Intelligent prediction method for coal mine and coal chemical industry high-salinity wastewater treatment effect based on machine learning

Abstract

The invention belongs to the field of wastewater treatment, and particularly relates to a coal mine and coal chemical industry high-salinity wastewater treatment effect intelligent prediction method based on machine learning. The method comprises the following steps: firstly, collecting multi-source data and preprocessing the multi-source data to cover multiple indexes of water inlet and water outlet and treatment process parameters; secondly, optimizing a BP neural network construction model by adopting an improved alpha evolutionary algorithm, wherein improvement comprises alpha operator, boundary limitation and strategy selection; meanwhile, a loss function is improved, and a catalyst activity attenuation factor is added. And then training and verifying the model, and outputting an optimal model for predicting the wastewater treatment effect of new data. According to the method, complex variables can be accurately considered, a local optimal solution is effectively avoided, the prediction precision is remarkably improved, and the high-precision prediction requirement of the coal mine and coal chemical industry high-salinity wastewater treatment effect is met.

Description

Technical Field

[0001] The present invention belongs to the field of wastewater treatment, and in particular relates to a method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning. Background Art

[0002] During coal mining and coal chemical production, a large amount of high-salt wastewater is generated. This type of wastewater has a complex composition and contains high concentrations of total dissolved solids, chemical oxygen demand, biological oxygen demand, heavy metal ions, and a high content of suspended solids. If it is discharged directly without effective treatment, it will cause serious pollution to the environment and threaten the ecological balance and human health. Traditional high-salt wastewater treatment methods are mostly based on experience and conventional monitoring methods to judge the treatment effect, and are unable to accurately and timely grasp the various indicators of the treated wastewater. With the increasingly stringent environmental protection standards, precise control of the treatment effect of high-salt wastewater has become a key requirement for the development of the industry. However, existing prediction methods have many shortcomings. On the one hand, manual monitoring and empirical judgment are highly subjective and have large errors, making it difficult to meet the requirements for high-precision prediction of wastewater treatment effects; on the other hand, some prediction methods based on simple models cannot fully consider the interaction of many complex variables in the treatment process, and the prediction accuracy is limited. Summary of the Invention

[0003] In response to the technical problems existing in the above-mentioned background technology, the present invention proposes an intelligent prediction method for the treatment effect of high-salt wastewater in coal mines and coal chemical industries based on machine learning.

[0004] In order to achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0005] S1. First, collect multi-source data from the treatment process of high-salt wastewater in coal mines and coal chemical industries, and pre-process the data;

[0006] S2. Using an improved alpha evolution algorithm to optimize the BP neural network to construct a machine learning model; the improvements to the alpha evolution algorithm include improvements to the alpha operator, improvements to the boundary constraints, and improvements to the strategy selection;

[0007] The improvements of the alpha operator are as follows:

[0008] Step 1: Improve the construction of the evolutionary matrix. First, calculate the fitness value of each individual, and then sort the individuals from best to worst according to the fitness value. Divide the population into k levels, and the number of individuals in the jth level is n. j The total number of individuals is N, and the sampling probability of individuals at different levels is From the jth layer with probability p j Extract n j ×p j ×N individuals form the evolution matrix E;

[0009] Step 2: Improve the calculation of the adaptive basis vector P to: where ω t is the dynamic weight, are the evolution paths of the last iteration, A and B represent matrices;

[0010] Step 3: Improve the random step size. First, improve the attenuation factor α: FEs and MaxFEs are the current function evaluation times and the maximum function evaluation times respectively. t is the population diversity index, calculated by the average Euclidean distance between individuals, D avg is the historical average of population diversity, then the random step size Δr = (ub-lb)·(2R1·R2-R2)·S·α, where ub and lb are the upper and lower bounds of the search space, S is the scaling factor, and R1 and R2 are real number matrices used to generate disturbances;

[0011] S3, improve the loss function and add the catalyst activity decay factor;

[0012] S4. Train and verify the constructed model and output the best model;

[0013] S5. Finally, the new data is imported into the optimal model to predict the wastewater treatment effect.

[0014] Preferably, the multi-source data are input data and predicted output data, wherein the input data are total dissolved solids in influent, chemical oxygen demand in influent, biological oxygen demand in influent, heavy metal ion concentration in influent, suspended solids content in influent, pH value in influent and treatment process parameters, including reaction temperature, reaction time, reagent content and wastewater residence time; the predicted output data include total dissolved solids in effluent, chemical oxygen demand in effluent, biological oxygen demand in effluent, heavy metal ion concentration in effluent, suspended solids content in effluent and pH value in effluent.

[0015] As an example, the number of individuals in the improvement of the alpha operator in step S2 is n j satisfy And the sampling probability p j satisfy

[0016] As a preference, the dynamic weight ω in the calculation of the improved adaptive basis vector P in step S2 is t It gradually decreases with the increase of the number of iterations, and the specific form is: Where T is the maximum number of iterations.

[0017] Preferably, the improvement of the boundary restriction in step S2 is to adopt an elastic boundary processing method: where e iis the value of the evolved individual in a certain dimension. When the individual crosses the boundary, it is not regenerated completely randomly, but is generated in an elastic area near the boundary.

[0018] Preferably, the improvement of the strategy selection in step S2 is a hybrid selection strategy, which adopts a hybrid selection strategy combining elite retention and tournament selection, and directly retains the m individuals with the best fitness in the population to the next generation. These individuals are called elite individuals, and then, for the remaining NM individuals, the tournament selection method is adopted.

[0019] Preferably, the improvement of the loss function in step S3 is: where N 样 is the total number of samples, and b is the activity of the catalyst.

[0020] Compared with the existing technology, the advantages and positive effects of the present invention lie in the use of an improved alpha evolution algorithm to optimize the BP neural network, the improved alpha operator to enhance population diversity and accelerate convergence; elastic boundary processing retains boundary information, and the hybrid selection strategy avoids premature convergence. The loss function incorporates the catalyst activity decay factor to better fit the actual treatment situation. These technologies work together to enable the model to accurately consider complex variables, effectively avoid local optimal solutions, and significantly improve prediction accuracy, meeting the industry demand for high-precision prediction of the treatment effect of high-salt wastewater in coal mines and coal chemical industries. DETAILED DESCRIPTION

[0021] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described below in conjunction with the embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0023] The treatment of high-salt wastewater has become a key problem that restricts the sustainable development of the industry. The composition of this type of wastewater is extremely complex, containing high concentrations of total dissolved solids, chemical oxygen demand, biological oxygen demand, heavy metal ions and a large amount of suspended solids. If it is not properly treated and discharged directly, it will cause serious pollution to the soil, water and atmospheric environment, endangering the ecological balance and human health. At the same time, with the continuous improvement of environmental protection standards, the traditional way of judging the effect of wastewater treatment by relying on experience and conventional monitoring methods can no longer meet the industry's urgent need for high-precision prediction of treatment effects. The present invention proposes a method for intelligent prediction of the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning.

[0024] First, multi-source data collection and preprocessing are carried out. Various data from the treatment process of high-salt wastewater from coal mines and coal chemical industries are collected, covering the total dissolved solids (TDS) in the influent, the chemical oxygen demand (COD) in the influent, the biological oxygen demand (BOD) in the influent, the heavy metal ion concentration in the influent, the suspended solids content in the influent, the pH value in the influent, and the treatment process parameters, including input data such as reaction temperature, reaction time, reagent content, and wastewater residence time. In addition, the effluent total dissolved solids (TDS), effluent COD, effluent BOD, effluent heavy metal ion concentration, effluent suspended solids content, and effluent pH value are used as predicted output data. These collected data are preprocessed to remove outliers, fill in missing values, and standardize the data to make different types of data comparable, providing a high-quality data foundation for subsequent model construction.

[0025] Then, considering that the existing alpha evolution algorithm will blindly explore in the search space, resulting in a slow convergence speed; and in the later stage of evolution, due to the reduction of population diversity, the algorithm is likely to wander around the local optimal solution, further affecting the convergence efficiency, the present invention improves the construction of the evolution matrix by improving the alpha operator, firstly calculates the fitness value of each individual, the fitness value is the value of the fitness function, and the fitness function is the inverse of the loss function, and then sorts the individuals from the best to the worst according to the fitness value, and divides the population into k levels, and the number of individuals in the jth level is n j The total number of individuals is N, and the sampling probability of individuals at different levels is From the jth layer with probability p j Extract n j ×p j ×N individuals form the evolution matrix E, where the number of individuals is n j satisfy And the sampling probability p j satisfy Improve the calculation of the adaptive basis vector P, which is improved to: where ω t is the dynamic weight, are the evolution paths of the last iteration, A and B represent matrices; to improve the random step size, first improve the attenuation factor α: FEs and MaxFEs are the current function evaluation times and the maximum function evaluation times respectively. t is the population diversity index, calculated by the average Euclidean distance between individuals, D avgIf is the historical average of population diversity, then the random step size Δr = (ub - lb) · (2R1 · R2 - R2) · S · α, where ub and lb are the upper and lower bounds of the search space, respectively, S is a scaling factor, and R1 and R2 are real matrices used to generate perturbations. Overall, the improved evolutionary matrix construction utilizes stratified sampling based on individual fitness ranking, fully utilizing information from individuals at different levels of the population. This significantly increases population diversity, significantly enhancing the algorithm's ability to escape local optima, thereby exploring a wider solution space and increasing the likelihood of finding the global optimal solution. The improved adaptive basis vector P calculation incorporates dynamic weights that vary with the number of iterations. This allows the algorithm to actively explore new solution spaces in the early stages of the iterations, while focusing on a refined search near the optimal solution in the later stages. This adaptive adjustment effectively accelerates convergence while ensuring search accuracy and reducing unnecessary waste of computational resources. The random step size is improved to be associated with the number of function evaluations and population diversity. In the early stages of the search, a larger step size can quickly locate possible areas. When approaching the optimal solution, the step size is reduced to avoid missing the optimal solution. The algorithm is like having intelligent navigation and can flexibly respond to different search stages, improving its ability to cope with complex optimization problems.

[0026] In addition, considering that in the existing Alpha Evolution Algorithm, the common practice of boundary restriction is to directly regenerate individuals completely randomly when individuals cross the boundary. Although this method is simple, it has obvious disadvantages. Generating individuals completely randomly will result in the loss of effective information accumulated by individuals who cross the boundary near the boundary, so that the algorithm may repeatedly explore the explored areas in the subsequent search process, reducing the search efficiency, and may destroy the overall evolutionary trend of the population, affecting the algorithm's convergence to the optimal solution. In terms of strategy selection, common practices are mostly single selection strategies, such as simple roulette selection or tournament selection. Roulette selection is easily affected by the distribution of individual fitness values. If the difference in fitness values ​​is too large, some individuals may be over-selected, causing the algorithm to converge prematurely; tournament selection is insufficient in maintaining population diversity. If the tournament scale is not set properly, it may lead to excessive concentration of excellent individuals in the population, which will also cause the algorithm to fall into a local optimum. Taking this situation into account, the improvement of the boundary restriction in the present invention is to adopt an elastic boundary processing method: where e iThe value of an evolved individual in a certain dimension. When an individual crosses a boundary, it is not regenerated completely randomly, but rather within an elastic region near the boundary. The improved strategy selection is a hybrid selection strategy that combines elite retention and tournament selection. The m individuals with the best fitness in the population are directly retained for the next generation. These individuals are called elite individuals. Then, tournament selection is used for the remaining NM individuals. This improvement uses an elastic boundary approach to handle boundary constraints. When an individual crosses a boundary, a new individual is generated in an elastic region near the boundary. This preserves valid information near the boundary, reduces repeated searches, maintains the population's evolutionary trend, and improves search efficiency. The strategy selection adopts a hybrid strategy that combines elite retention and tournament selection. Elite retention ensures the inheritance of excellent genes, while tournament selection maintains population diversity. The combination of the two prevents premature convergence of the algorithm and enhances its optimization capabilities for complex problems.

[0027] In addition, considering that most existing loss functions only focus on prediction accuracy, they measure model performance by calculating the difference between the predicted value and the true value. However, this model ignores some details and fails to reflect accurate changes, resulting in the model not accurately reflecting the actual processing situation, and the prediction results deviate significantly from the actual situation. The improvement of the loss function of this algorithm is: where N 样 is the total number of samples, and b is the activity of the catalyst. In the loss function (1-b) 2 As the catalyst is used, b decreases, (1-b) 2 Increase, then When the catalyst is just starting to be used, the activity is high and b is relatively large, the penalty of the loss function on the prediction error is relatively small; as the catalyst activity decays, b becomes smaller, As the value becomes smaller, the proportion of the same prediction error in the loss function will increase, which means that the requirement for prediction accuracy is getting higher and higher. The model will work harder to reduce the prediction error, which is in line with the demand that the prediction accuracy should be improved as the catalyst usage decays.

[0028] In the Alpha Evolution algorithm, improvements to the stratified sampling used in constructing the evolution matrix significantly increase population diversity, providing a richer set of initial solutions for the algorithm's search and preventing it from getting stuck in local optima. Improvements to adaptive basis vector calculation, leveraging dynamic weights that vary with the number of iterations, allow for proactive exploration of the solution space in the early stages and a more refined search in the later stages. This, in conjunction with the evolution matrix construction, improves search efficiency. Improvements to the random step size dynamically adjust based on the number of function evaluations and population diversity. This allows for rapid localization with large steps in the early stages of the search, followed by precise search with smaller steps as the optimal solution is approached. The flexible handling of boundary constraints preserves valid information near the boundaries, allowing the algorithm to better utilize this information for continuous optimization. The hybrid selection strategy ensures the inheritance of superior genes and population diversity, preventing premature convergence. These improvements, combined with improvements to the Alpha Evolution algorithm, optimize the BP neural network structure, determining the number of neurons in the input, hidden, and output layers, the number of hidden layers, connection weights, and biases. The loss function incorporates a catalyst activity decay factor, increasing the requirement for prediction accuracy as catalyst activity decreases. This improvement corresponds to the optimization process of the entire algorithm. The improvement of the algorithm's search capability creates conditions for the optimization of the loss function, and the strict requirements of the loss function prompt the algorithm to continuously seek better solutions, thereby comprehensively improving the accuracy and reliability of the prediction model.

[0029] Finally, the constructed model is trained and verified, the best model is output and the new data is imported into the best model to predict the wastewater treatment effect.

[0030] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any person skilled in the art may utilize the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes for application in other fields. However, any simple modification, equivalent change, and modification of the above embodiments made in accordance with the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industry based on machine learning, characterized in that: The following steps are involved: S1. First, collect multi-source data from the treatment process of high-salt wastewater in coal mines and coal chemical industries, and pre-process the data; S2. Using an improved alpha evolution algorithm to optimize the BP neural network to construct a machine learning model; the improvements to the alpha evolution algorithm include improvements to the alpha operator, improvements to the boundary constraints, and improvements to the strategy selection; The improvements of the alpha operator are as follows: Step 1: Improve the construction of the evolutionary matrix. First, calculate the fitness value of each individual, and then sort the individuals from best to worst according to the fitness value. Divide the population into k levels, and the number of individuals in the jth level is n. j The total number of individuals is N, and the sampling probability of individuals at different levels is From the jth layer with probability p j Extract n j ×p j ×N individuals form the evolution matrix E; Step 2: Improve the calculation of the adaptive basis vector P to: where ω t is the dynamic weight, are the evolution paths of the last iteration, A and B represent matrices; Step 3: Improve the random step size. First, improve the attenuation factor α: FEs and MaxFEs are the current function evaluation times and the maximum function evaluation times respectively. t is the population diversity index, calculated by the average Euclidean distance between individuals, D avg is the historical average of population diversity, then the random step size Δr = (ub-lb)·(2R1·R2-R2)·S·α, where ub and lb are the upper and lower bounds of the search space, S is the scaling factor, and R1 and R2 are real number matrices used to generate disturbances; S3, improve the loss function and add the catalyst activity decay factor; S4. Train and verify the constructed model and output the best model; S5. Finally, the new data is imported into the optimal model to predict the wastewater treatment effect.

2. The method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning according to claim 1, characterized in that: The multi-source data are input data and predicted output data, wherein the input data include the total dissolved solids in the influent, the chemical oxygen demand in the influent, the biological oxygen demand in the influent, the heavy metal ion concentration in the influent, the suspended solids content in the influent, the pH value in the influent, and the treatment process parameters, including the reaction temperature, reaction time, reagent content, and wastewater residence time; the predicted output data include the total dissolved solids in the effluent, the chemical oxygen demand in the effluent, the biological oxygen demand in the effluent, the heavy metal ion concentration in the effluent, the suspended solids content in the effluent, and the pH value in the effluent.

3. The method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning according to claim 1, characterized in that: The number of individuals in the improvement of the alpha operator in step S2 is n j satisfy And the sampling probability p j satisfy 4. The method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning according to claim 1, characterized in that: The dynamic weight ω in the calculation of the improved adaptive basis vector P in step S2 t It gradually decreases with the increase of the number of iterations, and the specific form is: Where T is the maximum number of iterations.

5. The method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning according to claim 1, characterized in that: The improvement of the boundary restriction in step S2 is to adopt an elastic boundary processing method: where e i is the value of the evolved individual in a certain dimension. When the individual crosses the boundary, it is not regenerated completely randomly, but is generated in an elastic area near the boundary.

6. The method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning according to claim 1, characterized in that: The improvement of the strategy selection in step S2 is a hybrid selection strategy, which adopts a hybrid selection strategy combining elite retention and tournament selection. The m individuals with the best fitness in the population are directly retained to the next generation. These individuals are called elite individuals. Then, for the remaining NM individuals, the tournament selection method is adopted.

7. The method for intelligently predicting the treatment effect of high-salt wastewater from coal mines and coal chemical industries based on machine learning according to claim 1, characterized in that: The improvement of the loss function in step S3 is: where N 样 is the total number of samples, and b is the activity of the catalyst.

Images