Ice period flow velocity prediction method based on WSO-ELM network model

Abstract

The invention relates to the technical field of laboratories, and discloses an ice period flow velocity prediction method based on a WSO-ELM network model, and the method is characterized in that the method comprises the following steps: building a dynamic ELM network model based on the WSO-ELM network model through employing a WSO optimization algorithm, employing the corresponding characteristics based on limited historical observation data and side sweeping Reida observation signals, and carrying out the prediction of the ice period flow velocity through employing a WSO algorithm; and carrying out optimization solution on model parameters by adopting a white shark optimization algorithm WSO, constructing a WSO-ELM network optimization model meeting the requirement of observing the ice-period river by the side-sweeping rada, and measuring the characteristics of the ice-period river by combining the real-time observation signal characteristics of the side-sweeping rada. Aiming at the difficulty of traditional ice period river measurement, in combination with a real-time observation result of a side sweeping rada system on an ice period river, an ELM network model based on a WSO optimization network coefficient is adopted, historical observation data is utilized, the network model is trained, and an optimal network structure meeting a training sample is obtained.

Description

Technical Field

[0001] The present invention relates to the field of algorithms, and in particular to a method for predicting flow velocity in an ice age based on a WSO-ELM network model. Background Art

[0002] Most rivers in my country north of 30 degrees north latitude will experience drifting ice and ice closures, and the flow measurements of hydrological stations on northern rivers are seriously affected by ice conditions in winter. The traditional velocity-area method is used to monitor the flow velocity and water level at the ice period point by contact, and then calculate part of the flow by combining the water level and the cross-section of the river. The observation efficiency is low by accumulating and extrapolating the cross-section flow. In addition, there are problems such as the contact monitoring equipment being easily damaged during the drifting ice period, the large workload and high risk of manual observation during the stable closure period, which cannot meet the high-quality development requirements of hydrological monitoring with "full coverage, all-weather, full process, full range, and full automation". Therefore, it is urgent to study new monitoring technology methods and complete sets of equipment to lay the foundation for realizing modern hydrological flow monitoring.

[0003] Develop non-contact flow measurement technology and equipment during the ice age, taking into account all-weather operation requirements, replacing manual operations, and realizing fully automatic monitoring. At present, the domestic side-scan radar flow measurement system measures the surface velocity of the river based on the Bragg scattering and Doppler principles, and calculates the cross-sectional flow rate by combining water level and cross-sectional data (depth distribution). It has achieved relatively ideal results in Wuzhou Station in Guangxi and Yunjinghong Hydrological Station on the Lancang River, realizing all-weather, fully automatic, and continuous flow velocity monitoring of southern rivers. The equipment is installed on the shore, which is convenient for daily maintenance. It is an online flow measurement technology with good application prospects. The application of side-scan radar to ice-age flow monitoring is a first at home and abroad. It requires customized development of radar equipment, and focuses on the accuracy of ice-age flow velocity monitoring, the establishment of a model for the relationship between radar-monitored flow velocity and cross-sectional average flow velocity, and calibration analysis.

[0004] To realize automatic monitoring of flow during glacial periods, it is also necessary to use hydrological and hydrodynamic methods to solve the problem of inverting the flow velocity of the cross-section after the side-scan radar measures the flow velocity during glacial periods. The radar equipment detects the electromagnetic scattering characteristics at the junction of the ice surface and the water surface by emitting electromagnetic waves of the corresponding frequency band, and measures the surface flow velocity of the river under the ice by processing the echo signal. Due to the random scattering of the ice surface and the possible electromagnetic interference and other characteristics in the surrounding environment, the measured flow velocity has the characteristics of random interference. In order to achieve stable measurement of glacial river characteristics, it is necessary to correct the flow velocity characteristics of glacial flow measurement.

[0005] In summary, the current technical defects can be summarized as follows: there are difficulties in traditional glacial river measurements, there are many problems in side-scan radar measurements of glacial river characteristics, and the sample capacity of glacial river measurement data is small, data modeling is complex, and optimization solutions are difficult. Summary of the invention

[0006] The present invention provides a method for predicting flow velocity in glacial period based on the WSO-ELM network model, which is used to solve the difficulties in traditional glacial river measurement and the many problems existing in the side-scan radar measurement of glacial river characteristics, and the difficulties such as the small sample capacity of glacial river measurement data, complex data modeling, and difficulty in optimization solution.

[0007] The present invention provides the following technical solution: a method for predicting flow velocity in glacial period based on WSO-ELM network model, comprising the following steps: adopting the great white shark optimization algorithm WSO based on the WSO-ELM network model, adopting the corresponding features based on limited historical observation data and side-scan radar observation signals, constructing a dynamic ELM network model, adopting the great white shark optimization algorithm WSO to optimize and solve the model parameters, constructing a WSO-ELM network optimization model that meets the requirements of side-scan radar observation of glacial rivers, and measuring the characteristics of glacial rivers in combination with the real-time observation signal characteristics of the side-scan radar.

[0008] Preferably, the method comprises the following steps: S1, initialize network parameters; S2. Build ELM network; S3, construct objective function; S4, solution parameter setting; S5, initialize population individuals; S6, initial population fitness evaluation; S7, setting relevant parameters of the Great White Shark algorithm calculation; S7. Iterate to find the global optimal value.

[0009] Preferably, when initializing the network parameters, the S1 adopts a forward single hidden layer structure for the ELM network training model to initialize the network parameters ;in, The number of nodes in the network's hidden layer; is the number of nodes in the network output layer. In the present invention, are the characteristic parameters of the side scan radar output signal, including six parameters: center Doppler shift, Doppler broadening, peak power, maximum signal amplitude, signal amplitude mean, and amplitude sequence variance; hidden layer The value range is ,common The output is the segmented flow velocity and average flow velocity obtained by measuring different distance units of the ice age river.

[0010] Preferably, when constructing the objective function, S3 has Different samples , ,in ; The mathematical model of constructing the ELM network model is as follows: ; In the formula, Represents the input weight vector connecting the network input layer nodes and the i-th hidden layer nodes; Represents the output weight vector connecting the i-th hidden layer node and the network output layer node; Represents the output of the network output value after the i-th sample passes through the ELM network; The target model function E of the ELM network is expressed as: ; In the formula, is the network input weight and hidden layer node threshold, is the estimated parameter of the objective function, and the WSO algorithm is used to find the optimal S, β, so that the error between the network output value and the corresponding actual value is minimized. , and obtain the optimal parameter estimates of the ELM network.

[0011] Preferably, when setting the solution parameters, S4 sets the parameters required for the solution, mainly including the following parameters: the number of individuals N of the initialized white shark group, the maximum number of iterations M, the dimension D of the search space for the number of solution parameters, the lower bound LB of the target solution space range, and the upper bound UB of the target solution space range; The solution parameters include all weights from the input layer to the hidden layer. , and all thresholds , and the weights from all hidden layers to the output layer ; Among them, the value range of all weights is 0~1; the weight range of all thresholds is -5~5.

[0012] Preferably, when initializing the population individuals, S5 uses uniformly distributed random numbers to initialize the initial position W_pos and initial speed V_pos of each white shark individual, and initializes the position and speed of each individual to , OK The uniformly distributed random number matrix of columns, where the individual position The value range of satisfies the constraints of the lower bound LB and upper bound UB of the target solution space range. The initial speed use , OK Initialize the matrix with all zero values ​​in its columns.

[0013] Preferably, when evaluating the fitness of the initial population, S6 writes a fitness calculation function, the evaluation function is determined according to the optimization goal of the problem, and the objective function is used as the fitness calculation function. The smaller the objective function value, the better the fitness. Each initial population is sequentially brought into the objective function to calculate the output target fitness; according to the calculated fitness, the fitness of each population is sorted by the sorting function, the population with the largest fitness and the minimum objective function value are selected, recorded as fmin0, and the individual corresponding to the optimal fitness value of the current iteration number is stored in In , the individual corresponding to the global optimal fitness value is saved in the global optimal value gbest.

[0014] Preferably, S6 and S7 include the following steps: The maximum frequency of shark movement fmax=0.75; the minimum frequency of shark movement fmin=0.07; calculation constant ; ; pmin=0.5; pmax=1.5; =6.250; =100; =0.0005.

[0015] Iterate to find the global optimal value: Assuming that the current iteration is the kth iteration, update the movement speed of the shark group and update the movement speed of the shark according to the following calculation formula: ; in, is a random number that follows a uniform distribution from 0 to 1. is a random number that follows a uniform distribution from 0 to 3; Update the position of the shark group: Update the position of the shark group according to the following calculation formula, ; if , the update formula is as follows: ; in, is a random number that follows a uniform distribution from 0 to 1; After generating a new population, constrain the individual position range of the new population, and the value range must meet the constraints of the lower bound LB and upper bound UB of the target solution space range. If it exceeds this range, the method of randomly generating a new population is used to replace the abnormal individuals. Optimize the individual selectivity of shark populations: Calculate the individual selection judgment factor, ; if , then the selected population individuals are optimized and updated; ; in, is a random number that follows a uniform distribution from 0 to 1. The definition is as follows: ; Calculate the fitness of the updated population, use the fitness calculation function to recalculate the fitness of the individuals in the generated new population, and sort the fitness of each population through the sorting function; update the contemporary optimal individual, calculate the sorting according to the fitness of each individual in the contemporary population, select the fitness of the optimal individual, and store the individual corresponding to the fitness into the contemporary optimal individual variable middle; If the fitness is less than fmin0, the individual corresponding to the fitness is stored in the global optimal value gbest; the population with the largest fitness is selected, and the individual corresponding to the optimal fitness value of the current iteration is stored in In the above equation, the individual corresponding to the global optimal fitness value is saved in the global optimal value gbest, otherwise, the global optimal value gbest remains unchanged.

[0016] Preferably, it also includes: judging whether the number of iterations is greater than the maximum number of iterations M set initially, and if so, outputting the optimal solution gbest and the corresponding optimal objective function value fmin0; updating the network reconstruction parameters , reconstruct the ELM network and repeat the calculation process; when all network reconstruction parameters are calculated, the optimal solution gbest with the best fitness is output, and the optimal ELM network construction parameters suitable for the river measurement are obtained , as well as the corresponding weights and thresholds; the ELM network is constructed using the above optimal weights, and the radar signals obtained by real-time measurement of the side-scan radar are processed to extract the real-time feature data of the Doppler frequency shift at the center of different range gates, Doppler broadening, peak power, maximum signal amplitude, signal amplitude mean, and amplitude series variance, and send them to the constructed optimal ELM network model to obtain the corrected measurement values ​​of the current glacial rivers.

[0017] The present invention has the following beneficial effects: The present invention aims at the difficulties of traditional glacial river measurement, combines the real-time observation results of glacial rivers by the side scanning radar system, adopts the ELM network model based on the WSO optimization network coefficient, uses historical observation data to complete the training of the network model, and obtains the optimal network structure that meets the training samples. In the ELM network training process, due to the complexity of glacial measurement data, the traditional nonlinear least squares estimation algorithm is used. In the process of calculating the ELM model parameters, the convergence time is relatively long, and the optimization process easily converges to the local optimal solution, resulting in the obtained network model parameters not reaching the optimal effect. In the present invention, the complexity of the optimization calculation of the parameters of the glacial network correction model is fully considered, the ELM network is improved, and the great white shark optimization algorithm (WSO) is used to dynamically model the ELM network model, so that the network model becomes a dynamic multi-parameter network optimization structure, and the WSO algorithm is fully utilized. The characteristics of strong global optimization ability and fast convergence speed, combined with the objective function solution, have high calculation accuracy and real-time performance, can effectively solve the time and accuracy problems in the calculation process of the time difference glacial river characteristics, and implement the observation data input with the side scanning radar to realize the prediction output of the current glacial river characteristic data. Currently, this algorithm has been successfully applied to the glacial river characteristics measurement project and achieved good experimental results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the glacial flow velocity prediction method based on the WSO-ELM network model.

[0019] Figure 2 This is the ELM network structure diagram constructed based on network parameters.

[0020] Figure 3 The original radar data analysis method was used to measure the characteristic map of river flow velocity during the glacial period.

[0021] Figure 4 This is a characteristic map of glacial river measurements at different times using the original radar algorithm.

[0022] Figure 5 Correlation diagram between the average flow velocity characteristics extracted by the radar raw data analysis algorithm and the measured average flow velocity characteristics.

[0023] Figure 6 is a schematic diagram of the objective function value changing with the evolutionary generation in the process of optimizing ELM network parameters using WSO.

[0024] Figure 7 shows the characteristic diagram of river flow velocity during the glacial period measured after optimization and calibration of the WSO-ELM network.

[0025] Figure 8 shows the characteristic diagram of glacial river measurements at different times after optimization and calibration using the WSO-ELM network. DETAILED DESCRIPTION

[0026] 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.

[0027] For example, see Figure 1-Figure 8 A method for predicting flow velocity in glacial period based on WSO-ELM network model is proposed. Aiming at the difficulties of traditional glacial river measurement and the many problems existing in the measurement of glacial river characteristics by side-scan radar, and combining the difficulties of small sample capacity of glacial river measurement data, complex data modeling, and difficult optimization solution, a WSO-ELM network model is proposed. The network model adopts the great white shark optimization algorithm (WSO). Based on the limited historical observation data and the corresponding characteristics of the side-scan radar observation signal, a dynamic ELM network model is constructed. The WSO is used to optimize and solve the model parameters to avoid the shortcomings of the transmission ELM network structure being fixed, the model coefficients being slow to converge, and being easy to converge to the local optimal value. The WSO is used to comprehensively improve the construction of the model structure and the optimization of the model parameters. The advantages of the WSO algorithm, such as strong global optimization ability and fast convergence speed, are used to construct a WSO-ELM network optimization model that meets the side-scan radar observation of glacial rivers. Combined with the real-time observation signal characteristics of the side-scan radar, the stable measurement of the characteristics of the glacial river is achieved.

[0028] This method effectively solves the accuracy and real-time problems in the glacial river measurement process by improving and combining traditional data modeling methods, and is of great significance to the glacial river measurement in northern my country.

[0029] The flow chart of the present invention is as follows Figure 1 shown.

[0030] The present invention is composed of the above steps, and the specific implementation of each step is as follows: (1) Initialize network parameters The ELM network training model adopts a forward single hidden layer structure and initializes the network parameters. .in, The number of nodes in the network input layer; The number of nodes in the network's hidden layer; is the number of nodes in the network output layer. In the present invention, are the characteristic parameters of the side scan radar output signal, including six parameters: center Doppler shift, Doppler broadening, peak power, maximum signal amplitude, signal amplitude mean, and amplitude sequence variance; hidden layer The value range is ,common The output is the segmented flow velocity and average flow velocity obtained by measuring different distance units of the glacial river, which is set according to the river width.

[0031] (2) Building the ELM network According to the network parameters, the ELM network is constructed. is the activation function of the hidden layer neurons, is the threshold value.

[0032] (3) Constructing the objective function Features Different samples , ,in

[0033] The mathematical model of constructing the ELM network model is as follows:

[0034] In the formula, Represents the input weight vector connecting the network input layer nodes and the i-th hidden layer nodes; Represents the output weight vector connecting the i-th hidden layer node and the network output layer node; Represents the output of the network output value after the i-th sample passes through the ELM network.

[0035] The target model function E of the ELM network can be expressed as

[0036] In the formula, It includes the network input weights and hidden layer node thresholds, and is the parameter to be estimated of the objective function. The WSO algorithm is used to find the optimal S, β, so that the error between the network output value and the corresponding actual value is minimized, that is, , we can get the optimal parameter estimates of the ELM network.

[0037] (2) Solution parameter setting Set the parameters required for the solution, mainly including the following parameters: the number of individuals in the initialized white shark group (N), the maximum number of iterations (M), the number of solution parameters (search space dimension D), the lower bound of the target solution space range (LB), and the upper bound of the target solution space range (UB).

[0038] In the present invention, the solution parameters include all the weights from the input layer to the hidden layer. , and all thresholds , and the weights from all hidden layers to the output layer Among them, the value range of all weights is 0~1; the weight range of all thresholds is -5~5; (3) Initialize population individuals The great white shark optimization algorithm performs iterative search in a given initialized population of individuals. In the present invention, uniformly distributed random numbers are used to initialize the initial position W_pos and initial velocity V_pos of each white shark individual, that is, the position and velocity of each individual are initialized to ( OK Columns) of uniformly distributed random numbers, where the individual positions The value range of satisfies the constraints of the lower bound (LB) and upper bound (UB) of the target solution space range. use ( OK columns) are initialized with all zero values.

[0039] (4) Initial population fitness evaluation Write a fitness calculation function, and the evaluation function is determined according to the optimization goal of the problem. In the present invention, since the optimal value of the problem to be solved is the parameter estimate corresponding to the minimum value of the objective function, the objective function is used as the fitness calculation function. The smaller the objective function value, the better the fitness. Each initial population is sequentially brought into the objective function, and the output objective function value, i.e., the fitness, is calculated.

[0040] According to the calculated fitness, the fitness of each population is sorted by the sorting function. The population with the largest fitness, that is, the minimum value of the objective function, is selected, denoted as fmin0, and the individual corresponding to the optimal fitness value of the current iteration number is stored in In , the individual corresponding to the global optimal fitness value is saved in the global optimal value gbest.

[0041] (5) Set the relevant parameters of the great white shark algorithm calculation. The maximum frequency of shark movement fmax = 0.75; the minimum frequency of shark movement fmin = 0.07; calculation constant . . pmin=0.5; pmax=1.5; =6.250; =100; =0.0005.

[0042] (6) Iterate to find the global optimal value Assuming that the current iteration is k, then Updated shark pack movement speed Use the following calculation formula to update the shark's moving speed

[0043] in, is a random number that follows a uniform distribution from 0 to 1. is a random number that follows a uniform distribution from 0 to 3.

[0044] Update shark group position Update the position of the shark group according to the following calculation formula

[0045] if , the update formula is as follows

[0046] in, is a random number that follows a uniform distribution from 0 to 1.

[0047] After generating a new population, constrain the individual position range of the new population, and the value range must satisfy the constraints of the lower bound (LB) and upper bound (UB) of the target solution space range. If it exceeds this range, the method of randomly generating a new population is used to replace the abnormal individuals.

[0048] Individual selective optimization of shark populations Calculate the individual judgment factor

[0049] if , then the selected population individuals are optimized and updated

[0050] in, is a random number that follows a uniform distribution from 0 to 1. Function The definition is as follows

[0051] Calculate the fitness after updating the population The fitness calculation function written in (4) is used to recalculate the fitness of the individuals in the new population. The fitness of each population is sorted by the sorting function.

[0052] Update the optimal individual of the current era According to the fitness of each individual in the contemporary population, the ranking is calculated and the fitness of the best individual is selected, and the individual corresponding to the fitness is stored in the contemporary best individual variable. middle.

[0053] If the fitness is less than fmin0, the individual corresponding to the fitness is stored in the global optimal value gbest.

[0054] Select the population with the largest fitness and store the individual corresponding to the optimal fitness value of the current iteration number in In the above equation, the individual corresponding to the global optimal fitness value is saved in the global optimal value gbest. Otherwise, the global optimal value gbest remains unchanged.

[0055] (7) Determine whether the number of iterations is greater than the maximum number of iterations M set initially. If so, output the optimal solution gbest and the corresponding optimal objective function value fmin0; (8) Update network reconstruction parameters , reconstruct the ELM network and repeat the calculation process in (2)-(7). When all network reconstruction parameters are calculated, the optimal solution gbest with the best fitness is output, and the optimal ELM network construction parameters suitable for the river measurement are obtained. , and the corresponding weights and thresholds.

[0056] (9) The ELM network is constructed using the above-mentioned optimal weights. The radar signals obtained by the real-time measurement of the side-scan radar are processed to extract real-time characteristic data such as the center Doppler shift of different range gates, Doppler broadening, peak power, maximum signal amplitude, signal amplitude mean, amplitude series variance, etc., and the data are sent to the constructed optimal ELM network model to obtain the corrected measurement values ​​of the current glacial rivers.

[0057] refer to Figures 3 to 8 , using the data of a site observation test to analyze, verify the superiority of the method provided by the present invention, and compare and analyze the data obtained by the traditional method to verify the superiority of the method proposed in this paper; Figures 3 to 8 The analysis of measured data shows that the WSO-ELM network model provided in the present invention adopts an optimization algorithm with extremely strong global optimization ability combined with the objective function solution, has high calculation accuracy and real-time performance, can effectively solve the problems of calculation time and accuracy in glacial river measurements, and has good practicality. After improving the objective function, the algorithm has good practical effects in other fields.

[0058] The beneficial effects of the present invention are as follows: The technical problem to be solved by the present invention is: in view of the difficulties in traditional glacial river measurements and the many problems existing in the side-scan radar measurement of glacial river characteristics, and in combination with the difficulties such as the small sample capacity of glacial river measurement data, complex data modeling, and difficulty in optimization and solution, a method based on the WSO-ELM network model is proposed. This method can efficiently utilize limited sample observation data, realize the correction of the real-time measurement feature data of the side-scan radar, and effectively solve the problem of glacial river characteristic measurement. Based on the WSO-ELM network model, the Great White Shark Optimization Algorithm (WSO) is used. Based on the limited historical observation data and the corresponding characteristics of the side-scan radar observation signal, a dynamic ELM network model is constructed. The model parameters are optimized and solved using WSO to avoid the shortcomings of the transmission ELM network structure being fixed, the model coefficients being slow to converge, and being easy to converge to the local optimal value. WSO is used to comprehensively improve the construction of the model structure and the optimization of the model parameters. The advantages of the WSO algorithm being strong in global optimization ability and fast in convergence speed are used to construct a WSO-ELM network optimization model that meets the side-scan radar observation of glacial rivers. Combined with the side-scan radar real-time observation signal characteristics, the stable measurement of glacial river characteristics is achieved. The present invention adopts an optimization algorithm with extremely strong global optimization ability combined with an objective function to solve a multi-parameter optimization model, has high calculation accuracy and real-time performance, and can effectively solve the problems of calculation time and accuracy in glacial river measurement. At present, the algorithm has been successfully used in related projects and has achieved good experimental results. This method effectively solves the accuracy and real-time problems in the glacial river measurement process by improving and combining traditional data modeling methods, and is of great significance to the glacial river measurement in northern my country.

[0059] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for predicting ice-age flow velocity based on the WSO-ELM network model, characterized by: The following steps are involved: Based on the WSO-ELM network model, the great white shark optimization algorithm WSO is used. The dynamic ELM network model is constructed based on limited historical observation data and the corresponding characteristics of the side-scan radar observation signals. The great white shark optimization algorithm WSO is used to optimize and solve the model parameters. A WSO-ELM network optimization model that meets the requirements of side-scan radar observation of glacial rivers is constructed, and the characteristics of glacial rivers are measured in combination with the real-time observation signal characteristics of the side-scan radar.

2. The method for predicting ice age flow velocity based on the WSO-ELM network model according to claim 1, characterized in that: The following steps are involved: S1, initialize network parameters; S2. Build ELM network; S3, construct objective function; S4, solution parameter setting; S5, initialize population individuals; S6, initial population fitness evaluation; S7, setting relevant parameters of the Great White Shark algorithm calculation; S7. Iterate to find the global optimal value.

3. The method for predicting ice age flow velocity based on the WSO-ELM network model according to claim 2 is characterized in that: When initializing the network parameters, S1 uses a forward single hidden layer structure for the ELM network training model to initialize the network parameters. ;in, The number of nodes in the network's hidden layer; is the number of nodes in the network output layer. In the present invention, are the characteristic parameters of the side scan radar output signal, including six parameters: center Doppler shift, Doppler broadening, peak power, maximum signal amplitude, signal amplitude mean, and amplitude sequence variance; hidden layer The value range is ,common The output is the segmented flow velocity and average flow velocity obtained by measuring different distance units of the ice age river.

4. The method for predicting ice age flow velocity based on the WSO-ELM network model according to claim 3 is characterized in that: When constructing the objective function, S3 has Different samples , ,in ; The mathematical model of constructing the ELM network model is as follows: ； In the formula, Represents the input weight vector connecting the network input layer nodes and the i-th hidden layer nodes; Represents the output weight vector connecting the i-th hidden layer node and the network output layer node; Represents the output of the network output value after the i-th sample passes through the ELM network; The target model function E of the ELM network is expressed as: ； In the formula, is the network input weight and hidden layer node threshold, is the estimated parameter of the objective function, and the WSO algorithm is used to find the optimal S, β, so that the error between the network output value and the corresponding actual value is minimized. , and obtain the optimal parameter estimates of the ELM network.

5. The method for predicting ice age flow velocity based on the WSO-ELM network model according to claim 4 is characterized in that: When setting the solution parameters, S4 sets the parameters required for the solution, mainly including the following parameters: the number of individuals N of the initialized white shark group, the maximum number of iterations M, the dimension D of the solution parameter number search space, the lower bound LB of the target solution space range, and the upper bound UB of the target solution space range; The solution parameters include all weights from the input layer to the hidden layer. , and all thresholds , and the weights from all hidden layers to the output layer ; Among them, the value range of all weights is 0~1; the weight range of all thresholds is -5~5.

6. The method for predicting ice age flow velocity based on the WSO-ELM network model according to claim 5, characterized in that: When initializing the population individuals, S5 uses uniformly distributed random numbers to initialize the initial position W_pos and initial speed V_pos of each white shark individual, and initializes the position and speed of each individual to , OK The uniformly distributed random number matrix of columns, where the individual position The value range of satisfies the constraints of the lower bound LB and upper bound UB of the target solution space range. The initial speed use , OK Initialize the matrix with all zero values ​​in its columns.

7. The method for predicting ice age flow velocity based on the WSO-ELM network model according to claim 6, characterized in that: When evaluating the fitness of the initial population, S6 writes a fitness calculation function. The evaluation function is determined according to the optimization goal of the problem. The objective function is used as the fitness calculation function. The smaller the objective function value, the better the fitness. Each initial population is sequentially brought into the objective function to calculate the output target fitness. According to the calculated fitness, the fitness of each population is sorted by the sorting function, and the population with the largest fitness and the minimum objective function value are selected, which is recorded as fmin0. The individual corresponding to the optimal fitness value of the current iteration number is stored in In , the individual corresponding to the global optimal fitness value is saved in the global optimal value gbest.

8. The method for predicting ice-age flow velocity based on the WSO-ELM network model according to claim 7, characterized in that: The steps S6 and S7 include the following steps: The maximum frequency of shark movement fmax=0.75; the minimum frequency of shark movement fmin=0.07; calculation constant ; ; pmin=0.5; pmax=1.5; =6.250; =100; =0.0005; Iterate to find the global optimal value: Assuming that the current iteration is the kth iteration, update the movement speed of the shark group and update the movement speed of the shark according to the following calculation formula: ； in, is a random number that follows a uniform distribution from 0 to 1. is a random number that follows a uniform distribution from 0 to 3; Update the position of the shark group: Update the position of the shark group according to the following calculation formula, ； if , the update formula is as follows: ； in, is a random number that follows a uniform distribution from 0 to 1; After generating a new population, constrain the individual position range of the new population, and the value range must meet the constraints of the lower bound LB and upper bound UB of the target solution space range. If it exceeds this range, the method of randomly generating a new population is used to replace the abnormal individuals. Optimize the individual selectivity of shark populations: Calculate the individual selection judgment factor, ； if , then the selected population individuals are optimized and updated; ； in, is a random number that follows a uniform distribution from 0 to 1. The definition is as follows: ； Calculate the fitness of the updated population, use the fitness calculation function to recalculate the fitness of the individuals in the generated new population, and sort the fitness of each population through the sorting function; update the contemporary optimal individual, calculate the sorting according to the fitness of each individual in the contemporary population, select the fitness of the optimal individual, and store the individual corresponding to the fitness into the contemporary optimal individual variable middle; If the fitness is less than fmin0, the individual corresponding to the fitness is stored in the global optimal value gbest; the population with the largest fitness is selected, and the individual corresponding to the optimal fitness value of the current iteration is stored in In the above equation, the individual corresponding to the global optimal fitness value is saved in the global optimal value gbest, otherwise, the global optimal value gbest remains unchanged.

9. The method for predicting ice-age flow velocity based on the WSO-ELM network model according to claim 8, characterized in that: Also includes: Determine whether the number of iterations is greater than the maximum number of iterations M set initially. If so, output the optimal solution gbest and the corresponding optimal objective function value fmin0; update the network reconstruction parameters , reconstruct the ELM network and repeat the calculation process; when all network reconstruction parameters are calculated, the optimal solution gbest with the best fitness is output, and the optimal ELM network construction parameters suitable for the river measurement are obtained , as well as the corresponding weights and thresholds; the ELM network is constructed using the above optimal weights, and the radar signals obtained by real-time measurement of the side-scan radar are processed to extract the real-time feature data of the Doppler frequency shift at the center of different range gates, Doppler broadening, peak power, maximum signal amplitude, signal amplitude mean, and amplitude series variance, and send them to the constructed optimal ELM network model to obtain the corrected measurement values ​​of the current glacial rivers.

Images