Intelligent prediction method and system for river transport ship arrival time

Abstract

The present invention discloses an intelligent prediction method and system for the arrival time of river transport ships, which belongs to the field of water transport. The method comprises: establishing an initial prediction information data set and performing initial feature selection; generating a new feature by multiplying any two features, calculating its information gain rate, screening it, and then adding it to the data set; normalizing the data set; constructing a ship arrival time prediction model; initializing parameters of an artificial bee colony method for optimization and determining a nectar content function; obtaining a new nectar source by an adaptive neighborhood search nectar source method; and mapping the generated nectar source to atoms in an atomic search method, iteratively updating the atomic position by calculating the acceleration of different atoms, and then comparing the nectar content of the atoms in the atomic search method with the nectar content of the nectar source in the adaptive neighborhood method, selecting a new nectar source until the optimized parameters are output after the round of iteration is completed; thus, real-time dynamic prediction of the entire process of the arrival time of river transport ships is realized.

Description

Technical Field

[0001] The present invention belongs to the field of water transportation, and more specifically, relates to a method and system for intelligently predicting the arrival time of river transport ships. Background Art

[0002] Among the many operational aspects of river transport, accurate prediction of ship arrival times is particularly critical. This not only significantly improves the operational efficiency of locks and ports and optimizes resource allocation, but also reduces vessel waiting time and overall operating costs. Therefore, developing an efficient and accurate method for predicting ship arrival times is crucial for improving the competitiveness of river transport and promoting coordinated economic development across regions.

[0003] Currently, the conventional method for predicting river vessel arrival times is empirical analysis, which estimates based on the ship's historical average speed. This method has low accuracy and struggles to identify the underlying coupling relationships that influence ship speed. It also fails to reflect the combined influence of multiple factors, including ship performance, water flow conditions, ship loading, and weather conditions. In recent years, as precise lock scheduling has gained increasing attention, several ship arrival time prediction methods have emerged for specific scenarios. For example, the patent "A Ship Scheduling and Control Method for Multi-Step Locks" (CN202310603307.X) proposes a ship arrival time prediction method suitable for situations where ship speeds vary little between cascade hubs, and this method is primarily used for static arrival time prediction before a ship departs. The patent "A Ship Lock Time Prediction Method and System" (CN202311055578.2) proposes classifying ship lock time into multiple factors, such as periodic and non-periodic trends. This method fails to account for differences in water flow conditions and individual characteristics of ships, impacting prediction accuracy. It is primarily used for ship lock time prediction in situations of unusually congested traffic, where navigation time accounts for a small proportion of lock time. The patent "A Double-Track Lock Automatic Scheduling Method" (CN202411413597.2) addresses the problem of predicting ship arrival times within 1-2 nautical miles of a lock and is less applicable for long-distance arrival time prediction. Overall, existing methods are unable to perform real-time dynamic prediction of river vessel arrival times under complex conditions. Therefore, it is necessary to use artificial intelligence methods to conduct in-depth analysis of the characteristics of ship navigation behavior, and to tap the inherent laws of ship arrival time on the basis of considering the synergistic effect of various characteristics, so as to realize real-time dynamic intelligent prediction of ship arrival time. Summary of the Invention

[0004] The present invention aims to provide a method and system for intelligently predicting river vessel arrival times, addressing the low prediction accuracy and inability to dynamically predict existing technologies. By filtering features using information gain and combining them with an artificial bee colony method for atomic search optimization to dynamically optimize model parameters, this method achieves real-time, dynamic, and intelligent prediction of ship arrival times throughout the entire process, improving both accuracy and practicality.

[0005] In view of the above defects or improvement needs of the prior art, as a first aspect of the present invention, the present invention provides an intelligent prediction method for the arrival time of river transport ships, comprising:

[0006] S1. Establish an initial prediction information dataset for river transport vessel arrival times;

[0007] S2. Perform feature selection by calculating the contribution of each feature in the initial prediction information dataset to the target quantity; then generate a new feature by multiplying any two features, calculate the corresponding information gain rate, filter it, and add it to the prediction information dataset; normalize the prediction information dataset obtained after filtering the initial features and adding the new features;

[0008] S3. Construct a ship arrival time prediction model based on a multi-layer perceptron based on the prediction information dataset obtained in S2;

[0009] S4. Initialize the parameters of the artificial bee colony method used to optimize the weights and biases of the multi-layer perceptron nodes in S3 and determine the honey content function of the bee position; obtain a new honey source through the adaptive neighborhood search method; and map the honey source generated by the adaptive neighborhood to the atoms in the atom search method. By calculating the interatomic forces, geometric constraints and atomic masses, the accelerations of different atoms are obtained, and the atomic positions are iteratively updated based on the acceleration. The honey content of the atoms in the atom search method is compared with the honey content of the honey source in the adaptive neighborhood method to select a new honey source; until the round of iteration is completed, the optimized weights and biases are output.

[0010] Furthermore, the initial prediction information data set in S1 includes: the arrival time of the river transport ship as the output vector, and the water level, flow, rainfall, wind speed, ship width, ship power, ship length, ship longitude, ship latitude, and sailing date as the input vector.

[0011] Furthermore, the specific calculation method of the contribution in S2 is:

[0012] Set up the first The samples are , a total of features, among which The first sample The feature is The model's predicted value for this sample is The predicted mean of the entire model is , then the The value obeys the following equation:

[0013] ,

[0014] in, Indicates the In the sample characteristic Value, that is, the final predicted value of this feature contribution.

[0015] Furthermore, the specific calculation method of the information gain rate in S2 is:

[0016] The two features The product of ; The set of new features formed by the product of any two features is called ;

[0017] New Features The contribution of The difference between the sum of the contributions of two features is the information gain rate, and its specific expression is:

[0018] ,

[0019] in, is the information gain rate; Characterized by Forecast results Contribution of 、 Characteristics 、 Forecast results Contribution of all new features that do not satisfy the above formula Eliminate from.

[0020] Furthermore, the honey content function in S4 is specifically:

[0021] ,

[0022] in, is the average prediction error of river transport ship arrival time, Represents the honey content function value.

[0023] Furthermore, the adaptive neighborhood search nectar source method in S4 is specifically as follows:

[0024] ,

[0025] in, A solution to the new nectar source for the neighborhood search nectar source method; The first method of searching for honey sources in the neighborhood A solution to a new honey source; is the adaptive coefficient; For the number The location of the nectar source; The randomly selected number is The location of the nectar source;

[0026] Sort the new nectar source locations in the neighborhood according to the nectar content, and count the nectar content of the new nectar source locations and the original nectar source locations. The statistical variable of the nectar content is recorded as , the initial value is 0; after sorting, it is in the The new nectar source at location is ; and the original nectar source is in the Location of nectar Perform bit-by-bit comparison; if , then let Increment by 1; if , then let Decrement by 1; if ,but The value does not change; after traversing all honey source locations, The final value of

[0027] When the statistical results of advantages and disadvantages When it is greater than 0, the adaptive coefficient becomes:

[0028] ,

[0029] Where, is the adaptive coefficient before the change;

[0030] When the statistical results of advantages and disadvantages When it is less than 0, the adaptive coefficient becomes:

[0031] ,

[0032] When the statistical results of advantages and disadvantages When it is equal to 0, the current adaptive coefficient does not need to be optimized and remains unchanged.

[0033] Furthermore, the specific calculation method of the acceleration in S4 is:

[0034] pass 、 as well as Calculate the acceleration of different atoms at the tth iteration :

[0035] ,

[0036] in, is the action on the tth iteration in the d-dimensional space The total force of atoms; is the global optimal atom pair in the d-dimensional space at the t-th iteration The geometric constraints of atoms; is the first The mass of an atom.

[0037] Furthermore, the 、 as well as The specific calculation method is:

[0038] ,

[0039] in, The honey sources that are in the top 50% of honey content in the artificial bee colony method; For atoms and The spatial characteristic distance of is a random number from 0 to 1;

[0040] ,

[0041] in, is the adaptive factor, which is adaptively adjusted with the number of iterations; They are the atom with the highest honey content in d-dimensional space, the atom with the lowest honey content, A vector of atoms;

[0042] ,

[0043] in, is the first An intermediate variable for calculating the mass of atoms; is the first The intermediate variable for calculating the mass of atoms; variable , Specifically:

[0044] ,

[0045] ,

[0046] in, Indicates the number of atoms; Respectively The amount of honey in atoms, the amount of honey in the atoms, the amount of honey in the atom with the highest honey content, and the amount of honey in the atom with the lowest honey content; is the natural base.

[0047] As a second aspect of the present invention, there is provided a system for intelligently predicting the arrival time of river transport vessels, comprising:

[0048] An initial prediction information data set construction unit is used to establish an initial prediction information data set for river transport ship arrival time;

[0049] The data preprocessing unit is used to calculate the importance of each feature in the initial prediction information data set to the target quantity through the information gain rate and perform feature selection; generate a new feature by multiplying any two features, and calculate the corresponding information gain rate, and then add it to the prediction information data set after screening; normalize the prediction information data set obtained after screening the initial features and adding the new features;

[0050] A prediction model building unit, configured to build a ship arrival time prediction model based on a multi-layer perceptron based on the prediction information data set obtained by the data preprocessing unit;

[0051] The parameter optimization unit is used to initialize the parameters of the artificial bee colony method used to optimize the weights and biases of the multi-layer perceptron nodes in the prediction model construction unit and determine the honey content function of the bee position; obtain a new honey source through the adaptive neighborhood search method; and map the honey source generated by the adaptive neighborhood to the atoms in the atom search method. By calculating the interatomic forces, geometric constraints and atomic masses, the acceleration of different atoms is obtained, and the atomic positions are iteratively updated based on the acceleration. The honey content of the atoms in the atom search method is then compared with the honey content of the honey source in the adaptive neighborhood method to select a new honey source; until the round of iteration is completed, the optimized weights and biases are output;

[0052] As a third aspect of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored, and the computer program is executed by a processor to execute any step of the above-mentioned method for intelligent prediction of river vessel arrival time.

[0053] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0054] 1. The intelligent prediction method for the arrival time of river-transported ships of the present invention constructs a ship arrival time prediction model with the help of a multi-layer perceptron, and uses its powerful nonlinear fitting ability to deeply explore the complex coupling relationship between multi-source features. When a ship is sailing, the impact of water level changes on the speed, the navigation status under the synergistic effect of different wind speeds and ship power, etc., are all complex nonlinear correlations. The multi-layer perceptron transforms these complex relationships into a learnable model through information transmission and transformation of multiple layers of neurons, takes the filtered features as input, and outputs the arrival time prediction results after hidden layer calculations, breaking through the limitations of traditional linear models, and more in line with the actual navigation scenarios of river-transported ships affected by multiple dynamic factors, providing model architecture support for accurate prediction.

[0055] 2. The present invention's intelligent method for predicting river vessel arrival times utilizes an artificial bee colony-atom search (ABC-ASO) method to collaboratively optimize model parameters. The artificial bee colony method initializes the optimization criteria for multi-layer perceptron weights, biases, and other parameters, and explores the parameter space through an adaptive neighborhood search for nectar sources. The atomic search method maps the nectar sources to atoms, simulates the interatomic forces, constraints, and mass relationships, and iteratively updates the atomic positions (i.e., parameter combinations). The "honey content" (an inverse indicator of the prediction error) of the solutions generated by the two methods is compared, and the nectar sources are updated preferentially until the optimization round is reached. This collaborative optimization leverages both the global exploratory power of the artificial bee colony and the local precision of the atomic search to efficiently find the optimal parameters for the multi-layer perceptron, significantly improving the accuracy and stability of the model's prediction of ship arrival times and ensuring that the prediction results are more closely aligned with actual navigation conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of an intelligent prediction method for river transport vessel arrival time according to an embodiment of the present invention;

[0057] Figure 2 Schematic diagram of the parameter optimization process of the multi-layer perceptron intelligent prediction model based on the artificial bee colony method according to an embodiment of the present invention;

[0058] Figure 3 A schematic diagram of the correlation relationship between factors in an embodiment of the present invention;

[0059] Figure 4 Schematic diagram of the contribution of different factors to the ship arrival time prediction result according to an embodiment of the present invention;

[0060] Figure 5 A statistical histogram of the prediction error of the ship arrival time for a certain section of a voyage according to an embodiment of the present invention;

[0061] Figure 6 This is a time series diagram of prediction error for a single ship in a certain section of a voyage according to an embodiment of the present invention;

[0062] Figure 7It is an iterative convergence diagram of the artificial bee colony method according to an embodiment of the present invention;

[0063] Figure 8 Schematic diagram comparing the iterative process of the improved ABC-ASO algorithm and the standard artificial bee colony algorithm according to an embodiment of the present invention;

[0064] Figure 9 2 is a diagram of system units according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0066] Example 1

[0067] Please refer to Figure 1 This embodiment 1 provides an intelligent prediction method for river vessel arrival time, comprising:

[0068] S1. Establish an initial prediction information dataset for river transport vessel arrival times;

[0069] S2. Perform feature selection by calculating the contribution of each feature in the initial prediction information dataset to the target quantity; then generate a new feature by multiplying any two features, calculate the corresponding information gain rate, filter it, and add it to the prediction information dataset; normalize the prediction information dataset obtained after filtering the initial features and adding the new features;

[0070] S3. Construct a ship arrival time prediction model based on a multi-layer perceptron based on the prediction information dataset obtained in S2;

[0071] S4. Initialize the parameters of the artificial bee colony method used to optimize the weights and biases of the multi-layer perceptron nodes in S3 and determine the honey content function of the bee position; obtain a new honey source through the adaptive neighborhood search method; and map the honey source generated by the adaptive neighborhood to the atoms in the atom search method. By calculating the interatomic forces, geometric constraints and atomic masses, the accelerations of different atoms are obtained, and the atomic positions are iteratively updated based on the acceleration. The honey content of the atoms in the atom search method is compared with the honey content of the honey source in the adaptive neighborhood method to select a new honey source; until the round of iteration is completed, the optimized weights and biases are output.

[0072] This embodiment 1 further explains the above steps.

[0073] (1) Construction of initial prediction information dataset

[0074] Establish an initial prediction information data set for the arrival time of river transport ships; the initial prediction information data set for the arrival time of river transport ships includes: taking the arrival time of river transport ships as an output vector, and taking water level, flow, rainfall, wind speed, ship width, ship power, ship length, ship longitude, ship latitude, and sailing date as input vectors.

[0075] (2) Data preprocessing

[0076] Perform feature selection on the initial prediction information dataset of river vessel arrival time. The specific steps include:

[0077] S201: The arrival time of the river transport vessel is used as the output vector, and the water level, flow, rainfall, wind speed, ship width, ship power, ship length, ship longitude, ship latitude, and sailing date are used as input vectors.

[0078] S202: In order to ensure the prediction accuracy and generalization ability of the model, the feature selection method is used to screen and optimize the input vector.

[0079] Furthermore, the contribution is based on The value is calculated to evaluate the impact of the ship prediction input vector on the prediction result (ship arrival time). The value is calculated as follows: Assuming The samples are , a total of features, among which The first sample The feature is The model's predicted value for this sample is The predicted mean of the entire model is ,So The value obeys the following equation:

[0080] ,

[0081] in Indicates the In the sample characteristic Value, that is, the final predicted value of the feature contribution.

[0082] In a preferred embodiment, The value can be directly calculated based on the Python extension package.

[0083] In a preferred embodiment, in order to explore the deep coupling relationship between the features, the information gain rate is used to evaluate the interaction between the features. The specific steps are as follows:

[0084] S20201: Combine two features The product of ; The set of new features formed by the product of any two features is called .

[0085] S20202: For any New features , its contribution is greater than The sum of the contributions of the two features indicates that there is a positive interaction between them, that is, it is believed that there is information gain in the interaction between the two features. The contribution of The difference between the sum of the contributions of two features is called the information gain rate. The specific expression is:

[0086] ,

[0087] in, is the information gain rate; Characterized by Forecast results Contribution of 、 Characteristics 、 Forecast results contribution.

[0088] S20203: Remove all new features that do not satisfy the S20202 formula from Eliminate from.

[0089] Furthermore, the original features in S101 and The new features in the dataset are normalized to eliminate the influence of features of different magnitudes and ensure the stability and efficiency of model training. Specifically:

[0090] ,

[0091] in, is the feature normalization quantity; , The input vectors are The maximum and minimum values ​​of .

[0092] (3) Prediction model construction

[0093] Based on the prediction information dataset proposed by S2, a ship arrival time prediction model based on a multi-layer perceptron is established; the following steps are included:

[0094] S301: Using the optimized input vector obtained in step 1 as the input variable of the multilayer perceptron;

[0095] S302: The input variables are trained in turn through a multi-layer feedforward neural network (preferably a 3-layer feedforward neural network) of a multi-layer perceptron from the input layer to the output layer. The activation function used in each layer of the feedforward neural network is Function, the output value of each layer of feedforward neural network can be calculated by the following formula: ,

[0096] ,

[0097] in, Is the current layer feedforward neural network node The output value of It is the feedforward neural network node of the previous layer The output value of is bias; is a node right The weight of is the activation function.

[0098] (4) Parameter optimization

[0099] For the ship arrival time prediction model established by S3, a parameter optimization method of the ship arrival time prediction model based on the artificial bee colony method is designed to explore the changes in the ship's speed characteristics along the way, including the gradual deceleration of the ship due to the rising water level of the river caused by precipitation, thereby optimizing the performance of the ship arrival time prediction model.

[0100] The parameters of the multilayer perceptron in the prior art are usually adjusted manually. This embodiment 1 proposes to automatically generate the optimal parameters of the multilayer perceptron based on the artificial bee colony method, thereby optimizing the performance of the ship arrival time prediction model based on the multilayer perceptron. Figure 2 , said S4 comprises the following steps:

[0101] In this embodiment, an artificial bee colony method is designed to optimize the node weights and biases of the multilayer perceptron in S3. The specific steps include:

[0102] S401: Initialize the parameters of the artificial bee colony method for optimizing the node weights and biases of the multilayer perceptron. The parameters of the artificial bee colony method include the number of bees in the bee colony (The number of random solutions used to optimize the node weights and biases of the multilayer perceptron. The recommended value is 100). The number of parameters to be tuned is (the number of node weights and biases of the multilayer perceptron), the maximum number of bee swarm optimization rounds (The maximum number of optimization rounds for the node weights and biases of the multilayer perceptron is recommended to be 200) and the initial solution generated by the swarm (The initial value of the random solution for parameter optimization of the node weights and biases of the multilayer perceptron is recommended to be a random number between 0 and 1).

[0103] S402: Calculate the honey content function of the bee position in the current round of artificial bee colony method , which is used to evaluate the quality of the parameters to be tuned for the multilayer perceptron. The honey content function of the bee position in the artificial bee colony method can be expressed as:

[0104] ,

[0105] in, is the average prediction error of river transport ship arrival time, Represents the honey content function value.

[0106] S403: Use the artificial bee colony adaptive neighborhood search nectar source method to obtain a new solution for the optimal parameters.

[0107] In order to improve the ability of the artificial bee colony method (ABC) to explore new solutions for the optimal parameters of the ship arrival time intelligent prediction model, this embodiment 1 further uses the following neighborhood search nectar source method:

[0108] ,

[0109] in, A solution to the new nectar source for the neighborhood search nectar source method; The solution for the i-th new nectar source of the neighborhood search nectar source method; The adaptive coefficient is 0.1 to 0.9, with an initial value of 0.9. As the number of iterations increases, it gradually decreases until it reaches 0.1. The single reduction amplitude is ((0.9-0.1) / ). For the number The location of the nectar source; The randomly selected number is The location of the nectar source.

[0110] S404: Sort the new nectar source locations in the neighborhood by nectar content, and count the nectar content of the new nectar source locations and the original nectar source locations. The statistical variable of the nectar content is recorded as , the initial value is 0; after sorting, it is in the The new nectar source at location is ; and the original nectar source is in the Location of nectar Perform bit-by-bit comparison; if , then let Increment by 1; if , then let Decrement by 1; if ,but The value does not change; after traversing all honey source locations, The final value of

[0111] When the statistical results of advantages and disadvantages When it is greater than 0, it means that the current adaptive coefficient can better explore new nectar sources, that is, it can better generate the parameters to be tuned for the multilayer perceptron. Therefore, the adaptive coefficient becomes

[0112] ,

[0113] Where, is the adaptive coefficient before the change.

[0114] When the statistical results of advantages and disadvantages When it is less than 0, it means that the current adaptive coefficient cannot better explore new nectar sources, that is, it cannot better generate the parameters to be tuned for the multilayer perceptron. Therefore, the adaptive coefficient becomes:

[0115] ,

[0116] When the statistical results of advantages and disadvantages When it is equal to 0, the current adaptive coefficient does not need to be optimized and remains unchanged.

[0117] S405: Design an atomic search method to select new nectar sources in the artificial bee colony method.

[0118] Considering that the artificial bee colony method in S403 uses a new nectar source selection method based on an adaptive neighborhood search formula, which has a high degree of randomness, in order to reduce the time it takes for the artificial bee colony method to explore better new nectar sources and increase the probability of the artificial bee colony method finding the optimal nectar source (i.e., the optimal values ​​of the multilayer perceptron node weights and bias parameters in S3), an artificial bee colony nectar source selection mechanism based on the atomic search algorithm (ASO) is designed.

[0119] In a preferred embodiment, the artificial bee colony nectar source selection mechanism based on atomic search method (ABC-ASO) has the following specific steps:

[0120] Based on the vector corresponding to the nectar source generated in step S403, atoms in the atom search method are formed. The spatial dimension d in the atom search method is the same as the parameter to be tuned in the artificial bee colony method.

[0121] Considering the interaction between atoms in the atomic search method, calculate the acceleration of different atoms at the tth iteration :

[0122] ,

[0123] in, is the action on the tth iteration in the d-dimensional space The total force of the atoms, that is, the k atoms with better honey content function value have a The randomly weighted sum of the forces acting on each atom is calculated as follows:

[0124] ,

[0125] in, The honey sources that are in the top 50% of honey content in the artificial bee colony method; For atoms and The spatial characteristic distance of A random number from 0 to 1.

[0126] acceleration In the formula is the global optimal atom pair in the d-dimensional space at the t-th iteration The geometric constraint of each atom can be expressed as:

[0127] ,

[0128] in, is the adaptive factor, which is adaptively adjusted with the number of iterations; They are the atom with the highest honey content in d-dimensional space, the atom with the lowest honey content, A vector of atoms.

[0129] acceleration In the formula is the first The mass of an atom can be expressed as:

[0130] ,

[0131] in, is the first An intermediate variable for calculating the mass of atoms; is the first The intermediate variable for calculating the mass of atoms; variable , Specifically:

[0132] ,

[0133] ,

[0134] in, Indicates the number of atoms; Respectively The amount of honey in atoms, the amount of honey in the atoms, the amount of honey in the atom with the highest honey content, and the amount of honey in the atom with the lowest honey content; is the natural base.

[0135] Based on different atoms in Iterate the atomic positions with the acceleration of iterations.

[0136] The nectar content of the atoms in the iterative atomic search method is compared with the nectar source nectar content of the artificial bee colony method. If the nectar content of the atoms in the iterative atomic search method is greater than the nectar source nectar content of the artificial bee colony method, the new nectar source position is adopted; otherwise, the current nectar source position is maintained.

[0137] S406: Determine whether the artificial bee colony method has reached the maximum number of bee colony optimization rounds If it is reached, the current honey source position is output, that is, the optimized weight and bias are output, thereby optimizing the performance of the multi-layer perceptron model used to predict the arrival time of river transport ships; if it is not reached, go to S402, at this time .

[0138] Based on the above method, this embodiment 1 is combined with specific experiments to illustrate the technical effects of this embodiment 1:

[0139] The water transport information and ship information of a certain section of the route were selected as the data set, including the arrival time data of 5,578 river transport ships. The following basic input features were selected based on the data set: water level, flow, rainfall, wind speed, ship width, ship power, ship length, ship longitude, ship latitude, and sailing date. Among them, 80% of the data were randomly selected as the training set, and the remaining 20% ​​of the data were used as the test set. During the operation of this section, the ship returned Beidou position information once a minute. The number of bees in the artificial bee colony method was set to 100, and the maximum number of bee colony optimization rounds was 200. The number of atoms in the atomic search method was 100, and the maximum iteration value was 100.

[0140] Please refer to Figure 3 , flow and rainfall have a strong correlation, at this time the two factors can be used to capture their synergistic effect in a multiplicative manner; Figure 4 It can be seen that the contribution of the product of flow and rainfall is about 15%, which is the third most contributing factor. Figure 4It can be seen that the contribution of the three factors of ship length, wind speed and sailing date does not exceed 5%. Therefore, the factors of the ship arrival time prediction model after screening are: water level, flow, rainfall, ship width, ship power, ship longitude, and ship latitude, and the product term of flow and rainfall is newly added as the input vector of the prediction.

[0141] Please refer to Figure 5 and Figure 6 , about 98% of the ships’ predicted arrival time errors are less than 75 minutes, accounting for less than 10% of the total sailing time. Figure 6 As can be seen, for a single vessel, it traveled a total of 700 minutes (11.6 hours) in this waterway. As sailing time accumulates, the prediction error within 60 minutes after leaving the starting lock is no more than 30 minutes (prediction accuracy: (700-30) / 700 = 95.7%), and the prediction error after 60 minutes of leaving the starting lock is no more than 20 minutes (prediction accuracy: (700-20) / 700 = 97.1%). This high prediction accuracy is achieved by the multi-layer perceptron continuously learning and mining navigation information of river vessels.

[0142] Please refer to Figure 7 The artificial bee colony method proposed in Example 1 can obtain the optimal solution of the intelligent prediction method through continuous iterative search. This is because the artificial intelligence method used in Example 1 can effectively and intelligently adjust the adaptive coefficient based on the current search results, thereby taking into account both the global search capability and the local optimization capability of the method.

[0143] Please refer to Figure 8 Under the artificial bee colony nectar source selection mechanism based on the atomic search method (ABC-ASO), the artificial bee colony method can converge to a better fitness value more quickly. This is because the introduction of the atomic search method enables the artificial bee colony method to explore better nectar sources based on the search computing power of the atomic search method.

[0144] Example 2

[0145] Please refer to Figure 9 This embodiment 2 provides an intelligent prediction system for river vessel arrival time, including:

[0146] An initial prediction information data set construction unit is used to establish an initial prediction information data set for river transport ship arrival time;

[0147] The data preprocessing unit is used to calculate the importance of each feature in the initial prediction information data set to the target quantity through the information gain rate and perform feature selection; generate a new feature by multiplying any two features, and calculate the corresponding information gain rate, and then add it to the prediction information data set after screening; normalize the prediction information data set obtained after screening the initial features and adding the new features;

[0148] A prediction model building unit, configured to build a ship arrival time prediction model based on a multi-layer perceptron based on the prediction information data set obtained by the data preprocessing unit;

[0149] The parameter optimization unit is used to initialize the parameters of the artificial bee colony method for optimizing the weights and biases of the multi-layer perceptron nodes in the prediction model construction unit and determine the honey content function of the bee position; obtain a new honey source through the adaptive neighborhood search method; and map the honey source generated by the adaptive neighborhood to the atoms in the atom search method, obtain the acceleration of different atoms by calculating the interatomic forces, geometric constraints and atomic masses, and iteratively update the atomic positions based on the acceleration, and then compare the honey content of the atoms in the atom search method with the honey content of the honey source in the adaptive neighborhood method to select a new honey source; until the round of iteration is completed, the optimized weights and biases are output.

[0150] Example 3

[0151] This embodiment 3 also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can implement any step of a method for intelligently predicting the arrival time of river-transported ships.

[0152] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which can store program codes.

[0153] For an introduction to the computer-readable storage medium provided in this application, please refer to the above method embodiment, and this application will not go into details here.

[0154] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for intelligently predicting the arrival time of river transport vessels, characterized in that: include: S1. Establish an initial prediction information dataset for river transport vessel arrival times; S2. Perform feature selection by calculating the contribution of each feature in the initial prediction information dataset to the target quantity; then generate a new feature by multiplying any two features, calculate the corresponding information gain rate, filter it, and add it to the prediction information dataset; normalize the prediction information dataset obtained after filtering the initial features and adding the new features; S3. Construct a ship arrival time prediction model based on a multi-layer perceptron based on the prediction information dataset obtained in S2; S4. Initialize the parameters of the artificial bee colony method used to optimize the node weights and biases of the multilayer perceptron in S3 and determine the honey content function of the bee position; obtain a new honey source through the adaptive neighborhood search method; map the honey source generated by the adaptive neighborhood to the atoms in the atom search method, calculate the acceleration of different atoms by calculating the interatomic forces, geometric constraints, and atomic masses, and iteratively update the atomic positions based on the acceleration. Then, compare the honey content of the atoms in the atom search method with the honey content of the honey source in the adaptive neighborhood method to select a new honey source; until the round of iteration is completed, output the optimized weights and biases; The specific calculation method of the acceleration in S4 is: pass 、 as well as Calculate the acceleration of different atoms at the tth iteration : ， in, is the action on the tth iteration in the d-dimensional space The total force of atoms; is the global optimal atom pair in the d-dimensional space at the t-th iteration The geometric constraints of atoms; is the first The mass of an atom; described 、 as well as The specific calculation method is: ， in, The honey sources that are in the top 50% of honey content in the artificial bee colony method; For atoms and The spatial characteristic distance of is a random number from 0 to 1; ， in, is the adaptive factor, which is adaptively adjusted with the number of iterations; They are the atom with the highest honey content in d-dimensional space, the atom with the lowest honey content, A vector of atoms; ， in, is the first An intermediate variable for calculating the mass of atoms; is the first The intermediate variable for calculating the mass of atoms; variable , Specifically: ， ， in, Indicates the number of atoms; Respectively The amount of honey in atoms, the amount of honey in the atoms, the amount of honey in the atom with the highest honey content, and the amount of honey in the atom with the lowest honey content; is the natural base.

2. The method for intelligently predicting the arrival time of river transport vessels according to claim 1, characterized in that: The initial prediction information data set in S1 includes: the arrival time of the river transport ship as the output vector, and the water level, flow, rainfall, wind speed, ship width, ship power, ship length, ship longitude, ship latitude, and sailing date as the input vector.

3. The method for intelligently predicting the arrival time of river transport vessels according to claim 1, characterized in that: The specific calculation method of the contribution in S2 is: Set up the first The samples are , a total of features, among which The first sample The feature is The model's predicted value for this sample is The predicted mean of the entire model is , then the The value obeys the following equation: ， in, Indicates the In the sample characteristic Value, that is, the final predicted value of this feature contribution.

4. The method for intelligently predicting the arrival time of river transport vessels according to claim 1, characterized in that: The specific calculation method of the information gain rate in S2 is: The two features The product of ; The set of new features formed by the product of any two features is called ; New Features The contribution of The difference between the sum of the contributions of two features is the information gain rate, and its specific expression is: ， in, is the information gain rate; Characterized by Forecast results Contribution of 、 Characteristics 、 Forecast results Contribution of all new features that do not satisfy the above formula Eliminate.

5. The method for intelligently predicting the arrival time of river transport vessels according to claim 1, characterized in that: The honey content function in S4 is specifically: ， in, is the average prediction error of river transport ship arrival time, Represents the honey content function value.

6. The method for intelligently predicting the arrival time of river transport vessels according to claim 1, characterized in that: The adaptive neighborhood search nectar source method in S4 is specifically as follows: ， in, A solution to the new nectar source for the neighborhood search nectar source method; The first method of searching for honey sources in the neighborhood A solution to a new honey source; is the adaptive coefficient; For the number The location of the nectar source; The randomly selected number is The location of the nectar source; Sort the new nectar source locations in the neighborhood according to the nectar content, and count the nectar content of the new nectar source locations and the original nectar source locations. The statistical variable of the nectar content is recorded as , the initial value is 0; after sorting, it is in the The new nectar source at location is ; and the original nectar source is in the Location of nectar Perform bit-by-bit comparison; if , then let Increment by 1; if , then let Decrement by 1; if ,but The value does not change; after traversing all honey source locations, The final value of When the statistical results of advantages and disadvantages When it is greater than 0, the adaptive coefficient becomes: ， Where, is the adaptive coefficient before the change; When the statistical results of advantages and disadvantages When it is less than 0, the adaptive coefficient becomes: ， When the statistical results of advantages and disadvantages When it is equal to 0, the current adaptive coefficient does not need to be optimized and remains unchanged.

7. An intelligent prediction system for river transport ship arrival time, characterized in that: include: An initial prediction information data set construction unit is used to establish an initial prediction information data set for river transport ship arrival time; The data preprocessing unit is used to calculate the importance of each feature in the initial prediction information data set to the target quantity through the information gain rate and perform feature selection; generate a new feature by multiplying any two features, and calculate the corresponding information gain rate, and then add it to the prediction information data set after screening; normalize the prediction information data set obtained after screening the initial features and adding the new features; A prediction model building unit, configured to build a ship arrival time prediction model based on a multi-layer perceptron based on the prediction information data set obtained by the data preprocessing unit; The parameter optimization unit is used to initialize the parameters of the artificial bee colony method used to optimize the weights and biases of the multi-layer perceptron nodes in the prediction model construction unit and determine the honey content function of the bee position; obtain a new honey source through the adaptive neighborhood search method; and map the honey source generated by the adaptive neighborhood to the atoms in the atom search method. By calculating the interatomic forces, geometric constraints and atomic masses, the acceleration of different atoms is obtained, and the atomic positions are iteratively updated based on the acceleration. The honey content of the atoms in the atom search method is then compared with the honey content of the honey source in the adaptive neighborhood method to select a new honey source; until the round of iteration is completed, the optimized weights and biases are output; The specific calculation method of acceleration in the parameter optimization unit is: pass 、 as well as Calculate the acceleration of different atoms at the tth iteration : ， in, is the action on the tth iteration in the d-dimensional space The total force of atoms; is the global optimal atom pair in the d-dimensional space at the t-th iteration The geometric constraints of atoms; is the first The mass of an atom; described 、 as well as The specific calculation method is: ， in, The honey sources that are in the top 50% of honey content in the artificial bee colony method; For atoms and The spatial characteristic distance of is a random number from 0 to 1; ， in, is the adaptive factor, which is adaptively adjusted with the number of iterations; They are the atom with the highest honey content in d-dimensional space, the atom with the lowest honey content, A vector of atoms; ， in, is the first An intermediate variable for calculating the mass of atoms; is the first The intermediate variable for calculating the mass of atoms; variable , Specifically: ， ， in, Indicates the number of atoms; Respectively The amount of honey in atoms, the amount of honey in the atoms, the amount of honey in the atom with the highest honey content, and the amount of honey in the atom with the lowest honey content; is the natural base.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement the intelligent prediction method for river vessel arrival time as described in any one of claims 1-6.

Images