Dangerous chemical transportation path screening method based on deep learning

Abstract

The application discloses a kind of based on deep learning's dangerous chemical transport path screening method, it is related to artificial intelligence data processing technical field, including: obtaining the road condition feature sequence and road environment feature sequence of candidate transport path;Simulate the running process of LNG tank car, obtain predicted vehicle motion response feature sequence;Based on storage tank structure parameter, LNG initial state parameter and vehicle motion response feature sequence, predict storage tank dynamic response feature sequence and liquid sloshing state feature sequence;According to road environment feature sequence, predict storage tank heat exchange state feature sequence;Based on the above prediction result, the comprehensive stability of candidate transport path is evaluated, select the path with the largest transport stability index as the target transport path.The prior art LNG tank car transport path screening process cannot consider the influence of transport path on the dynamic state of storage tank and LNG gas-liquid stability, which leads to inaccurate path safety evaluation.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence data processing technology, specifically to a method for screening hazardous chemical transportation routes based on deep learning. Background Technology

[0002] With the expanding application of liquefied natural gas (LNG) in energy supply, industrial production, and urban gas, LNG tanker transportation, as a crucial mode of land-based LNG transport, has garnered significant attention for its safety. Due to LNG's low temperature and high volatility, ensuring the structural safety of storage tanks and the stability of the internal medium during transportation is essential. Therefore, the appropriate selection of transportation routes is crucial for guaranteeing the safe operation of LNG tankers.

[0003] Existing methods for selecting LNG tanker transport routes typically plan routes based on factors such as road network information, vehicle traffic conditions, and transport distance. By avoiding road areas that do not meet vehicle traffic requirements, candidate routes that meet the transport conditions are identified. However, in actual transport, the operating status of LNG tankers is affected by various factors, including road conditions, transport environment, and changes in the vehicle's own operating status. Different transport routes may lead to varying degrees of dynamic changes in the vehicle's operation, thereby affecting the safety of the transport process.

[0004] Meanwhile, due to the continuous changes in vehicle operation, tank status, and LNG medium state during LNG tanker transportation, existing route selection methods struggle to comprehensively analyze the transportation process corresponding to different routes, leading to discrepancies between the route selection results and the actual safety conditions during transportation. Therefore, improving the accuracy and reliability of LNG tanker transportation route selection has become a pressing technical problem that needs to be solved. Summary of the Invention

[0005] This invention provides a deep learning-based method for screening hazardous chemical transportation routes, which addresses the problem that existing LNG tanker transportation route screening methods fail to comprehensively consider the impact of transportation routes on the dynamic state of storage tanks and the gas-liquid stability of LNG, leading to inaccurate route safety assessments.

[0006] In view of the above problems, the present invention provides a method for screening hazardous chemical transportation routes based on deep learning, the method comprising: Obtain the road condition feature sequence and road environment feature sequence for each candidate transportation route; Simulate the operation process of LNG tank trucks on each candidate path based on the road condition feature sequence, and obtain the predicted vehicle motion response feature sequence corresponding to each candidate path. Based on the structural parameters of the LNG tanker, the initial state parameters of LNG, and the predicted vehicle motion response feature sequence, the predicted tank dynamic response feature sequence and the predicted LNG liquid sloshing state feature sequence for each candidate path during transportation are obtained. Based on the structural parameters of the LNG tanker, a tank heat exchange prediction model is established, and the predicted tank heat exchange state feature sequence corresponding to each candidate path is obtained according to the initial state parameters of LNG and the road environment feature sequence. Based on the predicted tank dynamic response characteristic sequence, the predicted LNG liquid sloshing state characteristic sequence, and the predicted tank heat exchange state characteristic sequence, a comprehensive evaluation of LNG gas-liquid stability and tank dynamic stability is conducted on each candidate transportation route, and the candidate transportation route corresponding to the maximum transportation stability index is selected as the target transportation route.

[0007] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention establishes a mapping relationship between road transport conditions and LNG tanker storage conditions, enabling a joint assessment of the dynamic stability, liquid sloshing stability, and LNG gas-liquid stability of storage tanks under different transport routes. This allows the selected target transport routes to reduce operational risks caused by vehicle disturbance, liquid sloshing, and heat exchange during transport, thereby improving the safety and reliability of hazardous chemical transport. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the process of a method for screening hazardous chemical transportation routes based on deep learning, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the principle of tank dynamic response and liquid sloshing prediction provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the tank heat exchange prediction principle provided in an embodiment of the present invention. Detailed Implementation

[0009] This invention provides a deep learning-based method for screening hazardous chemical transportation routes, which specifically addresses the problem in existing LNG tanker transportation route screening processes that fail to comprehensively consider the impact of transportation routes on the dynamic state of storage tanks and the gas-liquid stability of LNG, leading to inaccurate route safety assessments.

[0010] Examples, such as Figure 1 As shown, this invention provides a method for screening hazardous chemical transportation routes based on deep learning, including: S100: Obtain the road condition feature sequence and road environment feature sequence for each candidate transportation route; Step S100 in the method provided in this embodiment of the invention includes: Based on the geographical location information corresponding to the starting point and destination of liquefied natural gas transportation, multiple passable road nodes are identified in the target area map, and passable roads are enumerated according to the connection relationship between the multiple passable road nodes to generate multiple candidate transportation routes. Each candidate transportation route is segmented according to the preset road segmentation rules to obtain multiple road segment units corresponding to each candidate transportation route. For each road segment unit, the corresponding road geometry information, road status information, and road environment information are obtained, and the road segment units are arranged in time sequence according to their arrangement order in the corresponding candidate transportation routes to generate road condition feature sequences and road environment feature sequences for each candidate transportation route.

[0011] In this embodiment, multiple candidate transportation routes are first determined based on the transportation start and end points of the liquefied natural gas (LNG) tanker trucks, and the road features corresponding to each candidate transportation route are extracted to provide input data for subsequent vehicle motion response prediction, tank dynamic response analysis, and LNG gas-liquid stability evaluation.

[0012] Specifically, the geographical location information of the origin and destination of the liquefied natural gas (LNG) transportation task is obtained. The origin is the LNG filling area, and the destination is the LNG receiving or storage area. The geographical location information includes the latitude and longitude of the origin and destination, map information of the area, and road network connection information.

[0013] Furthermore, a road network model is established in the target area map based on the geographical location information corresponding to the transportation origin and destination.

[0014] The road network model includes multiple passable road nodes and the road connections between these nodes. Passable road nodes include road intersection nodes, road entrance nodes, road exit nodes, and road transition nodes.

[0015] For each road node, the corresponding road attribute information is further obtained, including road width, road height limit, road slope, road type, and road risk area information.

[0016] Subsequently, based on the connection relationships between each passable road node, starting from the road node corresponding to the transportation origin, the search proceeds along the road network model to the road node corresponding to the transportation destination, obtaining multiple road connection schemes, and taking each road connection scheme that meets the transportation requirements as a candidate transportation path.

[0017] Furthermore, to ensure that the candidate transportation routes are suitable for LNG tanker transportation, the obtained candidate transportation routes are screened for safety adaptability.

[0018] Specifically, road traffic constraints are determined based on the LNG tanker vehicle size parameters, vehicle mass parameters, road transport regulations, and historical transport operation data.

[0019] The road traffic constraints include: road width constraints, used to determine whether the effective road width meets the width requirements of LNG tank trucks and the safety requirements for passing vehicles; road height restrictions, used to determine whether the road height restrictions meet the height requirements of the tank trucks; road slope constraints, used to determine whether there is insufficient power or braking risk when the vehicle is running on a continuous sloping road section; and dangerous area constraints, used to determine whether the transportation route passes through restricted traffic areas or high-risk areas.

[0020] If any segment of a candidate transportation route does not meet the above road traffic constraints, the transportation applicability of the corresponding candidate transportation route will be reduced, and routes with higher transportation risks will be eliminated.

[0021] The judgment range in the road traffic constraints is determined based on the LNG tanker vehicle structural parameters, the manufacturer's safety operation requirements, and historical transportation accident statistics to ensure that the selected candidate transportation routes meet the safety requirements for LNG tanker transportation.

[0022] After the above screening, several candidate transportation routes that meet the requirements for LNG tanker transportation were retained.

[0023] Furthermore, each candidate transportation route is segmented based on changes in road characteristics to obtain multiple road segment units.

[0024] Specifically, the pre-defined road segment division rules do not divide the road according to fixed distances, but rather determine the road segment boundary points based on changes in road characteristics. These changes include: abrupt changes in road slope, changes in road curvature, changes in road surface condition, and changes in the road environment.

[0025] When a significant change in road gradient is detected at a certain location in a candidate transportation route, such as when a gentle road transitions into a continuous uphill road, that location is taken as the road segment boundary.

[0026] When a change in road curvature is detected, such as when a straight road segment transitions into a continuous curved road segment, the corresponding location is taken as the new road segment boundary.

[0027] When a change in road surface condition is detected, such as transitioning from a regular asphalt road to a damaged, slippery, or special road surface area, the corresponding location will be used as the new road segment boundary.

[0028] By using the above method, each road segment unit corresponds to a relatively stable road characteristic state, avoiding the large differences in road characteristics within the same road segment caused by the traditional fixed-length division method, thereby improving the accuracy of subsequent vehicle motion response prediction.

[0029] For example, for an LNG tanker transport route of approximately 100km in length, if the first 20km of the road has a relatively gentle slope, the middle 30km has a continuous curved road section, and the last 50km has a slippery road surface area, then the route is divided into multiple road segment units according to the changes in road characteristics, rather than being divided into 10 road segments with a fixed length of 10km. This ensures that each road segment unit can truly reflect the changes in road excitation experienced by the vehicle during operation.

[0030] Furthermore, for each segment unit obtained, the corresponding road condition characteristics are acquired.

[0031] Among them, road condition features are used to describe the road's own attributes that affect the operating status of LNG tank trucks, including road geometry information and road condition information.

[0032] Road geometry information includes: road slope information, used to describe the impact of longitudinal road changes on vehicles during driving; road curvature information, used to describe the impact of lateral movement on vehicles during turning; road length information, used to describe the continuous driving distance of vehicles within the corresponding road area; and road elevation change information, used to describe the overall undulation of the road.

[0033] Road condition information includes: road surface smoothness information, used to describe the effects of periodic vibrations on vehicles during driving; road surface damage information, used to describe abnormal road surface conditions such as potholes and cracks; road surface slipperiness information, used to describe changes in vehicle adhesion caused by factors such as rainfall and water accumulation; road surface adhesion information, used to describe the contact state between vehicle tires and the road.

[0034] Specifically, road geometry and road condition information corresponding to each road segment unit are collected and arranged according to the order of each road segment unit in the candidate transportation path to form a road condition feature sequence for the corresponding candidate transportation path.

[0035] Furthermore, corresponding road environment information is obtained for each road segment unit.

[0036] Among them, road environment information is used to describe the external environmental conditions that affect the heat exchange process of LNG tank trucks, including: ambient temperature information, ambient humidity information, light intensity information, wind speed information, and rainfall status information.

[0037] Specifically, environmental parameters for each road segment unit within the corresponding time range are obtained through meteorological data interfaces, road environment monitoring equipment, and historical environmental databases.

[0038] Among them, ambient temperature and light intensity affect the process by which the outer wall of the storage tank absorbs ambient heat, ambient humidity and wind speed affect the heat exchange efficiency between the outer wall of the storage tank and the surrounding environment, and rainfall conditions affect the changes in the boundary conditions of the external environment of the storage tank.

[0039] Subsequently, based on the order of each road segment unit in the candidate transportation route, the corresponding road environment information is arranged in time sequence to generate a road environment feature sequence.

[0040] For example, an LNG tanker truck needs to be transported from a liquefied natural gas production base to a city gas receiving station. Based on the map information corresponding to the transportation origin and destination, a total of 8 road connection schemes were obtained. After filtering based on road width, height restrictions, slope, and hazardous area constraints, 3 candidate transportation routes that meet the transportation requirements were obtained. Among them, the first candidate transportation route is 95 kilometers long. Based on the changes in road slope, road curvature, and road surface condition, this route is divided into 12 road segment units. For one of these road segment units, the road condition characteristics are obtained: road length is 6 kilometers, average slope is 3.5 degrees, road curvature is relatively high, road surface smoothness is low, and there is a slight slippery condition. At the same time, the environmental characteristics of this road segment are obtained: ambient temperature is 30℃, ambient humidity is 60%, and light intensity is 500 watts per square meter. Subsequently, according to the vehicle's direction of travel, the road condition information and road environment information corresponding to the 12 road segment units are arranged to form the road condition feature sequence and road environment feature sequence corresponding to the candidate transportation route, and then input into the subsequent vehicle motion response prediction model to predict the vehicle vibration response and storage tank stability during the operation of LNG tank trucks under different transportation routes.

[0041] In summary, this step generates multiple candidate transportation routes based on the origin and destination of LNG tanker transport, and dynamically divides road segments by combining LNG tanker transport safety constraints and road feature changes. It further extracts road geometry information, road state information, and road environment information to form road condition feature sequences and road environment feature sequences that can reflect road changes throughout the transportation process. This provides a more accurate data foundation for subsequent vehicle motion response prediction, tank dynamic response prediction, and LNG gas-liquid stability evaluation, thereby improving the safety and reliability of LNG tanker transport route selection.

[0042] S200: Simulate the operation process of LNG tank trucks on each candidate path based on the road condition feature sequence, and obtain the predicted vehicle motion response feature sequence corresponding to each candidate path; Step S200 in the method provided in this embodiment of the invention includes: A vehicle motion response prediction model was constructed based on LNG tanker vehicle parameters; The road condition feature sequence corresponding to each candidate transportation route is input into the vehicle motion response prediction model. Based on the road change information in the road condition feature sequence, the operating status of LNG tank trucks on the corresponding candidate transportation routes is simulated to obtain vehicle motion response data of LNG tank trucks during transportation. The vehicle motion response data is arranged in chronological order during transportation to generate a predicted vehicle motion response feature sequence for each candidate transportation path.

[0043] In this embodiment, after obtaining the road condition feature sequence corresponding to each candidate transportation route, the vehicle motion response prediction model is further used to simulate the operation process of LNG tank trucks on different candidate transportation routes to obtain the corresponding predicted vehicle motion response feature sequence.

[0044] Specifically, a pre-built vehicle motion response prediction model is obtained.

[0045] Among them, the vehicle motion response prediction model is used to predict the dynamic response of LNG tank trucks during operation based on changes in road conditions. Its input is the road condition feature sequence corresponding to the candidate transportation route, and its output is the vehicle motion response data during the corresponding transportation process.

[0046] Furthermore, the road condition feature sequences corresponding to each candidate transportation route are input into the vehicle motion response prediction model.

[0047] Specifically, according to the arrangement order of road segment units in each candidate transportation route, the corresponding road condition characteristics are sequentially input into the vehicle motion response prediction model, so that the model can predict the vehicle motion response of LNG tankers at each time node based on the road characteristic change process.

[0048] Among them, the road slope change in the road condition feature sequence is used to reflect the change in the longitudinal motion response of the vehicle, the road curvature change is used to reflect the change in the lateral motion response of the vehicle, and the road surface condition change is used to reflect the change in the vertical vibration response of the vehicle.

[0049] Furthermore, in the process of predicting vehicle motion response, the vehicle operating status corresponding to each road segment unit is continuously predicted based on the time sequence corresponding to the road condition feature sequence.

[0050] Specifically, the corresponding time nodes are determined based on the vehicle's travel time on each road segment. The road segment travel time is determined based on the road segment length and the vehicle's speed. Road segment travel time = road segment length ÷ average vehicle speed.

[0051] Subsequently, based on the road condition characteristics corresponding to each time point, the road excitation is input into the vehicle motion response prediction model to obtain the vehicle motion response data at the corresponding time point.

[0052] The vehicle motion response data includes: longitudinal acceleration changes, which reflect the longitudinal motion changes of the LNG tanker as it passes through sloping road sections and during acceleration and deceleration; lateral acceleration changes, which reflect the lateral motion changes of the LNG tanker when it passes through turning sections; vertical acceleration changes, which reflect the vibration changes of the LNG tanker when it passes through different road surface conditions; and vehicle speed changes, which reflect the changes in vehicle operating speed during transportation.

[0053] Furthermore, based on the actual operating sequence of LNG tank trucks on the candidate transportation routes, the predicted vehicle motion response data is arranged in time sequence to generate the predicted vehicle motion response feature sequence for the corresponding candidate transportation routes.

[0054] Specifically, the moment when the LNG tanker enters the candidate transportation route is taken as the initial time node, and the vehicle motion response data at the corresponding time node is arranged according to the order in which the vehicle passes through each road segment unit along the transportation direction.

[0055] For example, after route screening, three candidate transportation routes are obtained for an LNG tanker transportation task. For one of the candidate routes, the information on changes in road slope, road curvature, and road surface condition corresponding to that route is input into a vehicle motion response prediction model. When the LNG tanker enters a continuous uphill section, the model predicts the change in longitudinal acceleration of the vehicle based on the change in road slope; when the LNG tanker enters a continuous curved section, the model predicts the change in lateral acceleration of the vehicle based on the change in road curvature; when the LNG tanker enters an uneven road surface area, the model predicts the change in vertical acceleration of the vehicle based on the change in road surface condition. Subsequently, according to the vehicle's running time sequence along the candidate transportation route, the above-mentioned predicted vehicle motion response data are arranged to form a predicted vehicle motion response feature sequence corresponding to the candidate transportation route, which is used for subsequent prediction of the storage tank dynamic response.

[0056] Step S200 in the method provided in this embodiment of the invention further includes: Based on the vehicle mass parameters, vehicle structure parameters, and vehicle driving parameters in the LNG tanker vehicle parameters, determine the vehicle dynamic characteristic parameters that affect the changes in the motion state of the LNG tanker. Based on the road slope information, road curvature information and pavement condition information in the road condition feature sequence, a road excitation feature sequence is constructed to characterize the impact of the road on LNG tank trucks. The vehicle dynamic characteristic parameters and road excitation characteristic sequences are fused to obtain the vehicle motion state input features. Based on the input features of vehicle motion state and the corresponding historical vehicle motion response data, a deep learning prediction model is trained until convergence, generating a vehicle motion response prediction model. The vehicle motion response data includes longitudinal acceleration changes, lateral acceleration changes, vertical acceleration changes, and vehicle speed changes.

[0057] Furthermore, in order to accurately predict the changes in the vehicle motion state of LNG tank trucks under different transportation routes, this embodiment constructs a vehicle motion response prediction model based on the vehicle parameters of LNG tank trucks.

[0058] Specifically, the vehicle parameter data corresponding to the LNG tanker truck is obtained, including vehicle mass parameters, vehicle structural parameters, and vehicle driving parameters.

[0059] Among them, vehicle mass parameters are used to characterize the vehicle's inertial response capability when subjected to road forces, including vehicle curb weight, LNG loading weight, and total vehicle mass; vehicle structural parameters are used to characterize the influence of vehicle structure on vehicle attitude changes and vibration transmission processes, including vehicle wheelbase, vehicle track width, vehicle center of gravity height, and suspension structural parameters; vehicle driving parameters are used to characterize the vehicle's current operating state, including vehicle speed, vehicle acceleration, and vehicle steering state.

[0060] Furthermore, based on the vehicle parameter data, the vehicle dynamic characteristic parameters that affect the changes in the motion state of the LNG tanker are determined.

[0061] Specifically, the characteristics of vehicle inertial influence are determined based on vehicle mass parameters.

[0062] Among them, the vehicle inertial influence feature is used to describe the degree of influence of changes in vehicle mass on the vehicle's motion response.

[0063] In this embodiment, the vehicle inertial influence characteristics are determined based on the ratio between the vehicle's current total mass and the reference mass under standard transportation conditions: Vehicle inertial influence characteristics = current total mass of vehicle ÷ reference mass under standard transportation conditions.

[0064] The standard transport condition reference mass is determined based on the design full load mass of the LNG tanker.

[0065] For example, if the standard full-load mass of a certain type of LNG tanker is 40t, and the total mass of the vehicle during actual transportation is 38t, then the vehicle mass influence characteristic is 0.95.

[0066] Furthermore, the influence characteristics of vehicle structure are determined based on vehicle structural parameters.

[0067] Specifically, the vehicle wheelbase, track width, center of gravity height, and suspension structural parameters are normalized to obtain vehicle structural influence characteristics under a unified scale.

[0068] Since different vehicle structural parameters have different physical dimensions, in order to avoid the impact of differences in the magnitude of different parameters on the model training results, this embodiment adopts an extreme value normalization method based on the statistical range of historical vehicle samples.

[0069] Specifically, for any vehicle structural parameter, based on the maximum and minimum values ​​of that parameter in historical LNG tanker transportation samples, the current vehicle structural parameter is converted into a dimensionless characteristic value: Vehicle structure normalization feature = (current vehicle structure parameter - historical minimum value of the parameter) ÷ (historical maximum value of the parameter - historical minimum value of the parameter).

[0070] The historical maximum and minimum values ​​were obtained by statistical analysis based on a preset number of historical operating samples of LNG tank trucks.

[0071] For example, the vehicle center of gravity height varies from 2.0m to 3.0m in historical transportation samples, and the current center of gravity height of the LNG tanker is 2.6m, so the normalized feature of the vehicle center of gravity height is 0.6.

[0072] The above method converts different vehicle structural parameters into feature data at a unified scale.

[0073] Furthermore, the vehicle's operating state characteristics are determined based on the vehicle's driving parameters. These characteristics include the vehicle's current speed, acceleration, and steering characteristics.

[0074] Vehicle acceleration characteristics are determined based on the changes in vehicle speed between consecutive sampling times: Vehicle acceleration characteristic = (Vehicle speed at current moment - Vehicle speed at previous sampling moment) ÷ Sampling time interval.

[0075] The sampling time interval was calibrated based on historical LNG tanker transportation data.

[0076] Specifically, based on vehicle speed change data and vehicle motion response data collected during historical transportation processes, the adjustment range of the sampling time interval is determined, and the sampling time interval is iterated and tested within the adjustment range according to a fixed step size.

[0077] Specifically, a vehicle motion response prediction model is trained based on vehicle motion feature data obtained at different sampling time intervals, and the validity of the corresponding sampling time interval is determined based on the error between the model prediction results and the actual vehicle motion response data.

[0078] For example, based on historical LNG tanker transportation data, the sampling time interval is determined to be within the range of 0.05 seconds to 0.2 seconds, with an adjustment step size of 0.01 seconds. The model prediction error is tested under different sampling time intervals. When the prediction error is lowest, the corresponding sampling time interval is determined as the vehicle state sampling interval. When the sampling time interval is 0.1 seconds, the corresponding prediction error is lowest, so 0.1 seconds is determined as the vehicle motion state sampling interval.

[0079] The above methods are used to obtain vehicle dynamic characteristic parameters, including vehicle inertial influence characteristics, vehicle structural influence characteristics, and vehicle operating state characteristics.

[0080] Furthermore, a road excitation feature sequence is constructed based on the road change information in the road condition feature sequence.

[0081] Specifically, the longitudinal excitation characteristics of the road are determined based on the road slope information, the lateral excitation characteristics of the road are determined based on the road curvature information, and the road vibration excitation characteristics are determined based on the road surface condition information.

[0082] Furthermore, the road excitation characteristics are fused based on the degree of influence of each road excitation factor on the vehicle motion response.

[0083] Specifically, the road excitation features are weighted and fused based on the degree of influence of road slope, road curvature, and pavement condition.

[0084] The weights of each influencing factor were determined based on historical LNG tanker transportation data.

[0085] Specifically, based on the correlation between changes in road factors and changes in vehicle motion response in historical data, the influence contribution of each road factor is calculated, and the influence contribution is normalized to obtain the corresponding weight.

[0086] The road slope influence weight is used to characterize the contribution of road slope changes to the vehicle's longitudinal motion response; the road curvature influence weight is used to characterize the contribution of road curvature changes to the vehicle's lateral motion response; and the pavement condition influence weight is used to characterize the contribution of pavement changes to the vehicle's vertical vibration response. The specific calculations are as follows: The weight of road factor influence = the contribution of the corresponding road factor influence ÷ the sum of the contribution of all road factors influence.

[0087] The contribution of the impact is determined based on the correlation between the road factor change sequence and the corresponding vehicle motion response change sequence.

[0088] For example, based on historical LNG tanker transportation data, the following contributions were obtained: road slope contribution is 0.72, road curvature contribution is 0.68, and road surface condition contribution is 0.94. After normalization, the following weights were obtained: road slope influence weight is approximately 0.31, road curvature influence weight is approximately 0.29, and road surface condition influence weight is approximately 0.40. Subsequently, the road slope influence feature, road curvature influence feature, and road surface condition influence feature were fused according to the above weights to obtain the road excitation feature sequence.

[0089] Furthermore, the vehicle dynamic characteristic parameters are fused with the road excitation characteristic sequence to obtain the vehicle motion state input features.

[0090] Specifically, vehicle dynamic characteristic parameters are used as features reflecting the LNG tanker's own response capability, and road excitation characteristic sequences are used as features reflecting the external road effect. These features are then combined to form the vehicle motion state input features.

[0091] The vehicle motion state input characteristics include: vehicle inertial influence characteristics, vehicle structural influence characteristics, vehicle operating state characteristics, road longitudinal excitation characteristics, road lateral excitation characteristics, and road vibration excitation characteristics.

[0092] Furthermore, based on the vehicle motion state input features and historical vehicle motion response data, a deep learning prediction model is trained to generate a vehicle motion response prediction model.

[0093] Specifically, since the vehicle motion response during LNG tanker transportation is affected by continuous road changes and exhibits significant time correlation, this embodiment employs a Long Short-Term Memory (LSTM) network to construct a vehicle motion response prediction model. The vehicle motion response prediction model includes: an input layer, a feature mapping layer, a temporal feature extraction layer, a response mapping layer, and an output layer.

[0094] The system consists of an input layer for receiving vehicle motion state input features; a feature mapping layer for performing nonlinear transformations on the input features to extract the correlation between vehicle dynamic features and road excitation features; a temporal feature extraction layer for learning the motion change patterns of vehicles under continuous road conditions; a response mapping layer for converting the extracted temporal features into vehicle motion response results; and an output layer for outputting predicted vehicle motion response data.

[0095] Specifically, in this embodiment, the temporal feature extraction layer is configured with two layers of long short-term memory network, with 64 hidden neural units in each layer.

[0096] The number of network layers and the number of hidden neural units are determined based on historical transport sample tests. When the model prediction error decreases by less than a preset improvement threshold after increasing the number of network layers or the number of hidden neural units, the current network structure is determined.

[0097] In this embodiment, when the prediction error improvement is less than 0.5% after the model structure is increased, a two-layer long short-term memory network structure is adopted.

[0098] The response mapping layer adopts a fully connected network structure.

[0099] The input data for the vehicle motion response prediction model includes: vehicle dynamic characteristic parameters and road excitation characteristic sequence. The output data for the vehicle motion response prediction model includes: longitudinal acceleration variation, lateral acceleration variation, and vehicle speed variation.

[0100] Furthermore, historical LNG tanker transportation data was obtained to train the vehicle motion response prediction model.

[0101] Specifically, historical vehicle parameter data is fused with corresponding road excitation data to form historical vehicle motion state input features.

[0102] The longitudinal acceleration changes, lateral acceleration changes, vertical acceleration changes, and vehicle speed changes collected in the corresponding time period were used as training labels.

[0103] During training, model parameters are adjusted based on the error between the model's predictions and the actual collected results. The model training loss is calculated using the mean squared error. Training loss = (Predicted vehicle motion response data - Actual vehicle motion response data) 2 The average value.

[0104] Furthermore, the model training process was tested based on historical LNG tanker transportation samples, the changes in model prediction error under different training rounds were statistically analyzed, and the model convergence condition was determined based on the stage when the error tends to stabilize.

[0105] Specifically, when the change in model prediction error within consecutive training rounds is lower than a preset error change threshold, it is considered that the update of model parameters has little impact on the prediction results, training is stopped and a vehicle motion response prediction model is generated.

[0106] For example, based on the test results of multiple sets of historical transportation samples, when the model training reaches 20 rounds, the prediction error enters a stable stage, and the change ratio of the prediction error during the 20 consecutive rounds of training is less than 0.5%. Therefore, the change of the prediction error for 20 consecutive rounds is determined to be less than 0.5% as the condition for model convergence.

[0107] In summary, this step constructs a vehicle motion response prediction model, fusing the dynamic characteristic parameters of LNG tank trucks with road excitation feature sequences, and trains a deep learning model using historical transportation data. This enables the prediction of longitudinal, lateral, and vertical motion responses and speed changes of vehicles based on road variations corresponding to different candidate transportation routes. By arranging the predicted vehicle motion response data chronologically according to the transportation process, a predicted vehicle motion response feature sequence is formed, providing a data foundation for subsequent tank dynamic response analysis and LNG transportation stability evaluation.

[0108] S300: Based on the structural parameters of the LNG tank truck storage tank, the initial state parameters of LNG and the predicted vehicle motion response feature sequence, predict the predicted tank dynamic response feature sequence and the predicted LNG liquid sloshing state feature sequence for each candidate path during transportation. Step S300 in the method provided in this embodiment of the invention includes: The dynamic response characteristic parameters of the LNG tanker under the influence of vehicle movement are determined based on the structural parameters of the LNG tanker. The dynamic response characteristic parameters include the structural dimensions of the tank, the installation position of the tank, and the support connection parameters of the tank. The dynamic response characteristic parameters, LNG initial state parameters, and predicted vehicle motion response characteristic sequences are input into the pre-built tank dynamic response prediction model to predict the force change state and vibration response state of the LNG tanker during transportation on each candidate transportation route. Based on the force change state and vibration response state, the predicted tank dynamic response characteristic sequence corresponding to each candidate transportation route is generated. Based on the predicted dynamic response feature sequence of the storage tank, the predicted LNG liquid sloshing state feature sequence corresponding to each candidate transportation path is obtained. The steps for constructing the tank dynamic response prediction model include: Acquire vehicle motion response characteristic sequences, LNG initial state parameters, and corresponding tank dynamic response sample data during the historical LNG tank truck transportation process. The tank dynamic response sample data includes tank force change characteristics and tank vibration response characteristics. Based on the historical LNG tanker structural parameters, the corresponding tank dynamic response characteristic parameters are determined. The tank dynamic response characteristic parameters, LNG initial state parameters, and vehicle motion response characteristic sequences are correlated to obtain the input features. Based on the input features and the corresponding dynamic response sample data of the storage tank, the deep learning prediction model is trained until the model output meets the preset training conditions, thus generating the dynamic response prediction model of the storage tank.

[0109] In this embodiment, after obtaining the predicted vehicle motion response feature sequence corresponding to each candidate transportation path, the LNG tank truck storage tank structure parameters and LNG initial state parameters are further combined to predict the dynamic response changes of the storage tank caused by the vehicle motion during transportation of different candidate transportation paths, thereby obtaining the predicted storage tank dynamic response feature sequence for the corresponding candidate transportation path.

[0110] Specifically, obtain the structural parameters of the LNG tanker and the initial state parameters of the LNG at the start of transportation.

[0111] Among them, the tank structure parameters are used to describe the structural characteristics of the LNG tanker itself, including the tank structure size parameters, tank installation location parameters, and tank support connection parameters.

[0112] The structural dimensional parameters of the storage tank include the tank length, tank diameter, tank wall thickness, and tank mass, which are used to characterize the influence of the tank structural dimensions on the transmission of vehicle motion loads. The tank installation position parameters include the tank center of gravity position, tank installation height, and the positional relationship of the tank relative to the vehicle chassis, which are used to characterize the changes in the torque on the tank during vehicle movement. The tank support connection parameters include the number of support structures, support connection stiffness, and connection damping parameters, which are used to characterize the attenuation characteristics of vehicle vibrations transmitted to the tank structure.

[0113] Furthermore, the dynamic response characteristic parameters of the storage tank are determined based on the structural parameters of the storage tank.

[0114] Specifically, feature transformation is performed on the structural dimensional parameters, installation location parameters, and support connection parameters of the storage tank to obtain dynamic response characteristic parameters that characterize the dynamic response capability of the storage tank.

[0115] Specifically, for tank structure parameters with different dimensions, this embodiment uses a normalization method based on the statistical range of historical LNG tank truck tank samples for feature transformation: Normalized feature of tank structure = (current tank structure parameters - minimum historical sample value) ÷ (maximum historical sample value - minimum historical sample value).

[0116] The historical maximum and minimum values ​​were obtained by statistically analyzing historical operating data from a preset number of LNG tank trucks. This method transforms the structural parameters of different tanks into dynamic response characteristic parameters on a unified scale.

[0117] Furthermore, the initial state parameters of the LNG are obtained.

[0118] Among them, the LNG initial state parameters are used to characterize the initial storage state of liquefied natural gas inside the tank when the LNG tanker starts transportation, including LNG thermodynamic state parameters, LNG loading state parameters, and LNG gas-liquid distribution state parameters.

[0119] Specifically, the thermodynamic state parameters of LNG include the initial temperature, initial pressure, and medium density of LNG, which are used to describe the internal thermodynamic state of LNG at the start of transportation; the loading state parameters of LNG include the loading mass, loading volume, and liquid level, which are used to describe the LNG filling state inside the storage tank; and the gas-liquid distribution state parameters of LNG include the gas phase space ratio and liquid phase ratio, which are used to describe the gas-liquid space distribution inside the storage tank.

[0120] Furthermore, the dynamic response characteristic parameters of the storage tank, the initial state parameters of LNG, and the predicted vehicle motion response characteristic sequence are input into the pre-constructed dynamic response prediction model of the storage tank.

[0121] Among them, the tank dynamic response prediction model is used to establish the mapping relationship between changes in vehicle motion state and changes in tank dynamic response. Its inputs include tank dynamic response characteristic parameters, LNG initial state parameters, and predicted vehicle motion response characteristic sequence. Its outputs include tank force change characteristics and tank vibration response characteristics.

[0122] Specifically, the stress variation characteristics of the storage tank are used to describe the changes in dynamic loads borne by the storage tank under the action of vehicle motion, including the longitudinal inertial load variation characteristics, the lateral inertial load variation characteristics, the vertical impact load variation characteristics, and the load variation characteristics at the support connection positions of the storage tank.

[0123] Among them, the longitudinal inertial load variation characteristics of the storage tank are used to reflect the dynamic action on the storage tank along the vehicle's direction of travel during the acceleration or deceleration of the LNG tanker; the lateral inertial load variation characteristics of the storage tank are used to reflect the lateral action on the storage tank during vehicle turning; the vertical impact load variation characteristics of the storage tank are used to reflect the impact generated when the vehicle passes over uneven roads; and the load variation characteristics of the storage tank support connection position are used to reflect the force changes in the connection area between the storage tank and the vehicle chassis.

[0124] The vibration response characteristics of storage tanks are used to describe the vibration state of the storage tank structure under the excitation of vehicle motion, including the characteristics of changes in storage tank vibration acceleration, changes in storage tank vibration amplitude, changes in storage tank vibration frequency, and changes in storage tank structural attitude.

[0125] Furthermore, a predictive model for the dynamic response of storage tanks is constructed. Specifically, vehicle motion response characteristic sequences, LNG initial state parameters, and corresponding sample data of the dynamic response of storage tanks during the transportation process are obtained from historical LNG tanker transportation processes.

[0126] Among them, the dynamic response sample data of the storage tank includes the actual collected characteristics of the stress change and vibration response of the storage tank.

[0127] Furthermore, based on the historical LNG tanker structural parameters, the corresponding tank dynamic response characteristic parameters are determined, and the tank dynamic response characteristic parameters, LNG initial state parameters, and vehicle motion response characteristic sequences are fused to obtain the model input features.

[0128] Specifically, the structural characteristics of the storage tank are used as inputs reflecting the dynamic response capability of the storage tank itself, the initial state parameters of LNG are used as inputs reflecting the internal medium state, and the vehicle motion response characteristic sequence is used as inputs reflecting the external excitation. The input data for the storage tank dynamic response prediction model are formed by feature splicing.

[0129] Furthermore, based on the input features and the corresponding tank dynamic response sample data, a deep learning prediction model is trained to generate a tank dynamic response prediction model. Specifically, this embodiment uses a temporal deep learning network to construct the tank dynamic response prediction model, which includes an input layer, a feature fusion layer, a temporal feature extraction layer, a response mapping layer, and an output layer.

[0130] The input layer receives the dynamic response characteristic parameters of the storage tank, the initial state parameters of LNG, and the vehicle motion response feature sequence; the feature fusion layer extracts the correlation features between different types of input parameters; the time-series feature extraction layer learns the dynamic response change law of the storage tank under continuous vehicle motion; the response mapping layer converts the extracted time-series features into the dynamic response result of the storage tank; and the output layer outputs the force change characteristics and vibration response characteristics of the storage tank.

[0131] In this embodiment, the temporal feature extraction layer adopts a long short-term memory network structure.

[0132] The number of network layers and the number of hidden neural units were calibrated based on historical LNG tanker transport samples.

[0133] Specifically, by adjusting different network structure parameters and comparing the changes in model prediction error, the current network structure is determined when the model prediction error decreases by less than a preset improvement threshold after increasing the number of network layers or hidden neural units.

[0134] For example, when the prediction error improvement is less than 0.5% after the model structure is adjusted, a two-layer long short-term memory network structure is adopted, with 64 hidden neural units in each layer.

[0135] Furthermore, during model training, the model parameters are adjusted based on the error between the predicted dynamic response of the storage tank output by the model and the actual dynamic response samples of the storage tank.

[0136] The model training loss is determined based on the mean square error between the predicted dynamic response data of the storage tank and the actual dynamic response data of the storage tank. Training loss = (Predicted tank dynamic response data - Actual tank dynamic response data) 2 The average value.

[0137] Furthermore, the convergence conditions of the model are determined based on the historical model training results.

[0138] Specifically, the loss reduction magnitude and the number of consecutive stable training rounds corresponding to the model reaching a stable prediction state in multiple historical training tasks are statistically analyzed, and the convergence judgment condition is determined based on the statistical results.

[0139] Among them, the number of consecutive stable training rounds is used to characterize the number of continuous training rounds in which the model parameters remain stable, and the loss change threshold is used to determine whether the model's predictive performance continues to improve.

[0140] For example, based on the statistical analysis of training results from multiple sets of historical LNG tanker transportation data, when the training loss decreases below a preset threshold for several consecutive training rounds during continuous model training, the model's prediction results no longer show significant improvement. Therefore, this condition is determined as the model convergence criterion. In this embodiment, when the model training loss changes by less than 0.5% during 20 consecutive training rounds, the model is considered to have reached a stable state, training is stopped, and a tank dynamic response prediction model is generated.

[0141] Furthermore, the predicted vehicle motion response feature sequence corresponding to each candidate transportation route is input into the tank dynamic response prediction model to obtain the tank force change characteristics and tank vibration response characteristics for the corresponding candidate transportation routes.

[0142] Subsequently, based on the running time sequence of LNG tank trucks along the candidate transportation routes, the predicted tank stress change characteristics and tank vibration response characteristics are arranged to form a sequence of predicted tank dynamic response characteristics for the corresponding candidate transportation routes.

[0143] For example, when an LNG tanker truck travels on a continuous uphill road, the vehicle's longitudinal motion response increases, and the longitudinal inertial load on the storage tank changes; when the vehicle travels on a continuous curved road section, the vehicle's lateral motion response increases, and the lateral force change on the storage tank increases; when the vehicle travels through an uneven road surface area, the vehicle's vertical vibration increases, and the storage tank's vibration response increases. Through the above methods, predicted dynamic response characteristic sequences of the storage tanks corresponding to different candidate transportation routes are obtained.

[0144] Step S300 in the method provided in this embodiment of the invention further includes: Based on the structural parameters of the LNG tanker, the internal structural parameters of the tank that affect the change of the LNG liquid movement state are determined. The internal structural parameters of the tank include the internal space dimensions of the tank, the internal anti-wave structure parameters of the tank, and the effective volume parameters of the tank. The initial motion state parameters of LNG liquid are determined based on the initial state parameters of LNG. The initial motion state parameters of LNG liquid include the initial liquid level parameters of LNG, the initial filling ratio parameters of LNG, and the gas-liquid spatial distribution parameters. The internal structural parameters of the storage tank, the initial motion parameters of the LNG liquid, and the predicted dynamic response characteristic sequence of the storage tank are input into the pre-constructed liquid sloshing state prediction model to predict the change in liquid surface fluctuation height, the maximum liquid surface offset distance, the liquid centroid offset, the liquid flow velocity, and the change in liquid impact pressure during the transportation of LNG liquid in each candidate transportation path, thereby generating the predicted LNG liquid sloshing state characteristic sequence corresponding to each candidate transportation path.

[0145] like Figure 2 As shown, further, after obtaining the predicted tank dynamic response feature sequence corresponding to each candidate transportation path, this embodiment combines the internal structural features of the LNG tanker tank and the initial state parameters of LNG to predict the motion state of the liquefied natural gas inside the tank after being subjected to dynamic action during vehicle operation, and obtains the predicted LNG liquid sloshing state feature sequence corresponding to each candidate transportation path.

[0146] Specifically, the internal structural parameters of the LNG tanker that affect the changes in the liquid's movement state are determined based on the structural parameters of the LNG tanker. These internal structural parameters include the internal space dimensions, wave-damping structure parameters, and effective volume parameters.

[0147] The internal space dimensional parameters of the storage tank are used to describe the spatial range of the LNG liquid's movable area, including the internal length, internal diameter, and internal cross-sectional dimensions of the storage tank; the wave-damping structure parameters are used to describe the damping effect of the internal structure of the storage tank on the liquid's movement, including the number of wave-damping plates, the installation position of the wave-damping plates, the spacing between the wave-damping plates, and the opening area; the effective volume parameters are used to describe the effective space inside the storage tank that can actually participate in the liquid's movement.

[0148] Furthermore, the parameters affecting liquid sloshing are determined based on the aforementioned internal structural parameters of the storage tank.

[0149] Specifically, firstly, based on the range of structural parameters in historical LNG tanker storage tank sample data, the internal structural parameters of each tank are normalized to obtain the corresponding structural feature values: Structural feature value = (current structural parameter - historical minimum value) ÷ (historical maximum value - historical minimum value).

[0150] Subsequently, based on the degree of influence of each structural parameter on the liquid sloshing state, the structural characteristic values ​​are fused to obtain the liquid sloshing influence parameters.

[0151] The structural influence weight is determined based on historical tank test data.

[0152] Specifically, multiple sets of liquid sloshing test data under different tank structural conditions are obtained. For any structural parameter, the correlation coefficient between the change sequence of that structural parameter and the corresponding liquid sloshing response change sequence is calculated, and the absolute value of the correlation coefficient is taken as the degree of correlation for that structural parameter. The degree of correlation is calculated as follows: Correlation degree = |Correlation coefficient between changes in structural parameters and changes in liquid sloshing response|.

[0153] Subsequently, the correlation between each structural parameter and the change in the maximum liquid surface offset distance and the change in liquid impact pressure were calculated and fused to obtain the comprehensive correlation of the corresponding structural parameters. The specific calculations are as follows: Overall correlation degree = (correlation degree of maximum liquid surface offset distance + correlation degree of liquid impact pressure) ÷ 2.

[0154] Furthermore, based on the comprehensive correlation of each structural parameter, normalization is performed to obtain the corresponding structural influence weights: Structural influence weight = Correlation degree of corresponding structural parameter ÷ Sum of correlation degrees of all structural parameters

[0155] Subsequently, the structural eigenvalues ​​are weighted and fused according to the obtained structural influence weights to obtain the liquid sloshing influence parameters: Liquid sloshing influence parameter = tank space size characteristic value × space size influence weight + wave-damping structure characteristic value × wave-damping structure influence weight + effective volume characteristic value × effective volume influence weight.

[0156] For example, based on historical tank experimental data, the correlation degrees for tank space size, wave-damping structure, and effective volume are calculated to be 0.70, 0.90, and 0.40, respectively. Therefore: the influence weight of tank space size = 0.70 ÷ (0.70 + 0.90 + 0.40) ≈ 0.35; the influence weight of wave-damping structure = 0.90 ÷ (0.70 + 0.90 + 0.40) ≈ 0.45; and the influence weight of effective volume = 0.40 ÷ (0.70 + 0.90 + 0.40) ≈ 0.20. Subsequently, the above weights are weighted and fused with the corresponding normalized structural eigenvalues ​​to obtain liquid sloshing influence parameters, which are used to characterize the changes in the LNG liquid motion state under different tank structure conditions.

[0157] Furthermore, the initial motion state parameters of the LNG liquid are determined based on the initial state parameters of the LNG.

[0158] The initial state parameters of LNG include initial temperature, initial pressure, loading mass, initial liquid level, and gas-liquid space ratio; the initial motion state parameters of LNG liquid include initial liquid level parameters, initial filling ratio parameters, and gas-liquid space distribution parameters.

[0159] The initial LNG loading ratio is determined based on the ratio of the initial LNG loading volume to the effective volume of the storage tank. LNG initial filling ratio parameter = LNG initial loading volume ÷ effective tank volume.

[0160] For example, when the effective volume of the storage tank is 50 cubic meters and the initial loading volume is 42 cubic meters, the initial LNG filling ratio parameter is 0.84.

[0161] Furthermore, the liquid sloshing influence parameters, LNG liquid initial motion state parameters, and predicted tank dynamic response characteristic sequences are input into the pre-constructed liquid sloshing state prediction model.

[0162] The liquid sloshing state prediction model is used to establish the mapping relationship between changes in the dynamic response of the storage tank and changes in the motion state of the LNG liquid. The output of the liquid sloshing state prediction model includes the change in liquid surface sway height, the maximum liquid surface offset distance, the liquid centroid offset, the liquid flow velocity, and the change in liquid impact pressure.

[0163] Among them, the change in liquid surface undulation height is used to characterize the degree of up-and-down undulation of the liquid surface; the maximum displacement distance of the liquid surface is used to characterize the maximum distance the free liquid surface moves relative to the initial stable position; the displacement of the liquid centroid is used to characterize the change in the center of liquid mass distribution; the liquid flow velocity is used to characterize the intensity of liquid motion; and the change in liquid impact pressure is used to characterize the dynamic pressure change generated by the liquid motion on the inner wall of the storage tank.

[0164] Furthermore, the model output results are arranged according to the transportation time sequence to generate a predicted LNG liquid sloshing state feature sequence for the corresponding candidate transportation paths.

[0165] Furthermore, a liquid sloshing state prediction model is constructed. Specifically, the dynamic response characteristic sequence of the storage tank, the initial state parameters of LNG, and the sample data of liquid sloshing state during historical LNG tanker transportation processes are obtained.

[0166] The sample data for liquid sloshing includes changes in liquid surface undulation height, maximum liquid surface offset distance, liquid centroid offset, liquid flow velocity, and changes in liquid impact pressure.

[0167] Among them, the change in liquid level fluctuation height and the maximum offset distance of the liquid level are obtained from liquid level detection data; the change in liquid impact pressure is obtained from tank inner wall pressure detection data; the liquid centroid offset is determined based on the change in liquid mass distribution; and the liquid flow velocity is obtained based on liquid motion detection data or experimental test data.

[0168] Furthermore, based on the historical tank structural parameters, the corresponding liquid sloshing influence parameters are determined, and the liquid sloshing influence parameters, LNG initial state parameters, and tank dynamic response characteristic sequences are fused to form the input features of the liquid sloshing state prediction model.

[0169] Specifically, the liquid sloshing state prediction model is constructed using a temporal prediction network based on a long short-term memory network, including an input layer, a feature fusion layer, a temporal feature extraction layer, a state mapping layer, and an output layer.

[0170] The input layer receives the model input features; the feature fusion layer extracts the correlation features between the tank structure, the initial state of the liquid, and the dynamic excitation; the temporal feature extraction layer learns the liquid motion change pattern during transportation; the state mapping layer converts the temporal features into liquid sloshing state parameters; and the output layer outputs the liquid sloshing state features.

[0171] Specifically, in this embodiment, the input dimension of the liquid sloshing state prediction model is determined according to the number of input features of the model, and the output dimension is set to 5 dimensions, corresponding to the above five types of liquid sloshing state parameters.

[0172] The temporal feature extraction layer adopts a two-layer long short-term memory network structure, with 64 hidden neural units in each layer; the feature fusion layer adopts a fully connected mapping structure to transform the input features into a unified feature space; and the state mapping layer adopts a fully connected layer to map the temporal hidden state to the liquid sloshing state prediction result.

[0173] Specifically, the parameters of liquid sloshing influence, LNG initial state parameters, and tank dynamic response feature sequence obtained during historical transportation are used as model inputs, and the changes in liquid surface fluctuation height, maximum liquid surface offset distance, liquid centroid offset, liquid flow velocity, and liquid impact pressure collected in the corresponding time period are used as model training labels.

[0174] During training, the network parameters are adjusted based on the deviation between the model's predicted output and actual liquid sloshing state samples. The model training loss is calculated using the mean squared error function. Training loss = (Predicted liquid sloshing state data - Actual liquid sloshing state data) 2 The average value.

[0175] Furthermore, the model structure parameters and training convergence conditions are determined based on the test results of historical transportation samples.

[0176] Specifically, the number of network layers and the number of hidden neurons in the temporal feature extraction layer are adjusted respectively. When the model prediction error decreases by less than the preset improvement threshold after increasing the number of network layers or the number of hidden neurons, the increase in network size is stopped and the current network structure is determined.

[0177] For example, during the historical LNG tanker transportation sample test, when the number of network layers increased to three or the number of hidden neural units increased to 128, the model prediction error decreased by less than 0.5%. Therefore, it was determined to adopt a two-layer long short-term memory network structure with 64 hidden neural units in each layer.

[0178] Furthermore, the convergence conditions are determined based on the changes in model loss during the historical training process.

[0179] Specifically, when the model loss change rate is less than 0.5% during 20 consecutive training rounds, the model parameter adjustment is considered to have stabilized, training is stopped, and a liquid sloshing state prediction model is generated.

[0180] The network structure parameters, number of training rounds, and convergence threshold were all determined based on the test results of historical LNG tanker transportation samples to ensure that the liquid sloshing state prediction model can adapt to the dynamic response changes in the actual transportation process.

[0181] For example, in an LNG tanker transportation task, the dynamic response feature sequence of the storage tank corresponding to the candidate route is obtained. Simultaneously, the internal space dimensions of the storage tank, the structure of the baffle plate, and the effective volume parameters are obtained, and combined with the LNG liquid level and filling ratio at the start of transportation to form the initial liquid motion state parameters. These parameters are then input into a liquid sloshing state prediction model. When the vehicle passes through a continuous curved road section, the model predicts an increase in the maximum liquid surface offset distance based on changes in lateral dynamic response; when the vehicle passes through an uneven road surface area, the model predicts changes in liquid flow velocity and impact pressure based on changes in the tank vibration response. Finally, the predicted LNG liquid sloshing state feature sequence for the corresponding candidate transportation route is obtained, providing a data foundation for subsequent comprehensive evaluation of LNG gas-liquid stability.

[0182] In summary, this step establishes a dynamic response prediction model for the LNG tanker and a liquid sloshing state prediction model by combining the structural parameters of the LNG tanker, the initial state parameters of the LNG, and the predicted vehicle motion response characteristic sequence. This enables the step-by-step prediction of the tank's dynamic response and the LNG liquid motion state under transportation conditions. The influence of the structure on the dynamic load transfer is characterized by the tank's dynamic response characteristic parameters. Combined with the vehicle motion excitation, the predicted stress changes, vibration response, and liquid sloshing state of the tank are predicted. This yields the predicted tank dynamic response characteristic sequence and the predicted LNG liquid sloshing state characteristic sequence for each candidate transportation route, providing a data foundation for subsequent transportation route stability evaluation.

[0183] S400: Based on the structural parameters of LNG tank truck storage tanks, a tank heat exchange prediction model is established. Based on the initial state parameters of LNG and the road environment characteristic sequence, the predicted tank heat exchange state characteristic sequence corresponding to each candidate path is obtained. Step S400 in the method provided in this embodiment of the invention includes: The parameters affecting the heat exchange of the LNG tank truck storage tank are determined based on the structural parameters of the storage tank. These parameters include the outer shell structural parameters, insulation structural parameters, and surface heat exchange parameters of the storage tank. The heat exchange influence parameters of the storage tank, the initial state parameters of LNG, and the road environment feature sequence are input into the pre-constructed heat exchange prediction model of the storage tank. Based on the changes in ambient temperature, solar radiation and humidity in the road environment feature sequence, the heat exchange state of the LNG tanker storage tank during transportation on each candidate transportation route is predicted, and the predicted heat exchange state feature sequence of the storage tank corresponding to each candidate transportation route is generated.

[0184] In this embodiment, after obtaining the road environment feature sequence corresponding to each candidate transportation route, the heat exchange process between the tank and the external environment during the transportation of LNG tank trucks is predicted by combining the structural parameters of the LNG tank truck storage tank and the initial state parameters of LNG, thereby obtaining the predicted heat exchange state feature sequence of the storage tank corresponding to each candidate transportation route.

[0185] Specifically, the structural parameters of the LNG tank truck storage tank are obtained, and the parameters affecting the heat exchange process of the storage tank are determined based on the structural parameters.

[0186] Among them, the heat exchange influence parameters of the storage tank are used to characterize the influence of the storage tank's own structural characteristics on the heat transfer process of the external environment, including the tank shell structural parameters, the storage tank insulation structural parameters, and the storage tank surface heat exchange parameters.

[0187] Specifically, the structural parameters of the tank shell are used to describe the ability of the tank's outer shell to impede the heat transfer process, including the type of tank shell material, the thickness of the tank shell, and the thermal conductivity parameters of the tank shell.

[0188] The insulation structure parameters of a storage tank are used to describe the tank's ability to insulate against external heat input, including the thickness of the insulation layer, the thermal conductivity of the insulation material, and the structural form of the insulation layer.

[0189] Tank surface heat transfer parameters are used to describe the heat exchange capacity between the outer surface of the tank and the external environment, including the outer surface area of ​​the tank, the external air flow state, and the convective heat transfer capacity of the tank surface.

[0190] Furthermore, the influence characteristics of the heat exchange structure of the storage tank are determined based on the aforementioned heat exchange influence parameters of the storage tank.

[0191] Specifically, since the tank shell structure parameters, tank insulation structure parameters, and tank surface heat transfer parameters have different physical dimensions, this embodiment determines the statistical range corresponding to each parameter based on historical LNG tank truck thermal performance test data, and uses extreme value normalization for feature transformation. The specific calculation is as follows: The impact characteristics of the storage tank structure = (current storage tank structure parameters - historical minimum value of corresponding structure parameters) ÷ (historical maximum value of corresponding structure parameters - historical minimum value of corresponding structure parameters).

[0192] The historical maximum and minimum values ​​of the corresponding structural parameters were obtained based on statistical analysis of historical LNG tank truck heat exchange test samples. This method transforms different structural parameters of the tank into heat exchange impact characteristics under a unified scale.

[0193] Furthermore, based on the degree of influence of different tank structural parameters on the changes in the internal thermal state of the tank, the heat exchange influence characteristics of each tank are fused to obtain the heat exchange influence parameters of the tank.

[0194] Specifically, internal temperature change data of LNG tank trucks during transportation is collected under different structural parameters. For any structural factor, the change sequence of that structural factor and the corresponding internal temperature change sequence of the tank are obtained, and the correlation coefficient between the two is calculated. The absolute value of the correlation coefficient is taken as the degree of correlation corresponding to that structural factor. The degree of correlation is calculated as follows: Correlation degree = |Correlation coefficient between changes in structural factors and changes in internal temperature of the storage tank|.

[0195] Subsequently, normalization was performed based on the correlation degree of each structural factor to obtain the influence weight of each structural factor: Structural influence weight = Correlation degree of corresponding structural factors ÷ Sum of correlation degrees of all structural factors.

[0196] Furthermore, based on the obtained structural influence weights, the structural characteristic values ​​corresponding to each structural factor are weighted and fused to obtain the heat exchange influence parameters of the storage tank.

[0197] For example, historical LNG tanker heat exchange test data were collected, and the correlation between the tank shell structure, insulation structure, surface heat exchange state and internal temperature change was calculated to be 0.30, 0.60, and 0.30, respectively. Then: the influence weight of the tank shell structure = 0.30 ÷ (0.30 + 0.60 + 0.30) = 0.25; the influence weight of the tank insulation structure = 0.60 ÷ (0.30 + 0.60 + 0.30) = 0.50; the influence weight of the surface heat exchange state = 0.30 ÷ (0.30 + 0.60 + 0.30) = 0.25.

[0198] Furthermore, the initial state parameters of the LNG are obtained.

[0199] The initial state parameters of LNG are used to describe the initial storage state of liquefied natural gas inside the tank when the LNG tanker begins transportation. These parameters include the initial temperature, initial pressure, initial loading mass, initial liquid level, and gas-liquid space ratio. Specifically, the initial thermal state characteristics of LNG are determined based on these initial state parameters.

[0200] Among them, the initial thermal state characteristics of LNG are used to characterize the impact of the thermal state of LNG itself at the start of transportation on the subsequent heat exchange process.

[0201] Specifically, the initial filling state is determined based on the initial LNG liquid level parameters and the effective volume of the storage tank, and the initial thermal state is determined based on the initial LNG temperature parameters and the initial pressure parameters.

[0202] For example, if the effective volume of an LNG tanker is 50 cubic meters and the LNG loading volume is 42 cubic meters at the start of transportation, then the initial filling ratio is 0.84; when the initial temperature and pressure of the LNG change, the driving force for heat exchange between the LNG inside the tank and the external environment changes accordingly.

[0203] Furthermore, the heat exchange influence parameters of the storage tank, the initial state parameters of LNG, and the road environment characteristic sequence are input into the pre-constructed heat exchange prediction model of the storage tank to predict the heat exchange state of the storage tank corresponding to different candidate transportation routes.

[0204] Among them, the road environment feature sequence is a time series data formed according to the arrangement order of each road segment unit in the candidate transportation route, including changes in ambient temperature, solar radiation, humidity, wind speed, and rainfall status.

[0205] Specifically, the tank heat exchange prediction model determines the tank's own heat transfer capacity based on the tank heat exchange influence parameters, determines the initial thermal state based on the LNG initial state parameters, determines the changes in the external thermal environment during transportation based on the road environmental characteristic sequence, and establishes the dynamic change relationship corresponding to the heat exchange process inside and outside the tank.

[0206] The predicted results of the heat exchange status of the storage tank include the characteristics of heat flow change in the storage tank, the characteristics of internal temperature change in the storage tank, the characteristics of temperature difference change between inside and outside the storage tank, and the characteristics of LNG thermal status change.

[0207] Specifically, the heat flow variation characteristics of the storage tank are used to describe the change in heat transferred from the environment to the storage tank per unit time; the internal temperature variation characteristics of the storage tank are used to describe the trend of LNG storage temperature change with transportation time; the temperature difference variation characteristics inside and outside the storage tank are used to describe the change in the driving force of heat exchange inside and outside the storage tank; and the thermal state variation characteristics of LNG are used to describe the temperature and pressure variation trends of LNG during transportation.

[0208] Furthermore, based on the time sequence corresponding to the candidate transportation paths, the predicted heat exchange states of the storage tanks are arranged to generate a feature sequence of predicted heat exchange states of the storage tanks for the corresponding candidate transportation paths.

[0209] Specifically, the time when the LNG tanker enters the candidate transportation route is taken as the starting time node, and the predicted results of the heat exchange status of the storage tank at the corresponding time node are arranged according to the order in which the vehicle passes through each road segment unit.

[0210] For example, an LNG tanker truck transports LNG from a liquefied natural gas production base to a city receiving terminal. After route screening, three candidate transport routes are obtained. For one of these candidate routes, the initial section passes through a high-temperature region, with the road environmental characteristic sequence showing an ambient temperature of 32°C and high solar radiation intensity. The middle section enters a mountainous area where the ambient temperature decreases to 20°C, while humidity increases. The final section enters an urban area where the ambient temperature remains stable. The aforementioned road environmental characteristic sequence, LNG initial state parameters, and tank heat exchange impact parameters are input into a tank heat exchange prediction model. The prediction results show that in high-temperature, high-radiation sections, the heat flow input to the tank increases, and the rate of internal temperature change increases; in low-temperature regions, the heat exchange rate of the tank decreases, and the internal temperature change tends to be gradual. Subsequently, the above prediction results are arranged according to the vehicle's transport time sequence to form the predicted tank heat exchange state characteristic sequence corresponding to this candidate transport route, which is used for subsequent comprehensive evaluation of LNG gas-liquid stability.

[0211] Step S400 in the method provided in this embodiment of the invention further includes: The project acquires the road environment feature sequence, LNG initial state parameters, and corresponding tank heat exchange state sample data during the historical LNG tank truck transportation process. The tank heat exchange state sample data includes the change in tank outer wall temperature, the change in temperature difference between inside and outside the tank, and the change in tank heat flux density. Based on the historical LNG tanker structural parameters, the corresponding tank heat exchange impact parameters are determined. The tank heat exchange impact parameters, LNG initial state parameters, and road environment feature sequences are correlated to obtain the model input features. Based on the model input features and the corresponding sample data of the heat exchange status of the storage tank, the deep learning prediction model is iteratively trained to generate the heat exchange prediction model of the storage tank.

[0212] In this embodiment, after obtaining the road environment feature sequence corresponding to each candidate transportation route, the heat exchange state between the tank and the external environment during transportation is predicted by combining the structural parameters of the LNG tank truck and the initial state parameters of LNG, thereby obtaining the predicted heat exchange state feature sequence of the tank corresponding to each candidate transportation route.

[0213] Specifically, the parameters affecting the heat exchange of the LNG tank truck storage tank are determined based on the structural parameters of the LNG tank truck storage tank.

[0214] Among them, the heat exchange influence parameters of the storage tank are used to characterize the influence of the structural characteristics of the storage tank on the heat transfer process, including the structural parameters of the storage tank shell, the structural parameters of the storage tank insulation, and the heat transfer parameters of the storage tank surface.

[0215] Specifically, the tank shell structure parameters include the tank shell thickness and the thermal conductivity of the material; the tank insulation structure parameters include the insulation layer thickness and the thermal conductivity of the insulation material; and the tank surface heat transfer parameters include the tank outer surface area and the surface convection heat transfer characteristics.

[0216] Since the above parameters have different physical dimensions, this embodiment normalizes each structural parameter to obtain structural influence characteristics at a unified scale. The specific calculations are as follows: Structural influence characteristics = (current structural parameters - historical sample minimum value) ÷ (historical sample maximum value - historical sample minimum value).

[0217] The historical maximum and minimum values ​​were obtained by statistical analysis based on historical LNG tanker storage tank structure data.

[0218] Furthermore, based on the degree of influence of different structural parameters on the heat exchange state of the storage tank, the influence characteristics of each structure are fused to obtain the heat exchange influence parameters of the storage tank.

[0219] Specifically, multiple sets of historical LNG tanker transportation data were collected to obtain data on changes in the tank's heat exchange state under different tank structural parameters. For any given structural parameter, the correlation coefficient between the sequence of changes in that structural parameter and the sequence of changes in the tank's heat flux density was calculated, and the absolute value of the correlation coefficient was used as the contribution of the corresponding structural parameter. The contribution was calculated as follows: Contribution of influence = Correlation coefficient between changes in structural parameters and changes in heat flux density of the storage tank.

[0220] Subsequently, the influence contribution of each structural parameter is normalized to obtain the structural influence weights: Structural influence weight = corresponding structural influence contribution ÷ sum of all structural influence contributions.

[0221] Furthermore, the normalized eigenvalues ​​corresponding to each structural parameter are weighted and fused according to the obtained structural influence weights to obtain the heat exchange influence parameters of the storage tank.

[0222] For example, based on historical LNG tanker heat exchange test data, the contribution values ​​of the tank shell structure, insulation structure, and surface heat exchange structure are calculated to be 0.4, 0.8, and 0.4, respectively. Therefore: the influence weight of the tank shell structure = 0.4 ÷ (0.4 + 0.8 + 0.4) = 0.25; the influence weight of the insulation structure = 0.8 ÷ (0.4 + 0.8 + 0.4) = 0.50; and the influence weight of the surface heat exchange structure = 0.4 ÷ (0.4 + 0.8 + 0.4) = 0.25.

[0223] Subsequently, the structural feature values ​​are fused according to the above weights to obtain structural influence parameters for predicting the heat exchange state of the storage tank.

[0224] Furthermore, the heat exchange influence parameters of the storage tank, the initial state parameters of LNG, and the road environment characteristic sequence are input into the pre-constructed heat exchange prediction model of the storage tank.

[0225] Among them, the tank heat exchange prediction model is used to establish the mapping relationship between the tank structural characteristics, initial LNG state, and changes in the transportation environment and changes in the tank heat exchange state.

[0226] Specifically, the inputs to the tank heat exchange prediction model include tank heat exchange influence parameters, LNG initial state parameters, and road environment characteristic sequence; the outputs include changes in tank outer wall temperature, changes in the temperature difference between inside and outside the tank, and changes in tank heat flux density.

[0227] Furthermore, based on changes in ambient temperature, solar radiation, and humidity in the road environment characteristic sequence, the heat exchange status of the storage tank during transportation is predicted.

[0228] Specifically, the model analyzes the road environment characteristics at each time point according to the time sequence corresponding to the candidate transportation routes, and predicts the changes in the heat exchange state of the storage tank at the corresponding time point by combining the heat exchange influence parameters of the storage tank and the initial state parameters of LNG. Among them, the changes in the heat exchange state of the storage tank include: the change in the temperature of the outer wall of the storage tank, which is used to characterize the change in the surface temperature of the storage tank caused by the external environment; the change in the temperature difference between the inside and outside of the storage tank, which is used to characterize the change in the driving force of heat exchange between the LNG inside the storage tank and the external environment; and the change in the heat flux density of the storage tank, which is used to characterize the change in heat transfer per unit area.

[0229] Furthermore, based on the predicted heat exchange status data of the storage tanks, the data is arranged in the order of the LNG tank trucks' running time along the candidate transportation routes to generate a predicted heat exchange status feature sequence for the corresponding candidate transportation routes.

[0230] Furthermore, the tank heat exchange prediction model is constructed in the following way: Acquire the road environment characteristic sequence, LNG initial state parameters, and corresponding sample data of tank heat exchange state during the historical LNG tanker transportation process.

[0231] The sample data on the heat exchange status of the storage tank includes the change in the temperature of the outer wall of the storage tank, the change in the temperature difference between the inside and outside of the storage tank, and the change in the heat flux density of the storage tank.

[0232] Subsequently, based on the historical LNG tanker storage tank structural parameters, the corresponding tank heat exchange impact parameters were determined, and the tank heat exchange impact parameters, LNG initial state parameters, and road environment feature sequences were fused to obtain the model input features.

[0233] Furthermore, the deep learning prediction model is trained based on the model input features and the corresponding sample data of the tank heat exchange status.

[0234] This embodiment uses a temporal deep learning network to construct a tank heat exchange prediction model. The model includes an input layer, a feature fusion layer, a temporal feature extraction layer, and an output layer.

[0235] The input layer receives the model input features; the feature fusion layer extracts the correlation between the tank structure, the initial state of LNG, and environmental changes; the time-series feature extraction layer learns the heat exchange change pattern caused by environmental changes during transportation; and the output layer outputs the predicted results of the tank's heat exchange state.

[0236] Specifically, the temporal feature extraction layer adopts a long short-term memory network structure, and the number of network layers and hidden units are determined based on the test results of historical transportation samples.

[0237] For example, when the model prediction error decreases by less than 0.5% after increasing the number of network layers, it is determined that the current network structure meets the prediction requirements. In this embodiment, a two-layer long short-term memory network structure is adopted, with 64 hidden units in each layer.

[0238] Furthermore, the model parameters are adjusted based on the error between the model prediction results and the actual heat exchange state samples of the storage tank.

[0239] The model training loss is determined based on the mean square error between the predicted and actual values.

[0240] When the model prediction error changes by less than 0.5% during 20 consecutive training rounds, the model is considered to have reached convergence, training is stopped, and the tank heat exchange prediction model is obtained.

[0241] For example, during the transportation of an LNG tanker, the changes in ambient temperature, solar radiation, and humidity corresponding to a candidate transportation route are input into the tank heat exchange prediction model. At the same time, combined with the tank structural parameters and the initial state parameters of LNG, the model outputs the results of the changes in the outer wall temperature, the internal and external temperature difference, and the heat flux density of the tank during transportation, and forms a predicted tank heat exchange state feature sequence for the corresponding candidate transportation route.

[0242] In summary, this step establishes a tank heat exchange prediction model by combining LNG tanker structural parameters, LNG initial state parameters, and road environment characteristic sequences corresponding to candidate transportation routes. This enables dynamic prediction of the impact of transportation environment changes on the tank heat exchange state. The tank heat exchange impact parameters characterize the influence of the tank structure on the heat transfer process. Combined with external factors such as ambient temperature, solar radiation, and humidity, the model predicts changes in the tank's outer wall temperature, internal and external temperature difference, and heat flux density during transportation. This forms a predicted tank heat exchange state characteristic sequence for each candidate transportation route, providing a data foundation for subsequent LNG gas-liquid stability evaluation and comprehensive transportation route selection.

[0243] S500: Based on the predicted dynamic response characteristic sequence of the storage tank, the predicted LNG liquid sloshing state characteristic sequence, and the predicted heat exchange state characteristic sequence of the storage tank, a comprehensive evaluation of the LNG gas-liquid stability and storage tank dynamic stability is conducted on each candidate transportation route, and the candidate transportation route corresponding to the maximum transportation stability index is selected as the target transportation route.

[0244] Step S500 in the method provided in this embodiment of the invention includes: Obtain the dynamic response safety constraints corresponding to the LNG tanker storage tank. The dynamic response safety constraints include the tank force constraints and the tank vibration constraints. Based on the dynamic response safety constraints, the constraint deviation is calculated for the predicted tank dynamic response characteristic sequence corresponding to each candidate transportation path, and the tank dynamic response deviation data corresponding to each time node is obtained. Based on the tank dynamic response deviation data, the cumulative deviation of the tank dynamic response from the safety constraints during transportation is calculated to obtain the tank dynamic deviation evaluation parameters corresponding to each candidate transportation path. Obtain the liquid sloshing safety constraints corresponding to the LNG tank truck storage tank. The liquid sloshing safety constraints include liquid surface fluctuation constraints and liquid impact pressure constraints. Based on the liquid sloshing safety constraints, the constraint deviation is calculated for the predicted LNG liquid sloshing state characteristic sequence corresponding to each candidate transportation path, and the liquid sloshing deviation data corresponding to each time node is obtained. Based on the liquid sloshing deviation data, the cumulative deviation of the liquid sloshing state from the safety constraints during transportation is calculated to obtain the liquid sloshing deviation evaluation parameters corresponding to each candidate transportation path. The dynamic stability index of the storage tank corresponding to each candidate transportation path is obtained by inverse transformation based on the evaluation parameters of the storage tank dynamic deviation and the evaluation parameters of the liquid sloshing deviation, and by fusion calculation of the transformation results. Based on the predicted heat exchange state characteristic sequence of the storage tank, the LNG gas-liquid stability of each candidate transportation route is evaluated, and the LNG gas-liquid stability index corresponding to each candidate transportation route is obtained. The transportation stability index corresponding to each candidate transportation route is obtained by weighted evaluation based on the tank dynamic stability index and the LNG gas-liquid stability index.

[0245] like Figure 3 As shown in this embodiment, after obtaining the predicted tank dynamic response characteristic sequence and the predicted LNG liquid sloshing state characteristic sequence corresponding to each candidate transportation route, the operational stability of the LNG tank truck storage tank under different candidate transportation routes is further evaluated based on the tank dynamic safety constraints and the liquid sloshing safety constraints. Combined with the evaluation results of the tank heat exchange state, the transportation stability index corresponding to each candidate transportation route is obtained.

[0246] Specifically, obtain the dynamic response safety constraints corresponding to the LNG tank truck storage tank.

[0247] Among them, the dynamic response safety constraint condition is used to characterize the safe response range that the storage tank can withstand under the action of vehicle motion, including the storage tank force constraint condition and the storage tank vibration constraint condition.

[0248] The force constraints on the storage tank are used to limit the dynamic load changes on the storage tank during vehicle movement, including the longitudinal force range, the lateral force range, and the force range at the support connection positions.

[0249] Vibration constraints on storage tanks are used to limit the vibration response of storage tanks after being excited by roads. These constraints include longitudinal vibration acceleration constraints, lateral vibration acceleration constraints, and vertical vibration acceleration constraints.

[0250] The safety constraints are determined based on the LNG tanker design parameters, structural strength test data, and historical transportation safety data.

[0251] Specifically, multiple sets of dynamic response data of storage tanks under normal transportation conditions are collected, and combined with the maximum allowable response range obtained from the structural safety test of the storage tank, the safety boundary of the corresponding response parameter is used as the dynamic response safety constraint condition.

[0252] For example, based on historical transportation test data of a certain type of LNG tank truck storage tank, the safe range of the tank vibration acceleration is determined to be no more than the design allowable vibration acceleration, and the safe range of the tank force is no more than the maximum allowable force value of the tank support structure.

[0253] Furthermore, based on the dynamic response safety constraints, constraint deviation calculations are performed on the predicted tank dynamic response characteristic sequences corresponding to each candidate transportation path to obtain tank dynamic response deviation data for each time node.

[0254] Specifically, the stress response value and vibration response value of the tank at each time point in the predicted dynamic response characteristic sequence of the tank are compared with the corresponding safety constraints.

[0255] Specifically, when the predicted response value does not exceed the safety constraint range, the dynamic response deviation at the corresponding time node is set to zero; when the predicted response value exceeds the safety constraint range, the corresponding dynamic response deviation is calculated based on the degree of exceedance. The specific calculation is as follows: Tank dynamic response deviation = (predicted dynamic response value - safety constraint boundary value) ÷ safety constraint boundary value.

[0256] Among them, the safety constraint boundary values ​​are determined based on the dynamic safety constraint conditions of the storage tank.

[0257] Furthermore, based on the tank dynamic response deviation data corresponding to each time point, the cumulative deviation of the tank dynamic response from the safety constraints during the entire transportation process is calculated.

[0258] Specifically, the dynamic response deviations at each time point are cumulatively calculated according to the LNG tanker transportation time sequence: The evaluation parameter for dynamic deviation of the storage tank is calculated as follows: Cumulative value of dynamic response deviation of the storage tank at each time point ÷ Number of transportation time points.

[0259] The above methods are used to obtain tank dynamic deviation evaluation parameters that can reflect the degree of accumulation of tank dynamic risks caused by candidate transportation routes.

[0260] Furthermore, obtain the liquid sloshing safety constraints corresponding to the LNG tanker storage tank.

[0261] Among them, the liquid sloshing safety constraint condition is used to characterize the safe range of LNG liquid movement state during transportation, including liquid surface fluctuation constraint condition and liquid impact pressure constraint condition.

[0262] Specifically, the liquid level fluctuation constraint is used to limit the maximum fluctuation range of the liquid level relative to the initial stable position; the liquid impact pressure constraint is used to limit the dynamic impact pressure generated on the inner wall of the storage tank during the liquid movement.

[0263] The safety constraints for liquid sloshing are determined based on the tank design parameters, internal anti-surge structure parameters, and historical transportation test data.

[0264] Furthermore, based on the liquid sloshing safety constraints, constraint deviations are calculated for the predicted LNG liquid sloshing state characteristic sequences corresponding to each candidate transportation path.

[0265] Specifically, the changes in liquid surface fluctuation height and liquid impact pressure in the predicted LNG liquid sloshing state characteristic sequence are compared with the corresponding safety constraint boundaries.

[0266] When the predicted liquid sloshing state does not exceed the safety constraints, the deviation at the corresponding time point is set to zero; when it exceeds the safety constraints, the liquid sloshing deviation data is calculated based on the degree of exceedance. The specific calculation is as follows: Liquid sloshing deviation = (predicted liquid sloshing state value - safety constraint boundary value) ÷ safety constraint boundary value.

[0267] Furthermore, based on the cumulative calculation of liquid sloshing deviation data at various time points during transportation, liquid sloshing deviation evaluation parameters are obtained: Liquid sloshing deviation evaluation parameter = cumulative value of liquid sloshing deviation at each time point ÷ number of transportation time points.

[0268] The above method is used to obtain the cumulative risk of LNG liquid sloshing under different candidate transportation routes.

[0269] Furthermore, the dynamic stability evaluation results of the storage tank are obtained by inversely converting the tank dynamic deviation evaluation parameters and the liquid sloshing deviation evaluation parameters.

[0270] Specifically, since a larger deviation evaluation parameter indicates worse stability, this embodiment uses a reverse conversion method to convert the risk evaluation parameter into a stability parameter. The specific calculation is as follows: Stability parameter = 1 ÷ (1 + deviation evaluation parameter).

[0271] Subsequently, the dynamic stability evaluation results of the storage tank and the liquid sloshing stability evaluation results are fused together to obtain the dynamic stability index of the storage tank. The specific calculation is as follows: Tank dynamic stability index = Tank dynamic stability parameter × first weight + Liquid sloshing stability parameter × second weight.

[0272] The first and second weights are determined based on the degree of correlation between different stability factors and transportation anomaly risks in historical LNG tanker transportation samples.

[0273] Specifically, multiple sets of historical LNG tanker transportation data were obtained, and the correlation between changes in tank dynamic stability parameters, changes in liquid sloshing stability parameters, and changes in transportation anomaly risk indicators was statistically analyzed. The absolute value of the corresponding correlation coefficients was used as the contribution factor. The contribution factor was calculated as follows: Impact contribution = |Correlation coefficient between changes in stability parameters and changes in transportation anomaly risk indicators|.

[0274] Subsequently, the influence contribution of each stability parameter is normalized to obtain the corresponding weights: First weight = Contribution of tank dynamic stability impact ÷ (Contribution of tank dynamic stability impact + Contribution of liquid sloshing stability impact); Second weight = Contribution of liquid sloshing stability to stability ÷ (Contribution of tank dynamic stability to stability + Contribution of liquid sloshing stability to stability).

[0275] For example, historical LNG tanker transportation sample data is obtained. The absolute value of the correlation coefficient between the change in the tank dynamic stability parameter and the change in the transportation anomaly risk index is calculated to be 0.55, and the absolute value of the correlation coefficient between the change in the liquid sloshing stability parameter and the change in the transportation anomaly risk index is 0.45. Based on the above influence contribution, normalization calculation is performed to obtain: First weight = 0.55 ÷ (0.55 + 0.45) = 0.55; Second weight = 0.45 ÷ (0.55 + 0.45) = 0.45.

[0276] Furthermore, the transportation stability index corresponding to each candidate transportation route is obtained by weighted fusion of the storage tank dynamic stability index and the LNG gas-liquid stability index.

[0277] Specifically, the transportation stability index is used to comprehensively evaluate the dynamic stability of storage tanks and the stability of LNG transportation status under different candidate transportation routes. The transportation stability index is calculated as follows: Transportation stability index = Tank dynamic stability index × Dynamic stability weight + LNG gas-liquid stability index × Gas-liquid stability weight.

[0278] Among them, the dynamic stability weight and the gas-liquid stability weight are determined based on the impact contribution of dynamic risk events and gas-liquid state abnormal events in historical transportation data.

[0279] Specifically, multiple sets of historical LNG tanker transportation data were obtained, and the correlation between changes in tank dynamic stability parameters, changes in LNG gas-liquid stability parameters, and changes in transportation anomaly risk indicators were calculated. The absolute value of the corresponding correlation coefficients was used as the contribution of influence. The contribution of influence was calculated as follows: Impact contribution = |Correlation coefficient between changes in stability parameters and changes in transportation anomaly risk indicators|.

[0280] Subsequently, the weights were obtained by normalizing the impact contribution of each stability factor: Dynamic stability weight = contribution of storage tank dynamic stability impact ÷ (contribution of storage tank dynamic stability impact + contribution of LNG gas-liquid stability impact); Gas-liquid stability weight = LNG gas-liquid stability impact contribution ÷ (tank dynamic stability impact contribution + LNG gas-liquid stability impact contribution).

[0281] For example, based on historical transportation samples, the contribution of the storage tank's dynamic stability is calculated to be 0.55, and the contribution of the LNG's gas-liquid stability is calculated to be 0.45. Therefore, the dynamic stability weight = 0.55 ÷ (0.55 + 0.45) = 0.55, and the gas-liquid stability weight = 0.45 ÷ (0.55 + 0.45) = 0.45.

[0282] Finally, based on the comparison of the transportation stability indices corresponding to each candidate transportation route, the candidate transportation route with the largest transportation stability index is selected as the target transportation route.

[0283] Step S500 in the method provided in this embodiment of the invention further includes: Obtain the gas-liquid state safety constraints corresponding to the LNG tank truck storage tank. The gas-liquid state safety constraints include LNG temperature constraints, LNG pressure constraints, and BOG production constraints. Based on the predicted heat exchange state characteristic sequence of the storage tank and the initial state parameters of LNG, the LNG temperature change state, LNG pressure change state and BOG production change state during the transportation process of each candidate transportation route are determined. Based on the gas-liquid state safety constraints, the constraint deviations of LNG temperature change, LNG pressure change and BOG production change for each candidate transportation path are calculated to obtain gas-liquid state deviation data for each time point during transportation. Based on the gas-liquid state deviation data, the cumulative deviation of LNG gas-liquid state from the gas-liquid state safety constraints during transportation is calculated to obtain the LNG gas-liquid deviation evaluation parameters corresponding to each candidate transportation route. The LNG gas-liquid deviation evaluation parameters are then reverse-converted to obtain the LNG gas-liquid stability index corresponding to each candidate transportation route.

[0284] In this embodiment, after obtaining the predicted heat exchange state characteristic sequence of the storage tank corresponding to each candidate transportation path, the gas-liquid state change of the LNG storage tank during transportation is predicted by combining the initial state parameters of LNG. The stability of each candidate transportation path is evaluated according to the gas-liquid state safety constraints to obtain the corresponding LNG gas-liquid stability index.

[0285] Specifically, obtain the gas-liquid state safety constraints corresponding to the LNG tanker storage tank.

[0286] Among them, the gas-liquid state safety constraints are used to characterize the temperature, pressure and gas-liquid conversion range allowed for LNG to maintain a stable storage state during transportation, including LNG temperature constraints, LNG pressure constraints and BOG production constraints.

[0287] LNG temperature constraints are used to limit the range of LNG liquid temperature changes during transportation to prevent the LNG temperature from rising too quickly due to external heat input, which could cause changes in the gas-liquid state inside the storage tank. LNG pressure constraint conditions are used to limit the range of pressure changes inside the storage tank, so as to prevent the safety valve from opening frequently or affecting the safe operation of the storage tank due to continuous pressure increase. BOG production constraints are used to limit the amount of gaseous natural gas produced by the heating and vaporization of liquefied natural gas during transportation, in order to avoid excessive gas-liquid conversion that could lead to abnormal changes in tank pressure.

[0288] Specifically, the safety constraints for the gas-liquid state are determined based on the LNG storage tank design parameters, the thermal properties of the LNG medium, and historical transportation safety data.

[0289] Specifically, by collecting data on changes in LNG temperature, tank pressure, and BOG production under different ambient temperatures, transportation times, and loading states under normal transportation conditions, the safe range of variation for each state parameter is determined.

[0290] For example, for a certain type of LNG tanker, based on historical transportation test data, if the following conditions are determined: the LNG temperature rise rate during transportation does not exceed the preset temperature change range, the tank pressure change does not exceed the design allowable pressure range, and the BOG generation per unit time does not exceed the historical safe operating maximum value, then the corresponding range will be used as the gas-liquid state safety constraint conditions.

[0291] Furthermore, based on the predicted heat exchange state characteristic sequence of the storage tank and the initial state parameters of LNG, the gas-liquid state changes of LNG during the transportation process of each candidate transportation route are determined.

[0292] Specifically, the predicted heat exchange state feature sequence of the storage tanks corresponding to each candidate transportation route is used as the external heat input feature, and combined with the initial state parameters of LNG, the changes in the internal thermal state of LNG during transportation are predicted.

[0293] The initial state parameters of LNG include initial temperature, initial pressure, initial loading volume, and initial gas-liquid ratio, which are used to characterize the basic state of LNG inside the storage tank at the start of transportation.

[0294] Specifically, the cumulative heat absorbed by LNG during transportation is calculated based on the heat exchange changes in the heat exchange state characteristic sequence of the storage tank. The cumulative heat is determined according to the following formula: Cumulative heat = heat flux density at each time point × corresponding heat exchange area of ​​the storage tank × cumulative value of time intervals.

[0295] The heat flux density is obtained by predicting the heat exchange state characteristic sequence of the storage tank, and the heat exchange area of ​​the storage tank is determined according to the structural parameters of the storage tank.

[0296] Furthermore, based on the accumulated heat and the initial state parameters of the LNG, the LNG temperature change state is determined. The specific calculation is as follows: LNG temperature change = cumulative heat ÷ (LNG mass × LNG specific heat capacity).

[0297] The LNG mass is determined based on the initial loading condition, and the LNG specific heat capacity is obtained based on the LNG thermophysical parameters.

[0298] Subsequently, the pressure change inside the storage tank was determined based on the LNG temperature change.

[0299] Specifically, based on the relationship between LNG temperature changes and saturation pressure changes, the LNG pressure change status at the corresponding time point is obtained.

[0300] At the same time, the changes in BOG production during transportation are determined based on the changes in LNG temperature and gas-liquid balance.

[0301] Specifically, when LNG absorbs heat, causing some of the liquefied natural gas to vaporize, the corresponding BOG production is determined based on the latent heat of vaporization and the cumulative heat absorbed. The specific calculation is as follows: BOG generation = cumulative absorbed heat ÷ latent heat of LNG vaporization.

[0302] Using the above methods, the LNG temperature change status, LNG pressure change status, and BOG production change status corresponding to each candidate transportation route are obtained.

[0303] Furthermore, based on the gas-liquid state safety constraints, constraint deviations are calculated for the LNG temperature change, LNG pressure change, and BOG production change states corresponding to each candidate transportation path, to obtain gas-liquid state deviation data for each time node during transportation.

[0304] Specifically, the LNG temperature, LNG pressure, and BOG production at each time point are compared with the corresponding safety constraint boundaries.

[0305] When the corresponding gas-liquid state parameters do not exceed the safety constraints, the deviation at that time point is set to zero; when the corresponding gas-liquid state parameters exceed the safety constraints, the gas-liquid state deviation is calculated based on the degree to which it exceeds the safety boundary. The specific calculation is as follows: Gas-liquid state deviation = (predicted gas-liquid state parameters - safety constraint boundary value) ÷ safety constraint boundary value.

[0306] The safety constraint boundary values ​​are determined based on the gas-liquid state safety constraint conditions.

[0307] Furthermore, the deviations for temperature, pressure, and BOG production were calculated separately and then fused to obtain the comprehensive gas-liquid state deviation. The specific calculations are as follows: Overall gas-liquid state deviation = temperature deviation × temperature weight + pressure deviation × pressure weight + BOG generation deviation × BOG weight.

[0308] The weights were determined based on historical LNG transportation risk data. Specifically, the impact of abnormal temperature, abnormal pressure, and abnormal BOG events on safety risks in historical transportation samples was statistically analyzed, and the contribution levels were normalized to obtain the corresponding weights.

[0309] For example, based on historical data, the following statistics are obtained: the contribution of temperature anomalies is 0.5; the contribution of pressure anomalies is 0.8; and the contribution of BOG generation anomalies is 0.7. The corresponding weights are calculated as follows: Temperature weight = 0.5 ÷ (0.5 + 0.8 + 0.7) ≈ 0.25; Pressure weight = 0.8 ÷ (0.5 + 0.8 + 0.7) ≈ 0.40; BOG weight = 0.7 ÷ (0.5 + 0.8 + 0.7) ≈ 0.35.

[0310] Furthermore, based on the gas-liquid state deviation data, the cumulative deviation of the LNG gas-liquid state from the safety constraints during the entire transportation process is calculated.

[0311] Specifically, the comprehensive gas-liquid state deviations at each time point are cumulatively calculated according to the LNG tanker transportation time sequence: LNG gas-liquid deviation evaluation parameter = cumulative value of comprehensive gas-liquid state deviation at each time point ÷ number of transportation time points.

[0312] The larger the LNG gas-liquid deviation evaluation parameter, the higher the degree to which the LNG gas-liquid state deviates from the safe range in the corresponding candidate transportation path.

[0313] Furthermore, the LNG gas-liquid deviation evaluation parameters are inversely transformed to obtain the LNG gas-liquid stability index for the corresponding candidate transportation route. The specific calculation is as follows: LNG gas-liquid stability index = 1 ÷ (1 + LNG gas-liquid deviation evaluation parameter).

[0314] By performing the above reverse conversion, candidate transport routes with smaller gas-liquid state deviations can obtain higher LNG gas-liquid stability indices.

[0315] For example, in an LNG tanker transport mission, three candidate transport routes were obtained. The first candidate route, due to its lower ambient temperature and shorter transport time, is predicted to have smaller LNG temperature changes, stable pressure changes, and lower BOG generation, corresponding to an LNG gas-liquid deviation evaluation parameter of 0.08, resulting in an LNG gas-liquid stability index of approximately 0.93. The second candidate route passes through a high-temperature region, predicting increased heat absorption in the storage tank, with a corresponding LNG gas-liquid deviation evaluation parameter of 0.25, resulting in an LNG gas-liquid stability index of approximately 0.80. Therefore, based on the LNG gas-liquid stability index corresponding to each candidate transport route, the impact of different transport routes on the gas-liquid stability of LNG can be evaluated.

[0316] In summary, this step integrates the characteristic sequences of tank dynamic response, LNG sloshing, and heat exchange states to comprehensively evaluate candidate routes from two dimensions: tank dynamics and gas-liquid stability. Stability deviation is calculated by establishing safety constraints, and historical data is used to determine index weights to quantify route risk. Finally, the transportation stability index is obtained by merging the dynamic stability index and the gas-liquid stability index. The route corresponding to the maximum value is selected to optimize transportation safety and stability.

Claims

1. A method for selecting hazardous chemical transportation routes based on deep learning, characterized in that, The method is applied to the screening of LNG tanker transport routes, including: Obtain the road condition feature sequence and road environment feature sequence for each candidate transportation route; Simulate the operation process of LNG tank trucks on each candidate path based on the road condition feature sequence, and obtain the predicted vehicle motion response feature sequence corresponding to each candidate path. Based on the structural parameters of the LNG tanker, the initial state parameters of LNG, and the predicted vehicle motion response feature sequence, the predicted tank dynamic response feature sequence and the predicted LNG liquid sloshing state feature sequence for each candidate path during transportation are obtained. Based on the structural parameters of the LNG tanker, a tank heat exchange prediction model is established, and the predicted tank heat exchange state feature sequence corresponding to each candidate path is obtained according to the initial state parameters of LNG and the road environment feature sequence. Based on the predicted tank dynamic response characteristic sequence, the predicted LNG liquid sloshing state characteristic sequence, and the predicted tank heat exchange state characteristic sequence, a comprehensive evaluation of LNG gas-liquid stability and tank dynamic stability is conducted on each candidate transportation route, and the candidate transportation route corresponding to the maximum transportation stability index is selected as the target transportation route.

2. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 1, characterized in that, Obtain the road condition feature sequence and road environment feature sequence for each candidate transportation route, including: Based on the geographical location information corresponding to the starting point and destination of liquefied natural gas transportation, multiple passable road nodes are determined in the target area map, and passable roads are enumerated according to the connection relationship between the multiple passable road nodes to generate multiple candidate transportation routes. Each candidate transportation path is segmented according to a preset road segmentation rule to obtain multiple road segment units corresponding to each candidate transportation path. For each road segment unit, the corresponding road geometry information, road state information, and road environment information are obtained, and the road segment units are arranged in a time sequence according to their arrangement order in the corresponding candidate transportation routes to generate road condition feature sequences and road environment feature sequences for each candidate transportation route.

3. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 1, characterized in that, Based on the road condition feature sequence, the operation process of LNG tank trucks on each candidate path is simulated to obtain the predicted vehicle motion response feature sequence corresponding to each candidate path, including: A vehicle motion response prediction model was constructed based on LNG tanker vehicle parameters; The road condition feature sequence corresponding to each candidate transportation route is input into the vehicle motion response prediction model. Based on the road change information in the road condition feature sequence, the operating status of the LNG tanker on the corresponding candidate transportation route is simulated to obtain the vehicle motion response data of the LNG tanker during transportation. The vehicle motion response data is arranged in chronological order during transportation to generate a predicted vehicle motion response feature sequence for each candidate transportation path.

4. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 3, characterized in that, A vehicle motion response prediction model is constructed based on LNG tanker vehicle parameters, including: Based on the vehicle mass parameters, vehicle structure parameters, and vehicle driving parameters in the LNG tanker vehicle parameters, determine the vehicle dynamic characteristic parameters that affect the changes in the motion state of the LNG tanker. Based on the road slope information, road curvature information and pavement condition information in the road condition feature sequence, a road excitation feature sequence is constructed to characterize the impact of the road on LNG tank trucks. The vehicle dynamic characteristic parameters and the road excitation characteristic sequence are fused to obtain the vehicle motion state input features. Based on the vehicle motion state input features and the corresponding historical vehicle motion response data, the deep learning prediction model is trained until convergence, generating a vehicle motion response prediction model. The vehicle motion response data includes longitudinal acceleration changes, lateral acceleration changes, vertical acceleration changes, and vehicle speed changes.

5. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 1, characterized in that, Based on the structural parameters of the LNG tanker, the initial state parameters of the LNG, and the predicted vehicle motion response feature sequence, the predicted tank dynamic response feature sequence and the predicted LNG liquid sloshing state feature sequence for each candidate route during transportation are obtained, including: The dynamic response characteristic parameters of the LNG tanker under the influence of vehicle movement are determined based on the structural parameters of the LNG tanker. The dynamic response characteristic parameters include the structural dimensions of the tanker, the installation position of the tanker, and the support connection parameters of the tanker. The dynamic response characteristic parameters, LNG initial state parameters, and predicted vehicle motion response characteristic sequence are input into the pre-constructed tank dynamic response prediction model to predict the force change state and vibration response state of the LNG tanker during transportation on each candidate transportation route. Based on the force change state and vibration response state, the predicted tank dynamic response characteristic sequence corresponding to each candidate transportation route is generated. Based on the predicted tank dynamic response feature sequence, the predicted LNG liquid sloshing state feature sequence corresponding to each candidate transportation path is obtained. The steps for constructing the tank dynamic response prediction model include: Acquire vehicle motion response characteristic sequences, LNG initial state parameters, and corresponding tank dynamic response sample data during the historical LNG tanker transportation process. The tank dynamic response sample data includes tank force change characteristics and tank vibration response characteristics. Based on the historical LNG tanker structural parameters, the corresponding tank dynamic response characteristic parameters are determined. The tank dynamic response characteristic parameters, LNG initial state parameters, and vehicle motion response characteristic sequences are correlated to obtain input features. Based on the input features and the corresponding tank dynamic response sample data, the deep learning prediction model is trained until the model output meets the preset training conditions, thus generating the tank dynamic response prediction model.

6. The method for selecting hazardous chemical transport routes based on deep learning according to claim 5, characterized in that, Based on the predicted tank dynamic response feature sequence, the predicted LNG liquid sloshing state feature sequence corresponding to each candidate transportation path is obtained, including: Based on the structural parameters of the LNG tanker, the internal structural parameters of the tank that affect the change of the LNG liquid movement state are determined. The internal structural parameters of the tank include the internal space size parameters of the tank, the internal anti-wave structure parameters of the tank, and the effective volume parameters of the tank. The initial motion parameters of LNG liquid are determined based on the initial state parameters of LNG liquid, which include the initial liquid level parameters of LNG liquid, the initial filling ratio parameters of LNG liquid, and the gas-liquid spatial distribution parameters. The internal structural parameters of the storage tank, the initial motion state parameters of the LNG liquid, and the predicted dynamic response characteristic sequence of the storage tank are input into a pre-constructed liquid sloshing state prediction model to predict the change in liquid surface fluctuation height, the maximum liquid surface offset distance, the liquid centroid offset, the liquid flow velocity, and the change in liquid impact pressure during the transportation of LNG liquid in each candidate transportation path, thereby generating the predicted LNG liquid sloshing state characteristic sequence corresponding to each candidate transportation path.

7. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 1, characterized in that, A tank heat exchange prediction model is established based on the LNG tanker storage tank structural parameters. Based on the LNG initial state parameters and road environment characteristic sequences, a predicted tank heat exchange state characteristic sequence is obtained for each candidate path, including: The heat exchange impact parameters of the LNG tanker are determined based on the structural parameters of the LNG tanker. These parameters include the outer shell structural parameters, the insulation structural parameters, and the surface heat exchange parameters of the tanker. The heat exchange influence parameters of the storage tank, the initial state parameters of LNG, and the road environment feature sequence are input into the pre-constructed heat exchange prediction model of the storage tank. Based on the changes in ambient temperature, solar radiation, and humidity in the road environment feature sequence, the heat exchange state of the LNG tanker storage tank during transportation on each candidate transportation route is predicted, and the predicted heat exchange state feature sequence of the storage tank corresponding to each candidate transportation route is generated.

8. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 7, characterized in that, The steps for constructing the tank heat exchange prediction model include: The system acquires the road environment feature sequence, LNG initial state parameters, and corresponding tank heat exchange state sample data during the historical LNG tanker transportation process. The tank heat exchange state sample data includes the change in tank outer wall temperature, the change in temperature difference between inside and outside the tank, and the change in tank heat flux density. Based on the historical LNG tanker storage tank structural parameters, the corresponding tank heat exchange impact parameters are determined. The tank heat exchange impact parameters, LNG initial state parameters, and road environment feature sequences are correlated to obtain the model input features. Based on the input features of the model and the corresponding sample data of the heat exchange status of the storage tank, the deep learning prediction model is iteratively trained to generate the heat exchange prediction model of the storage tank.

9. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 1, characterized in that, Based on the predicted tank dynamic response characteristic sequence, the predicted LNG liquid sloshing state characteristic sequence, and the predicted tank heat exchange state characteristic sequence, a comprehensive evaluation of LNG gas-liquid stability and tank dynamic stability is performed on each candidate transportation route, including: Obtain the dynamic response safety constraints corresponding to the LNG tanker storage tank, including the tank force constraints and the tank vibration constraints. Based on the dynamic response safety constraints, the constraint deviation is calculated for the predicted tank dynamic response feature sequence corresponding to each candidate transportation path, and the tank dynamic response deviation data corresponding to each time node is obtained. Based on the tank dynamic response deviation data, the cumulative deviation of the tank dynamic response from the safety constraints during transportation is calculated to obtain the tank dynamic deviation evaluation parameters corresponding to each candidate transportation path. Obtain the liquid sloshing safety constraints corresponding to the LNG tanker storage tank, the liquid sloshing safety constraints including liquid surface fluctuation constraints and liquid impact pressure constraints; Based on the liquid sloshing safety constraints, the constraint deviation is calculated for the predicted LNG liquid sloshing state characteristic sequence corresponding to each candidate transportation path, and the liquid sloshing deviation data corresponding to each time node is obtained. Based on the liquid sloshing deviation data, the cumulative deviation of the liquid sloshing state from the safety constraints during transportation is calculated to obtain the liquid sloshing deviation evaluation parameters corresponding to each candidate transportation path. The tank dynamic deviation evaluation parameters and liquid sloshing deviation evaluation parameters are reverse-converted, and the conversion results are fused and calculated to obtain the tank dynamic stability index corresponding to each candidate transportation path. Based on the predicted heat exchange state characteristic sequence of the storage tank, the LNG gas-liquid stability of each candidate transportation route is evaluated, and the LNG gas-liquid stability index corresponding to each candidate transportation route is obtained. The transportation stability index corresponding to each candidate transportation route is obtained by weighted evaluation based on the storage tank dynamic stability index and the LNG gas-liquid stability index.

10. The method for selecting hazardous chemical transportation routes based on deep learning according to claim 9, characterized in that, The gas-liquid stability of LNG is evaluated for each candidate transportation route based on the predicted heat exchange state characteristic sequence of the storage tank, including: Obtain the gas-liquid state safety constraints corresponding to the LNG tanker storage tank, including LNG temperature constraints, LNG pressure constraints, and BOG production constraints. Based on the predicted heat exchange state characteristic sequence of the storage tank and the initial state parameters of LNG, the LNG temperature change state, LNG pressure change state and BOG production change state during the transportation process of each candidate transportation route are determined. Based on the gas-liquid state safety constraints, constraint deviations are calculated for the LNG temperature change, LNG pressure change, and BOG production change states corresponding to each candidate transportation path, and gas-liquid state deviation data corresponding to each time node during transportation are obtained. Based on the gas-liquid state deviation data, the cumulative deviation of the LNG gas-liquid state from the gas-liquid state safety constraints during transportation is calculated to obtain the LNG gas-liquid deviation evaluation parameters corresponding to each candidate transportation path. The LNG gas-liquid deviation evaluation parameters are then reverse-converted to obtain the LNG gas-liquid stability index corresponding to each candidate transportation path.

Images