Search

CN-121189968-B - Liquefied natural gas transportation scheduling method and system based on safety maximization

CN121189968BCN 121189968 BCN121189968 BCN 121189968BCN-121189968-B

Abstract

The invention provides a liquefied natural gas transportation scheduling method and system based on safety maximization, which relate to the technical field of safety management and comprise the steps of extracting characteristics of multidimensional information of a transportation vehicle through deep learning, and generating a safety characteristic vector by utilizing a time sequence convolution network and an attention mechanism; and carrying out multi-objective optimization on the route through a genetic algorithm, and generating an optimal scheduling instruction by integrating security situation scores and route attributes. The invention improves the safety and efficiency of liquefied natural gas transportation.

Inventors

  • ZHU YING

Assignees

  • 北京博达顺源天然气有限公司

Dates

Publication Date
20260512
Application Date
20250911

Claims (9)

  1. 1. The liquefied natural gas transportation scheduling method based on the safety maximization is characterized by comprising the following steps of: the method comprises the steps of obtaining multidimensional information of a liquefied natural gas transportation vehicle, carrying out feature extraction on the multidimensional information by adopting a deep learning algorithm, extracting time sequence features by adopting a time sequence convolution network, merging feature weights of different data sources by an attention mechanism, generating feature vectors reflecting safety states, calculating based on the feature vectors to obtain safety situation scores, carrying out time sequence segmentation on the multidimensional information by adopting a sliding window, obtaining preprocessing data by normalization processing, carrying out feature extraction on the preprocessing data by adopting the time sequence convolution network, obtaining time sequence features by adopting causal convolution operation and multi-layer expansion convolution, constructing a mode correlation matrix based on the time sequence features, representing the correlation strength among different mode features, inputting the mode correlation matrix into a plurality of parallel feature extraction channels with different perception field sizes to obtain multi-scale feature representations, constructing a time sequence state memory unit to dynamically merge the multi-scale feature representations with the historical feature and calculate the abnormality scores, carrying out self-adaption adjustment on the feature based on the abnormality scores, obtaining a dynamic matrix by adopting a self-adaption function of temperature coefficient adjustment, carrying out feature conversion on the self-adaption value, carrying out causal convolution operation and multi-layer expansion operation on the self-adaption value, carrying out self-adaption operation on the dynamic vector and the self-adaption value, carrying out the self-adaption operation on the dynamic vector and the normalized value, carrying out the self-adaption operation on the weighted vector and the weighted vector, and the weighted value, and the weighted vector has the self-adaption value, and the initial value is obtained, and the self-adaption value matrix is obtained, and the self-adaption state matrix after the key value is obtained, calculating the fusion characteristics through a multi-layer perceptron to obtain characteristic weights, wherein the multi-layer perceptron comprises a plurality of full-connection layers, carrying out Hadamard product operation on the characteristic weights and hidden layer characteristics output by a middle layer of the multi-layer perceptron, and obtaining security situation scores through nonlinear transformation; Establishing a dynamic graph structure of a road transportation network, wherein the road transportation network comprises nodes and road sections, and each road section is provided with weight information of traffic attributes; And carrying out multi-objective optimization on the plurality of candidate transportation routes, taking the security situation scores and the route attributes as optimization targets, solving by adopting a genetic algorithm to obtain a plurality of groups of optimized solutions, selecting an optimal transportation route according to preset weights, and generating a transportation scheduling instruction based on the optimal transportation route.
  2. 2. The method of claim 1, wherein the step of establishing a dynamic graph structure of a road transportation network, the road transportation network including nodes and road segments, each road segment having weight information of traffic attributes, comprises: Establishing a dynamic graph structure of a road transportation network, mapping road nodes into graph nodes, mapping road segments into graph edges, wherein feature vectors of the graph nodes comprise node geographic position coordinates, node types and node traffic capacity, and the initial weights of the graph edges are obtained through calculation of the road segment lengths, average vehicle speeds and vehicle flow; acquiring real-time traffic flow data and historical statistical data of the road transportation network, and performing space-time alignment and standardization processing to obtain standardized data; combining the standardized data with the feature vector of the graph node to calculate the dynamic association strength between the graph nodes, wherein the dynamic association strength is calculated through a learnable parameter matrix and LeakyReLU activation functions to obtain a node representation with enhanced time sequence; Calculating the space association degree between the nodes of the graph based on a space query matrix and a space key value matrix, wherein each row of the space query matrix represents a query vector of one node, each row of the space key value matrix represents a key vector of one node, and the sizes of the space query matrix and the space key value matrix are the node number multiplied by the characteristic dimension; Extracting space-time fusion characteristics among nodes of a connection graph, processing through a full connection network to obtain a weight adjustment factor, multiplying an initial weight by the weight adjustment factor to obtain an updated graph edge weight, calculating a deviation score of the updated graph edge weight and a historical graph edge weight, calculating a weight attenuation factor based on the deviation score when the deviation score exceeds a preset deviation threshold, and adjusting the updated graph edge weight by using the weight attenuation factor to obtain an adjusted graph edge weight.
  3. 3. The method of claim 2, wherein the step of calculating a plurality of candidate transportation routes using a graph neural network based on the dynamic graph structure of the road transportation network comprises: Performing feature aggregation and transmission based on the feature vector of the graph node and the adjusted graph edge weight to obtain global features; Obtaining dynamic association weights among nodes through similarity calculation of query vectors and key vectors of the nodes, and enhancing the global features based on the dynamic association weights to obtain enhanced features; constructing a gate control graph rolling unit comprising an updating gate and a resetting gate, calculating parameters of the updating gate and the resetting gate according to the enhanced features and the historical features, and carrying out iterative updating on the enhanced features by utilizing the parameters of the updating gate and the resetting gate to obtain a road network representation with time sequence memory capacity; Generating an initial transportation route based on the road list representation, extracting road list representation vectors of two nodes connected with the graph edges in the initial transportation route, calculating cosine similarity of the two road list representation vectors as a compatibility score, multiplying the compatibility score by a safety coefficient of the graph edges to obtain a time sequence memory characteristic score, calculating the product of the time sequence memory characteristic score and the position weight of each graph edge in the initial transportation route to obtain a graph edge weighted score, accumulating the weighted scores of all the graph edges in the initial transportation route to obtain a comprehensive score, and selecting the transportation route with the highest comprehensive score as a reference transportation route; And carrying out branch expansion on the reference transportation route, and carrying out replacement search on the graph edges in the reference transportation route based on preset expansion depth parameters to generate a plurality of candidate transportation routes.
  4. 4. The method of claim 1, wherein the step of performing multi-objective optimization on the plurality of candidate transportation routes, solving by using a genetic algorithm with the security situation score and the route attribute as optimization objectives to obtain a plurality of sets of optimization solutions, and selecting an optimal transportation route according to a preset weight comprises: mapping the plurality of candidate transportation routes into gene coding sequences, wherein the graph edges in each route are mapped into gene positions in the gene sequences, and calculating the safety weight of each gene position based on the safety situation scores; Constructing a multi-objective optimization function, combining the security situation score with the graph edge weight in the dynamic graph structure of the road transport network to obtain a security target value, and combining the time consumption and the distance consumption of the route to obtain an efficiency target value; Generating an initial coding population with a scale of N, calculating a safety target value and an efficiency target value of each individual in the population as fitness values, and performing non-dominant sorting on the initial coding population based on the multi-objective optimization function; Calculating the safety weight similarity of adjacent gene positions based on the initial coding population, determining a crossing interval according to the safety weight similarity, and adjusting the crossing probability based on a non-dominant grade; calculating a safe weight deviation value and a fluctuation value of a gene position to determine a variation position, and adjusting variation probability according to iteration times to perform variation operation; And carrying out weighted calculation on the comprehensive scores on the individuals in the child population based on the preset weights on the safety target values and the efficiency target values, and selecting the transportation route corresponding to the individual with the highest comprehensive score as the optimal transportation route.
  5. 5. The method of claim 4, wherein calculating safe weight similarity of adjacent gene loci based on the initial coding population, determining crossover intervals based on the safe weight similarity and adjusting crossover probabilities based on non-dominant levels, calculating safe weight bias values and fluctuation values of gene loci, determining mutation positions, and adjusting mutation probabilities based on iteration times comprises: Calculating a safety weight difference value of adjacent gene positions in a gene sequence, dividing the safety weight difference value by a larger safety weight value in the adjacent gene positions, and determining the gene positions with the safety weight similarity larger than a dynamic similarity threshold and the gene position spacing smaller than a preset interval length constraint as a crossing interval; adding a lower limit of the cross probability and a regulating quantity of the cross probability to obtain the cross probability, wherein the regulating quantity of the cross probability is obtained by multiplying a difference value between an upper limit of the cross probability and a lower limit of the cross probability by a regulating coefficient of a non-dominant grade, and the regulating coefficient of the non-dominant grade is a ratio of a difference value between a maximum non-dominant grade and a current individual non-dominant grade to the maximum non-dominant grade; The method comprises the steps of taking the ratio of the difference value of the population average safety weight and the current position safety weight to the population average safety weight as a safety weight deviation value, taking the ratio of the fluctuation range of the current position safety weight to the maximum safety weight as a safety weight fluctuation value, taking the weighted sum of the safety weight deviation value and the safety weight fluctuation value as a genetic locus assessment score, selecting a variation position based on the genetic locus assessment score, adjusting variation probability according to the current iteration times, and carrying out variation operation on the selected position.
  6. 6. The method of claim 4, wherein weighting the safety target value and the efficiency target value based on preset weights to calculate a composite score comprises: Calculating a route risk coefficient based on the liquefied natural gas loading capacity, the container protection level and the emergency resource distribution characteristics along the route, and taking the product of the route risk coefficient and the safety target value as a corrected safety target value; normalizing the corrected safety target value and the corrected efficiency target value, calculating the target Euclidean distance between individuals on the pareto front, clustering based on a distance threshold value, and determining a clustering center as a representative individual; Calculating a main score for each representative individual based on the preset safety target value and the preset weight of the efficiency target value, screening candidate individuals based on the road section overlapping rate, calculating candidate scores, and taking the weighted sum of the main score and the candidate scores as the comprehensive score of the individuals; And selecting an optimal representative individual and a corresponding alternative individual according to the comprehensive score, respectively serving as a main recommended route and an alternative route, and setting a dynamic switching threshold value based on the safety target value.
  7. 7. Lng transportation scheduling system based on safety maximization for implementing the method according to any of the previous claims 1-6, characterized in that it comprises: The system comprises a first unit, a second unit, a third unit and a fourth unit, wherein the first unit is used for acquiring multidimensional information of the liquefied natural gas transportation vehicle, performing feature extraction on the multidimensional information by adopting a deep learning algorithm, extracting time sequence features by a time sequence convolution network, fusing feature weights of different data sources by an attention mechanism, generating feature vectors reflecting safety states, and calculating based on the feature vectors to obtain a safety situation score; The system comprises a second unit, a first unit, a second unit and a third unit, wherein the second unit is used for establishing a dynamic graph structure of a road transportation network, the road transportation network comprises nodes and road sections, and each road section is provided with weight information of traffic attributes; And the third unit is used for carrying out multi-objective optimization on the plurality of candidate transportation routes, taking the security situation scores and the route attributes as optimization targets, adopting a genetic algorithm to solve the security situation scores and the route attributes to obtain a plurality of groups of optimized solutions, selecting an optimal transportation route according to preset weights, and generating a transportation scheduling instruction based on the optimal transportation route.
  8. 8. An electronic device, comprising: A processor; A memory for storing processor-executable instructions; Wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any of claims 1 to 6.
  9. 9. A computer readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the method of any of claims 1 to 6.

Description

Liquefied natural gas transportation scheduling method and system based on safety maximization Technical Field The invention relates to a safety management technology, in particular to a liquefied natural gas transportation scheduling method and system based on safety maximization. Background Lng is an important component of clean energy and its land transportation scale continues to increase. The liquefied natural gas has the characteristics of inflammability, explosiveness, low temperature and the like in the transportation process, has high safety risk, and can cause serious casualties and property loss once accidents occur. Therefore, how to ensure the safety of the road transportation of the liquefied natural gas becomes an important problem to be solved in the field of energy logistics. The current liquefied natural gas transportation scheduling is mainly based on the shortest path or the lowest cost principle to carry out route planning, and a static path planning algorithm is generally adopted to carry out scheduling decision by combining manual experience. With the development of information technology, part of enterprises introduce a geographic information system and an electronic map technology, so that the accuracy of transportation path planning is improved, but the method still has obvious defects in a safety control aspect. The existing dispatching system lacks comprehensive analysis capability of vehicle and environment multidimensional information, and cannot effectively utilize multi-source heterogeneous data such as vehicle states, driving behaviors, meteorological conditions and the like to carry out security risk assessment, so that security situation perception is insufficient, and potential risks are difficult to predict. The existing scheduling optimization algorithm is often focused on economic indexes such as transportation cost and time efficiency, and the quantitative consideration of safety factors is insufficient, so that a multi-objective optimization mechanism taking a safety situation score as a core optimization target is lacking, the balance of transportation efficiency and cost is difficult to realize on the premise of ensuring safety, and the fundamental requirement of 'safe first' in dangerous goods transportation cannot be met. Disclosure of Invention The embodiment of the invention provides a liquefied natural gas transportation scheduling method and system based on safety maximization, which can solve the problems in the prior art. According to a first aspect of an embodiment of the present invention, there is provided a lng transportation scheduling method based on safety maximization, including: The method comprises the steps of obtaining multidimensional information of a liquefied natural gas transportation vehicle, carrying out feature extraction on the multidimensional information by adopting a deep learning algorithm, extracting time sequence features by a time sequence convolution network, fusing feature weights of different data sources by an attention mechanism, generating feature vectors reflecting safety states, and calculating to obtain a safety situation score based on the feature vectors; Establishing a dynamic graph structure of a road transportation network, wherein the road transportation network comprises nodes and road sections, and each road section is provided with weight information of traffic attributes; And carrying out multi-objective optimization on the plurality of candidate transportation routes, taking the security situation scores and the route attributes as optimization targets, solving by adopting a genetic algorithm to obtain a plurality of groups of optimized solutions, selecting an optimal transportation route according to preset weights, and generating a transportation scheduling instruction based on the optimal transportation route. In an alternative embodiment of the present invention, The method comprises the steps of adopting a deep learning algorithm to extract characteristics of the multidimensional information, extracting time sequence characteristics through a time sequence convolution network, fusing characteristic weights of different data sources through an attention mechanism, generating a characteristic vector reflecting a safety state, and calculating based on the characteristic vector to obtain a safety situation score, wherein the step of calculating comprises the following steps: Performing time sequence segmentation on the multidimensional information by adopting a sliding window, and obtaining preprocessing data through normalization processing; performing feature extraction on the preprocessed data by adopting a time sequence convolution network, and obtaining time sequence features through causal convolution operation and multilayer expansion convolution; The method comprises the steps of establishing a time sequence feature, establishing a modal correlation matrix based on the time sequence feature, wherein th