a multi-objective particle swarm optimization algorithm, a multi-objective genetic algorithm and a polynomial regression algorithm; The specific implementation includes the following steps: According to the data information obtained in step S3, the air duct hanging height, the distance between the air duct and the side wall, and the distance between the air duct and the working face are used as ventilation parameters and optimization objects, the arrangement range of the air duct hanging height, the arrangement range of the distance between the air duct and the side wall, and the arrangement range of the distance between the air duct and the working face are used as constraint conditions, and the dust concentration at the driver's position in the dead-end tunnel, the dust concentration at the breathing height of the pedestrian side, and the dust concentration at the breathing height of the tunnel are used as optimization targets, and a multi-objective particle swarm optimization algorithm is used to optimize the ventilation parameters of the target excavation tunnel; During the optimization process, a combination of multi-objective genetic algorithm and polynomial regression algorithm was used to calculate the dust concentration at the driver's side, the pedestrian's breathing height, and the breathing height of the lane in the dead-end lane. The method of combining the multi-objective genetic algorithm and the polynomial regression algorithm specifically comprises the following steps: (1) Population initialization: According to the set range of the initial learning rate, a random strategy is used to generate the initial population, and the number of iterations is set to 0; (2) Based on the initial population, a polynomial regression algorithm is used to calculate the fitness value of each individual in the initial population; the fitness value is the first evaluation index RMSE and the second evaluation index ; (3) Perform fast non-dominated sorting on the initial population to divide it into several levels; in each level, calculate the crowding distance of each individual; (4) Selecting parent individuals from the initial population, performing genetic recombination on the parent individuals to generate new offspring individuals; performing genetic mutation on the new offspring individuals to enhance the diversity of the population; (5) The number of iterations increases by 2; (6) Merge the parent and child populations, and use the polynomial regression algorithm to calculate the fitness value of each individual in the merged population; the fitness value is the first evaluation index RMSE and the second evaluation index ; (7) Perform fast non-dominated sorting on the merged population to divide the merged population into several levels; in each level, calculate the crowding distance of each individual; (8) Generate new parents, perform genetic recombination on the parents to generate new offspring; perform genetic mutation on the new offspring to enhance the diversity of the population; (9) Determine the number of iterations: If the number of iterations reaches the set value, the iteration is judged to be over and the optimal parameters are output; Otherwise, the number of iterations increases by 1, and returns to step (6) for the next round of iterations; S5. According to the optimization solution set obtained in step S4, based on the Pareto front solution set, the optimal ventilation parameters of the target tunneling tunnel are obtained, and the ventilation parameter optimization of the target tunneling tunnel is completed.

2. The method for optimizing ventilation parameters of a tunnel according to claim 1, characterized in that The step S2 comprises the following steps: Convert the data information obtained in step S1 into a design matrix; A polynomial regression algorithm is used to fit the data information obtained in step S1, with the mean square error between the predicted value and the actual value as the loss function.

3. The method for optimizing ventilation parameters of a tunnel according to claim 2, characterized in that The step S2 specifically includes the following steps: The sample data of the single ventilation parameters obtained in step S1 is converted into the following design matrix X: In the formula is the sample data of the mth ventilation parameter; n is the highest degree of the polynomial; The polynomial regression algorithm is used to fit the data information obtained in step S1, with the mean square error between the predicted value and the actual value as the loss function; specifically, the following steps are included: A. Initialize the first-order moment vector and the second-order moment vector 0; initialize the weight vector is a random value that conforms to the standard normal distribution Column vector of ; initialize stale weight values for ; Initialize the number of iterations i to 1; B. For the i-th iteration process: The loss function value is calculated using the following formula: : In the formula is the actual value; is the predicted value, and , is the j-th row vector of the design matrix X; Calculate the loss function value The gradient value ; Update first moment estimate for ,in is the first-order moment attenuation coefficient; Update the second moment estimate for ,in is the second-order moment attenuation coefficient; Update first-order moment bias correction for ; Update second-order moment bias correction for ; Update the weight vector for ,in is the learning rate, is the first positive number less than the set value; C. Determine the result obtained in step B: like , then stop the loop and use the obtained parameters as the optimal parameters of the polynomial regression algorithm; where, is the set threshold; like , then increase the value of i by 1, The value of is updated to , and return to step B for the next round of iteration; D. Use the following formula to calculate the evaluation index corresponding to the optimal parameters: Where RMSE is the first evaluation indicator; SSE is the sum of squared errors, and ; is the second evaluation indicator; SSR is the regression sum of squares, and , is the mean of the actual values; SST is the total sum of squares, and .

4. The method for optimizing ventilation parameters of a tunnel according to claim 3, characterized in that The step S3 specifically includes the following steps: The difference between the fitting result obtained in step S2 and the actual data is used as the detected data set XX. ; The z-score method is used to detect the data in the detected data set XX: Calculate the mean of the detected data set XX for , calculate the standard deviation of the detected data set XX for ; is the value of the jjth element in the detected data set XX; Calculate the z-score value of the jjth element in the detected data set XX for ; Setting Thresholds , and make a judgment: like , then the data corresponding to the jjth element in the detected data set XX is considered as abnormal data and is directly removed; like , then the data corresponding to the jjth element in the detected data set XX is considered normal data and is retained; Finally, all normal data is retained and data cleaning is completed.

5. The method for optimizing ventilation parameters of a tunnel according to claim 4, characterized in that The step S4 specifically comprises the following steps: Build optimization goals: Optimization problem A: The optimization variable is , the optimization goal is , the fixed parameters are and ; Optimization problem B: The optimization variables are , the optimization goal is , the fixed parameters are and ; Optimization Problem C: The optimization variables are , the optimization goal is , the fixed parameters are and ; in, is the hanging height of the wind tube, is the distance between the wind tube and the side wall, is the distance between the wind tube and the working surface, is the dust concentration at the driver's side in the dead-end tunnel, is the dust concentration at the breathing height of pedestrians, is the dust concentration at the breathing height of the tunnel; Build constraints: The multi-objective particle swarm optimization algorithm is used to optimize the ventilation parameters of the target tunneling tunnel, which specifically includes the following steps: a. Initialize the model parameters and set the number of iterations to 0; b. Initialize the population: Use a random generation strategy to generate the initial particle swarm position and initial particle velocity, and use them as the current particle swarm position and current particle velocity; c. According to the current particle group position and current particle speed, a method combining a multi-objective genetic algorithm and a polynomial regression algorithm is used to calculate the dust concentration at the driver's side, the pedestrian's breathing height, and the breathing height of the lane in the dead-end lane, and the calculated dust concentration is used as the fitness value of each particle; d. Perform non-dominated sorting on the fitness values ​​of each particle in the current population, and store the examples in the first layer in the REP set; e. Adopt the adaptive grid algorithm to divide the particles in the REP set into different grids and calculate the probability of each grid being selected; f. According to the probability obtained in step e, a grid is selected by the roulette method, and the individuals in the selected grid are used as the gBest candidate set; one of the individuals in the gBest candidate set is selected by random selection as the global optimal gBest; g. Update the speed and position of all particles, and set the local optimal pBest of the particles in the current population to the particle itself; h. Perform non-dominated sorting on the fitness values ​​of each particle in the current population, and store the examples in the first layer in the REP set; use an adaptive grid algorithm to divide the particles in the REP set into different grids, and calculate the probability of each grid being selected; To achieve the update and maintenance of the REP set; i. Update the local optimal pBest according to the dominance relationship: If the newly generated particle dominates the parent particle, the local optimum pBest is updated; If the newly generated particle is dominated by the parent particle, it will not be updated; If the newly generated example is complementary to the parent particle, the local optimum pBest is updated according to the set probability; j. The number of iterations increases by 1; k. Determine the number of iterations: If the number of iterations reaches the set maximum value, the iteration ends, and the Pareto frontier solution set and the corresponding fitness value are output according to the current REP set; Otherwise, return to step g for the next iteration.

6. The method for optimizing ventilation parameters of a tunneling tunnel according to claim 5, characterized in that The fast non-dominated sorting specifically comprises the following steps: 1) Setup and , For individuals The set of all solutions dominated by To dominate individuals The number of solutions, and Initialized to an empty set, Initialized to 0; 2) Traverse every pair of individuals in the population :like Dominate Then join in ,like Dominate but The value of is increased by 1; 3) Get all The individuals are 0, and the acquired individuals are classified into the first layer , the first layer As the current layer; 4) For each individual in the current layer, check Solution , and each The corresponding The value of is reduced by 1; 5) Yes Make a judgment: If is 0, then the corresponding Into a new layer, and the new layer as the current layer; 6) Return to step 4) for the next round of iteration until all individuals in the population are stratified; The calculation of the crowding distance of each individual specifically includes the following steps: For each objective function , sort the objective function values ​​in ascending order, and set the crowding distances of the first and last solutions after sorting to infinity; the objective function includes RMSE and ; For the remaining solutions after sorting, the corresponding crowding distance is calculated using the following formula: In the formula is the crowding distance of the kkth solution; is the objective function value of the kk+1th solution; is the objective function value of the kk-1th solution; is the objective function The maximum value of is the objective function The minimum value of The step of selecting a parent individual from the initial population specifically includes the following steps: Randomly select two individuals A and B; Compare the non-dominated levels of individuals A and B: If A and B are of different levels, the individual with the lower level is selected; If A and B are at the same level, the individual with a larger crowding distance is selected; Repeat the above steps until a set number of individuals are selected as parent individuals; The genetic recombination of the parent individuals specifically comprises the following steps: Randomly select two parents and ; Randomly generate crossover probabilities , and make a judgment: If the crossover probability If it is greater than the set value, proceed to the next step; If the crossover probability If it is less than or equal to the set value, there is no crossover and the algorithm ends; The crossover ratio is calculated using the following formula: : In the formula For interval Random numbers in ; is the distribution index; The following formula is used to generate offspring individuals and : For the generated offspring individuals and Make a judgment: like and If both are within the set range, the generation of offspring individuals is completed; like and If it exceeds the set range, the excess value will be deleted to within the set range to complete the generation of offspring individuals; The said gene mutation of the new offspring individuals specifically comprises the following steps: Generate mutation probability , and make a judgment: like If it is greater than the set value, proceed to the next step; like If it is less than or equal to the set value, no mutation is performed and the algorithm ends; The disturbance is calculated using the following formula: : In the formula is the current gene value; for The lower limit of for The upper limit value of is a first parameter for controlling the range of variation, and , r is the interval The random numbers in is the variation distribution index; is a second parameter used to control the range of variation, and ; Generate new gene values for , is the original gene value; Generate new gene values Make a judgment: like Within the set range, the generation of new genes is completed; like If it is not within the set range, the exceeding value will be deleted to the set range to complete the generation of new genes; The generation of a new parent generation specifically includes the following steps: In order of levels, from the first level At the beginning, go down layer by layer and select individuals in turn until the number of selected individuals meets the set requirement, completing the generation of the new parent generation; In the same level, individuals with larger crowding distance are preferred; The adaptive grid algorithm specifically comprises the following steps: Determine the maximum and minimum bounds for each dimension in the target space; The modulus of the grid is calculated using the following formula: In the formula is the size of a grid; is the maximum boundary of the i-th dimension; is the minimum boundary of the i-th dimension; M is the number of grids to be divided in the i-th dimension; Count the number of particles in the grid; Traverse the particles in the external archive and calculate the grid number where the particle is located; The grid position of the nth particle is calculated as ,in is the objective function value of the nth particle, To take integer value; Determine the grid position of the particle by the particle number and calculate the probability of the nth grid being selected for , is the number of particles in the nth grid; The non-dominated sorting specifically includes the following steps: For every pair of individuals in the population , make a dominance relationship judgment; If individual mm is not dominated by any other individual, then individual mm is classified into the first layer ; Finally, The individuals in are the Pareto frontier solution set; The dominance relationship is defined as: Individual mm dominates individual nn if and only if individual mm is not inferior to individual nn in all objectives and individual mm is superior to individual nn in at least one objective.

7. The method for optimizing ventilation parameters of a tunnel according to claim 6, characterized in that The step S5 specifically includes the following steps: According to the optimization solution set obtained in step S4, the corresponding Pareto front solution set is obtained, and the optimal solution is obtained by using the ideal unreachable point method, so as to obtain the optimal ventilation parameters of the target tunneling tunnel and complete the ventilation parameter optimization of the target tunneling tunnel.

8. A dust concentration control method comprising the ventilation parameter optimization method for a tunneling tunnel according to any one of claims 1 to 7, characterized in that The following steps are also included: S6. Using the optimal ventilation parameters obtained in step S5, the dust concentration in the tunneling tunnel is controlled to complete the dust concentration control in the tunneling tunnel.

Images