Intelligent wharf rail-mounted gantry crane operation time prediction method and system and electronic equipment

Abstract

The invention relates to an intelligent wharf rail-mounted gantry crane operation time prediction method and system and electronic equipment. The method comprises the steps that historical operation data and weather environment data of an intelligent wharf rail-mounted gantry crane are obtained; data cleaning and feature engineering extraction are carried out on the obtained data, a data set is formed, feature engineering extraction comprises extraction of coordinates of starting and ending points of rail-mounted gantry crane operation, extraction of container information and matching of historical operation data and weather environment data, and statistical test and sensitivity analysis are carried out on all extracted features; determining features for constructing the data set; and a rail-mounted gantry crane operation time matrix is constructed based on the data set, and a rail-mounted gantry crane operation time prediction result is given by using a Monte-Carlo algorithm according to an input rail-mounted gantry crane operation instruction and a current weather environment condition, and comprises operation time obeying distribution and a confidence interval. Compared with the prior art, the method has the advantages of high-precision and real-time rail-mounted gantry crane operation time prediction, good robustness and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent container terminals, and particularly to a method, system and electronic device for predicting the operation time of an intelligent terminal rail-mounted gantry crane. Background Art

[0002] A rail-mounted gantry crane is a container handling tool operating in the yard of an intelligent container terminal, undertaking the task of handling containers in and out of the yard, and is a hub connecting the automatic guided vehicle (AGV) at the terminal front and the external container truck at the gate. Its operation time is related to important scheduling processes at the container terminal front, such as the integrated scheduling of the rail-mounted gantry crane and AGV, and the stowage plan, and has great research significance. To achieve accurate prediction of the operation time of the rail-mounted gantry crane, several existing technical solutions are as follows.

[0003] Zheng Renhao, "Research on the Scheduling of 'Crane - AGV' and Yard Stacking Location Allocation Considering Sea - Rail Intermodal Transport", Dalian Jiaotong University, 2024, proposed a rail-mounted gantry crane operation time prediction system based on a physical model. This system divides the working process of the rail-mounted gantry crane into several processes, including the horizontal movement of the rail-mounted gantry crane, the time for the spreader to grasp and release the container, and the vertical movement of the spreader. The system adopts the idea of linear fitting, regarding the horizontal and vertical movement processes as uniform motion throughout, and the speed comes from the set value of the simulation software. In the technical solution proposed in this literature, the model considering the rail-mounted gantry crane movement as uniform linear motion is different from the actual movement mode of the rail-mounted gantry crane, and is not accurate enough; the horizontal and vertical movement speeds of the rail-mounted gantry crane come from the set value of the simulation software, without actual data support, and it is difficult to be applied to the production process of the container terminal.

[0004] Jin Xin, "Research on Automated Yard Scheduling Based on Deep Reinforcement Learning", Shandong University, 2024, proposed an operation time prediction system considering the movement of the rail-mounted gantry crane along the bay direction. This system considers the movement speed of the rail-mounted gantry crane along the bay direction, the time for lifting and lowering the container, and assumes that the lifting time of the rail-mounted gantry crane for all containers is equal to the lowering time, both set as the time for the rail-mounted gantry crane to move 3 units along the bay direction. In the technical solution proposed in this literature, only the time for the rail-mounted gantry crane to move along the bay direction is considered, without considering the time for the rail-mounted gantry crane to move along the row direction within the bay and the time for the spreader to move vertically along the layer. The physical model construction is incomplete; assuming that the grasping time of the rail-mounted gantry crane for all containers is equal to the lowering time, it does not consider the difference in the interaction time between the spreader and the AGV and the support under different working conditions; there is also the problem of lack of actual data support, and the implementation effect in the container terminal is limited.

[0005] Fan Houming et al., "Joint Scheduling Optimization of AGV and Yard Crane in Automated Terminals with Synchronized Loading and Unloading", Industrial Engineering and Management, 2024, 29(1): 41-51, proposed a static quay crane operation time prediction system combined with actual production data. This system sets the time for moving one bay, completing one lifting or lowering of a container operation, and completing one loading / unloading operation as fixed values, and the time comes from the field statistical data of a certain port. The technical solution proposed in this literature generally still belongs to a linear model and does not match the actual situation of quay cranes well; although there is actual production data as support, the unit moving time of the quay crane is set as a static statistical mean, and the dynamic factors in the data are not fully explored; at the same time, the problem that there are statistical differences in the operation time between different container blocks is ignored, and the influencing factors considered are not comprehensive enough.

[0006] Zhou C et al., "Integrated Optimization on Yard Crane Scheduling and Vehicle Positioning at Container Yards", Advanced Engineering Informatics, 2022, 51: 101477, proposed a quay crane operation time prediction system considering different working conditions. The system considers the differences in the grasping and lowering operation times in the loading and unloading directions, and the operation times under the two working conditions are calculated separately. The prediction of the moving time of the quay crane is still based on a linear model and is calculated from the single-bay moving time. Although the technical solution proposed in this literature takes into account the fact that there are statistical differences in the time during the loading and unloading processes, it still cannot get rid of the limitations of the linear model, and the accuracy of predicting the quay crane operation time is limited.

[0007] Chen X et al., "A Data-Driven Genetic Programming Heuristic for Real-World Dynamic Seaport Container Terminal Truck Dispatching", 2020 IEEE Congress on Evolutionary Computation (CEC). 2020: 1-8, proposed a data-driven quay crane operation time prediction system. In the system, the starting and ending positions of the quay crane movement are regarded as nodes, and the operation time is connected between the two nodes. The operation time follows a probability distribution formed by historical data statistics and is a random value extracted from this probability distribution. The technical solution proposed in this literature is supported by historical data and fully explores the dynamic factors contained in the historical data. However, only the starting and ending positions are used as the indexes of the operation time, different working conditions are not distinguished, and business influencing factors such as container weight, lane ID, double-container operation mode, and environmental influencing factors such as wind speed and precipitation are not considered, so there is still room for optimization.

[0008] In summary, although the existing technical solutions comprehensively use physical models and data-driven research methods to achieve the prediction of quay crane operation time in container terminals, there are still problems such as lack of actual data support, failure to consider the influence of different working conditions, insufficient exploration of dynamic factors in historical data, and incomplete consideration of business and environmental influencing factors, making it difficult for the existing technical solutions to accurately predict the quay crane operation time under various working conditions.

[0009] How to achieve accurate intelligent quay crane operation time prediction in the terminal has become a technical problem to be solved. Summary of the Invention

[0010] The purpose of the present invention is to provide an intelligent quay crane operation time prediction method, system and electronic device to overcome the above-mentioned defects existing in the prior art.

[0011] The purpose of the present invention can be achieved by the following technical solutions:

[0012] According to one aspect of the present invention, there is provided an intelligent quay crane operation time prediction method, the method comprising:

[0013] Obtain the historical operation data and weather environment data of the intelligent quay crane in the terminal;

[0014] Clean the acquired data and perform feature engineering extraction to form a dataset. The feature engineering extraction includes extracting the starting and ending point coordinates of the gantry crane operation, extracting container information, matching historical operation data with weather environment data, and performing statistical tests and sensitivity analysis on each extracted feature to determine the features used to construct the dataset.

[0015] Based on the dataset, construct a gantry crane operation time matrix that stores the historical distribution of gantry crane operation times. For the input gantry crane operation instructions and the current weather environment conditions, use the Monte-Carlo algorithm to give the gantry crane operation time prediction results, including the distribution and confidence interval that the operation time follows.

[0016] Preferably, classify the gantry crane working conditions according to the gantry crane interaction object type and the gantry crane business process type. The gantry crane interaction objects include AGVs and brackets; the gantry crane business process types include storing or retrieving containers. The data cleaning eliminates invalid values and outliers in the operation times under different gantry crane working conditions, and uses the interquartile range method to identify the lower and upper bounds of outliers. Data smaller than the lower bound or larger than the upper bound will be eliminated. Specifically:

[0017] LB = Q 1 - 1.5×IQR

[0018] UB = Q 3 + 1.5×IQR

[0019] IQR = Q 3 - Q 1

[0020] where LB is the lower bound of the outlier, UB is the upper bound of the outlier, Q 1 and Q 3 are the first quartile and the third quartile respectively, and IQR is the interquartile range.

[0021] Preferably, the feature engineering extraction includes:

[0022] Extracting the starting and ending point coordinates of the gantry crane operation: Different interaction types use different coding methods. The block area type, lane number, and bracket number extracted from the strings of their respective coding methods are characterized by numbers for classification, and one-hot codes are formed according to different values; Store the extracted coordinate information within the block area and the classification information processed by the one-hot code respectively to form features. The interaction types include: yard, AGV, and bracket. The coordinate information within the block area includes the coordinates of the block area, bay, row, and tier.

[0023] Extraction of container information: Extract information on container weight, whether the container is a refrigerated container, and the double-container operation status of the container on the AGV from the container information, classify it using numbers, and form a one-hot code according to different values; for the container weight information with continuous values that cannot be used as matrix elements, discretize it into labels; store the container weight information and its discretized labels together with the classified information processed by the one-hot code to form features respectively.

[0024] Matching of historical operation data and weather environment data: By searching for weather data points in the nearest hour corresponding to the start time of a single quay crane operation, match the original values of continuous precipitation and wind speed at the corresponding time points to the operation data; discretize precipitation and wind speed into labels for use as matrix elements; store the original values of precipitation and wind speed together with the discretized labels to form environmental data features.

[0025] Preferably, the statistical test is specifically: Through the T-test, verify whether there is a statistical difference in the quay crane operation time for each feature under different values.

[0026] The sensitivity analysis is specifically: Use the random forest model to evaluate the importance of each feature and score it to determine the features that have a significant impact on the quay crane operation time; and draw a partial dependence plot based on the evaluation results of the random forest model to evaluate the change trend of each feature under the non-linear impact on the quay crane operation time, so as to determine the features used to construct the dataset.

[0027] Preferably, each dimension of the quay crane operation time matrix is a feature of the dataset, including the interaction type and business process of the quay crane operation, and the starting and ending point coordinates of the quay crane operation; the matrix element of the quay crane operation time matrix represents the statistical distribution formed by the historical quay crane operation time under the condition that each dimension is determined.

[0028] Preferably, the method further includes inputting the features of the quay crane operation time matrix into a neural network to output the quay crane operation time prediction value in real time.

[0029] The neural network includes several layers of neurons, divided into an input layer, a hidden layer, and an output layer; input the features of the quay crane operation time matrix into the input layer, each layer of neurons receives the weighted input from the previous layer, passes it to the next layer after passing through the activation function, and finally outputs the quay crane operation time prediction value at the output layer.

[0030] According to another aspect of the present invention, there is provided an intelligent dock rail crane operation time prediction system, the system comprising a data processing function module and an algorithm prediction function module, wherein the data processing function module performs data cleaning and feature extraction on the acquired dock rail crane operation data and the environmental weather data of the corresponding time to form a data set; the algorithm prediction function module constructs a rail crane operation time matrix based on the data set formed by the data processing function module, executes a prediction algorithm on the input rail crane operation instructions and the current weather environment conditions, and outputs a rail crane operation time prediction result;

[0031] The prediction algorithms include Monte-Carlo algorithm and neural network.

[0032] Preferably, the data processing function module includes a data cleaning submodule and a feature engineering submodule;

[0033] The data cleaning submodule is used to remove missing or abnormal invalid values ​​in different rail crane working conditions in the acquired data, and remove outliers by statistical methods. The rail crane working conditions are classified according to the types of rail crane interaction objects and the business process types of the rail crane;

[0034] The feature engineering submodule includes rail crane working condition classification, extraction of the starting and ending point coordinates of rail crane operations, extraction of container information, matching of historical operation data with weather environment data, and statistical tests and sensitivity analysis of the extracted features to determine the features used to construct the data set.

[0035] Preferably, the training process of the neural network is accelerated by a GPU to output a predicted value of the rail crane operation time in real time; and the Monte-Carlo algorithm outputs a distribution and a confidence interval obeyed by the rail crane operation time.

[0036] According to a third aspect of the present invention, there is provided an electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the method described above is implemented when the processor executes the program.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1) For the first time, the present invention combines the actual situation of the terminal, specifically differentiates the types of interaction objects of the rail-mounted gantry crane and the direction of the business process of the rail-mounted gantry crane according to specific working conditions; factors other than coordinates are taken into consideration, including operation characteristics such as container weight, special container types, lane ID, double-container operation mode, etc., and at the same time, real-time dynamic weather environment influencing factors are considered, including wind speed and precipitation; the extracted multiple features are used to construct the operation time matrix of the rail-mounted gantry crane, and the Monte-Carlo algorithm is executed to output the distribution and confidence interval that the operation time follows, which greatly improves the prediction accuracy compared with the existing methods. At the same time, the confidence interval helps the system to have high fault tolerance prediction in case of emergencies and has good robustness.

[0039] 2) The present invention combines the Monte-Carlo algorithm with a neural network, proposes a framework in which the confidence interval and the predicted value of the working time of the rail-mounted gantry crane are parallel, and uses GPU to accelerate the neural network training, which has excellent real-time performance and can achieve real-time, high-precision, and dynamic prediction.

[0040] 3) The two prediction algorithms of the present invention are combined and used, which can be applied to the prediction of the operation time of the rail-mounted gantry crane under various working conditions, provide effective data support for important scheduling processes such as the integrated scheduling of the rail-mounted gantry crane and AGV and the stowage plan at the front of the container terminal, and help to improve the utilization rate of system resources at the front of the intelligent terminal, which has great research significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flow chart of the data processing function module in the prediction system of the present invention;

[0042] Figure 2 It is a schematic diagram of the working condition classification of the rail-mounted gantry crane in the prediction system of the present invention;

[0043] Figure 3 It is a schematic flow chart of the prediction algorithm of the present invention;

[0044] Figure 4 It is a schematic diagram of the error percentile of the prediction results between the method of the present invention and the prior art;

[0045] Figure 5 It is a schematic flow chart of the prediction method of the present invention;

[0046] Figure 6 It is a schematic diagram of the structure of the electronic device of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] In view of the limitations of the above solutions, the present invention application combines the actual interaction situation between the rail-mounted gantry crane and other machinery in the terminal, divides the working process of the rail-mounted gantry crane into multiple working conditions according to the types of starting and ending points and different business processes. For each working condition, operation characteristics such as container weight, special container types, lane ID, double-container operation mode, etc. are considered, and at the same time, environmental impact factors such as wind speed and precipitation are considered to achieve the integration of static and dynamic factors. A multi-dimensional matrix of rail-mounted gantry crane operation time prediction is used to statistically represent the data distribution to fully explore the characteristics of historical data. And considering the problem of fault tolerance that needs to be considered in the actual terminal scheduling process, a prediction implementation plan with two parallel methods of Monte-Carlo algorithm and neural network is proposed. The Monte-Carlo algorithm gives the distribution and confidence interval that the operation time follows, and the neural network prediction model gives the real-time prediction value, realizing high-precision prediction that meets the actual needs of the terminal. To further improve the accuracy of the rail-mounted gantry crane operation time prediction in the intelligent terminal and provide technical support for the scheduling system of the intelligent terminal, thereby optimizing the operation efficiency and resource utilization rate of the system, the present invention proposes an intelligent terminal rail-mounted gantry crane operation time prediction method and system.

[0049] This embodiment relates to an intelligent terminal rail-mounted gantry crane operation time prediction method, such as Figure 5 , the method includes:

[0050] Step 1, obtain the historical operation data and weather environment data of the intelligent terminal rail-mounted gantry crane;

[0051] Step 2, perform data processing on the obtained data to form a data set. The data processing includes data cleaning and feature engineering extraction, writing to a storage medium, and using T-test, random forest model, and partial dependence graph to perform sensitivity analysis on the influencing factors of the rail-mounted gantry crane operation time, and the feature selection has good interpretability.

[0052] The process of data cleaning includes:

[0053] 1-1, rail-mounted gantry crane working condition classification, including classification according to the type of interaction object of the rail-mounted gantry crane (such as AGV or support) and the type of business process of the rail-mounted gantry crane (such as storing containers or retrieving containers).

[0054] 1-2. For different working conditions of the rail-mounted crane, identify and obtain the invalid values in the data, remove the outliers caused by special circumstances such as human factors, equipment failures, or bad weather through statistical methods, and retain the valuable data, including the following steps: For the scatter plots of the working hours of the rail-mounted crane under each working condition, draw histograms respectively, and statistically analyze the distribution followed by the working hours, as well as their interval and frequency characteristics; Based on the statistical data, use the Interquartile Range (IQR) method to identify and remove the outliers. The basic idea of the IQR method is: Based on the sorting position of the data from small to large, without depending on the specific form of the data distribution, to process non-normal distributed data. Let the first quartile of the data be Q 1 , and the third quartile be Q 3 , which respectively represent that 25% and 75% of the entire data are less than this number. The interquartile range IQR is defined as the difference between the third quartile and the first quartile:

[0055] IQR = Q 3 - Q 1

[0056] Identify the lower bound LB and upper bound UB of the outliers, which are respectively expressed as:

[0057] LB = Q 1 - 1.5×IQR, UB = Q 3 + 1.5×IQR

[0058] Data less than the lower bound LB or greater than the upper bound UB will be removed.

[0059] The feature extraction process includes:

[0060] 2-1. Extraction of the starting and ending point coordinates of the rail-mounted crane operation, including the following steps: Extract information from the strings of the respective coding methods of different interaction types (such as yard, AGV, support). For the yard-AGV combination, the information extracted is the block coordinates (block, bay, row, tier), block type (two-end type or cantilever type), and lane number. For the yard-support combination, the information extracted is the block coordinates (block, bay, row, tier) and lane number; For the information such as block type and lane number in the string that is characterized by numbers for classification, form one-hot codes according to their different values; Store the directly extracted coordinate information and the classification information processed by the one-hot code respectively to form features.

[0061] 2-2. Extraction of container information, including the following steps: extracting information such as container weight, whether the container is a refrigerated container, and the double-container operation status of the container on the AGV from the container information; for information classified by numbers such as whether the container is a refrigerated container and the double-container operation status of the container on the AGV, generating one-hot codes according to their different values; for the continuously valued container weight information that cannot be used as matrix elements, discretizing it into labels; storing the container weight information and its discretized labels together with the classified information processed by one-hot codes to form features respectively. The corresponding relationship between the container weight and its discretized label is:

[0062]

[0063] 2-3. Extraction of weather environment data, including the following steps: matching historical operation data with weather environment data, and accurately finding the nearest weather data time point corresponding to the start time of a single quay crane operation to the hour; matching the original values of continuously valued precipitation and wind speed at the corresponding time point to the operation data; discretizing precipitation and wind speed into labels according to the national standards of the China Meteorological Administration GB / T 28591-2012 and GB / T 28592-2012 for use as matrix elements; storing the original values of precipitation and wind speed and the discretized labels to form environmental data features.

[0064] 2-4. The statistical test uses the T-test to verify whether there are statistical differences in the quay crane operation time under different values of each feature. The sensitivity analysis process includes the following steps: using the random forest model to evaluate the importance of features and score them to determine which features have a significant impact on the quay crane operation time; drawing a partial dependence plot according to the evaluation results of the random forest model to evaluate the change trend of each feature under the non-linear impact on the quay crane operation time, and determining the features used to construct the dataset to ensure the accuracy and interpretability of feature engineering. The basic idea of the random forest model is: measuring the importance of a feature by evaluating the degree of decline in the performance of the prediction model after randomly shuffling the values of a certain feature. Using the original features to pre-train the prediction model, and recording the original error Error original , then the values of the j-th feature are randomly permuted one by one, and the prediction is made again and the new error is calculated to calculate the importance score Importance(j) of the j-th feature:

[0065]

[0066] Step 3. Execute the prediction algorithm based on the dataset and output the prediction result. It includes: constructing a quay crane operation time matrix driven by historical data, and through two implementation schemes of the Monte-Carlo algorithm and neural network, giving the operation time prediction value and confidence interval for a certain quay crane working condition.

[0067] This embodiment also relates to an intelligent quay rail crane operation time prediction system, which mainly includes a data processing module and an algorithm prediction module, as Figure 5 .

[0068] As Figure 1 , the data processing module first obtains the historical operation data of the quay rail crane and the weather environment conditions at the corresponding time through the storage medium to obtain the original data. The data processing module includes a data cleaning sub-module and a feature engineering sub-module.

[0069] The data cleaning sub-module is used to identify invalid values in the original data, remove outliers caused by special situations such as human factors, equipment failures, or bad weather through statistical methods, and retain valuable data.

[0070] The feature engineering sub-module extracts the influencing factors existing in the original data, and analyzes the statistical differences of each influencing factor and its impact on the operation time of the rail crane through T-tests, random forest models, and partial dependence plots, screens out the features in the original data, forms a data set available for the algorithm, and writes it into the storage medium. The feature engineering sub-module classifies the working conditions of the rail crane according to different interaction objects and business processes, extracts the starting and ending point coordinate information of the rail crane operation, as well as container-related information, and matches and extracts it in combination with historical operation data and environmental data. Through statistical tests and sensitivity analysis, the feature set for the algorithm prediction module is finally determined, and the final data set is formed and written into the storage medium.

[0071] The feature extraction process includes:

[0072] 1) Extraction of the starting and ending point coordinates of the rail crane operation: Extract information from strings with their respective coding methods for different types (such as yard, AGV, support). For the yard-AGV combination, the extracted information is the coordinates in the container area (container area, bay, row, tier), container area type (two-end type or cantilever type), and lane number. For the yard-support combination, the extracted information is the coordinates in the container area (container area, bay, row, tier) and support number; for the information such as container area type and lane number in the string, which are characterized by numbers for classification, one-hot codes are formed according to their different values; the directly extracted coordinate information and the classified information processed by one-hot codes are stored to form features respectively.

[0073] 2) Extraction of container information: Extract information such as container weight, whether the container is a refrigerated container, and the double-container operation status of the container on the AGV from the container information; for information classified by numbers such as whether the container is a refrigerated container and the double-container operation status of the container on the AGV, form one-hot codes according to their different values; for the container weight information with continuous values that cannot be used as matrix elements, discretize it into labels; store the container weight information and its discretized labels together with the classified information processed by one-hot codes to form features respectively. The correspondence between the container weight and its discretized label is as follows:

[0074]

[0075] 3) Extraction of weather environment data: Match the historical operation data with the weather environment data, and find the nearest weather data time point corresponding to the start time of each single gantry crane operation, accurate to the hour; match the original values of the continuously valued precipitation and wind speed at the corresponding time point to the operation data; discretize the precipitation and wind speed into labels according to the national standards GB / T 28591-2012 and GB / T 28592-2012 of the China Meteorological Administration for use as matrix elements; store the original values of the precipitation and wind speed and the discretized labels to form environmental data features.

[0076] 4) Statistical test: Through the T-test, verify whether there are statistical differences in the gantry crane operation time under different values of each feature. Sensitivity analysis evaluates the importance of each feature by using the random forest model and draws partial dependence plots (PDPs) to analyze the non-linear impact of each feature on the gantry crane operation time to ensure the accuracy and interpretability of the feature engineering. The basic idea of the random forest model is as follows: Measure the importance of a feature by evaluating the degree of decline in the performance of the prediction model after randomly shuffling the values of a certain feature. Pre-train the prediction model with the original features and record the original error Error original , then randomly permute the values of the j-th feature one by one, predict again and calculate the new error to calculate the importance score Importance(j) of the j-th feature:

[0077]

[0078] Based on the dataset generated by the data processing module and written into the storage medium, the algorithm prediction module constructs a quay crane operation time matrix on a computer electronic device, which serves as the basis for the prediction algorithm. The quay crane operation time matrix is used to store the historical distribution of quay crane operation time in matrix form. Each dimension of the matrix represents the key influencing factors identified in the data processing module, including the interaction type and business process of quay crane operation, as well as the starting and ending point coordinates of quay crane operation. Each element (i.e., matrix element) in the matrix represents the statistical distribution formed by the historical operation time of the quay crane under the determined dimensions.

[0079] The prediction algorithm includes two implementation schemes: the Monte-Carlo algorithm and the neural network model. The GPU is used to accelerate the training process of the neural network prediction model to output the predicted value of quay crane operation time in real time; for each input quay crane operation instruction and the current environmental conditions, the Monte-Carlo algorithm gives the distribution and confidence interval that the operation time follows.

[0080] The basic idea of the Monte-Carlo algorithm is: simulate a large number of samples of a certain random variable, and construct the empirical distribution followed by this random variable based on these samples. For a large number of independent and identically distributed samples, according to the central limit theorem, the distribution of the sample mean will be close to a normal distribution. If the standard deviation σ or the standard error of the sample can be known or estimated (where N is the sample size), then a confidence interval with a confidence level of 1-α can be constructed, following the formula below, where z α / 2 is the quantile with a tail area of α / 2 under the standard normal distribution:

[0081]

[0082] The quay crane operation time matrix considers the historical operation times of quay cranes with the same dimensions in each dimension as random variables following the same distribution. The Monte-Carlo algorithm is used to conduct a number of independent random trials on the probability distribution of each matrix element in the quay crane operation time matrix to construct a probability model of the quay crane operation time of this type as the matrix element. The results of the statistical random trials are calculated for their mathematical expectation and standard deviation, which are used as the estimated parameters of the normal distribution, thereby giving the confidence interval of the quay crane operation time. Based on the matrix element, after inputting the characteristics of each dimension of the quay crane operation, the Monte-Carlo algorithm can give the mean, variance, and confidence interval of the quay crane operation time as the prediction result.

[0083] The basic idea of a neural network is as follows: learn the mapping relationship between the input and output through multiple layers of non-linear transformations. A neural network consists of several layers of neurons, which are divided into an input layer, a hidden layer, and an output layer. Each layer of neurons receives the weighted input from the previous layer, passes it through an activation function, and then transmits it to the next layer, finally generating a result at the output layer. During the training process, the backpropagation algorithm is used to adjust the weights according to the loss function to minimize the prediction error. The training process can be accelerated using a GPU. The above process can be expressed by the following formula:

[0084] Calculation of neuron output, where z [l] is the linear combination of the l-th layer, W [l] and b [l] are the weight matrix and bias vector of this layer respectively, a [l-1] is the activation output of the previous layer, and g(·) is the activation function:

[0085] a [l] = g(z [l] )

[0086] z [l] = W [l] a [l-1] + b [l]

[0087] Loss function (mean squared error), where L(·) represents the loss function, is the predicted value, and y is the true value:

[0088]

[0089] The historical operation data of the rail-mounted gantry crane is divided into a training set and a test set. The input layer of the neural network is the operation characteristics of each dimension of the rail-mounted gantry crane, and the output layer is the operation time of the rail-mounted gantry crane under this characteristic. During the training process, the backpropagation algorithm is used to adjust the weights according to the loss function to minimize the prediction error until the loss function converges, and the test set is used for error verification to ensure that the output value has high precision. Finally, the operation characteristics of each dimension of the rail-mounted gantry crane are input to the neural network, and the high-precision predicted value of the operation time of the rail-mounted gantry crane is output in real time.

[0090] In this embodiment, all the historical operation data of the rail-mounted gantry crane in a certain intelligent container terminal in China in May 2024, as well as the environmental data including wind speed and precipitation, are read from the storage medium, and the time is accurate to the hour.

[0091] The data processing function module is as Figure 1 shown, differentiates the combination of the start point and the end point types, and obtains four working condition combinations, such as Figure 2Shown: From the yard (ASC) to the AGV, from the AGV to the yard, from the yard to the support, and from the support to the yard. The difference between the start time and the end time of the operation is calculated to obtain the time taken for each operation of the gantry crane.

[0092] Data cleaning sub-module: Excludes missing or abnormal invalid values in the original data, and removes outliers affected by special circumstances such as human factors, equipment failures, and bad weather through statistical methods, retaining data with analytical significance.

[0093] The feature engineering sub-module is used for sensitivity analysis such as coordinate extraction, container information extraction, environmental data matching, T-test, and partial dependence plots.

[0094] Coordinate extraction: Different coding methods are adopted according to different interaction types, and relevant information such as the container area type, coordinates within the container area, support number, and lane number is obtained to extract the start and end point coordinates of the gantry crane operation. Interaction types include: yard, AGV, support.

[0095] Container information extraction: Extract information such as the weight of the container, whether the container is a refrigerated container, and the double-container operation status of the container on the AGV from the container information; for information such as whether the container is a refrigerated container and the double-container operation status of the container on the AGV, which is characterized by numbers and classified, one-hot codes are formed according to their different values; for the container weight information with continuous values that cannot be used as matrix elements, it is discretized into labels; the container weight information and its discretized labels are stored together with the classified information processed by one-hot codes to form features respectively.

[0096] Environmental data matching: Match the historical operation data of the gantry crane with the environmental data according to the timestamp to obtain the wind speed and precipitation during the operation accurate to the hour;

[0097] Perform a T-test on the gantry crane operation time under each value of each influencing factor, evaluate the importance of the features using a random forest model and score them, and draw partial dependence plots to determine which data to select as features, as Figure 3 shown.

[0098] After passing through the data processing module, features are extracted from the historical operation data and environmental data of the gantry crane, a data set is formed, and written to the storage medium for the algorithm prediction function module to read and use.

[0099] In the algorithm prediction function module, as Figure 3As shown in the figure, based on features such as the block area and bay position of the dataset, an operation time matrix of the rail-mounted gantry crane is defined. Each dimension of the matrix is the features that affect the operation time of the rail-mounted gantry crane in the dataset, and each matrix element is the probability distribution of the operation time of the rail-mounted gantry crane under the value of this feature. The prediction algorithm can adopt two implementation methods: the Monte-Carlo algorithm and the neural network prediction model. The Monte-Carlo algorithm is based on the probability distribution of the operation time matrix of the rail-mounted gantry crane. Through multiple independent random trials, the results are statistically analyzed and the mathematical expectation and standard deviation are calculated as the estimation parameters of the normal distribution, so as to give the confidence interval of the operation time of the rail-mounted gantry crane. For the neural network method, a deep neural network with an input layer, a hidden layer, and an output layer is trained using the dataset. The network weights are iteratively optimized through the error backpropagation algorithm, and finally a prediction model is constructed to give the predicted value of the operation time of the rail-mounted gantry crane in real time under the condition of determining the feature values. Through the algorithm prediction module, the present invention can accurately predict the operation time of the rail-mounted gantry crane under any working conditions and give the confidence interval to ensure high robustness and real-time performance.

[0100] Figure 4 Shows the prediction results of the intelligent terminal rail-mounted gantry crane proposed by the present invention for transporting containers from the yard to the support for operation. To verify the accuracy of the prediction, the historical operation data of the rail-mounted gantry crane is divided into a training set and a test set at a ratio of 80% and 20%. The training set is used to construct the operation time matrix of the rail-mounted gantry crane, conduct Monte-Carlo algorithm simulation experiments, and neural network training, while the test set is used to calculate the relative error and verify the accuracy of the prediction results.

[0101] The average relative error (Mean Relative Error, MRE) on the test set is calculated by the following formula, where y pred represents the predicted value, y true represents the true value, n represents the number of samples in the test set, and the subscript i represents the i-th sample:

[0102]

[0103] The test results show that the MRE of the existing technology is 21.3%; the MRE of the Monte-Carlo algorithm is 4.3%, reducing the error by 79.8%; the MRE of the neural network algorithm is 5.3%, reducing the error by 75.1%. It can be seen from the error percentage diagram that for the points with an error of 50%, there are still many misjudgment cases in the existing technology, with a proportion as high as 6.0%; while for the Monte-Carlo algorithm and the neural network algorithm used in the present invention, there are basically no misjudgments under this fault tolerance scale. This difference shows that the present invention significantly reduces the occurrence probability of large-error misjudgment events.

[0104] Driven by the historical operation data and weather data of the rail-mounted crane, the system of this embodiment has formed a complete data processing method. Through two implementation schemes of the Monte-Carlo algorithm and neural network, the operation time of the rail-mounted crane is predicted quickly and accurately, and the error is significantly lower than that of the prior art. Moreover, it covers various working conditions of the intelligent terminal rail-mounted crane, takes into account various dynamic factors, and has good fault tolerance for emergencies in prediction, which is beneficial to improving the robustness of the system. It can be seen that the present invention significantly improves the prediction accuracy of the operation time of the intelligent terminal rail-mounted crane, is beneficial to optimizing the scheduling scheme of the intelligent terminal system, and thus improves the operation efficiency and resource utilization rate.

[0105] The present invention is applicable to the prediction of the operation time of the rail-mounted crane under various working conditions, and has the advantages of real-time, high-precision, and dynamic. At the same time, it takes into account the uncertain factors and has good robustness. The present invention can use GPU to accelerate the neural network training and has excellent real-time performance. The present invention provides effective data support for important scheduling processes at the forefront of container terminals such as the integrated scheduling of rail-mounted cranes and AGVs and the stowage plan, which helps to improve the system resource utilization rate at the forefront of intelligent terminals and has great research significance.

[0106] Such as Figure 6 , the electronic device of the present invention includes a central processing unit (CPU), which can execute various appropriate actions and processes according to the computer program instructions stored in the read-only memory (ROM) or the computer program instructions loaded from the storage unit into the random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus. It also includes a graphics processing unit (GPU) for processing the training of the neural network.

[0107] Multiple components in the device are connected to the I / O interface, including: an input unit, such as a keyboard, a mouse, etc.; an output unit, such as various types of displays, speakers, etc.; a storage unit, such as a disk, an optical disc, etc.; and a communication unit, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit allows the device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0108] The processing unit executes the various methods and processes described above. For example, in some embodiments, the method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed onto the device via the ROM and / or the communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute the method by any other suitable means (e.g., by means of firmware).

[0109] The functions described above herein may be performed at least in part by one or more hardware logic components. By way of example and not limitation, the types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system on a chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0110] The program code for implementing the methods of the present invention may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, a special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine, or entirely on the remote machine or server.

[0111] In the context of the present invention, a machine-readable medium may be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0112] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An intelligent method for predicting the operation time of a dock rail crane, characterized in that: The method comprises: Obtain historical operation data and weather environment data of intelligent terminal rail crane; The acquired data is cleaned and feature-engineered to form a data set. Feature-engineering extraction includes extracting the coordinates of the start and end points of the rail crane operation, extracting container information, matching historical operation data with weather and environmental data, and performing statistical tests and sensitivity analysis on each extracted feature to determine the features used to construct the data set; Based on the data set, a rail crane operation time matrix is ​​constructed to store the historical distribution of rail crane operation time. According to the input rail crane operation instructions and the current weather environment conditions, the Monte-Carlo algorithm is used to give the rail crane operation time prediction results, including the distribution and confidence interval of the operation time.

2. The method for predicting the operation time of an intelligent dock rail crane according to claim 1, characterized in that: The working conditions of the rail crane are classified according to the types of the rail crane interaction objects and the types of the rail crane business processes, where the rail crane interaction objects include AGV and brackets; the business process types of the rail crane include box storage or box retrieval; the data cleaning is performed to remove invalid values ​​and outliers of the operation time under different rail crane working conditions, and the interquartile range method is used to identify the lower and upper bounds of the outliers. Data less than the lower bound or greater than the upper bound will be removed, specifically: LB=Q1-1.5×IQR UB=Q3+1.5×IQR IQR=Q3-Q1 Among them, LB is the lower bound of the outlier, UB is the upper bound of the outlier, Q1 and Q3 are the first quartile and the third quartile respectively, and IQR is the interquartile range.

3. The method for predicting the operation time of an intelligent dock rail crane according to claim 1, characterized in that: The feature engineering extraction includes: Extraction of coordinates of the start and end points of rail crane operations: Different interaction types use different encoding methods. The container area type, lane number, and bracket number extracted from the character string of each encoding method are classified with digital representations, and a unique hot code is formed according to different values; the extracted coordinate information in the container area and the classification information processed by the unique hot code are stored to form features respectively; the interaction types include: yard, AGV, and bracket, and the coordinates in the container area include the coordinate information of the container area, bay, row, and layer; Extraction of container information: Extract the container weight, whether the container is a refrigerated container, and the double-box operation status of the container on the AGV from the container information, use digital representation to classify, and form a one-hot code according to different values; For the container weight information with continuous values ​​that cannot be used as matrix elements, discretize it into labels; store the container weight information and its discretized labels with the classification information processed by the one-hot code to form features respectively; Matching of historical operation data and weather environment data: by finding the weather data point of the nearest hour corresponding to the start time of a single rail crane operation, the original values ​​of the continuous precipitation and wind speed at the corresponding time point are matched to the operation data; the precipitation and wind speed are discretized into labels for use in matrix elements; the original values ​​of precipitation and wind speed are stored with the discretized labels to form environmental data features.

4. The method for predicting the operation time of an intelligent dock rail crane according to claim 1, characterized in that: The statistical test is specifically: through T test, verify whether each feature has a statistical difference in the operation time of the rail crane under different values; The sensitivity analysis is specifically as follows: using a random forest model to evaluate the importance of each feature and score it, and determining the features that have a significant impact on the operation time of the rail crane; And according to the evaluation results of the random forest model, a partial dependence diagram is drawn to evaluate the changing trend of each feature under the nonlinear influence of the rail crane operation time, so as to determine the features used to construct the data set.

5. The method for predicting the operation time of an intelligent dock rail crane according to claim 1, characterized in that: The dimensions of the rail crane operation time matrix are the characteristics of the data set, including the interaction type and business process of the rail crane operation, and the starting and ending point coordinates of the rail crane operation; the matrix elements of the rail crane operation time matrix represent the statistical distribution of the historical operation time of the rail crane when the dimensions are determined.

6. The method for predicting the operation time of an intelligent dock rail crane according to claim 1, characterized in that: The method further includes inputting each feature of the rail crane operation time matrix into a neural network and outputting a predicted value of the rail crane operation time in real time. The neural network includes several layers of neurons, which are divided into an input layer, a hidden layer and an output layer; the various features of the rail crane operation time matrix are input into the input layer, each layer of neurons receives the weighted input of the previous layer, passes it to the next layer after the activation function, and finally outputs the predicted value of the rail crane operation time in the output layer.

7. A prediction system for operation time of intelligent dock rail crane, the system executing the method as claimed in any one of claims 1 to 7, characterized in that: The system includes a data processing function module and an algorithm prediction function module, wherein the data processing function module performs data cleaning and feature extraction on the acquired dock rail crane operation data and the environmental weather data of the corresponding time to form a data set; the algorithm prediction function module constructs a rail crane operation time matrix based on the data set formed by the data processing function module, executes a prediction algorithm on the input rail crane operation instructions and the current weather environment conditions, and outputs a rail crane operation time prediction result; The prediction algorithms include Monte-Carlo algorithm and neural network.

8. The prediction system for intelligent terminal AGV operation time according to claim 7 is characterized in that: The data processing function module includes a data cleaning submodule and a feature engineering submodule; The data cleaning submodule is used to remove missing or abnormal invalid values ​​in different rail crane working conditions in the acquired data, and remove outliers by statistical methods. The rail crane working conditions are classified according to the types of rail crane interaction objects and the business process types of the rail crane; The feature engineering submodule includes extracting the coordinates of the start and end points of the rail crane operation, extracting container information, matching historical operation data with weather environment data, and performing statistical tests and sensitivity analysis on the extracted features to determine the features used to construct the data set.

9. The prediction system for intelligent terminal AGV operation time according to claim 7 is characterized in that: The training process of the neural network is accelerated by GPU to output the predicted value of the operation time of the rail crane in real time; the Monte-Carlo algorithm outputs the distribution and confidence interval of the operation time of the rail crane.

10. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

Images