Waste steel stockpiling volume data prediction method based on unmanned aerial vehicle three-dimensional collection

Through the improved Red-tailed Kite optimization algorithm and BP neural network, combined with the three-dimensional point cloud data collected by the drone, a WRTH-BP prediction model was established, which solved the problems of low efficiency and poor accuracy of traditional scrap steel storage volume measurement, and achieved high efficiency , accurate volume prediction and management.

CN119068041APending Publication Date: 2024-12-03OUYE LIANJIN RENEWABLE RESOURCES CO LTD +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411074148.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Traditional scrap steel storage volume measurement relies on manual measurement, with low efficiency and large errors. The existing drone technology mainly focuses on point cloud data acquisition and preliminary processing, and lacks in-depth volume prediction and management research, especially in complex environments. Accuracy and stability are difficult to guarantee.

Method used

Combining the improved Red-tailed Kite optimization algorithm and BP neural network, the drone collects three-dimensional point cloud data for preprocessing, feature extraction and model training, establishes a WRTH-BP prediction model, optimizes hidden layer weights and biases, and realizes scrap steel stacks Efficient and accurate prediction of volume storage.

Benefits of technology

It improves the accuracy and stability of scrap steel storage volume prediction, enhances the algorithm's global search ability and convergence performance, solves the problems of inefficiency and large errors of traditional methods, and achieves more efficient volume management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068041A_ABST
    Figure CN119068041A_ABST
Patent Text Reader

Abstract

The invention discloses a waste steel stockpiling volume data prediction method based on unmanned aerial vehicle three-dimensional collection, and relates to the technical field of data prediction, and the method comprises the steps: S1, employing an unmanned aerial vehicle to collect three-dimensional point cloud data of waste steel stockpiling, carrying out the preprocessing of the data, and enabling the preprocessing to comprise noise filtering, missing point complementation and point cloud alignment, high-quality three-dimensional point cloud data is obtained; s2, effective characteristic data are extracted from the preprocessed three-dimensional point cloud data, the effective characteristic data comprise the density, the distance, the angle, the height and the like of waste steel stockpiling volume points, and the effective characteristic data serve as input of a BP neural network; s3, adjusting the hidden layer weight beta and offset alpha of the BP neural network through an improved red tail plinth optimization algorithm, and establishing a waste steel stockpiling volume WRTH-BP prediction model; and S4, performing volume prediction on the newly collected three-dimensional point cloud data by using the WRTH-BP prediction model, and outputting an accurate volume value of the steel scrap stockpiling.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data prediction, and in particular to a method for predicting scrap steel stockpile volume data based on three-dimensional acquisition by unmanned aerial vehicles. Background Art

[0002] In scrap steel management, accurate prediction of scrap steel stockpile volume is of great significance for optimizing resource utilization and reducing environmental impact. Traditional methods rely on manual measurement, which is inefficient and has large errors. The development of drone technology has made it possible to automatically measure the volume of scrap steel stockpile. By acquiring three-dimensional data through drones, a three-dimensional model of scrap steel stockpile can be quickly and accurately constructed.

[0003] As an important recyclable resource, the storage and management of scrap steel are of great significance to the reuse of resources. However, the volume measurement and management of scrap steel stockpiles has always been a complex and critical issue. Traditional volume measurement methods mainly rely on manual measurement and estimation, which have disadvantages such as low accuracy, low efficiency, and high labor intensity. In recent years, with the development of drone technology, the use of drones to measure the volume of scrap steel stockpiles has become a new trend. Drones can quickly and accurately obtain three-dimensional data of scrap steel stockpiles. Combined with advanced point cloud processing technology, accurate measurement and dynamic monitoring of the volume of scrap steel stockpiles can be achieved. However, current research and applications mainly focus on the collection and preliminary processing of point cloud data, and there is still a lack of in-depth research on how to use these data for accurate volume prediction and management. In particular, how to ensure the accuracy and stability of measurement in a complex environment is still a problem that needs to be solved urgently.

[0004] BP neural network (Back Propagation Neural Network, BP) is a classic artificial neural network model, widely used in data prediction, classification and regression tasks. BP neural network adjusts network weights through the back propagation algorithm to achieve learning and prediction of complex data patterns. Compared with other neural network models, BP neural network has a simple structure, is easy to implement and has high computational efficiency, and is suitable for prediction tasks of various data types. In the industrial field, BP neural network is used to predict key indicators in the production process, such as equipment failure rate, product quality and energy consumption.

[0005] The Red-tailed Hawk Optimization (RTH) algorithm is an emerging nature-inspired optimization algorithm that simulates the intelligent behavior of red-tailed kites in the hunting process in nature. The algorithm achieves a balance between global search and local search by simulating the process of red-tailed kites searching for prey, including the stages of preliminary search, tracking, and final capture of prey. The Red-tailed Hawk Optimization algorithm performs well in solving complex optimization problems, especially in dealing with nonlinear, multi-peak, and high-dimensional optimization problems, with strong adaptability and robustness. At present, the algorithm has been widely used in engineering optimization, image processing, machine learning parameter optimization and other fields. However, the application of the Red-tailed Hawk Optimization algorithm in the field of UAV three-dimensional data processing and volume prediction requires further research and practical verification to improve its effectiveness and efficiency in practical applications.

[0006] Publication No. CN117452806A discloses an improved red-tailed kite optimization algorithm, which improves the gravity factor of the red-tailed kite algorithm. In the improvement process, the gravity factor is improved nonlinearly by using the relationship between the current number of iterations and the maximum number of iterations. The relationship between the current number of iterations and the maximum number of iterations can reduce the risk of the algorithm falling into the local optimum to a certain extent. Simply relying on the relationship between the current number of iterations and the maximum number of iterations does not take into account the diversity of the group. Diversity is an important factor in preventing the optimization algorithm from falling into the local optimal solution. If the diversity of the group is too low, the algorithm may converge to the local optimal solution in the early stage and fail to find the global optimal solution; secondly, due to the lack of a feedback mechanism for the group state, simply relying on the number of iterations to adjust the gravity factor may cause the algorithm to easily fall into the local optimal solution when encountering complex or multi-peak optimization problems. Summary of the invention

[0007] The present invention aims to provide a method for predicting scrap steel stockpile volume data based on three-dimensional acquisition by unmanned aerial vehicles, and realizes efficient and accurate prediction of scrap steel stockpile volume by combining an improved red-tailed kite optimization algorithm (WRTH) and a BP neural network (BP).

[0008] The technical solution adopted by the present invention to achieve the above-mentioned objectives is: optimizing the BP neural network model based on the improved red-tailed kite optimization algorithm, which is used to process the three-dimensional point cloud data collected by the drone and predict the volume of the scrap steel stockpile; a method for predicting the scrap steel stockpile volume data based on three-dimensional collection by the drone includes four main steps: data preprocessing, feature extraction, model training and volume prediction, and the specific steps are as follows.

[0009] S1. Use a drone to collect three-dimensional point cloud data of scrap steel stockpiles, and preprocess the data. The preprocessing includes noise filtering, missing point completion and point cloud alignment to obtain high-quality three-dimensional point cloud data.

[0010] S2. Extracting effective feature data from the preprocessed three-dimensional point cloud data, wherein the effective feature data includes density, distance, angle, height, etc. of the scrap steel stockpile volume points, and using the effective feature data as input of a BP neural network for training.

[0011] S3, adjusting the hidden layer weight β and bias α of the BP neural network through an improved red-tailed kite optimization algorithm to establish a scrap steel stockpile volume WRTH-BP prediction model; the improved red-tailed kite optimization algorithm includes:

[0012] Step 1: Improve the transition factor of the Red-tailed Kite Optimization Algorithm. Use different forms of transition factors at different stages of the Red-tailed Kite Optimization Algorithm to improve the position update mathematical model of the global search stage of the standard Red-tailed Kite Optimization Algorithm.

[0013] Step 2: Improve the gravity factor of the red-tailed kite optimization algorithm, and combine it with the improved transition factor γ(t) to improve the position update mathematical model of the local development stage of the standard red-tailed kite optimization algorithm;

[0014] S4. Use the WRTH-BP prediction model to predict the volume of the newly collected 3D point cloud data and output the accurate volume value of the scrap steel stockpile.

[0015] Furthermore, in step S1, the mathematical model of data preprocessing is as follows:

[0016] P clean =PC;

[0017] Where P is the point cloud data, and the mathematical model is: P = {p1, p2, p3, ..., p q}, C is the noise point cloud, and the mathematical model is: C = {c1, c2, c3, ..., c s}, q is the number of point cloud data, s is the number of noise point clouds, P clean The 3D point cloud data of the pre-processed scrap steel stockpile.

[0018] Furthermore, the missing points are completed using a neighborhood interpolation algorithm. Suppose the missing points of the three-dimensional point cloud data of the scrap steel stockpile are M = {m1, m2, m3, ..., m k}, for each missing point in the 3D point cloud data of the scrap steel stockpile, set the neighborhood size and find the nearest known points around it; use the k-nearest neighbor algorithm (k-NN) to find the missing points in the neighborhood, and use the weighted average method to interpolate the missing points in the 3D point cloud data of the scrap steel stockpile, where the weight is determined according to the distance between the missing point and the yth neighborhood point. The mathematical model is:

[0019]

[0020] In the formula, w yis the weight of the y-th domain point, a y is the distance between the kth missing point and the yth neighboring point in its neighborhood, ∈ is a minimum value, and its value is 0.001;

[0021] The interpolation mathematical model is:

[0022]

[0023] In the formula, u c is the coordinates of the missing points in the 3D point cloud data of the scrap steel stockpile, u y is the coordinate of the yth neighborhood point;

[0024] The least square method is used to fit a plane, and the coordinates of the missing points of the three-dimensional point cloud data of the scrap steel stockpile are projected onto the fitting plane to obtain the interpolation value, where the mathematical model of the projection is:

[0025]

[0026] In the formula, u proj are the coordinates after projection, v, b, z are the plane parameters, o, r, z are the horizontal axis, vertical axis and vertical axis.

[0027] Furthermore, in step S2, effective feature data is extracted from the preprocessed three-dimensional point cloud data. The effective feature data is extracted using the Kendall rank correlation coefficient method, and the mathematical model is:

[0028]

[0029] Where n is the number of predicted values ​​of the three-dimensional point cloud data of scrap steel storage, x i and x j For the two rankings in the first dataset, y i and j is the corresponding ranking in the second dataset, sgn(x i -x j ) is a symbolic function, if x i -x j >0, then the sgn function value is 1, x i -x j If sgn is less than 0, the function value is -1; otherwise, the function value is 0.

[0030] Furthermore, the scrap steel stockpile volume RTH-BP prediction model in step S2 includes an improved red-tailed kite optimization algorithm module, a three-dimensional point cloud data input module and an output module of the scrap steel stockpile, a forward propagation module, and a backward propagation module; the forward propagation module includes a hidden layer and an output layer, and the weight β and bias α in the hidden layer are adjusted using the improved red-tailed kite optimization algorithm;

[0031] The mathematical model for the calculation from the input module to the hidden layer is:

[0032]

[0033] In the formula, h j is the output value of the jth neuron of the hidden layer of the three-dimensional point cloud data of the scrap steel stockpile; F is the activation function; x ir is the 3D point cloud data value of the irth scrap steel stockpile input into the RTH-BP prediction model, β ir,j is the weight from the irth feature of the input layer to the jth neuron of the hidden layer, α j is the bias of the jth neuron in the hidden layer;

[0034] The computational mathematical model from the hidden layer to the output module is:

[0035]

[0036] In the formula, vr is the predicted value of scrap steel stockpile volume output by the RTH-BP prediction model; j The weight from the jth neuron in the hidden layer to the output layer, cr is the bias of the output layer.

[0037] Furthermore, the step S3, the improved red-tailed kite optimization algorithm, improves the position update mathematical model of the global search phase of the standard red-tailed kite optimization algorithm by improving the transition factor of the red-tailed kite optimization algorithm, and the specific steps are:

[0038] Different forms of transition factors are used in different stages of the Red Kite Optimization Algorithm, and the adaptive fitness value feedback mechanism and random disturbance intensity control parameters are introduced to improve the transition factor TF(t) of the Red Kite Optimization Algorithm;

[0039]

[0040] In formula (1), γ(t) is the improved transition factor, t is the current iteration number, T is the maximum iteration number, is the minimum fitness value of the tth iteration, is the average fitness value of the tth iteration, f best is the global minimum fitness value, λ(t) is the random disturbance intensity control parameter, and the mathematical model is:

[0041]

[0042] In formula (2), λ0 is the initial value of the random disturbance intensity control parameter, which is 2, and σ f,max is the maximum value of the standard deviation of the group fitness of the red-tailed kite optimization algorithm, and σf(t) is the standard deviation of the group fitness at the tth iteration;

[0043] D2. By using formula (3), the transition factor γ(t) of the improved red-tailed kite optimization algorithm is used to improve the mathematical model of position update in the global search phase of the red-tailed kite optimization algorithm;

[0044] X(t)=X best +(X mean -X(t+1))×Levy(dim)×γ (3);

[0045] In formula (3), X(t) is the latest position of the red-tailed kite at the tth iteration, X mean is the average position of the red-tailed kite population, X best is the global optimal position of the red-tailed kite, Levy is the flight distribution function, γ is the transition factor of the tth iteration improvement, and dim is the problem dimension.

[0046] Furthermore, by combining the number of iterations and the difference in group fitness, the improved transition factor can enhance the exploration ability in the early stage and the development ability in the later stage, so as to better balance exploration and development; in the early stage, the transition factor combines the difference in fitness to ensure that when the group diversity is high, a larger step size is maintained to enhance the exploration ability; in the later stage, by introducing random perturbations and standard deviation adjustment mechanisms, appropriate diversity is maintained to prevent premature convergence to the local optimum; at the same time, the dynamic adjustment of λ enables the random perturbation intensity to be adaptively adjusted according to the fitness distribution of the group, further enhancing the adaptability and robustness of the algorithm; in the later stage, by introducing random perturbation terms, the algorithm is guaranteed to have a certain degree of randomness when it is close to convergence, so as to prevent it from falling into the local optimal solution.

[0047] Furthermore, the step S3 improves the position update mathematical model of the local development phase of the standard red-tailed kite optimization algorithm by improving the gravity factor of the red-tailed kite optimization algorithm and combining the improved transition factor γ(t), and the specific steps are as follows:

[0048] Y1. Calculate the distribution density H(t) of the red-tailed kite population through the distribution of fitness. The mathematical model is:

[0049]

[0050] In formula (4), is the fitness value of the i-th red-tailed kite position in the t-th iteration, and N is the maximum size of the red-tailed kite;

[0051] Y2, combining the distribution density H(t) and the number of iterations t to improve the gravity factor of the red-tailed kite optimization algorithm, the improved gravity factor g(t);

[0052]

[0053] In formula (5), H(t) is the density of the red-tailed kite population at the tth iteration, and Hmax is the maximum value of the distribution density;

[0054] Y3. Improve the mathematical model of position update of the red-tailed kite optimization algorithm in the local development stage by using the improved gravity factor g(t) and the improved transition factor γ(t);

[0055] X(t)=a(t)·X best +x(t)·StepSize1(i)+y(t)·StepSize2(t) (6);

[0056] In formula (6), a(t) is the acceleration coefficient of the tth iteration, StepSize1(t) is the first step length, and the mathematical model is:

[0057] StepSize1(t)=X(t)-γ(t)·X mean ;

[0058] StepSize2(t) is the second step size, and the mathematical model is:

[0059] StepSize2(t)=g(t)×X(t)-γ(t)·X best ; g(t) is the improved gravity factor.

[0060] Furthermore, by introducing the entropy value H(t), the diversity of the group can be effectively measured. When the group diversity is high, the gravity factor is larger, which enhances the exploration ability; when the group diversity is low, the gravity factor is smaller, which enhances the development ability and prevents premature convergence to the local optimal solution; at the same time, the dynamically adjusted gravity factor helps to enhance the exploration ability in the early stage of optimization and the development ability in the later stage, thereby better balancing global search and local optimization, converging to high-quality solutions faster, reducing the number of iterations, and improving the solution accuracy.

[0061] Furthermore, the specific steps of adjusting the hidden layer weight β and bias α of the BP neural network using the improved red-tailed kite optimization algorithm are as follows:

[0062] S31, encoding the hidden layer weight β and bias α of the BP neural network for scrap steel stockpile volume prediction into a space vector U, and establishing a two-dimensional mapping between the space vector U and the individual position of the red-tailed kite of the red-tailed kite optimization algorithm;

[0063] S32, using formula (7) to initialize the population position of the red-tailed kite optimization algorithm, setting parameters including population size N, problem dimension dim, algorithm search space upper bound UB, algorithm search space lower bound LB, maximum number of iterations T, initial number of iterations t=0, and fitness function;

[0064]

[0065] In formula (7), X(0) is the initial position of the red-tailed kite, η i,j is the position of the i-th red-tailed kite individual in the j-th dimension, j = 1, 2, 3..., dim, rand (0, 1) is a random number between 0 and 1;

[0066] S33, judging whether the current number of iterations t satisfies t=T, if so, then exiting the setting of the hidden layer weight β and bias α of the BP neural network, outputting the individual position of the red-tailed kite corresponding to the global minimum fitness value and parsing it into the space vector of the hidden layer weight β and bias α of the BP neural network, and obtaining the best hidden layer weight β and bias α parameter value of the BP neural network for predicting the scrap steel stockpile volume; otherwise, executing step S34;

[0067] S34, using the fitness function to calculate the individual position fitness value of the current iteration red-tailed kite population, the minimum fitness value is And compared with the fitness value of the last iteration Compare and take the smaller fitness value as the current best fitness value;

[0068] S35, establishing a population position update formula of the red-tailed kite optimization algorithm in the global search phase, the local development phase, and the transition phase, and updating the individual positions of the red-tailed kites, the specific steps are:

[0069] S351, in the high-flying stage, the red-tailed kite will search for the best position of the prey in the air, and establish the mathematical model of the position update of the red-tailed kite optimization algorithm in the global search stage according to formula (3);

[0070] S352, in the low-flying stage, the red-tailed kite will fly in a spiral around the prey position, select the best attack time and position, and establish a mathematical model for the position update of the red-tailed kite optimization algorithm in the transition stage according to formula (8);

[0071] X(t)=X best +(x(t)+y(t))·StepSize(t) (8):

[0072] In formula (8), x(t) and y(t) are direction coordinates, the mathematical model of x(t) is: x(t) = R(t)·sin(θ(t)); the mathematical model of y(t) is y(t) = R(t)·cos(θ(t)); R((t) is the spiral flight radius, a random number between 0 and 1, θ(t) is the angle gain, StepSize(t) is the step size at the current iteration t, and the mathematical model is: StepSize(t) = X(t)-X mean ;

[0073] S353, in the diving stage, the red-tailed kite suddenly dives from the best position to attack the prey, and the mathematical model of the position update of the red-tailed kite optimization algorithm in the local development stage is established according to formula (6);

[0074] S36, execute t=t+1 for the current iteration number t, and return to step S33.

[0075] Furthermore, the fitness function f is:

[0076]

[0077] Where n′ is the number of three-dimensional data samples of scrap steel storage, W i′ is the true value of the three-dimensional data sample of the i′th scrap steel stockpile, Predicted value of the 3D data sample of the i′th scrap stockpile.

[0078] The beneficial effects of the present invention are as follows: the improved transition factor makes the red-tailed kite optimization algorithm more efficient and stable in solving complex optimization problems by combining fitness differences, introducing random disturbances and dynamic adjustment mechanisms. The improved model can effectively improve the global search capability and convergence performance of the algorithm by better balancing exploration and development, increasing randomness and considering population diversity; by introducing distribution density and dynamic adjustment mechanisms, the improved gravity factor model has significant advantages in maintaining population diversity, balancing exploration and development, and improving convergence speed and accuracy; compared with the existing gravity factor model, the improved model is more flexible and adaptable, and can better cope with complex optimization problems. The improved red-tailed kite optimization algorithm is used to adjust the hidden layer weights β and bias α of the BP neural network, and a scrap steel stockpile volume WRTH-BP prediction model is established to solve the problem that the traditional method of scrap steel stockpile volume prediction relies on manual measurement, which is inefficient and has large errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 Diagram of the steps for predicting scrap steel stockpile volume based on the improved Red Kite optimization algorithm.

[0080] Figure 2 Flowchart for adjusting the hidden layer weights β and bias α of the BP neural network based on the improved red-tailed kite optimization algorithm.

[0081] Figure 3 This is a training set comparison chart of the Red-tailed Kite optimization algorithm-BP prediction model and the improved Red-tailed Kite optimization algorithm-BP prediction model for scrap steel stockpile volume prediction.

[0082] Figure 4 This is a comparison chart of the fitness of the Red-tailed Kite optimization algorithm-BP prediction model and the improved Red-tailed Kite optimization algorithm-BP prediction model for scrap steel stockpile volume prediction.

[0083] Figure 5 This is a comparison chart of the errors in predicting scrap steel stockpile volume using the Red-tailed Kite optimization algorithm-BP prediction model and the improved Red-tailed Kite optimization algorithm-BP prediction model.

[0084] Figure 6 This is a comparison chart of the effects of the Red-tailed Kite optimization algorithm-BP prediction model and the improved Red-tailed Kite optimization algorithm-BP prediction model on the prediction of scrap steel stockpile volume. DETAILED DESCRIPTION

[0085] 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.

[0086] See also Figure 1-Figure 6 The present invention provides a technical solution: a method for predicting scrap steel storage volume data based on three-dimensional acquisition by unmanned aerial vehicles includes four main steps: data preprocessing, feature extraction, model training and volume prediction. The specific steps are as follows: Figure 1 shown.

[0087] S1. Use a drone to collect three-dimensional point cloud data of scrap steel stockpiles, and preprocess the data. The preprocessing includes noise filtering, missing point completion and point cloud alignment to obtain high-quality three-dimensional point cloud data.

[0088] More specifically, the mathematical model of data preprocessing is as follows:

[0089] P clean = PC;

[0090] Where P is the point cloud data, and the mathematical model is: P = {p1, p2, p3, ..., p q}, C is the noise point cloud, and the mathematical model is: C = {c1, c2, c3, ..., c s}, q is the number of point cloud data, s is the number of noise point clouds, P clean The 3D point cloud data of the pre-processed scrap steel stockpile.

[0091] More specifically, the missing points are completed using a neighborhood interpolation algorithm. Assume that the missing points of the three-dimensional point cloud data of the scrap steel stockpile are M = {m1, m2, m3, ..., m k}, for each missing point in the 3D point cloud data of the scrap steel stockpile, set the neighborhood size and find the nearest known points around it; use the k-nearest neighbor algorithm (k-NN) to find the missing points in the neighborhood, and use the weighted average method to interpolate the missing points in the 3D point cloud data of the scrap steel stockpile, where the weight is determined according to the distance between the missing point and the yth neighborhood point. The mathematical model is:

[0092]

[0093] In the formula, w y is the weight of the y-th domain point, d y is the distance between the kth missing point and the yth neighboring point in its neighborhood, ∈ is a minimum value, and its value is 0.001;

[0094] The interpolation mathematical model is:

[0095]

[0096] In the formula, u c is the coordinates of the missing points in the 3D point cloud data of the scrap steel stockpile, u y is the coordinate of the yth neighborhood point;

[0097] The least square method is used to fit a plane, and the coordinates of the missing points of the three-dimensional point cloud data of the scrap steel stockpile are projected onto the fitting plane to obtain the interpolation value, where the mathematical model of the projection is:

[0098]

[0099] In the formula, u proj are the coordinates after projection, v, b, z are the plane parameters, o, r, z are the horizontal axis, vertical axis and vertical axis.

[0100] S2. Extracting effective feature data from the preprocessed three-dimensional point cloud data, wherein the effective feature data includes density, distance, angle and height of the scrap steel stockpile volume points, and using the effective feature data as input of a BP neural network for training.

[0101] More specifically, effective feature data is extracted from the preprocessed three-dimensional point cloud data. The effective feature data extraction adopts the Kendall rank correlation coefficient method, and the mathematical model is:

[0102]

[0103] Where n is the number of predicted values ​​of the three-dimensional point cloud data of scrap steel storage, x i and x j For the two rankings in the first dataset, y i and j is the corresponding ranking in the second dataset, sgn(xi -x j ) is a symbolic function, if x i -x j >0, then the sgn function value is 1, x i -x j If sgn is less than 0, the function value is -1; otherwise, the function value is 0.

[0104] More specifically, the scrap steel stockpile volume RTH-BP prediction model includes an improved red-tailed kite optimization algorithm module, a three-dimensional point cloud data input module and an output module of the scrap steel stockpile, a forward propagation module, and a back propagation module; the forward propagation module includes a hidden layer and an output layer, and the weight β and bias α in the hidden layer are adjusted using the improved red-tailed kite optimization algorithm;

[0105] The mathematical model for the calculation from the input module to the hidden layer is:

[0106]

[0107] In the formula, h j is the output value of the jth neuron of the hidden layer of the three-dimensional point cloud data of the scrap steel stockpile; F is the activation function; x ir is the 3D point cloud data value of the irth scrap steel stockpile input into the RTH-BP prediction model, β ir,j is the weight from the irth feature of the input layer to the jth neuron of the hidden layer, α j is the bias of the jth neuron in the hidden layer;

[0108] The computational mathematical model from the hidden layer to the output module is:

[0109]

[0110] In the formula, vr is the predicted value of scrap steel stockpile volume output by the RTH-BP prediction model; j The weight from the jth neuron in the hidden layer to the output layer, cr is the bias of the output layer.

[0111] S3, adjusting the hidden layer weight β and bias α of the BP neural network through an improved red-tailed kite optimization algorithm to establish a scrap steel stockpile volume WRTH-BP prediction model; the improved red-tailed kite optimization algorithm includes:

[0112] Step 1: Improve the transition factor of the Red-tailed Kite Optimization Algorithm. Use different forms of transition factors at different stages of the Red-tailed Kite Optimization Algorithm to improve the position update mathematical model of the global search stage of the standard Red-tailed Kite Optimization Algorithm. The specific steps are:

[0113] D1. Use different forms of transition factors at different stages of the red-tailed kite optimization algorithm, introduce an adaptive fitness value feedback mechanism and random perturbation intensity control parameters to improve the transition factor TF(t) of the red-tailed kite optimization algorithm;

[0114]

[0115] In formula (1), γ(t) is the improved transition factor, t is the current iteration number, T is the maximum iteration number, is the minimum fitness value of the tth iteration, is the average fitness value of the tth iteration, f best is the global minimum fitness value, λ(t) is the random disturbance intensity control parameter, and the mathematical model is:

[0116]

[0117] In formula (2), λ0 is the initial value of the random disturbance intensity control parameter, which is 2, and σ f,max is the maximum value of the standard deviation of the group fitness of the red-tailed kite optimization algorithm, and σf(t) is the standard deviation of the group fitness at the tth iteration;

[0118] D2. By using formula (3), the transition factor γ(t) of the improved red-tailed kite optimization algorithm is used to improve the mathematical model of position update in the global search phase of the red-tailed kite optimization algorithm;

[0119] X(t)=X best +(X mean -X(t+1))×Levy(dim)×γ (3);

[0120] In formula (3), X(t) is the latest position of the red-tailed kite at the tth iteration, X mean is the average position of the red-tailed kite population, X best is the global optimal position of the red-tailed kite, Levy is the flight distribution function, γ is the transition factor improved in the tth iteration, and dim is the problem dimension.

[0121] Step 2: Improve the gravity factor of the red-tailed kite optimization algorithm, and combine it with the improved transition factor γ(t) to improve the position update mathematical model of the local development stage of the standard red-tailed kite optimization algorithm; the specific steps are:

[0122] Y1. Calculate the distribution density H(t) of the red-tailed kite population through the distribution of fitness. The mathematical model is:

[0123]

[0124] In formula (4), is the fitness value of the i-th red-tailed kite position in the t-th iteration, and N is the maximum size of the red-tailed kite;

[0125] Y2, combining the distribution density H(t) and the number of iterations t to improve the gravity factor of the red-tailed kite optimization algorithm, the improved gravity factor g(t);

[0126]

[0127] In formula (5), H(t) is the density of the red-tailed kite population at the tth iteration, and Hmax is the maximum value of the distribution density;

[0128] Y3. Improve the mathematical model of position update of the red-tailed kite optimization algorithm in the local development stage by using the improved gravity factor g(t) and the improved transition factor γ(t);

[0129] X(t)=a(t)·X best +x(t)·StepSize1(t)+y(t)·StepSize2(t) (6);

[0130] In formula (6), a(t) is the acceleration coefficient of the tth iteration, StepSize1(t) is the first step length, and the mathematical model is:

[0131] StepSize1(t)=X(t)-γ(t)·X mean ;

[0132] StepSize2(t) is the second step size, and the mathematical model is:

[0133] StepSize2(t)=g(t)×X(t)-γ(t)·X best ; g(t) is the improved gravity factor.

[0134] More specifically, the hidden layer weights β and bias α of the BP neural network are adjusted by the improved red-tailed kite optimization algorithm, as shown in Figure 2 As shown, the specific steps are:

[0135] S31, encoding the hidden layer weight β and bias α of the BP neural network for scrap steel stockpile volume prediction into a space vector U, and establishing a two-dimensional mapping between the space vector U and the individual position of the red-tailed kite of the red-tailed kite optimization algorithm;

[0136] S32, using formula (7) to initialize the population position of the red-tailed kite optimization algorithm, setting parameters including population size N, problem dimension dim, algorithm search space upper bound UB, algorithm search space lower bound LB, maximum number of iterations T, initial number of iterations t=0, and fitness function;

[0137]

[0138] In formula (7), X(0) is the initial position of the red-tailed kite, η i,j is the position of the i-th red-tailed kite individual in the j-th dimension, j=1, 2, 3..., dim, rand(0, 1) is a random number between 0 and 1;

[0139] S33, judging whether the current number of iterations t satisfies t=T, if so, then exiting the setting of the hidden layer weight β and bias α of the BP neural network, outputting the individual position of the red-tailed kite corresponding to the global minimum fitness value and parsing it into the space vector of the hidden layer weight β and bias α of the BP neural network, and obtaining the best hidden layer weight β and bias α parameter value of the BP neural network for predicting the scrap steel stockpile volume; otherwise, executing step S34;

[0140] S34, using the fitness function to calculate the individual position fitness value of the current iteration red-tailed kite population, the minimum fitness value is And compared with the fitness value of the last iteration Compare and take the smaller fitness value as the current best fitness value;

[0141] S35, establishing a population position update formula of the red-tailed kite optimization algorithm in the global search phase, the local development phase, and the transition phase, and updating the individual positions of the red-tailed kites, the specific steps are:

[0142] S351, in the high-flying stage, the red-tailed kite will search for the best position of the prey in the air, and establish the mathematical model of the position update of the red-tailed kite optimization algorithm in the global search stage according to formula (3);

[0143] S352, in the low-flying stage, the red-tailed kite will fly in a spiral around the prey position, select the best attack time and position, and establish a mathematical model for the position update of the red-tailed kite optimization algorithm in the transition stage according to formula (8);

[0144] X(t)=X best +(x(t)+y(t))·StepSize(t) (8);

[0145] In formula (8), x(t) and y(t) are direction coordinates, the mathematical model of x(t) is: x(t) = R(t) sin(θ(t)); the mathematical model of y(t) is y(t) = R(t) cos(θ(t)); R(t) is the spiral flight radius, a random number between 0 and 1, θ(t) is the angle gain, StepSize(t) is the step size at the current iteration t, and the mathematical model is: StepSize(t) = X(t) - X mean ;

[0146] S353, in the diving stage, the red-tailed kite suddenly dives from the best position to attack the prey, and the mathematical model of the position update of the red-tailed kite optimization algorithm in the local development stage is established according to formula (5);

[0147] S36, execute t=t+1 for the current iteration number t, and return to step S33.

[0148] Furthermore, the fitness function f is:

[0149]

[0150] Where n′ is the number of three-dimensional data samples of scrap steel storage, W i′ is the true value of the three-dimensional data sample of the i′th scrap steel stockpile, Predicted value of the 3D data sample of the i′th scrap stockpile.

[0151] S4. Use the WRTH-BP prediction model to predict the volume of the newly collected 3D point cloud data and output the accurate volume value of the scrap steel stockpile.

[0152] More specifically, a standard red-tailed kite optimization algorithm (RTH) and an improved red-tailed kite optimization algorithm (WRTH) were used for comparison. The common parameters of RTH and WRTH were set as follows: population size N=100, dimension dim=2, boundary 1b=[0,0], ub=[20,20], maximum number of iterations T=40, and minimum fitness value of 0.1. The main function of the algorithm and the optimization algorithm were implemented using the m function written in Matlab. A BP neural network prediction model program was written, and the initial value of the number of input layer nodes was set to 10, the number of hidden layer nodes was set to 15, and the number of output layer nodes was set to 10. The training parameters were set, with the number of training times being 50, the minimum prediction error of the scrap steel stockpile volume being 1e-4, and the initial value of the learning rate being 0.01.

[0153] More specifically, the three-dimensional scrap steel stockpile data collected by the drone is input into the MATLAB program in XLSX format, and divided into training set and test set. Eighty percent of the three-dimensional scrap steel stockpile data is used as the training set, and twenty percent is used as the test set. The test set is used for real prediction. The total number of data samples is 115. The data set is shuffled, and the input feature dimension is 4. The program is run, and the three-dimensional scrap steel stockpile training set data is predicted using the WRTH-BP prediction model and the RTH-BP prediction model, as shown in Figure 2. Figure 3 As shown in the figure, the prediction results of the training set, it can be seen that the WRTH-BP prediction model is closer to the true value and more stable than the RTH-BP prediction model.

[0154] Input the test set data into the trained WRTH-BP prediction model and RTH-BP model, and the output is as follows Figure 4 , Figure 5 , Figure 6 .

[0155] More specifically, if Figure 4 As shown in the figure, the fitness change comparison diagram of the WRTH-BP prediction model and the RTH-BP model for scrap steel stockpile volume prediction, the fitness value adopts the absolute error design, the smaller the fitness value, the higher the prediction accuracy of the model used for scrap steel stockpile volume; the initial fitness value of RTH-BP for scrap steel stockpile volume prediction is 0.19, and the initial fitness value of WRTH-BP for scrap steel stockpile volume prediction is 0.18. With the increase of the number of iterations, the fitness value of WRTH-BP decreases rapidly. Compared with the RTH-BP model, the fitness value decreases faster. Within the same range of iterations, the fitness value of IRTH decreases more obviously. The results show that the improved algorithm has a more significant effect on improving fitness in the initial iteration; in the middle of the iteration, during the 5th to 10th iteration, the fitness value of RTH-BP began to gradually decrease, and dropped to about 0.15 at the 10th iteration, while the fitness value of WRTH-BP continued to decrease, and dropped to about 0.13 at the 10th iteration, continuing to outperform RTH-BP. From the final fitness value, the fitness value of RTH-BP is about 0.13, while the improved IRTH fitness value is even lower, about 0.11, which shows that WRTH-BP has a better effect in predicting the scrap steel stockpile volume.

[0156] More specifically, if Figure 5 As shown in the figure, the output errors of the WRTH-BP model and the RTH-BP model in the prediction of scrap steel stockpile volume change with the prediction samples. The prediction error range of the RTH-BP model is larger and the fluctuation is more violent. The error range is approximately between -20 and 20, while the prediction error of the WRTH-BP model is relatively small and the fluctuation range is narrow. The error range is approximately between -5 and 5. From the trend of the curve, the error mean of the RTH-BP model deviates greatly from 0, indicating that the deviation of the model in the prediction of scrap steel stockpile volume is large; the error mean of the WRTH-BP model is close to 0, indicating that the model is more accurate in the prediction of scrap steel stockpile volume and has a smaller deviation.

[0157] More specifically, if Figure 6As shown in the figure, the output values ​​of the WRTH-BP model and the RTH-BP model for predicting the scrap steel stockpile volume. The prediction result of the RTH-BP model is unstable. With the change of samples, the predicted value fluctuates greatly, and there are obvious high and low errors; the prediction result of the WRTH-BP model is relatively stable, the difference between the predicted value and the true value is small, and the fluctuation is also relatively gentle; in summary, the performance of the WRTH-BP model in the prediction of the scrap steel stockpile volume is significantly better than that of the original RTH-BP model. The WRTH-BP model is not only closer to the true value in the overall trend, but also has smaller errors and higher stability, which verifies its effectiveness and superiority in the actual scrap steel stockpile volume prediction application.

[0158] It should be noted that, in the description of the invention, the directions or positional relationships indicated by the terms "up", "down", "left", "right", "front", "back", etc. are descriptions of the structure of the present invention based on the accompanying drawings, and are only for the convenience of describing the present invention, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0159] The "first" and "second" in this technical solution are only used to distinguish the names of the same or similar structures, or corresponding structures with similar functions, and are not an arrangement of the importance of these structures, nor do they have a ranking, comparison of size, or other meanings.

[0160] In addition, unless otherwise clearly specified and limited, the terms "installation" and "connection" should be understood in a broad sense. For example, the connection can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be a connection between the two structures. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood based on the overall idea of ​​the present invention and the specific context of the present solution.

Claims

1. A method for predicting scrap steel stockpile volume data based on three-dimensional acquisition by unmanned aerial vehicles, comprising four steps of data preprocessing, feature extraction, model training and volume prediction, characterized in that: The specific steps are: S1. Using a drone to collect three-dimensional point cloud data of scrap steel stockpiles, and preprocessing the data, wherein the preprocessing includes noise filtering, missing point completion, and point cloud alignment; S2, extracting effective feature data from the preprocessed three-dimensional point cloud data, wherein the effective feature data includes density, distance, angle and height of the scrap steel stockpile volume points, and using the effective feature data as input of the BP neural network for training; S3, adjusting the hidden layer weight β and bias α of the BP neural network through an improved red-tailed kite optimization algorithm to establish a scrap steel stockpile volume WRTH-BP prediction model; the improved red-tailed kite optimization algorithm includes: Step 1: Improve the transition factor of the Red-tailed Kite Optimization Algorithm. Use different forms of transition factors at different stages of the Red-tailed Kite Optimization Algorithm to improve the position update mathematical model of the global search stage of the standard Red-tailed Kite Optimization Algorithm. Step 2: Improve the gravity factor of the red-tailed kite optimization algorithm, and combine it with the improved transition factor γ(t) to improve the position update mathematical model of the local development stage of the standard red-tailed kite optimization algorithm; S4. Use the WRTH-BP prediction model to predict the volume of the newly collected 3D point cloud data and output the accurate volume value of the scrap steel stockpile.

2. The method for predicting scrap steel stockpile volume data based on three-dimensional acquisition by unmanned aerial vehicles according to claim 1, characterized in that: In step S1, the mathematical model of data preprocessing is as follows: P clean =P-C; Where P is the point cloud data, and the mathematical model is: P = {p1, p2, p3, ..., p q }, C is the noise point cloud, and the mathematical model is: C = {c1, c2, c3, ..., c s }, q is the number of point cloud data, s is the number of noise point clouds, P clean The 3D point cloud data of the pre-processed scrap steel stockpile.

3. The method for predicting scrap steel storage volume data based on three-dimensional acquisition by unmanned aerial vehicles according to claim 1, characterized in that: The step S3, the improved red-tailed kite optimization algorithm, improves the position update mathematical model of the global search phase of the standard red-tailed kite optimization algorithm by improving the transition factor of the red-tailed kite optimization algorithm, and the specific steps are: D1. Use different forms of transition factors at different stages of the red-tailed kite optimization algorithm, introduce an adaptive fitness value feedback mechanism and random perturbation intensity control parameters to improve the transition factor TF(t) of the red-tailed kite optimization algorithm; In formula (1), γ(t) is the improved transition factor, t is the current iteration number, T is the maximum iteration number, is the minimum fitness value of the tth iteration, is the average fitness value of the tth iteration, f best is the global minimum fitness value, λ(t) is the random disturbance intensity control parameter, and the mathematical model is: In formula (2), λ0 is the initial value of the random disturbance intensity control parameter, which is 2, and σ f,max is the maximum value of the standard deviation of the group fitness of the red-tailed kite optimization algorithm, and σf(t) is the standard deviation of the group fitness at the tth iteration; D2. By using formula (3), the transition factor γ(t) of the improved red-tailed kite optimization algorithm is used to improve the position update mathematical model of the global search phase of the red-tailed kite optimization algorithm; X(t)=X best +(X mean -X(t+1))×Levy(dim)×γ (3); In formula (3), X(t) is the latest position of the red-tailed kite at the tth iteration, X mean is the average position of the red-tailed kite population, X best is the global optimal position of the red-tailed kite, Leυy is the flight distribution function, γ is the transition factor improved in the tth iteration, and dim is the problem dimension.

4. The method for predicting scrap steel stockpile volume data based on three-dimensional acquisition by unmanned aerial vehicles according to claim 2, characterized in that: The step S3 improves the position update mathematical model of the local development phase of the standard red-tailed kite optimization algorithm by improving the gravity factor of the red-tailed kite optimization algorithm and combining the improved transition factor γ(t). The specific steps are as follows: Y1. Calculate the distribution density H(t) of the red-tailed kite population through the distribution of fitness. The mathematical model is: In formula (4), is the fitness value of the i-th red-tailed kite position in the t-th iteration, and N is the maximum size of the red-tailed kite; Y2, combining the distribution density H(t) and the number of iterations t to improve the gravity factor of the red-tailed kite optimization algorithm, the improved gravity factor g(t); In formula (5), H(t) is the density of the red-tailed kite population at the tth iteration, and Hmax is the maximum value of the distribution density; Y3. Improve the mathematical model of position update of the red-tailed kite optimization algorithm in the local development stage by using the improved gravity factor g(t) and the improved transition factor γ(t); X(t)=a(t)·X best +x(t)·StepSize1(i)+y(t)·StepSize2(t)(6); In formula (6), a(t) is the acceleration coefficient of the tth iteration, StepSize1(t) is the first step length, and the mathematical model is: StepSizel(t)=X(t)-γ(t)·X mean ; StepSize2(t) is the second step size, and the mathematical model is: StrpSize2(t)=g(t)×X(t)-γ(t)·X best ; g(t) is the improved gravity factor.

5. A method for predicting scrap steel stockpile volume data based on three-dimensional acquisition by unmanned aerial vehicles according to any one of claims 1 to 4, characterized in that: The specific steps of adjusting the hidden layer weight β and bias α of the BP neural network using the improved red-tailed kite optimization algorithm are as follows: S31, encoding the hidden layer weight β and bias α of the BP neural network for scrap steel stockpile volume prediction into a space vector U, and establishing a two-dimensional mapping between the space vector U and the individual position of the red-tailed kite of the red-tailed kite optimization algorithm; S32, using formula (7) to initialize the population position of the red-tailed kite optimization algorithm, setting parameters including population size N, problem dimension dim, algorithm search space upper bound UB, algorithm search space lower bound LB, maximum number of iterations T, initial number of iterations t=0, and fitness function; In formula (7), X(0) is the initial position of the red-tailed kite, η i,j is the position of the i-th red-tailed kite individual in the j-th dimension, j = 1, 2, 3..., dim, rand (0, 1) is a random number between 0 and 1; S33, judging whether the current number of iterations t satisfies t=T, if so, then exiting the setting of the hidden layer weight β and bias α of the BP neural network, outputting the individual position of the red-tailed kite corresponding to the global minimum fitness value and parsing it into the spatial vector of the hidden layer weight β and bias α of the BP neural network, and obtaining the best hidden layer weight β and bias α parameter values ​​of the BP neural network for predicting the scrap steel stockpile volume; Otherwise, execute step S34; S34, using the fitness function to calculate the individual position fitness value of the current iteration red-tailed kite population, the minimum fitness value is And compared with the fitness value of the last iteration Compare and take the smaller fitness value as the current best fitness value; S35, establishing a population position update formula of the red-tailed kite optimization algorithm in the global search phase, the local development phase, and the transition phase, and updating the individual positions of the red-tailed kites, the specific steps are: S351, in the high-flying stage, the red-tailed kite will search for the best position of the prey in the air, and establish the mathematical model of the position update of the red-tailed kite optimization algorithm in the global search stage according to formula (3); S352, in the low-flying stage, the red-tailed kite will fly in a spiral around the prey position, select the best attack time and position, and establish a mathematical model for the position update of the red-tailed kite optimization algorithm in the transition stage according to formula (8); X(t)=X best +(x(t)+y(t))·StepSize(t) (8); In formula (8), x(t) and y(t) are direction coordinates, the mathematical model of x(t) is: x(t) = R(t) sin(θ(t)); the mathematical model of y(t) is y(t) = R(t) cos(θ(t)); R(t) is the spiral flight radius, a random number between 0 and 1, θ(t) is the angle gain, StepSize(t) is the step size at the current iteration t, and the mathematical model is: StepSize(t) = X(t) - X mean ; S353, in the diving stage, the red-tailed kite suddenly dives from the best position to attack the prey, and the mathematical model of the position update of the red-tailed kite optimization algorithm in the local development stage is established according to formula (6); S36. Execute t=t+1 for the current iteration number t, and return to step S33.

Citation Information

Cited By

  • Intelligent environment monitoring method, system and device and readable storage medium

    CN119249367A