CN117094460A

CN117094460A - Multi-target travel business control method, system and medium based on OD data

Info

Publication number: CN117094460A
Application number: CN202311086400.4A
Authority: CN
Inventors: 胡天宇; 王康晟; 于淏辰; 马惠敏
Original assignee: University of Science and Technology Beijing USTB
Current assignee: University of Science and Technology Beijing USTB
Priority date: 2023-08-28
Filing date: 2023-08-28
Publication date: 2023-11-21
Anticipated expiration: 2043-08-28
Also published as: CN117094460B

Abstract

The invention belongs to the technical field of information processing, and discloses a multi-target travel business control method, a system and a medium based on OD data, wherein M paths are automatically planned by a computer by means of a new generation information technology according to the data packet transfer request statistics condition of each information point at the moment T; the M data carriers are transferred between the information points in corresponding paths, so that all data packets generated at time T can reach their destination information points at time t+1. The OD data is called Origin-Destination data, is a data type with the information flow direction relation carried by the data, and is used as large-scale data, and the data has a clear graph structure and records space-time modes and trends. Macroscopically, the data on each temporal section can be characterized as a directed graph, and microscopically, each piece of data is a directed edge that forms the directed graph.

Description

Multi-target travel business control method, system and medium based on OD data

Technical Field

The invention belongs to the technical field of information processing, and particularly relates to a multi-target travel business control method, system and medium based on OD data.

Background

Currently, the traveller's problem (Traveling Salesman Problem, TSP) is a classical problem in the field of optimization, the objective of which is to solve a shortest path so that the traveller starts from the departure point, returns to the departure point after traversing all sites without great loss. Common methods for solving the TSP problem include greedy algorithm, simulated annealing, genetic algorithm, tabu search, particle swarm optimization, and the like. However, these methods are not new generation information technology methods which firstly adopt a graph neural network to predict and solve the multi-objective travel business problem based on the prediction result, and have the limitations of low calculation speed, low convergence speed, easy sinking into a local optimal solution, difficult obtaining of a global optimal solution and the like.

The Multi-objective traveller problem (Multi-objective Traveling Salesman Problem M-TSP) is an extension of the well-known traveller problem (TSP) in that it requires a set of routes to be determined for M travellers (here M is also included in the optimization objective) to start from a fixed initial site and return to the initial site. The nature of this problem makes it more suitable for practical application scenarios and by adding additional boundary constraints it can be applied to various vehicle path problems (Vehicle Routing Problem, VRP). However, although TSP and VRP have been widely studied, the study on M-TSP is relatively limited. Currently, there have been some studies focused on this problem and various solutions have been proposed, including binary programming-based solutions, two-stage methods of combining p-median problems (PMP) with TSP, and new algorithms based on Ant Colony Optimization (ACO), but there is much unexplored room in this field, and more research on M-TSP is necessary.

Meanwhile, the existing solving method of the traveling salesman problem, in particular to the solving method of the multi-target traveling salesman problem, is not focused on the data type which is very easy to generate in the production and operation process of an enterprise, namely OD data, is designed to be converted into the multi-target traveling salesman problem from the data type, and the research of the solving end-to-end business process solving method is carried out. The OD data is totally called Origin-Destination data, and is a data type with the information flow relation representation carried by the data. The departure place and departure time of the information carried by the data, the flow direction of the information, the destination and arrival time of the information are included. As a large-scale data, the data has a clearer graph structure and records space-time patterns and trends. Macroscopically, the data on each temporal section can be characterized as a directed graph, and microscopically, each piece of data is a directed edge that forms the directed graph.

Through the above analysis, the problems and defects existing in the prior art are as follows:

the calculation speed is slow: the traditional TSP solving method and some M-TSP solving methods have lower calculating speed due to higher complexity when processing large-scale problems. This results in difficulty in handling large-scale data sets or real-time problems in practical applications.

The convergence speed is slow: some conventional optimization algorithms, such as simulated annealing and genetic algorithms, require a long time to reach a better solution. This results in a slow convergence of the algorithm and a fast finding of a high quality solution.

Easily falls into a locally optimal solution: some optimization algorithms are easy to fall into a local optimal solution, and a global optimal solution cannot be found. This may result in relatively poor solution results that do not meet the actual requirements.

It is difficult to find a globally optimal solution: for complex TSP and M-TSP problems, finding a globally optimal solution is a challenging task. The existing method often cannot ensure that a globally optimal solution is found, but a suboptimal solution is found through a heuristic method.

Solutions focused on OD data are lacking: existing methods lack a specific solution for the OD data types in the enterprise production run when dealing with M-TSPs. These methods do not fully exploit the characteristics of the OD data and may not fully take into account the actual traffic demands and constraints.

Aiming at the defects and problems, the technical problems to be solved include:

efficient solution algorithm: there is a need to develop efficient algorithms that can handle large-scale TSP and M-TSP problems in a reasonable time, increasing computation speed and convergence speed.

Search strategy for global optimal solution: new search strategies need to be researched and designed to overcome the problem that the traditional algorithm is easy to fall into a local optimal solution, and to provide a better method for solving a global optimal solution.

OD data based solution: solutions to the OD data need to be focused on, converting it to a multi-objective traveler problem, and taking into account the actual business needs and constraints. This requires a combination of data preprocessing, specific model design and solution algorithms.

High performance computation and parallelization: and by utilizing a high-performance calculation and parallelization technology, the solving efficiency is improved, and the solving process of TSP and M-TSP problems is accelerated.

The multi-objective optimization method comprises the following steps: for the M-TSP problem, a multi-objective optimization method needs to be studied to balance route quality between different travelers and provide a number of efficient solutions for selection.

In the industrial field, the above-mentioned drawbacks and problems lead to the following specific consequences and challenges:

1) The production efficiency is reduced: if these methods are used for path optimization in manufacturing, logistics or supply chains, slow computation speeds can lead to delays in production decisions, thereby affecting overall production efficiency.

2) Increasing the cost: slow convergence or sinking to a locally optimal solution may result in production and logistics paths that are not optimal, thereby increasing transportation and production costs.

3) Challenges to meeting order requirements: if the industrial production relies on these algorithms for production planning and logistics optimization, the above problems may lead to production planning mismatch with actual order requirements, thereby affecting the timeliness of order delivery.

4) Waste of resources: the lack of solutions focusing on OD data may result in waste of resources (e.g., raw materials, machines, employees, etc.) because the optimization potential of the enterprise is not fully utilized with OD data.

5) Influence customer satisfaction: instability of the logistics and supply chain can lead to delays or errors in delivery, thereby affecting customer satisfaction and reputation of the business.

6) Losing market competitiveness: in a highly competitive market, efficient production and logistics optimization is critical. If an enterprise fails to effectively deal with these problems, it may lose the competitive advantage of an adversary.

Disclosure of Invention

Aiming at the problems existing in the prior art, the invention provides a multi-target tourist control method, a system and a medium based on OD data.

The invention is realized in such a way, a multi-objective travel business control method based on OD data, which automatically plans M paths by a computer by means of information technology according to the data packet transfer request statistics condition of each information point at the moment T; the M data carriers are transferred between the information points in corresponding paths, so that all data packets generated at time T can reach their destination information points at time t+1.

The multi-objective travel business control method based on the OD data further comprises the following steps:

step one, carrying out space-time sequence prediction on OD data, weighting each information point, according to observation on data transfer history of each information point, predicting data departure distribution rules and data arrival distribution rules of the information point by taking statistics of each information point at the moment T as a reference, and defining different weights for each information point; through the weight, each information point is characterized, the information points are divided into information points with more data flows and information points with more data flows according to the positive and negative weight values and the absolute value values, and the information points are divided into data processing priorities according to the absolute value values;

predicting the most likely OD data measurement in the next H time steps given the first M OD data of the window;

wherein v is _t ∈R ⁿ Is an observation vector of n information points with time step of t, and each element in the vector is all historical observation data of one information point;

dividing the information point area to be processed based on the prediction result obtained in the first step, selecting the departure point and the path point of the data carrier, and dividing each information point into categories K _l ，K ₂ ，...，K _m The goal is to minimize the square error F:

Wherein mu _i Is category K _m Is a mean vector of (a);

step three, based on the starting and ending points and the path points obtained in the step two, converting into a multi-objective travel business problem and solving the multi-objective travel business problem, and firstly abstracting the multi-objective travel business problem into the following mathematical model:

the data carrier number, a, is denoted by i=0, 1, …, n, m _i D, as an auxiliary intermediary variable _ij Representing the topological distance between any two information points, the multi-objective travel business problem can be expressed by the following definition:

A _i -A _j +1≤(n-1)(1-R _i，j )2≤i≠j≤n；

0≤A _i ≤n 2≤i≤n；

R _i，j ∈{0，l}i，j＝l，...，n；

R _i ∈Z i＝2，...，n。

further, the first step specifically includes:

(1) Preprocessing the original OD data, including basic processing such as outlier removal, blank value filling or removal, data merging and the like;

(2) The OD data space-time dimension is separated and mapped, and for the space dimension of the OD data, the attribute of each information point is defined as follows: whether data transfer is permitted is noted X or data transfer is permitted is noted Y, i.e., each information point is denoted (X, Y); the topological distance between any two information points i and j is denoted as d _ij The space dimension information w between any two information points i and j of the information points is represented by the following mapping method _ij ：

After mapping is a matrix of size (N, N) because each information point does not have a spatial dimension with itself So that the elements of the diagonal of the matrix are zero;sigma is w _ij D represents d _ij E represents natural logarithm, and e=0.1 is set to adjust the sparseness of W; through the transformation, w is a matrix with the size of (N, N), w _ij ∈[0，l)；

For the time dimension of the OD data, the primary data packet inflow at each moment T is marked as 1, the data packet outflow is marked as-1, and the data packet inflow is set as Null when the data packet does not flow in or flows out; converting the data grid with the numerical value into normal distribution between (0, 1), and setting zero for the left cell, so as to realize the two situations of no record and equal inflow and outflow; the processed data is marked as v;

the calculation method comprises the following steps: dividing each value in the matrix w by the largest integer in w and subtracting 1.0 to convert the integer value to between-1 and 1; after conversion, sequencing the converted values, scaling the values to a range of 0 and 1, mapping the values to standard normal distribution with an average value of 0 and a standard deviation of 1, traversing the original data, scaling the whole values to a range between 0 and 1 through the same mapping relation, and calculating a cumulative distribution function of the normal distribution;

(3) Designing a neural network structure to predict the distribution of OD data T+1 time separated according to space and time dimensions; the network structure consists of two space-time convolution layers and a fully connected output layer, each space-time convolution layer comprises a time convolution layer and a space-diagram convolution layer, and the input stream v _t-M+l ，...，v _t Calculating to obtain output flowEach space-time convolution layer comprises two time gating convolution layers, and residual error connection and bottleneck strategies are adopted in each space-time convolution layer; finally, a fully connected output layer integrates various features to generate a final prediction result +.>Providing high quality predictions when processing OD data with spatiotemporal characteristics;

stacking the time-gating convolution layer and the space-diagram convolution layer according to the time-space-time sequence, reducing the parameter number by reducing the channel number C, and inhibiting overfitting by layer normalization;

the input and output of the spatio-temporal convolution layer are both 3-dimensional tensors, for the input of block lOutput ofCalculated by the following formula:

wherein (1)>Is the upper and lower temporal layers of block I; theta (theta) ^l Is a graph convolution spectrum core; g represents a graph characteristic coefficient; reLU (·) represents a ReLU activation function, using the fully connected layer as the output layer;

finally, a final output Zε R can be obtained from the model ^n×c w+b, where w.epsilon.R ^c Is the weight vector, b is the bias, and the model performance is evaluated using the L2 penalty.

Further, the second step specifically includes:

(1) Clustering information points to be processed according to similarity distances, randomly selecting m information points as initial clustering centers, and regarding each information point sample x in the information point set _i Calculating the similarity distance to the clustering center, wherein the similarity distance comprises the historical change amplitude of the data packet quantity of the information point relative to the data packet quantity of all other information points, the data packet relative quantity of the information point at the current moment and the data packet relative quantity of the information point at the next moment; for each category C _i Recalculating the cluster centerUntil a termination condition is reached; the termination condition is to compare the updated cluster center with the cluster center of the previous iteration, and if the difference between the updated cluster center and the cluster center of the previous iteration is smaller than a preset threshold value or reaches a preset maximum iteration number, the algorithm is terminated; otherwise, returning to the distance calculation step, and continuing iteration;

(2) Judging the category of each category obtained by clustering, and extracting the departure point and the approach point of the data carrier;

the constraint conditions are as follows:

wherein O is _m Is a time interval (t ₀ ，t ₀ +T) the number of overall packet outputs for the mth cluster region, I _m Is a time interval (t ₀ ,t ₀ +t), the overall packet input number of the mth cluster region, L, the data carrier start threshold, N _m,t Is the relative number of data packets of the mth cluster area at the time point t, d _m A lower limit on the relative number of data packets representing the mth cluster region, U _m Upper limit of data packet relative quantity for representing mth cluster area, N _m,max Representing a time interval (t ₀ ,t ₀ +T) maximum number of packets relative to the mth cluster region, N _m,min Representing a time interval (t ₀ ,t ₀ +t) the relative number of packets for the mth cluster region is the minimum;

(3) And according to the data packet input data and the data packet output data of each area, subtracting the data packet input data from the data packet output data to obtain a difference, wherein the difference is positive, the data packet input or negative exists in the area at the time T+1, the data packet output exists in the area at the time T+1, the data carrier transfer starting and ending point is the data carrier transfer by the information point with the negative difference, the positive difference is the path point of the data carrier transfer, and the information point to be processed for carrying out the data packet transfer is determined.

Further, the third step specifically includes setting the following constraint conditions:

constraint 1: each data carrier starts from information point i=0 and finally returns information point i=0;

constraint 2: each of the remaining information points i=1, …, n, m except for the information point i=0 is accessed exactly once;

constraint 3: the disallowed information point sends a data packet pointing to itself;

constraint 4: each information point is listed in a traversal sequence of the data carrier;

Constraint 5: the access sequence of the starting and ending information point i=0 is 1;

constraint 6: any paths are not allowed to occur starting from the start information point i=0 and ending to the information point i=0, ensuring that constraint 1 holds.

Further, the method for controlling the multi-objective tourist based on the OD data comprises the following steps: in accordance with R for all present _ij D corresponding to each _ij D with the smallest absolute value _ij Preferentially listing the alternative path segments and traversing d with larger absolute value _ij Logic proceeds of (a); according to the result of the clustering of the information points in the second step, each class is allocated with a data carrier, and for the situation in each class, the following processing is carried out: checking whether the listed alternative path segments can form end-to-end loops each time a new path segment is introduced, allocating a data carrier to each formed end-to-end loop for packet transfer, and taking part in the alternatives forming the loopThe path segments are removed from the candidate set; the iteration is circulated until all the candidate sets formed by all the information points to be processed in the category selected in the step two are completely emptied; and finally, carrying out the processing on all the categories calculated in the second step to obtain a solving result of the multi-objective travel business problem.

It is a further object of the present invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the OD data based multi-objective traveller control method.

It is another object of the present invention to provide a computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the OD data-based multi-objective travel itinerary control method.

Another object of the present invention is to provide an information data processing terminal for implementing the OD data-based multi-objective carrier control method.

Another object of the present invention is to provide an OD data-based multi-target traveller control system based on the OD data-based multi-target traveller control method, the OD data-based multi-target traveller control system comprising:

the request statistics module is used for automatically planning M paths by a computer according to the data packet transfer request statistics condition of each information point at the moment T by means of a new generation information technology;

And the path transfer module is used for transferring M data carriers according to the corresponding paths among the information points, so that all data packets generated at the time T can reach the destination information point at the time T+1.

In combination with the technical scheme and the technical problems to be solved, the technical scheme to be protected has the following advantages and positive effects:

first, based on the above technical advances, the advantages and positive effects in industry mainly include:

1) Production and logistics efficiency are improved: more accurate data prediction means that resources and paths can be better planned, thereby reducing inefficient or redundant operations. This not only increases production and logistics efficiency, but also reduces the associated costs.

2) Resources are saved: efficient problem scaling and accurate path planning may reduce unnecessary waste of resources, such as manpower, equipment, and time. In this way, the enterprise may reallocate these resources to other critical tasks or projects.

3) Increasing system reliability: by applying technological advances, the probability of system failure or delay can be reduced. The higher system reliability not only reduces maintenance and repair costs, but also improves customer satisfaction.

4) The adaptability is strong: the multi-objective travel business control method brought by technical progress can better adapt to complex network environment and data requirements, so that the method has high adaptability in various different industrial application scenes.

5) The overall operation cost is reduced: more optimized solutions may not only reduce direct logistics and production costs, but also reduce potential costs due to errors or delays.

6) The decision quality is improved: more accurate and efficient data analysis and prediction may provide better insight to the management layer, making more informed and informed decisions.

7) Increase market competitiveness: enterprises can utilize the technical advances to improve the production and logistics efficiency of the enterprises, thereby better meeting the demands of clients and improving the competitiveness of the enterprises in the market.

8) Environmental protection: more accurate path planning and resource allocation means that unnecessary energy consumption and emissions are reduced, thereby reducing the impact on the environment.

9) Customer satisfaction improves: faster, efficient and accurate service increases customer satisfaction, which helps to build long-term customer relationships and increase customer loyalty.

The technical progress brings direct economic benefit and has positive influence on the sustainability and long-term development of enterprises.

Secondly, the invention provides an end-to-end business process solving method for converting the OD data into a multi-target travel business problem and solving the multi-target travel business problem, wherein the solving target of the multi-target travel business problem is a traversing path for completing information synchronization among different information points under the view angle of the OD data. For example, information point a generates 1 data packet at time T and requires transmission via the data carrier to information point B, which would create a large load on the data link and may occur if the transfer request for each data packet is performed in real time. Therefore, it is easily conceivable that the data packets may be buffered and accumulated for a certain time before being transmitted via the data link.

Third, for the three steps of the provided multi-objective traveler control method based on OD data, the following is a possible significant technical improvement made for each step:

step one: space-time sequence prediction of OD data

1) Enhancing prediction accuracy: by means of the time-space sequence prediction of the OD data, future data flow modes can be predicted more accurately, and a more robust basis is provided for further decisions.

2) Flexible information point weight allocation: different information points are given different weights according to the data outflow or inflow conditions, and a novel method is provided for classifying and prioritizing the information points.

3) Real-time data stream optimization: and adjusting the data flow in real time according to the weight of the information points and the historical data flow condition thereof so as to meet the actual requirements.

Step two: dividing the information point area to be processed

1) Data partition optimization: the information point areas can be more effectively divided through the guidance of the prediction result, so that convenience is provided for subsequent data processing.

2) Dynamic data carrier selection: based on the prediction and the region division results, suitable data carrier departure points and route points can be dynamically selected, so that the data flow is smoother.

3) Definition of objective function: a well-defined objective function is introduced, i.e. minimizing the square error F, making the solution of the problem more well-defined and directional.

Step three: conversion to a multi-objective travel business problem and solving 1) commonality of problem conversion: the real problem is abstracted and converted into the multi-objective travel business problem, and the conversion mode has universality and can be applied to various real scenes.

2) The solving efficiency is improved: after the multi-objective travel business problem is converted, the multi-objective travel business problem can be solved by using the existing algorithm or method, so that the overall solving efficiency is improved.

3) Expansibility: the provided mathematical model can be expanded or modified according to actual requirements, and has strong expansibility.

In general, the OD data-based multi-objective trip control method combines a spatio-temporal data analysis, an optimization algorithm and a multi-objective trip problem, and provides a new and efficient solution to the data flow and processing problems in practical applications.

Fourth, for each claim, the significant technical advances they bring can be discussed from the following point of view:

space-time sequence prediction based on OD data: the claims relate to the use of neural networks and other techniques to predict spatiotemporal sequences that can more accurately predict future OD data distributions. This predictive capability may provide a more accurate input for resolution of the traveler problem, thereby improving the accuracy and efficiency of path planning.

Clustering and region division: the claim relates to clustering according to similarity distances of information points to be processed and dividing areas of the information points to be processed. Through clustering and region division, the problem scale can be effectively reduced, and the complexity of path planning is simplified. This enables the algorithm to more quickly process large-scale information point data, improving computational efficiency.

Data carrier selection and path planning: the claims relate to selecting starting and destination points for a data carrier and converting the problem into a multi-objective traveler problem for solution. The multi-objective traveller problem is a classical combined optimization problem, and by applying corresponding mathematical modeling and solving algorithms, an optimal or near optimal path scheme can be found. The path planning method can improve the efficiency of data packet transfer, reduce network congestion and delay, and improve the overall performance of the system.

The technical advances in these claims are mainly manifested in the following aspects: more accurate data prediction, efficient problem scaling down, accurate path planning and optimization solutions. The advances enable the multi-target travel business control method based on the OD data to be better suitable for complex network environments and data requirements, improve the efficiency and reliability of data transmission, and bring remarkable technical progress for practical application scenes.

Drawings

FIG. 1 is a flow chart of a method for controlling a multi-objective trip based on OD data according to an embodiment of the present invention;

FIG. 2 is a logic diagram of the steps of embodiment 1 provided in an embodiment of the present invention;

FIG. 3 is a second logic diagram of step 1 provided in an embodiment of the present invention;

FIG. 4 is a step three logic diagram of example 1 provided by an embodiment of the present invention;

FIG. 5 is a first effect chart of step two of embodiment 1 provided in the present embodiment;

FIG. 6 is a second effect chart of the second step of the embodiment 1 provided by the embodiment of the present invention;

fig. 7 is a third effect diagram of the second step of embodiment 1 according to the present invention;

FIG. 8 is a first effect diagram of step three of embodiment 1 provided by the embodiment of the present invention;

fig. 9 is a second effect diagram of step three of embodiment 1 provided in the embodiment of the present invention;

Fig. 10 is a third effect diagram of the third step of embodiment 1 provided in the embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

As shown in fig. 1, the method for controlling a multi-objective tourist based on OD data according to the embodiment of the present invention includes the following steps:

s101: according to the data packet transfer request statistics condition of each information point at the moment T, automatically planning M paths by a computer by means of a new generation information technology;

s102: the M data carriers are transferred between the information points in corresponding paths, so that all data packets generated at time T can reach their destination information points at time t+1.

The specific implementation of the combined hardware is as follows:

1. hardware requirements

1) And the Central Processing Unit (CPU) is a high-performance multi-core processor and is used for processing complex data analysis and path planning algorithms.

2) Random Access Memory (RAM) is a memory capacity sufficient to store and process large amounts of OD data and intermediate calculations.

3) A data calculation unit: high performance graphics cards (GPUs) are used to undertake data computation functions.

3) And the storage device is a rapid Solid State Disk (SSD) for storing an OD data set, a prediction model, a path planning algorithm and the like.

4) Network Interface Cards (NICs) for receiving and transmitting OD data packets and communicating with other system modules or devices.

5) Sensors, such as GPS, infrared sensors, ultrasonic sensors, etc., are used to collect OD data in real time and monitor the position and status of the data carrier.

6) Mobile data carriers, such as drones, mobile robots, autonomous vehicles, etc., are used to transfer data packets between information points.

2. Detailed description of the preferred embodiments

1) And (3) data collection:

OD data for each information point is collected in real time using a sensor.

The NIC is used to receive OD packets from other systems or devices.

The collected data is stored in the SSD for subsequent processing.

2) Data preprocessing:

the OD data was cleaned, anomaly detected and missing value filled using the CPU.

Normalization or normalization of the data may also be required, as desired.

3) Data prediction:

and loading a pre-trained neural network model by using the GPU, and inputting the cleaned OD data into the model for prediction.

The prediction results are stored in the SSD for use by the path planning algorithm.

4) Path planning:

the path planning algorithm is loaded and the prediction results are used as input data.

And calculating by using a CPU to obtain M paths.

The path data are sent to the respective mobile data carrier.

5) And (3) data transfer:

according to the planned path, M data carriers are started.

The data carrier transfers data packets between the information points until all data packets arrive at the destination at time T + 1.

6) Monitoring and feedback:

the position and status of the data carrier is monitored in real time using sensors and this information is fed back to the central control system.

Real-time adjustment or optimization of the path may also be performed as needed.

7) Data archiving and reporting:

all OD data, prediction results, path planning results, etc. are stored in the SSD for subsequent analysis or backup.

And generating a report by using the CPU, and displaying indexes such as efficiency, accuracy and the like of data transfer.

The scheme fully combines hardware and algorithm, and ensures the high-efficiency execution of the multi-objective tourist control method of OD data in practical application.

For example: packets B (to be sent to point B), C (to be sent to point C), D (to be sent to point D), E (to be sent to point E) are generated at point a. After traversing from point a through the data carrier to point B, C, D, E and back, each packet arrives at its destination.

Example 1:

the multi-target travel business control method based on OD data provided by the embodiment of the invention comprises the following steps:

step one, carrying out space-time sequence prediction on OD data and giving weight to each information point. As shown in fig. 2; the purpose is as follows: according to the observation of the data transfer history of each information point, the data departure distribution rule and the data arrival distribution rule of the information point are predicted by taking statistics of each information point at the moment T as a reference, and different weights are defined for each information point. And (3) characterizing each information point by the weight, dividing each information point into information points with more data flows (the weight value is negative) and information points with more data flows (the weight value is positive) according to the positive and negative weight values and the absolute value of each information point, and processing the data for the information point division according to the priority.

Problem definition: i.e. the most likely OD data measurement in the next H time steps given the first M OD data of the window.

Wherein v is _t ∈R ⁿ Is an observation vector of n information points with a time step of t, and each element in the vector is all historical observation data of one information point.

The treatment method comprises the following steps:

(1) And preprocessing the original OD data, and performing basic processing such as outlier removal, blank value filling or removal, data merging and the like. Particular attention is paid to the need to remove the original data where long tail distribution exists. The prediction process focuses on the whole of all information points, and if the data transfer histories of some information points are too short or no data transfer histories exist, the data transfer relation matrix forms a sparse matrix, which is not beneficial to computer solution. And, for such information points, there is significantly no need for large-scale data-intensive transfer by means of a data transfer carrier.

(2) The OD data space-time dimension separation and mapping, for the space dimension of the OD data (i.e., connectivity relationship and transmission distance between information points). The attribute of each information point is defined as follows: whether data transfer is allowed is noted X or data transfer is allowed is noted Y, i.e. each information point may be denoted (X, Y). The topological distance between any two information points i and j is denoted as d _ij Characterizing the space dimension information w between any two information points i, j of the information points by the following mapping method _ij ：

After mapping is a matrix of size (N, N), the diagonal elements of the matrix are zero because each information point has no information in spatial dimensions with itself.Sigma is w _ij D represents d _ij E represents a natural logarithm, and e=0.1 is set to adjust the degree of sparseness of W. Through the transformation, w is a matrix with the size of (N, N), w _ij ∈[0，l)。

For the time dimension of OD data (i.e., the time series change of packets at each point in time), the primary packet ingress at each time T is denoted as 1, the packet egress is denoted as-1, and Null (Null) is set when there is neither ingress nor egress. And converting the data cells with numerical values into normal distribution among (0, 1), and setting zero for the left cells to realize the two conditions of no record and equal inflow and outflow. The processed data is denoted v.

The calculation method comprises the following steps: each value in the matrix w is divided by the largest integer in w and subtracted by 1.0 to convert the integer value to between-1 and 1. After conversion, the converted values are ordered and scaled to be within the range of 0 and 1, the values are mapped to standard normal distribution with the average value of 0 and the standard deviation of 1, the original data are traversed finally, the whole values are scaled to be between 0 and 1 through the same mapping relation, and the cumulative distribution function of the normal distribution is calculated.

(3) The neural network structure is designed to predict the distribution of OD data T+1 time after separation according to space and time dimensions.

The network structure consists of two space-time convolution layers and a fully connected output layer, and aims to process data with space and time dependence simultaneously, and each space-time convolution layer comprises a time convolution layer and a space diagram convolution layer in order to capture space-time characteristics of input data more efficiently. For input stream v _t-M+l ，…，v _t Calculating to obtain output flowEach spatio-temporal convolution layer comprises two time-gated convolution layers, and in order to optimize network performance and maintain stability, a residual connection and bottleneck strategy are adopted inside each spatio-temporal convolution layer. Such residual connection helps to prevent the gradient vanishing problem, making the network more stable during training. And the bottleneck strategy reduces the calculation complexity by reducing the number of parameters and improves the efficiency. Finally, a fully connected output layer integrates various features to generate a final prediction result +. >This makes it possible to provide high quality predictions when processing OD data with spatiotemporal characteristics.

The time-gated convolutional layers and the spatial-map convolutional layers are stacked in time-space-time order, the number of parameters is reduced by reducing the number of channels C, and the overfitting is suppressed by layer normalization.

wherein (1)>Is the upper and lower temporal layers of block I; theta (theta) ^l Is a graph convolution spectrum core; g represents a graph characteristic coefficient; reLU (·) represents the ReLU activation function, with the fully connected layer as the output layer.

Finally, a final output Zε R can be obtained from the model ^n×c w+b, where w.epsilon.R ^c Is a weight vector and b is a bias. And, the model performance was evaluated using the L2 loss.

Step two: dividing the information point area to be processed and selecting the departure point and the path point of the data carrier based on the prediction result obtained in the step 1. As shown in fig. 3;

the purpose is as follows: how to accurately identify which information points need to be prioritized during operation of the data carrier, so that each data packet can reach its destination at time t+1, while ensuring the operation efficiency of the overall system.

Problem definition: dividing each information point into categories K _l ，K ₂ ，…，K _m The goal is to minimize the square error F:

wherein mu _i Is category K _m Is a mean vector of (c).

The treatment method comprises the following steps:

(1) And clustering the information points to be processed according to the similarity distance. Firstly randomly selecting m information points as an initial clustering center, and for each information point sample x in the information point set _i Calculate the similarity distance to the cluster center (atThe similarity of the space-time characteristic distribution of the site is referred to as the invention). The similarity distance includes the historical change amplitude of the number of data packets of the information point relative to the number of data packets of all other information points, the relative number of data packets of the information point at the current moment and the relative number of data packets of the information point at the next moment. For each category C _i Recalculating the cluster centerUntil a termination condition is reached. The termination condition is to compare the updated cluster center with the cluster center of the previous iteration, and if the difference between them is smaller than a preset threshold or the preset maximum number of iterations is reached, the algorithm is terminated. Otherwise, returning to the distance calculation step, and continuing iteration.

(2) And carrying out category judgment on each category obtained by the clustering, and extracting the departure point and the path point of the data carrier.

The constraint conditions are as follows:

Wherein O is _m Is a time interval (t ₀ ，t ₀ +T) the number of overall packet outputs for the mth cluster region, I _m Is a time interval (t ₀ ，t ₀ +t), the overall packet input number of the mth cluster region, L, the data carrier start threshold, N _m，t At time point tThe relative number of data packets of the m-th clustering area, d _m A lower limit on the relative number of data packets representing the mth cluster region, U _m Upper limit of data packet relative quantity for representing mth cluster area, N _m，max Representing a time interval (t ₀ ，t ₀ +T) maximum number of packets relative to the mth cluster region, N _m，min Representing a time interval (t ₀ ，t ₀ +t) the relative number of packets for the mth cluster region is the minimum.

Finally, according to the data packet input data and the data packet output data of each area, the data packet output data is subtracted by the data packet input data to be used as a difference, the difference can be positive (which indicates that the area has data packet input at the time T+1) or negative (which indicates that the area has data packet output at the time T+1), the information point with the negative difference is the starting and ending point of data carrier transfer, the information point with the positive difference is the path point of data carrier transfer, and the information point to be processed for data packet transfer is determined.

Step three: based on the starting and ending points and the path points obtained in the second step, converting into a multi-objective travel business problem and solving the multi-objective travel business problem, as shown in fig. 4;

The purpose is as follows: the invention relates to a multi-target travel business problem solving method based on OD data, so that a final foothold point needs to convert a starting point and a path point which are calculated and obtained before into the multi-target travel business problem and solve the multi-target travel business problem, and the end-to-end whole process from the OD data to the multi-target travel business problem solving is finally realized.

Problem definition: first abstract the multi-objective travel business problem into the following mathematical model:

the data carrier number (i.e. the presence of a plurality of data carrier transport data in the whole of information points and several data paths, corresponding to a "multi-objective" in the multi-objective traveller problem) is denoted by i=0, 1, …, n, m, a _i D, as an auxiliary intermediary variable _ij Representing the topological distance between any two information points (and in step 1d _ij Meaning consistent). The multi-objective travel itinerary question may be expressed by the following definition:

A _i -A _j +1≤(n-1)(1-R _i，j )2≤i≠j≤n；

0≤A _i ≤n 2≤i≤n；

R _i，j ∈{0，l}i，j＝1，...，n；

R _i ∈Z i＝2，...，n.

the treatment method comprises the following steps:

on the basis of the foregoing problem modeling, the following constraint conditions are set:

constraint 1: each data carrier starts from information point i=0 and finally returns information point i=0

Constraint 2: each of the remaining information points i=1, …, n, m except for the information point i=0 is accessed exactly once.

Constraint 3: the point of information is not allowed to send data packets directed to itself.

Constraint 4: each information spot is listed in a traversal sequence of the data carrier.

Constraint 5: the access order of the origination and termination information point i=0 is 1 (i.e., the first access)

The solving process comprises the following steps: in accordance with R for all present _ij D corresponding to each _ij D with the smallest absolute value _ij Preferentially listing the alternative path segments and traversing d with larger absolute value _ij Is performed by logic of (a). According to the result of the clustering of the information points in the second step, each class is allocated with a data carrier, and for the situation in each class, the following processing is carried out: each time a new path segment is introduced, it is checked whether the listed alternative path segments are able to form end-to-end loops, a data carrier is allocated for each formed end-to-end loop for packet transfer, and the alternative path segments that have participated in the formed loop are removed from the alternative set. And (3) repeating the steps until all the candidate sets formed by all the information points to be processed in the category selected in the step two are completely emptied. And finally, carrying out the processing on all the categories calculated in the second step to obtain a solving result of the multi-objective travel business problem.

The multi-target travel business control system based on OD data provided by the embodiment of the invention comprises:

It should be noted that the embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or special purpose design hardware. Those of ordinary skill in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device of the present invention and its modules may be implemented by hardware circuitry, such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., as well as software executed by various types of processors, or by a combination of the above hardware circuitry and software, such as firmware.

The embodiment of the invention has a great advantage in the research and development or use process, and has the following description in combination with data, charts and the like of the test process.

In example 1, the effect of step one is shown in table 1;

table 1 RMSE statistics based on OD dataset and PeMSD7 dataset test set ablation experiments

The RMSE referred to herein is root mean square error and is calculated as follows:

step 2 effect: as shown in fig. 5-7;

step 3 effects as shown in fig. 8-10.

Based on the above-mentioned claim description, the following are two specific embodiments:

example 1: urban traffic planning

1) Background and application scenario:

in a large city, taxis, buses, sharing bicycles, etc. may be used as data carriers. OD data represents travel demand from one location to another. By using the method, future traffic flow can be predicted to optimize the urban traffic system.

2) The specific implementation scheme is as follows:

data preprocessing: first, OD data of various vehicles in a city, such as start-stop data of taxis, get-on and get-off data of buses, etc., are collected. These data are cleaned, such as removing abnormal data points, filling missing data, etc.

Space-time prediction: and predicting the traffic flow of the city by using the neural network structure, and predicting the future travel demand.

Path planning: and according to the prediction result, an optimal path is established for buses, taxis and the like, so that the predicted travel demand can be met, and the total travel time or distance is minimized.

Example 2: intelligent warehouse system

1) Background and application scenario:

in a large warehouse, there are multiple robots for the handling of goods. The warehouse is internally provided with a plurality of shelves and goods placement points. The robot needs to carry goods according to the appointed path, so that the requirements of each order are met.

2) The specific implementation scheme is as follows:

data preprocessing: OD data in the warehouse is collected, including the starting location and target location of the cargo. And cleaning the data, removing abnormal data and filling missing data.

Space-time prediction: using the neural network architecture described above, the handling requirements of the good are predicted, such as predicting which goods will likely be ordered within a time period in the future.

Path planning: and planning a carrying path for the robot according to the prediction result, so as to ensure that the requirements of all orders can be met in the shortest time.

Both the above embodiments are based on principles of spatio-temporal data prediction and path planning, but exhibit different specific implementation methods in different application scenarios. The multi-target travel business control method has wide application prospect, can play the advantages in multiple fields, improves the efficiency and reduces the cost.

The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the invention is not limited thereto, but any modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention will be apparent to those skilled in the art within the scope of the present invention.

Claims

1. The multi-target travel business control method based on the OD data is characterized in that M paths are automatically planned by a computer according to the data packet transfer request statistics condition of each information point at the moment T and the flow of 'OD data prediction-information point clustering-transfer path planning'; the M data carriers are transferred between the information points in corresponding paths, so that all data packets generated at time T can reach their destination information points at time t+1.

2. The OD data-based multi-destination traveler control method as in claim 1, wherein the OD data-based multi-destination traveler control method comprises the steps of:

dividing the information point area to be processed based on the prediction result obtained in the first step, selecting the departure point and the path point of the data carrier, and dividing each information point into categories K and K ₂ ，…，K _{Measuring amount} The goal is to minimize the square error F:

wherein mu _i Is category K _{Measuring amount} Is a mean vector of (a);

A _i -A _j +1≤(n-1)(1-R _i，j ) 2≤i≠j≤n；

0≤A _i ≤n 2≤i≤n；

R _i，j ∈{0，} i，j＝，…，n；

R _i ∈Z i＝2，...，n。

3. the method for controlling a multi-objective carrier based on OD data according to claim 2, wherein said step one specifically comprises:

(1) Preprocessing the original OD data, removing abnormal values, filling or removing blank values, and merging data to perform basic processing;

(2) Separating and mapping the space-time dimension of the OD data, and defining the attribute of each information point for the space dimension of the OD data;

(3) Designing a neural network structure to predict the distribution of OD data T+1 time separated according to space and time dimensions; the network structure consists of two space-time convolution layers and a fully connected output layer, each space-time convolution layer comprises a time convolution layer and a space-diagram convolution layer, and the input stream v _t-M+1 ，...，v _t Calculating to obtain output flowEach space-time convolution layer comprises two time-gating convolution layers, and each space-time convolution layer is internally provided withAdopting residual connection and bottleneck strategy; finally, a fully connected output layer integrates various features to generate a final prediction result +.>Providing high quality predictions when processing OD data with spatiotemporal characteristics;

4. The method for controlling a multi-objective carrier based on OD data according to claim 2, wherein the step two specifically comprises:

the constraint conditions are as follows:

wherein O is _{Measuring amount} Is a time interval (t ₀ ，t ₀ T) the number of overall packet outputs for the first cluster region, I _{Measuring amount} Is a time interval (t ₀ ，t ₀ T) the number of overall packet inputs for the first number of cluster regions, L being the data carrier start threshold, N _{Quantity t} Is the relative number of data packets of the first clustering area at the time point t, d _{Measuring amount} A lower limit on the relative number of data packets representing a first number of cluster regions, U _{Measuring amount} Upper limit of data packet relative quantity representing first quantity clustering area, N _{Quantity, quantity ax} Representing a time interval (t ₀ ，t ₀ T) maximum number of data packets relative to the first cluster region, N _{Quantity, quantity in} Representing a time interval (t ₀ ，t ₀ T) a relative number of packets for the first cluster region is a minimum;

5. The method for controlling a multi-objective carrier based on OD data according to claim 2, wherein said step three comprises the following constraint conditions:

constraint 2: each of the other information points i=1, …, n, m except for the information point i=0 is accessed exactly once

6. The OD data-based multi-destination traveler control method as in claim 2, wherein the OD data-based multi-destination traveler control method comprises: in accordance with R for all present _ij D corresponding to each _ij D with the smallest absolute value _ij Preferentially listing the alternative path segments and traversing d with larger absolute value _ij Logic proceeds of (a); according to the result of the clustering of the information points in the second step, each class is allocated with a data carrier, and for the situation in each class, the following processing is carried out: checking whether the listed alternative path segments can form end-to-end loops or not when introducing new path segments each time, distributing a data carrier for each formed end-to-end loop to carry out data packet transfer, and removing the alternative path segments which are participated in forming the loop from the alternative set; the iteration is circulated until all the candidate sets formed by all the information points to be processed in the category selected in the step two are completely emptied; and finally, carrying out the processing on all the categories calculated in the second step to obtain a solving result of the multi-objective travel business problem.

7. The OD data-based multi-objective trip control method according to claim 1, comprising:

1) And (3) data collection:

collecting OD data of each information point in real time by using a sensor;

using the NIC to receive OD packets from other systems or devices;

storing the collected data in an SSD for subsequent processing;

2) Data preprocessing:

cleaning OD data by using a CPU, detecting abnormality and filling missing values;

normalization or normalization of the data may also be required as needed;

3) Data prediction:

loading a pre-trained neural network model by using a GPU, and inputting the cleaned OD data into the model for prediction;

storing the prediction result in SSD for use by a path planning algorithm;

4) Path planning:

loading a path planning algorithm and using a prediction result as input data;

calculating by using a CPU to obtain M paths;

transmitting path data to each mobile data carrier;

5) And (3) data transfer:

starting M data carriers according to the planned path;

the data carrier transfers data packets between the information points until all the data packets reach the destination at the time of T+1;

6) Monitoring and feedback:

monitoring the position and state of the data carrier in real time by using a sensor and feeding back the information to a central control system;

Real-time adjustment or optimization of the path may also be performed as needed;

7) Data archiving and reporting:

storing all OD data, prediction results, path planning results and the like in SSD for subsequent analysis or backup;

8. A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the OD data-based multi-objective carrier control method according to any one of claims 1 to 6.

9. An information data processing terminal for implementing the OD data-based multi-destination hotel control method according to any one of claims 1 to 6.

10. An OD data-based multi-destination traveller control system based on the OD data-based multi-destination traveller control method of any one of claims 1 to 6, comprising: