Mine filling body strength prediction method based on IWOA-Elman neural network

Through the improved whale optimization algorithm, the Elman neural network is optimized and the IWOA-Elman model is built, which solves the problems of low efficiency and poor accuracy of traditional mine filling strength prediction, and realizes high-precision and efficient filling strength prediction, which improves the mine filling engineering stability and safety.

CN119066975APending Publication Date: 2024-12-03NORTHEASTERN UNIV CHINA
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411200916.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Traditional mine fill strength prediction methods are inefficient and difficult to accurately predict. Traditional neural networks have problems such as slow convergence speed and easy to fall into local optimality during training, which limits their application effect.

Method used

The improved whale optimization algorithm (IWOA) is used to optimize the Elman neural network, and optimize the weight and threshold of the network by introducing chaotic mapping, adaptive weights and Levy flight strategy, and build an IWOA-Elman regression prediction model for mine fill intensity prediction .

Benefits of technology

It significantly improves the accuracy and efficiency of filling strength prediction, enhances the stability and safety of mine filling projects, reduces dependence on traditional experimental tests, reduces resource consumption, and provides a scientific basis for the design of mine filling materials.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066975A_ABST
    Figure CN119066975A_ABST
Patent Text Reader

Abstract

The invention relates to a mine filling body strength prediction method based on an IWOA-Elman neural network, and the method comprises the steps: importing and preprocessing data, building an Elman network, solving the optimal weight and threshold value of the network based on an improved IWOA optimization algorithm, predicting and analyzing a result, and the like. By introducing improved strategies such as chaotic mapping and adaptive weight, the accuracy and robustness of prediction are remarkably improved. According to the method, time and cost of a traditional experimental method are reduced, scientific basis is provided for design and optimization of mine filling materials, important reference value and guiding significance are achieved for intelligent mine filling, and the intelligent level of mine safety production can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of intelligent mine filling, and in particular to a mine filling body strength prediction method based on an IWOA-Elman neural network. Background Art

[0002] As an important part of the national pillar industry, mines play an important role in resource exploitation and environmental protection. With the increase in resource mining depth and the increasingly severe environmental problems, the treatment of waste materials such as tailings has become an issue that cannot be ignored in the mining industry. The backfill mining method has been widely used in mining production because it can effectively improve resource utilization and conform to the development concept of environmental protection and safety.

[0003] The uniaxial compressive strength of the filling body is a key indicator for evaluating its performance, which directly affects the stability and safety of mine filling projects. Traditional filling body strength prediction methods mainly rely on time-consuming and inefficient experimental tests. In addition, as a multi-phase composite mechanical medium, the strength of the filling body is highly complex and nonlinearly related to various factors such as density, ash-sand ratio, and curing time, making accurate prediction more difficult. In recent years, with the rapid development of artificial intelligence technology, artificial neural networks have been introduced into the prediction of filling material strength, and their nonlinear dynamic processing capabilities have been used to improve the accuracy and speed of prediction. However, traditional neural networks such as BP neural networks have problems such as slow convergence speed and easy to fall into local optimality during the training process, which limits their application effect in the prediction of filling body strength.

[0004] In order to solve the above problems, the present invention aims to propose a mine filling strength prediction method based on improved whale optimization algorithm (IWOA) and Elman neural network to improve the accuracy and efficiency of the prediction, while solving the limitations of the existing technology and providing more scientific and reliable technical support for the design and optimization of mine filling materials. Summary of the invention

[0005] In order to solve the problems of the prior art, the present invention provides a mine filling strength prediction method based on IWOA-Elman neural network.

[0006] In order to solve the above technical problems, the present invention is implemented by the following technical solutions: 1. A mine filling strength prediction method based on IWOA-Elman neural network, comprising the following steps:

[0007] Step 1: Improve the Elman neural network by introducing the IWOA algorithm to optimize the network weights and thresholds;

[0008] Step 2: Collect and organize mine filling ratio test data as a training data source, where the input parameters include but are not limited to the mass fractions of cement, fly ash and tailings, which directly reflect the material composition of the filling body;

[0009] Step 3: Set the compressive strength of the filling body as the output parameter of the model, i.e. the prediction target;

[0010] Step 4: Based on the weights and thresholds optimized in step 1, an IWOA-Elman regression prediction model is constructed, and the model is trained using the training data in step 2 to learn the complex mapping relationship between input parameters and output parameters;

[0011] Step 5: Evaluate the performance of the model and compare it with the existing model;

[0012] Step 6: Verify the prediction accuracy of the model through actual data.

[0013] In this implementation, the whale optimization algorithm is first improved by introducing strategies such as chaotic mapping, adaptive weights and Levy flight, and is used to optimize the weights and biases of the hidden layer of the Elman neural network, and an IWOA-Elman regression prediction model is established; then, the initial data set of mine filling is preprocessed using normalization and shuffle processing, and the mass fractions of cement, fly ash and tailings are set as input parameters, and the compressive strength of the filling body is set as the output parameter. The IWOA-Elman neural network completes the feature extraction work, maps the original features to the new feature space, and maximizes the correlation between the filling body strength and the components. It is used for actual prediction, and the prediction results of different models are verified and compared to verify the effectiveness of the method proposed in this scheme.

[0014] In a specific implementation, the IWOA algorithm is improved by introducing strategies such as chaotic mapping, adaptive weights and Levy flight.

[0015] In a specific embodiment, the training data is also preprocessed, including but not limited to normalization and shuffling, to enhance the generalization ability of the model and avoid overfitting.

[0016] In a specific implementation, the performance of the IWOA-Elman model is evaluated by constructing a comprehensive evaluation system, including but not limited to comparative analysis with the Elman, PSO-Elman and WOA-Elman models.

[0017] In a specific embodiment, the evaluation indicators include root mean square error (RMSE), mean absolute percentage error (MAPE), mean absolute error and coefficient of determination (R2).

[0018] In a specific implementation, the method can achieve high-precision prediction of the strength of mine filling bodies and has important engineering application value.

[0019] The beneficial effects of the present invention are:

[0020] 1. Improve prediction accuracy: The Elman neural network is optimized using the improved whale optimization algorithm (IWOA), which significantly improves the prediction accuracy of filling strength, thereby enhancing the stability and safety of mine filling projects;

[0021] 2. Enhance algorithm efficiency and robustness: Through strategies such as chaos mapping and adaptive weights, the IWOA algorithm performs well in global search capability and convergence speed, effectively avoiding local optimal solutions and improving the robustness of the model;

[0022] 3. Reduce costs and resource consumption: Reduce dependence on traditional experimental tests, reduce time and economic costs, and reduce resource consumption during the experiment.

[0023] 4. Significant engineering application value: The method of the present invention provides a scientific basis for the design and optimization of mine filling materials, and has important engineering application value and practical guiding significance. It can also promote the development of mine intelligence and support mine filling towards intelligent direction, help realize the automated and intelligent filling process, and improve the efficiency of mine production. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the overall system process and technical route of the present invention.

[0025] Figure 2 It is a schematic diagram of Cubic chaos mapping of the improved strategy of the algorithm used in the present invention.

[0026] Figure 3 It is a schematic diagram of the nonlinear improvement of the convergence coefficient of the improvement strategy of the algorithm used in the present invention.

[0027] Figure 4 It is a schematic diagram of the test results of the optimal number of hidden layer nodes of the Elman neural network of the present invention.

[0028] Figure 5 This is a graph of the running results of the improved algorithm used in the present invention in the test function.

[0029] Figure 6 It is a comparison chart of the results of the prediction model of the present invention in the training set and the test set.

[0030] Figure 7 It is a comparison diagram of the error curves of the prediction model of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] like Figure 1-7 A mine filling strength prediction method based on IWOA-Elman neural network is shown.

[0033] 1. Improved whale optimization algorithm

[0034] The traditional whale optimization algorithm is a meta-heuristic optimization algorithm developed based on the whale clustering behavior. It uses random or optimal search agents to simulate hunting behavior, and uses spirals to simulate the bubble net attack mechanism of humpback whales. It has the advantages of fast convergence speed, strong global search performance, few parameters and simple mechanism. The algorithm execution process mainly includes three stages: prey capture, bubble net attack and search and predation. The traditional algorithm falls into the defects of local optimality and slow convergence speed. At the same time, in order to ensure its adaptability, stability and accuracy in practical engineering applications, this paper introduces improvement strategies from the following aspects to improve the performance and application effect of the algorithm:

[0035] (1) Chaotic Mapping-Cubic Mapping

[0036] Chaos is a random nonlinear dynamic system that is sensitive to initial values. The common fixed initialization strategy in intelligent optimization algorithms will limit the performance of the algorithm. In order to improve the convergence speed while ensuring the diversity of solutions, it is proposed to use chaos mapping as the population initialization strategy of the WOA algorithm to improve the optimization performance of the algorithm. This paper uses the classic Cubic Map chaos system, and its mapping formula is as follows:

[0037] x n+1 =ρx n (1-x n 2 ) (1)

[0038] Where x n ∈(0,1), ρ is the control parameter, which determines the chaos of the Cubic mapping. When its value is 2.595, the histogram and distribution diagram of the Cubic chaotic mapping are shown in the attached figure. Figure 2 As shown, there is a better chaotic traversal effect.

[0039] (2) Adaptive parameters

[0040] ① Nonlinear improvement of convergence coefficient a

[0041] The coefficient vector A1 determines the complementary and restrictive relationship between the three behaviors of individual whales, maintaining the balance between local search and global search, and the value of A1 depends on the convergence factor a. In the traditional algorithm, it decreases linearly from 2 to 0 with the number of iterations. In the early stage of iteration, the algorithm has a strong global search capability; as the iteration proceeds, a gradually decreases, the local search capability is enhanced and the convergence speed is faster, but it is easy to fall into the local optimum. The ideal situation is that in the later stage of iteration, while ensuring the local optimum, the search range is expanded, and the probability of the algorithm jumping out of the local optimum will also increase, achieving a balance between global development and local exploration. Therefore, consider using a slope change curve instead of the original linearly decreasing straight line to enhance the optimization ability in the later stage. The curve formula and schematic diagram used in this article are shown in the attached figure. Figure 3 As shown, g represents the current number of iterations, and a0 is 2:

[0042]

[0043] ② Heuristic Probability

[0044] In order to simulate the shrinking encirclement and spiral rise in the whale hunting process at the same time, the random probability p∈[0,1] is used to determine which behavior to take. In order to balance the global search and local search capabilities in the standard WOA algorithm, this paper uses heuristic probability to improve the algorithm. The formula is as follows:

[0045]

[0046] ③Adaptive inertia weight

[0047] Inertia weight is an important parameter of particle swarm algorithms. When its value is large, the algorithm has a stronger search capability; when its value is small, the algorithm has a stronger later search capability, which can achieve enhanced search near the optimal solution. Therefore, in order to make full use of the optimal solution and improve the algorithm's optimization accuracy, an improved formula for adaptive inertia weight is introduced.

[0048]

[0049] In specific applications, when p <p h When p≥p, the shrinking and surrounding mechanism is adopted, and then the value of the convergence vector A1 is used to determine whether to search randomly. The specific formula is as follows: when |A1|≥1, the Levy strategy is used to update the global position. When |A1|<1, the shrinking and surrounding mechanism is continued. h When |A1|<1, the spiral rise mechanism is adopted, and the judgment is made again. When |A1|<1, the spiral rise mechanism is adopted around the optimal solution:

[0050] X new (g) = ω·X(g) + D·e bl cos(2πl) |A1|≥1 (5)

[0051] X new (g) = (1-ω)·X * (g)+D·e bl cos(2πl) |A1|< 1 (6)

[0052] (3) Global search improvement - Levy flight

[0053] Levy flight is a random behavior strategy with small-range search and large-range jump. It is used to simulate the step size and direction in the random walk or search process. The specific update formula is as follows:

[0054]

[0055] In the formula Indicates the updated position of the individual whale after the current iteration using the Levy flight strategy. A1 and C1 are the same as described above. Represents point product, step is a random step size, generated according to the Mantegna method:

[0056]

[0057] where μ=(μ1,μ1...μ d ) and v=(v1,v1...v d ) are all d-dimensional space vectors, β is taken as 1.5, μ and v obey and The normal distribution of

[0058]

[0059] σ v =1 (10)

[0060] 2. IWOA-Elman Optimization Model

[0061] Like the BP neural network, the Elman neural network uses the "gradient descent" algorithm to update parameters and train the network, which has the disadvantages of slow training speed and easy to fall into the local optimal solution. Therefore, this paper introduces the improved whale optimization algorithm and uses the final calculated individual global optimal position as the optimal weight and threshold of the Elman neural network to improve the overall prediction accuracy of the model and make it more robust. The implementation steps of IWOA-Elman are summarized as follows:

[0062] 1) Obtain data and preprocess it;

[0063] 2) Set network parameters and build a basic Elman neural network;

[0064] 3) Initialize IWOA parameters, initialization of individual whale positions representing the Elman neural network weights ω and threshold b

[0065] X(k)=[ω1,ω2,ω3,b1,b2]

[0066] 4) Calculate the fitness of each individual. The fitness function is as follows:

[0067] min h(ω1,ω2,ω3,b1,b2)=min{mean(MSE train +MSE test )}

[0068] In the formula, mean means to find the average value, MSE train and MSE test Respectively represent the mean square error of the training set and the test set, and the calculation method is:

[0069]

[0070] M and N represent the number of samples in the training set and the test set respectively. i Represents the network prediction result, Represents the actual result.

[0071] 5) Using the IWOA algorithm to iteratively update, the position of the optimal fitness value is expressed as:

[0072]

[0073] 6) Analyze the position of the optimal individual, extract the optimal weights and thresholds and assign them to the Elman neural network, and perform network training operations until the error threshold is reached.

[0074] The overall technical route is as follows Figure 1 As shown:

[0075] 3. Filling strength prediction model based on IWOA-Elman neural network

[0076] 3.1 Data Preprocessing

[0077] The training data comes from a certain mine filling slurry ratio test data set. Three ratio factors that affect the filling strength are selected, namely cement mass fraction, fly ash mass fraction and tailings mass fraction. Then the corresponding compressive strength of the filling is used as the standard for judging the strength, that is, the output result;

[0078] Data normalization: In order to eliminate the dimensionality impact of singular samples, the training samples need to be normalized so that the sample data range is stabilized between [0,1], which helps to improve the performance of the model. Correspondingly, the prediction results obtained by using the trained network need to be denormalized in the subsequent model evaluation to obtain the true prediction results.

[0079] The purpose of data out-of-order processing is to break the orderliness of the data, eliminate the correlation of data samples, ensure the randomness of data distribution during training, improve the generalization ability of the model, and avoid the phenomenon of "overfitting" of the network.

[0080] The training set is selected from the preprocessed data set at a ratio of 0.85, and the remaining data is used as the test set. At the same time, in order to avoid overfitting or underfitting in the model building process, a part of the training set is divided out as a validation set to test the generalization ability of the model.

[0081] 3.2 Model structure and parameter design

[0082] 3.2.1 Elman Neural Network Structure

[0083] The Elman neural network is selected as the basic prediction model, and three ratio parameters are selected as input variables. The number of neurons in the input layer of the corresponding neural network is 3; the compressive strength of the filling body is used as the output variable, and the number of neurons in the output layer of the Elman neural network is 1. Too many or too few neurons in the hidden layer will lead to under-fitting or over-fitting problems, which determine the overall performance of the network. The present invention estimates the value range of the number of hidden layer nodes through an empirical formula, and tries to find the optimal number of hidden layer neurons in turn. The formula is as follows:

[0084]

[0085] Where nums in The number of neurons in the input layer is 3, nums out The number of neurons in the output layer is 1, and a is a constant between 1 and 10, so the range of values ​​that can be calculated is [3,12]. In order to determine the best one, different Elman neural networks are established according to the sample data, and predictions and simulations are performed 50 times, with 1000 iterations each time. The best number of nodes in the hidden layer is determined based on the different average errors obtained. The results are shown in the attached figure. Figure 4 As shown. Comprehensively analyzing the errors of the training set and the test set, and considering the network performance and time complexity, the number of hidden layer nodes is finally set to 5, and the number of neurons in the hidden layer of the Elman neural network is determined to be 5. The number of neurons in the follow-up layer should be consistent with the hidden layer, so the number of neurons in the follow-up layer is also 5. In summary, the structure of the Elman neural network is determined to be "3-5-5-1".

[0086] 3.2.2 IWOA parameter design

[0087] In the improved whale algorithm, the whale population size is set to 30, the maximum evolutionary generation is 50, and the number of variables to be optimized, that is, the dimension of the solution space, needs to be determined according to the Elman neural network structure. The number of neurons in the input layer is 3, the number of neurons in the hidden layer is 5, the number of neurons in the receiving layer is 5, and the number of neurons in the output layer is 1. Therefore, the dimension of the solution space of IWOA is: 3*5+5 2 +5+5+1 is 51, which represents the weight from the input layer to the hidden layer, the weight from the hidden layer to the receiving layer, the weight from the hidden layer to the output layer, the threshold of the hidden layer and the threshold of the output layer respectively.

[0088] After the algorithm is executed, the corresponding optimal weights and thresholds are extracted. The dimension of IWOA is , and the operation results are saved in a 1×dim matrix. The weight matrix from the input layer to the receiving layer is 3×5; the weight matrix from the hidden layer to the receiving layer is 5×5; the weight matrix from the hidden layer to the output layer is 5×1; the threshold matrix of the hidden layer is 5×1; and the threshold matrix of the output layer is 1×1. In summary, the elements are extracted according to the corresponding indexes in the 1×dim optimal vector output by the final algorithm, and reconstructed into the weight and threshold matrices of the corresponding dimensions.

[0089] 3.3 Model Evaluation Indicators

[0090] In order to verify the fitting effect and advantages of the IWOA-Elman model, this paper intends to establish a variety of different regression prediction models for comparative analysis. In order to compare the advantages and disadvantages of each model and analyze the effectiveness and accuracy of the prediction model, the following evaluation indicators are used: mean absolute error, root mean square error, mean absolute percentage error and determination coefficient.

[0091] The first three can better reflect the error between the predicted value and the true value. The smaller the result, the better the prediction performance of the model for the filling strength. The larger the determination coefficient (close to 1), the better the model fitting effect and the higher the prediction accuracy.

[0092] 4. Algorithm simulation and example prediction analysis

[0093] 4.1 Algorithm Performance Test

[0094] Through the experiment, 8 benchmark test functions were finally selected to test the optimization ability of the improved algorithm in terms of convergence speed, convergence accuracy and global search ability. In order to more intuitively demonstrate the advantages of the improved algorithm, genetic algorithm, particle swarm algorithm, gray wolf algorithm and standard whale algorithm were introduced for comparative testing. In order to ensure the reliability of the results, each algorithm was independently repeated 30 times, with a population of 30, a function dimension of 30, and an iteration number of 500. The corresponding IWOA and other four algorithms have a convergence curve comparison chart in the attached figure. Figure 5 ,Analysis shows that the traditional algorithm has premature ,maturing phenomenon in most of the test functions, that is, it falls into the local optimal solution, ,and the convergence curve basically presents a stable straight line, ,and the optimization effect is not ideal. The ,improved algorithm represented by the red curve has excellent ,performance in convergence accuracy, convergence speed and stability. ,By introducing strategies such as inertia weight and Levy ,flight, inflection points appear multiple times in the convergence process, ,which effectively improves the premature ,maturing phenomenon and verifies the effectiveness and stability of the ,improved algorithm.

[0095] 4.2 Analysis of filling strength prediction results

[0096] Four prediction models, Elman, PSO-Elman, WOA-Elman and IWOA-Elman, were tested. Each model was run 10 times independently and the final prediction results were averaged. The analysis and summary of the experiment are shown in the attached Figure 6 and 7 , we can see that the last one is the IWOA-Elman model proposed by the present invention. First of all, in the training set, all indicators are better than the other models. While ensuring high fitting degree during training, it also has excellent performance in prediction, which proves that the algorithm used in this paper has a significant improvement effect on the Elman neural network and can well predict the strength of the filling body.

[0097] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A mine filling strength prediction method based on IWOA-Elman neural network, characterized in that: The following steps are involved: Step 1: Improve the Elman neural network by introducing the IWOA algorithm to optimize the network weights and thresholds; Step 2: Collect and organize mine filling ratio test data as a training data source, where the input parameters include but are not limited to the mass fractions of cement, fly ash and tailings, which directly reflect the material composition of the filling body; Step 3: Set the compressive strength of the filling body as the output parameter of the model, i.e. the prediction target; Step 4: Based on the weights and thresholds optimized in step 1, an IWOA-Elman regression prediction model is constructed, and the model is trained using the training data in step 2 to learn the complex mapping relationship between input parameters and output parameters; Step 5: Evaluate the performance of the model and compare it with the existing model; Step 6: Verify the prediction accuracy of the model through actual data.

2. The mine filling strength prediction method based on IWOA-Elman neural network according to claim 1 is characterized in that: The IWOA algorithm is improved by introducing strategies such as chaos mapping, adaptive weight and Levy flight.

3. The method for predicting the strength of mine filling body based on IWOA-Elman neural network according to claim 1 is characterized in that: It also includes preprocessing of training data, including but not limited to normalization and shuffling, to enhance the generalization ability of the model and avoid overfitting.

4. The mine filling strength prediction method based on IWOA-Elman neural network according to claim 1 is characterized in that: The performance of the IWOA-Elman model is evaluated by constructing a comprehensive evaluation system, including but not limited to comparative analysis with the Elman, PSO-Elman and WOA-Elman models.

5. The mine filling strength prediction method based on IWOA-Elman neural network according to claim 1 is characterized in that: The evaluation indicators include root mean square error (RMSE), mean absolute percentage error (MAPE), mean absolute error and coefficient of determination (R 2 ).

6. A mine filling strength prediction method based on IWOA-Elman neural network according to any one of claims 1 to 5, characterized in that: The method can achieve high-precision prediction of the strength of mine filling bodies and has important engineering application value.

Citation Information

Cited By

  • Rock burst prediction method based on improved WOA optimization FTWSVM

    CN120429736A

  • Performance prediction and parameter optimization method and system for proton exchange membrane fuel cell

    CN120473010A

  • Early-stage monitoring data-based filling body later-stage strength prediction method

    CN121705666A

  • A method for predicting the late strength of a fill based on early monitoring data

    CN121705666B