Multi-objective path dynamic optimization method and system based on cold chain logistics

Abstract

This invention discloses a multi-objective path dynamic optimization method and system for cold chain logistics, belonging to the field of logistics technology. The method includes data acquisition and initialization, cargo damage prediction, multi-objective optimization solution, dynamic reprogramming, temperature prediction and refrigeration control, and data archiving steps. The multi-objective optimization targets delivery cost, customer satisfaction, and carbon emissions, and is solved using an improved non-dominated sorting genetic algorithm with adaptive weighted congestion distance. Dynamic reprogramming is triggered by weighted indices of temperature deviation, delay time, and road condition changes, and employs an adaptive large-scale neighborhood search algorithm for local re-optimization. Temperature control uses a long short-term memory neural network to predict the compartment temperature, achieving feedforward PID control. The system includes modules for data acquisition, data processing, cargo damage prediction, multi-objective optimization, dynamic reprogramming, temperature prediction and control, human-computer interaction, and data storage.

Description

Technical Field

[0001] This invention relates to the field of logistics technology, and more specifically to a multi-objective path dynamic optimization method and system based on cold chain logistics. Background Technology

[0002] Cold chain logistics is a logistics method that maintains a suitable temperature environment throughout the production, storage, transportation, and sales of temperature-sensitive products. It is mainly used in the fields of fresh agricultural products, pharmaceuticals, and chemical raw materials. With the improvement of residents' consumption levels and the rapid development of e-commerce, the market size of cold chain logistics continues to expand, placing higher demands on delivery efficiency and service quality. The core task of cold chain logistics distribution is to rationally plan delivery routes while meeting temperature control requirements, achieving the overall optimization of cost, timeliness, and service quality.

[0003] Traditional cargo damage prediction models often employ simple linear time decay models or Arrhenius equations based on absolute temperature values. These models only consider the static impact of temperature on cargo quality, failing to adequately account for the nonlinear accelerated decay effects of temperature deviations from the optimal value and frequent temperature fluctuations during actual delivery. In real-world cold chain delivery, vehicles need to open their doors when loading and unloading goods at customer nodes, causing a brief increase in the vehicle's temperature. The intermittent operation of refrigeration equipment also causes periodic temperature fluctuations. These unstable temperature factors can have a greater impact on the quality of certain sensitive goods than continuous slight temperature deviations. Existing models struggle to accurately quantify the additional damage caused by these temperature fluctuations, resulting in low cargo damage prediction accuracy and consequently affecting the rationality of delivery route optimization decisions. Regarding multi-objective optimization algorithms, non-dominated sorting genetic algorithms are commonly used to solve multi-objective optimization problems in cold chain logistics delivery routes, but standard algorithms have limitations in maintaining the uniformity of the Pareto solution set distribution. Standard algorithms use fixed weights to calculate congestion distance. When the dimensions and numerical ranges of the objective function values ​​differ significantly, some objectives may dominate the calculation results, leading to a dense distribution of the solution set in some objective directions and a sparse distribution in others. This makes it difficult to provide decision-makers with diverse solutions that uniformly cover the Pareto front. Furthermore, existing route optimization methods are mostly static optimizations, generating delivery plans based on predicted data during the planning phase. They do not consider changes in dynamic factors such as traffic conditions and weather conditions during delivery execution, lacking real-time monitoring and dynamic adjustment mechanisms. When the actual situation deviates significantly from expectations, the effectiveness of the original plan will decrease markedly. Regarding temperature control, existing refrigerated truck refrigeration systems mostly use feedback control based on measured temperatures. Adjustment is only initiated when sensors detect that the temperature inside the truck exceeds the set range. Due to the thermal inertia and response delay of the refrigeration system, feedback control struggles to respond promptly to temperature changes, easily leading to temperature overshoot and continuous fluctuations. This is detrimental to maintaining cargo quality, and frequent high-power adjustments also increase energy consumption.

[0004] Therefore, designing a multi-objective path dynamic optimization method and system based on cold chain logistics to solve the above problems is of great significance. Summary of the Invention

[0005] To address the problems existing in the background technology, this invention provides a multi-objective path dynamic optimization method based on cold chain logistics, comprising the following steps: S1. Data Acquisition and Initialization: Collect data on distribution center location, customer node location, customer demand, customer time window, temperature-sensitive parameters of goods category, vehicle configuration parameters, real-time traffic network data, and meteorological data. Preprocess the collected data to generate a delivery task dataset. S2. Cargo Damage Prediction: Based on the temperature-sensitive parameters of cargo categories and real-time temperature data, a cargo damage prediction model that considers temperature deviation and temperature fluctuation is adopted to calculate the quality decay rate of various types of cargo and the predicted cargo damage rate when they arrive at each customer node. S3. Multi-objective optimization solution: With the optimization objectives of minimizing total delivery cost, maximizing customer satisfaction, and minimizing carbon emissions, and with constraints of vehicle capacity, time window, and temperature, an improved non-dominated sorting genetic algorithm based on adaptive weighted congestion distance is used to solve the problem, generate a Pareto optimal solution set, and select a delivery route plan from it to be sent to the vehicle terminal for execution. S4. Dynamic Replanning: During the delivery process, real-time data on vehicle location, compartment temperature, and road traffic status are collected to calculate the dynamic replanning trigger index value. When the trigger index value exceeds the preset threshold, an adaptive large-scale neighborhood search algorithm is used to perform incremental local re-optimization on the remaining undelivered nodes to update the delivery route plan. S5. Temperature Prediction and Cooling Control: Based on historical temperature data and real-time operating data, the system uses a long short-term memory neural network to predict the cabin temperature within a set time period in the future. The system calculates the optimal cooling power output value based on the predicted temperature deviation to achieve feedforward control of the cooling system. S6. Data Archiving: After the delivery task is completed, record the actual delivery route, the temperature curve throughout the process, the actual damage to goods and energy consumption data, and store them in the historical database for model parameter optimization.

[0006] Furthermore, step S2 includes the following sub-steps: S21. Obtain the category-sensitive parameters of the currently transported goods, including the activation energy of quality degradation. Quality degradation pre-exponential factor Optimal storage temperature Temperature deviation sensitivity coefficient Temperature fluctuation sensitivity coefficient ; S22. Obtain the current time from the vehicle temperature sensor. Real-time temperature of the carriage Calculate the standard deviation of temperature data within a set time window prior to the current moment. ; S23. Calculate the quality decay rate at the current moment. The calculation formula is: ;in, for The rate of quality degradation at any given time (1 / h); The quality degradation pre-exponential factor (1 / h); It is a natural exponential function; The activation energy for quality decay (J / mol); The universal gas constant is 8.314 (J / (mol·K)). for Real-time temperature (K) of the carriage at any given moment; This refers to the temperature deviation sensitivity coefficient. For real-time temperature and optimal storage temperature The absolute difference (K); The optimal storage temperature (K); This is the temperature fluctuation sensitivity coefficient; for Set the standard deviation (K) of the temperature data within the time window before the specified time. S24. Based on the estimated arrival time of each customer node in the delivery route plan. Integrate the quality decay rate over time to calculate the arrival time of the goods. Loss rate at each customer node The calculation formula is: ;in, For the goods to arrive at the Loss rate (%) at each customer node; It is a natural exponential function; Represents the time from the start time 0 to the arrival time. Definite integral operations; for The rate of quality degradation at any given time (1 / h); To reach the Time (h) for each customer node; S25. Output the predicted loss rate of each customer node to the multi-objective optimization module for calculating the objective function of loss cost.

[0007] Furthermore, step S3 includes the following sub-steps: S31. Construct the first objective function With the goal of minimizing total delivery costs, the calculation formula is as follows: ;in, Total delivery cost (RMB); Transportation cost (yuan); Refrigeration cost (RMB); Cost of damaged goods (yuan); Cost of penalty for breach of contract within the time window (RMB); S32. Constructing the second objective function With the goal of maximizing customer satisfaction, the calculation formula is as follows: ;in, Average customer satisfaction; Total number of customers; For the first Customer satisfaction score; Indicates to From 1 to Summation; S33. Constructing the third objective function With the goal of minimizing carbon emissions, the calculation formula is as follows: ;in, Total carbon emissions (kg); Fuel carbon emission coefficient (kg / L); The total number of vehicles; This represents the total number of road segments; Fuel consumption per unit distance when unloaded (L / km); Fuel consumption increment per unit load per unit distance (L / (kg·km)); For vehicles Current load capacity (kg); For the first Distance of the road segment (km); As a decision variable, vehicle After the first The value is 1 if the road segment is valid, otherwise the value is 0. The carbon emission coefficient for refrigeration power consumption (kg / kWh); For vehicles Total power consumption for cooling (kWh); Indicates to From 1 to Summation; Indicates to From 1 to Summation; S34, Randomly generated A feasible solution is used as the initial population, and each individual uses an integer code to represent the delivery path; S35. Calculate three objective function values ​​for each individual in the population, and perform fast non-dominated sorting to divide the individuals into different Pareto rank levels. S36. Calculate the adaptive weighted crowding distance for each individual within the same Pareto level. The calculation formula is: ;in, For the first The crowded distance between individuals; This represents the summation of three objective functions; For the first Adaptive weights for each objective; For the first After sorting the target, the first The next neighbor of each individual One target value; For the first After sorting the target, the first The first neighbor of each individual One target value; For the current population The maximum value of each objective; For the current population The minimum value of each objective; S37. Use a binary tournament selection operator to select parent individuals based on Pareto rank and crowding distance, and perform simulated binary crossover and polynomial mutation to generate offspring individuals; S38. Merge the parent and offspring populations, selecting populations based on Pareto rank priority, and then prioritizing those with greater crowding distance within the same rank. Individuals enter the next generation of the population; S39. Determine if the maximum number of iterations has been reached. If it has, output the Pareto optimal solution set; otherwise, return to step S35 to continue iterating. S310. Use the TOPSIS method to select the compromise optimal solution from the Pareto optimal solution set as the delivery route scheme.

[0008] Furthermore, step S4 includes the following sub-steps: S41. Collect the current vehicle location and passenger compartment temperature from the vehicle terminal every preset sampling period. The system collects data on external ambient temperature and vehicle speed, and obtains the real-time traffic congestion index of the road ahead from a traffic data interface. ; S42. Calculate the normalized temperature deviation. The calculation formula is: ;in, Normalized temperature deviation; Current temperature of the carriage With the target temperature setpoint The absolute difference (°C); The upper limit of the permissible temperature for goods (°C); The lower limit of the permissible temperature for goods (°C); S43. Calculate the estimated arrival time based on the current location and real-time traffic conditions. Calculate normalized delay time The calculation formula is: ;in, Normalization caused delays; This is a function to find the maximum value. Estimated arrival time (min); The latest arrival time (in minutes) requested by the customer; The allowable delay tolerance time (min); S44. Calculate the Normalized Road Condition Change Index The calculation formula is: ;in, This is a normalized road condition change index; This represents the current actual traffic congestion index; The traffic congestion index preset during the planning stage; This represents the maximum value of the traffic congestion index. S45. Calculate the dynamic replanning trigger index value. The calculation formula is: ;in, The replanning trigger indicator value; Temperature deviation weighting coefficient; This is the weighting coefficient for delay time; Road condition weighting coefficient; Normalized temperature deviation; Normalization caused delays; This is a normalized road condition change index; S46. Determine the trigger indicator value Does it exceed the preset threshold? If the duration exceeds the preset minimum number of cycles, then proceed to step S47; otherwise, return to step S41 to continue monitoring. S47. Mark the nodes that have completed delivery as fixed nodes, and construct subproblems for the remaining undelivered nodes; S48. Adaptive large-scale neighborhood search algorithm is used to solve the subproblem: select a destruction operator from the set of destruction operators in a roulette wheel manner to remove some nodes, select a repair operator from the set of repair operators in a roulette wheel manner to re-insert the removed nodes, determine whether to accept the new solution according to the simulated annealing criterion, and output the re-optimized path scheme after iterating until the termination condition is met. S49. Update the optimized route plan to the vehicle terminal to guide the driver to perform delivery according to the new route.

[0009] Furthermore, step S5 includes the following sub-steps: S51. Constructing the input feature vector for the temperature prediction model The input feature vector Including history The time step includes the internal temperature sequence of the carriage, the external ambient temperature sequence, the cooling power sequence, the vehicle speed sequence, the door opening and closing status sequence, and the current cargo load. S52, Input feature vector Input a pre-trained long short-term memory neural network model, output the future The predicted temperature of the carriage at each time step is taken as the predicted temperature. ; S53, Calculate the predicted temperature deviation The calculation formula is: ;in, To predict temperature deviation (°C); To predict the temperature (°C); Set the target temperature (°C); S54. Computational Environment and Vehicle Speed ​​Correction Coefficient The calculation formula is: ;in, This is a correction factor; The environmental temperature influence coefficient; The current ambient temperature (°C); For reference ambient temperature (°C); This is the coefficient affecting vehicle speed; Current vehicle speed (km / h); For reference vehicle speed (km / h); S55. Calculate the optimal cooling power output value. The calculation formula is: ;in, The optimal cooling power output value (kW); Based on the cooling capacity (kW); This is the proportional control coefficient; To predict temperature deviation (°C); These are integral control coefficients; The time integral value of the temperature deviation (°C·s); These are the differential control coefficients; The rate of change of temperature deviation (°C / s); This is a correction factor; S56, Optimal cooling power output value Send the signal to the refrigeration unit controller to adjust the actual output power of the refrigeration unit; S57, If the predicted temperature If the temperature exceeds the permissible range for the cargo, a temperature warning message will be sent to the dispatch center and the driver's terminal.

[0010] This invention also provides a multi-objective path dynamic optimization system based on cold chain logistics, comprising: The data acquisition module includes an on-board data acquisition unit and an external data interface unit. The on-board data acquisition unit is installed on the delivery vehicle and is electrically connected to a temperature sensor inside the vehicle compartment, an ambient temperature sensor outside the vehicle compartment, a GPS positioning module, a vehicle speed sensor, and a door status sensor, respectively, for collecting data on vehicle compartment temperature, ambient temperature, vehicle position, vehicle speed, and door status. The external data interface unit is communicatively connected to a traffic data server and a meteorological data server, respectively, for acquiring road network data, real-time traffic data, and meteorological data. The data processing module is communicatively connected to the data acquisition module and is used to perform missing value imputation, outlier removal, and format unification processing on the acquired raw data to generate a standardized delivery task dataset. The cargo damage prediction module, which is communicatively connected to the data processing module, includes a parameter storage unit, a temperature statistics unit, and a decay calculation unit. The parameter storage unit is used to store temperature-sensitive parameters for each category of cargo. The temperature statistics unit is communicatively connected to the data acquisition module and is used to calculate the standard deviation of temperature data. The decay calculation unit is communicatively connected to both the parameter storage unit and the temperature statistics unit and is used to calculate the quality decay rate and cargo damage rate. The multi-objective optimization module, communicatively connected to the data processing module and the cargo loss prediction module, includes an objective function calculation unit, a non-dominated sorting unit, a crowding distance calculation unit, and a genetic operation unit. The objective function calculation unit calculates three objective function values: total delivery cost, customer satisfaction, and carbon emissions. The non-dominated sorting unit, communicatively connected to the objective function calculation unit, performs rapid non-dominated sorting of the population. The crowding distance calculation unit, also communicatively connected to the non-dominated sorting unit, calculates adaptive weighted crowding distance. The genetic operation unit, communicatively connected to the crowding distance calculation unit, performs selection, crossover, and mutation operations and outputs a Pareto optimal solution set. The dynamic replanning module, communicatively connected to the data acquisition module, includes an index calculation unit, a trigger judgment unit, and a local reoptimization unit. The index calculation unit is used to calculate the normalized temperature deviation, normalized delay time, normalized road condition change index, and replanning trigger index value. The trigger judgment unit is communicatively connected to the index calculation unit and is used to determine whether the trigger condition is met. The local reoptimization unit is communicatively connected to the trigger judgment unit and the multi-objective optimization module, respectively, and is used to perform incremental local reoptimization on the remaining undelivered nodes. The temperature prediction and control module, communicatively connected to the data acquisition module, includes a feature extraction unit, a neural network inference unit, and a power calculation unit. The feature extraction unit extracts input feature vectors from historical data. The neural network inference unit, communicatively connected to the feature extraction unit, incorporates a long short-term memory neural network model to output predicted temperature values. The power calculation unit, communicatively connected to the neural network inference unit, calculates the optimal cooling power output value. The power calculation unit, communicatively connected to the refrigeration unit controller, sends cooling power control commands. The human-computer interaction module, which is communicatively connected to the multi-objective optimization module and the dynamic replanning module, includes a dispatcher management terminal and a driver mobile terminal; the dispatcher management terminal is used to display delivery route plans and real-time monitoring information; the driver mobile terminal is located in the cab of the delivery vehicle and is wirelessly connected to the dynamic replanning module to receive navigation routes and early warning information. The data storage module is communicatively connected to the data processing module, the cargo damage prediction module, and the temperature prediction and control module, respectively, and is used to store historical delivery data and model parameters.

[0011] In the preferred embodiment, the cargo damage prediction module includes: The parameter storage unit is communicatively connected to the data storage module and reads the activation energy of quality decay, the pre-exponential factor of quality decay, the optimal storage temperature, the temperature deviation sensitivity coefficient, and the temperature fluctuation sensitivity coefficient of each type of goods from the data storage module. The temperature statistics unit receives real-time temperature data from the vehicle-mounted data acquisition unit, calculates the temperature standard deviation according to a set time window, and outputs it to the attenuation calculation unit. The attenuation calculation unit calculates the quality attenuation rate and the damage rate based on the received temperature-sensitive parameters, real-time temperature, and temperature standard deviation, and outputs the damage rate to the objective function calculation unit of the multi-objective optimization module.

[0012] In the preferred embodiment, the multi-objective optimization module includes: The objective function calculation unit receives road network distance data and customer time window data from the data processing module, and receives cargo loss rate data from the cargo loss prediction module. It calculates the objective function value of total delivery cost, the objective function value of customer satisfaction, and the objective function value of carbon emissions, and outputs the three objective function values ​​to the non-dominated sorting unit. The non-dominated sorting unit performs fast non-dominated sorting of all individuals in the population according to three objective function values, and outputs the sorting results to the crowding distance calculation unit. The crowding distance calculation unit calculates the adaptive weighted crowding distance of each individual within the same Pareto level based on the sorting results, and outputs the calculation results to the genetic operation unit. The genetic operation unit performs selection operations based on Pareto level and crowding distance, performs crossover and mutation operations on the selected individuals to generate offspring population, and outputs the final Pareto optimal solution set to the human-computer interaction module.

[0013] In the preferred embodiment, the dynamic replanning module includes: The index calculation unit receives vehicle temperature and vehicle location data from the vehicle data acquisition unit, receives real-time traffic congestion index from the external data interface unit, calculates normalized temperature deviation, normalized delay time, and normalized road condition change index, and calculates replanning trigger index value according to preset weight coefficients, and outputs the trigger index value to the trigger judgment unit. The triggering judgment unit compares the received triggering index value with a preset threshold and determines whether the duration of exceeding the threshold exceeds a preset minimum number of cycles. When the triggering condition is met, it sends a replanning start signal to the local re-optimization unit. After receiving the replanning start signal, the local re-optimization unit obtains the current delivery route plan from the multi-objective optimization module, performs incremental local re-optimization on the remaining undelivered nodes using an adaptive large-scale neighborhood search algorithm, and sends the re-optimized route plan to the driver's mobile terminal through the human-machine interaction module.

[0014] In the preferred embodiment, the temperature prediction and control module includes: The feature extraction unit reads historical temperature data from the data storage module and receives real-time operating data from the vehicle data acquisition unit. It extracts input feature vectors including the compartment temperature sequence, ambient temperature sequence, cooling power sequence, vehicle speed sequence, door status sequence, and cargo load, and outputs the input feature vectors to the neural network inference unit. After receiving the input feature vector, the neural network inference unit performs inference calculations through the built-in long short-term memory neural network model and outputs the temperature prediction value to the power calculation unit. The power calculation unit calculates the predicted temperature deviation based on the received temperature prediction value and the target temperature set value, calculates the correction coefficient by combining the ambient temperature and vehicle speed data received from the vehicle data acquisition unit, calculates the optimal cooling power output value based on the predicted temperature deviation and the correction coefficient, and converts the optimal cooling power output value into a control command and sends it to the refrigeration controller. The power calculation unit is also connected to the human-machine interaction module. When the temperature prediction value exceeds the allowable range, it pushes temperature warning information to the dispatcher management terminal and the driver mobile terminal.

[0015] The beneficial effects achieved by this invention are as follows: This invention addresses the shortcomings of existing models in predicting cargo damage by introducing temperature deviation and temperature fluctuation correction terms into the classic Arrhenius equation. This constructs a cargo damage prediction model that considers both temperature deviation and temperature fluctuation. The temperature deviation correction term quantifies the accelerating effect of temperature deviation on quality degradation by calculating the relative deviation between the real-time temperature and the optimal storage temperature. The temperature fluctuation correction term quantifies the additional damage to quality caused by temperature instability by calculating the standard deviation of temperature data within a set time window. This model can more accurately reflect the actual impact of temperature instability caused by door opening and handling, equipment fluctuations, etc., on cargo quality during cold chain distribution, improving the accuracy of cargo damage prediction and providing a more reliable decision-making basis for distribution route optimization.

[0016] In the field of multi-objective optimization algorithms, this invention addresses the problem of uneven solution set distribution in standard non-dominated sorting genetic algorithms by proposing an adaptive weighted crowding distance calculation method. This method dynamically calculates the weight of each objective based on the distribution of objective function values ​​in the current population, assigning greater weight to objective directions with sparse solution sets. This encourages the algorithm to expand its search in those directions, thereby improving the uniformity of the Pareto solution set distribution across the entire frontier. Compared to fixed-weight methods, the adaptive weight mechanism allows the algorithm to automatically adapt to differences in the dimensions and numerical ranges of different objective functions, obtaining a uniformly distributed set of diverse solutions without manual parameter adjustment, facilitating selection by decision-makers based on actual preferences.

[0017] In terms of dynamic replanning, this invention addresses the lack of dynamic adaptability in existing static optimization methods by designing a dynamic replanning trigger mechanism based on multi-indicator fusion. This mechanism comprehensively considers three dimensions: temperature deviation, time delay, and road condition changes. It calculates the replanning trigger index value through normalization and weighted summation. When this index value consistently exceeds a preset threshold, local reoptimization is triggered. Compared to single-indicator triggering, the multi-indicator fusion mechanism can more comprehensively assess the deviation of delivery execution status, reducing false triggers caused by short-term fluctuations of a single factor, while also avoiding overlooking complex anomalies caused by the superposition of multiple factors. The incremental local reoptimization strategy only optimizes the remaining undelivered nodes, resulting in lower computational load and faster response speed compared to global replanning, meeting the timeliness requirements of real-time scheduling.

[0018] In terms of temperature control, this invention addresses the issue of lag in traditional feedback control by employing a Long Short-Term Memory (LSTM) neural network to predict future temperature changes in the vehicle compartment. The predicted temperature deviation is then used as input to a PID controller, enabling feedforward control of the refrigeration system. The LSM network captures the temporal patterns of temperature changes through a gating mechanism, predicting future temperature trends based on multi-dimensional features such as historical temperature sequences, ambient temperature, refrigeration power, vehicle speed, and door status. This feedforward control strategy based on predicted temperature adjusts the refrigeration power before actual temperature deviations occur, reducing temperature fluctuations and mitigating the risk of cargo damage caused by these fluctuations. Furthermore, the introduction of ambient temperature and vehicle speed correction coefficients allows the control strategy to adapt to different operating conditions, avoiding unnecessary over-cooling and reducing refrigeration energy consumption. Attached Figure Description

[0019] Figure 1 This is a flowchart of the multi-objective path dynamic optimization method based on cold chain logistics of the present invention; Figure 2 This is a structural diagram of the multi-objective path dynamic optimization system based on cold chain logistics of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Reference Figures 1-2 This invention provides a multi-objective path dynamic optimization system based on cold chain logistics, including a data acquisition module, a data processing module, a cargo damage prediction module, a multi-objective optimization module, a dynamic replanning module, a temperature prediction and control module, a human-computer interaction module, and a data storage module.

[0022] The data acquisition module is responsible for collecting various types of data required for system operation, including an onboard data acquisition unit and an external data interface unit. The onboard data acquisition unit is installed on the delivery vehicle and is electrically connected to a temperature sensor inside the vehicle compartment to collect compartment temperature data, to an ambient temperature sensor outside the compartment to collect ambient temperature data, to a GPS positioning module to collect the vehicle's real-time location coordinates, to a vehicle speed sensor to collect vehicle speed, and to a door status sensor to detect door opening / closing status. The onboard data acquisition unit uploads the collected data to a cloud server via a wireless communication network. The external data interface unit communicates with a traffic data server to obtain road network topology and real-time traffic information, and with a meteorological data server to obtain ambient temperature and weather forecast data for the delivery area.

[0023] The data processing module communicates with the data acquisition module, receiving the raw data collected by the acquisition module. It performs missing value imputation, outlier removal, and format standardization on the data, generating a standardized delivery task dataset for use by other modules. The data processing module also controls data quality to ensure the accuracy and reliability of the data used in subsequent calculations.

[0024] The cargo damage prediction module communicates with the data processing module to calculate the quality decay rate and damage rate of the cargo. The cargo damage prediction module includes a parameter storage unit, a temperature statistics unit, and a decay calculation unit. The parameter storage unit communicates with the data storage module to store temperature-sensitive parameters for each type of cargo, including the activation energy of quality decay, the pre-exponential factor of quality decay, the optimal storage temperature, the temperature deviation sensitivity coefficient, and the temperature fluctuation sensitivity coefficient. The temperature statistics unit communicates with the on-board data acquisition unit to receive real-time temperature data and calculate the standard deviation of the temperature data according to a set time window, reflecting the degree of temperature fluctuation. The decay calculation unit communicates with both the parameter storage unit and the temperature statistics unit, receiving the temperature-sensitive parameters, real-time temperature, and temperature standard deviation. It calculates the quality decay rate based on the improved Arrhenius equation, calculates the damage rate through time integration, and outputs the damage rate to the multi-objective optimization module as input for the objective function calculation.

[0025] The multi-objective optimization module communicates with the data processing module and the cargo loss prediction module to construct and solve the multi-objective optimization model. The multi-objective optimization module includes an objective function calculation unit, a non-dominated sorting unit, a congestion distance calculation unit, and a genetic operation unit. The objective function calculation unit receives road network distance data and customer time window data from the data processing module, and cargo loss rate data from the cargo loss prediction module, calculating the objective function values ​​for total delivery cost, customer satisfaction, and carbon emissions, respectively. The non-dominated sorting unit communicates with the objective function calculation unit, receives the objective function values ​​for each individual, and executes a fast non-dominated sorting algorithm to divide the population into different Pareto levels. The congestion distance calculation unit communicates with the non-dominated sorting unit, receives the sorting results, and calculates the adaptive weighted congestion distance for each individual within the same Pareto level. The genetic operation unit communicates with the congestion distance calculation unit, performs selection operations based on Pareto level and congestion distance to determine parent individuals, performs crossover and mutation operations on the parent individuals to generate offspring, and iterates until the termination condition is met, outputting the Pareto optimal solution set to the human-computer interaction module for decision-makers to select.

[0026] The dynamic replanning module communicates with the data acquisition module to monitor delivery execution status in real time and trigger route replanning when necessary. The dynamic replanning module includes an indicator calculation unit, a trigger judgment unit, and a local reoptimization unit. The indicator calculation unit receives vehicle temperature and vehicle location data from the onboard data acquisition unit and real-time traffic congestion index from the external data interface unit. It calculates normalized temperature deviation, normalized delay time, and normalized road condition change index, calculates replanning trigger indicator values ​​based on preset weighting coefficients, and outputs the trigger indicator values ​​to the trigger judgment unit. The trigger judgment unit communicates with the indicator calculation unit, compares the received trigger indicator values ​​with preset thresholds, and determines whether the duration of exceeding the threshold exceeds a preset minimum number of cycles. When the trigger condition is met, it sends a replanning start signal to the local reoptimization unit. The local reoptimization unit communicates with the trigger judgment unit and the multi-objective optimization module. After receiving the replanning start signal, it obtains the current delivery route plan from the multi-objective optimization module, performs incremental local reoptimization on the remaining undelivered nodes using an adaptive large-scale neighborhood search algorithm, and sends the reoptimized route plan to the driver's mobile terminal through the human-machine interaction module to guide delivery execution.

[0027] The temperature prediction and control module communicates with the data acquisition module to predict changes in the vehicle compartment temperature and output cooling power control commands. The temperature prediction and control module includes a feature extraction unit, a neural network inference unit, and a power calculation unit. The feature extraction unit reads historical temperature data from the data storage module and receives real-time operating data from the onboard data acquisition unit. It extracts input feature vectors including the compartment temperature sequence, ambient temperature sequence, cooling power sequence, vehicle speed sequence, door status sequence, and cargo load, and outputs these input feature vectors to the neural network inference unit. The neural network inference unit communicates with the feature extraction unit and has a pre-trained long short-term memory neural network model. After receiving the input feature vectors, it performs inference calculations through the model and outputs the predicted temperature value to the power calculation unit. The power calculation unit communicates with the neural network inference unit. It calculates the predicted temperature deviation based on the received predicted temperature value and the target temperature setpoint, calculates a correction coefficient using the ambient temperature and vehicle speed data received from the onboard data acquisition unit, and calculates the optimal cooling power output value using a PID control algorithm based on the predicted temperature deviation and the correction coefficient. The optimal cooling power output value is then converted into a control command and sent to the refrigeration unit controller to adjust the cooling power. The power calculation unit is also connected to the human-machine interaction module. When the temperature prediction value exceeds the allowable temperature range of the goods, it pushes temperature warning information to the dispatcher management terminal and the driver's mobile terminal.

[0028] The human-machine interface (HMI) module is communicatively connected to the multi-objective optimization module and the dynamic replanning module, respectively, and is used to display system operation information and receive user decision input. The HMI module includes a dispatcher management terminal and a driver mobile terminal. The dispatcher management terminal displays delivery route planning results, real-time delivery monitoring maps, vehicle status information, temperature monitoring curves, and early warning information, and receives route selection and parameter adjustment inputs from the dispatcher. The driver mobile terminal is located in the driver's cab of the delivery vehicle and is wirelessly connected to the dynamic replanning module. It receives and displays navigation routes, delivery task lists, and temperature early warning information to guide the driver in executing delivery tasks.

[0029] The data storage module is communicatively connected to the data processing module, cargo damage prediction module, and temperature prediction and control module, respectively, and is used to store various types of data required for system operation. The data storage module stores historical delivery execution records, temperature curves, energy consumption data, and performance indicators, providing a data foundation for operational analysis and decision support. The data storage module also stores the temperature-sensitive parameters of the cargo damage prediction model and the weight parameters of the temperature prediction neural network model, which are available for use by the cargo damage prediction module and the temperature prediction and control module. The system can periodically use historical data to calibrate, optimize, and iteratively train the model parameters, continuously improving prediction accuracy and control effectiveness.

[0030] The system of this invention enables data exchange and collaborative operation among its various modules via communication connections. The data acquisition module provides raw data input to other modules; the data processing module preprocesses the data and distributes it to each calculation module; the damage prediction module and the multi-objective optimization module collaborate to plan the delivery route; the dynamic replanning module and the temperature prediction and control module enable real-time monitoring and dynamic adjustment during the delivery process; the human-computer interaction module facilitates human-computer collaboration; and the data storage module provides persistent data storage and model parameter management. Each module performs its specific function while cooperating with others to jointly achieve multi-objective dynamic optimization of the cold chain logistics delivery route.

[0031] This invention designs a multi-objective path dynamic optimization method based on cold chain logistics, including steps such as data acquisition and initialization, cargo damage prediction, multi-objective optimization solution, dynamic replanning, temperature prediction and refrigeration control, and data archiving. The specific implementation process is as follows: Step S1 involves data acquisition and initialization. Before the delivery task begins, the system first collects data on the distribution center location, customer node location, customer demand, customer time window, temperature-sensitive parameters for each product category, vehicle configuration parameters, real-time traffic network data, and meteorological data. The distribution center location and customer node location are represented by geographic coordinates, which can be obtained through a geographic information system (GIS). Customer demand represents the quantity of goods to be delivered to each customer node. The customer time window represents the earliest and latest acceptable arrival time for each customer; exceeding this time range will result in service quality degradation or penalty costs. The temperature-sensitive parameters for each product category are used for subsequent damage prediction calculations, including the quality degradation characteristics of that product category under different temperature conditions. Vehicle configuration parameters include the vehicle's maximum load capacity, cargo compartment volume, temperature zone configuration type, refrigeration unit rated power, and power adjustment range. Real-time traffic network data is obtained through a traffic data service interface, including the road network topology and real-time travel times for each road segment. Meteorological data is obtained through a meteorological data service interface, including the current ambient temperature of the delivery area and temperature forecasts for future periods. The system performs missing value imputation, outlier removal, and format standardization on the collected raw data to generate a standardized delivery task dataset, providing a data foundation for subsequent calculations.

[0032] Step S2 is cargo damage prediction. Goods transported in cold chain logistics are typically temperature-sensitive products, whose quality deteriorates with changes in transportation time and temperature conditions. Accurately predicting the quality status of goods upon arrival at each customer node is crucial for optimizing delivery routes. This invention employs a cargo damage prediction model that considers both temperature deviation and temperature fluctuation. Compared to traditional models that only consider absolute temperature values, this model better reflects the impact of temperature instability on cargo quality during actual cold chain transportation.

[0033] Step S21 obtains the category-sensitive parameters of the currently transported goods. These parameters include the activation energy of quality decay. Quality degradation pre-exponential factor Optimal storage temperature Temperature deviation sensitivity coefficient and temperature fluctuation sensitivity coefficient Activation energy for quality degradation The activation energy reflects the sensitivity of goods to temperature changes; different types of goods have different activation energy values, which can be obtained through experimental measurement or by consulting relevant literature. Quality degradation pre-exponential factor. Related to the inherent characteristics of the goods. Optimal storage temperature. This is the ideal storage temperature for this category of goods, at which the rate of quality degradation is lowest. Temperature deviation sensitivity coefficient. This characterizes the sensitivity of cargo quality to temperature deviations from optimal values. Temperature fluctuation sensitivity coefficient. It characterizes the sensitivity of cargo quality to frequent temperature fluctuations. Even if the average temperature is suitable, the quality of some cargo will decline more rapidly when the temperature fluctuates greatly.

[0034] Step S22: Obtain the current time from the vehicle temperature sensor. Real-time temperature of the carriage And calculate the standard deviation of the temperature data within the set time window prior to the current time. The standard deviation reflects the degree of temperature fluctuation within the time window; a larger standard deviation indicates more severe temperature fluctuations. The length of the time window can be set according to actual needs. A shorter time window reflects recent temperature fluctuations, while a longer time window reflects overall temperature stability.

[0035] Step S23: Calculate the quality decay rate at the current moment. The quality decay rate represents how quickly the quality of goods deteriorates per unit time, and is a core computational factor in cargo damage prediction. The quality decay rate calculation formula used in this invention is as follows: ;in, for The rate of quality degradation at any given time, expressed in units of 1 / h; This is the pre-exponential factor for quality degradation, in units of 1 / h; It is a natural exponential function; The activation energy for quality decay is expressed in J / mol. This is the universal gas constant, with a value of 8.314, and the unit is J / (mol・K); for Real-time temperature of the carriage at any given moment, in Kelvin (K). The temperature deviation sensitivity coefficient is a dimensionless parameter. For real-time temperature and optimal storage temperature The absolute difference, in units of K; The optimal storage temperature is expressed in Kelvin (K). is the temperature fluctuation sensitivity coefficient, which is a dimensionless parameter; for The standard deviation of temperature data within a time window is set before the specified time, in K. In this formula, Partly derived from the classic Arrhenius equation, it describes the effect of temperature on the rate of chemical reactions and is widely used in food science and medicine to describe the degradation characteristics of product quality with temperature changes. This is a temperature deviation correction term. When the real-time temperature deviates from the optimal temperature, this term is greater than 1, which increases the decay rate. The term is a temperature fluctuation correction term. When temperature fluctuations are large, this term is greater than 1, further increasing the decay rate. The introduction of these two correction terms makes the model more consistent with the actual situation of cold chain transportation, because there are temperature instability factors such as door opening and unloading, and equipment fluctuations during actual delivery.

[0036] Step S24: Based on the estimated arrival time of each customer node in the delivery route plan. Integrate the quality decay rate over time to calculate the arrival time of the goods. Loss rate at each customer node The damage rate represents the percentage loss in quality of goods relative to their initial quality. The calculation formula is as follows: ;in, For the goods to arrive at the Loss rate per customer node, in percentage. It is a natural exponential function; Represents the time from the start time 0 to the arrival time. Definite integral operations; for The rate of quality degradation at any given time, expressed in units of 1 / h; To reach the The time for each client node is expressed in hours (h). In actual calculations, since temperature data are discrete sampled values, definite integrals can be approximated using numerical integration methods, such as the trapezoidal rule or Simpson's rule.

[0037] Step S25 outputs the predicted damage rate for each customer node to the multi-objective optimization module as input data for calculating the objective function of damage cost. A higher damage rate means a greater damage cost, and the optimization algorithm will tend to select delivery routes that can reduce the damage rate.

[0038] Step S3 involves solving a multi-objective optimization problem. Cold chain logistics delivery route optimization involves multiple conflicting objectives, such as the contradiction between reducing costs and improving service quality, and the trade-off between reducing carbon emissions and shortening delivery time. Traditional single-objective optimization methods struggle to balance multiple objectives. This invention employs a multi-objective optimization method, simultaneously considering three objectives: total delivery cost, customer satisfaction, and carbon emissions. By solving for the Pareto optimal solution set, it provides decision-makers with multiple alternatives.

[0039] Step S31: Construct the first objective function The objective is to minimize total delivery costs. Total delivery costs consist of four components: transportation costs, refrigeration costs, damage costs, and time window penalty costs. The calculation formula is as follows: ;in, Total delivery cost, in yuan; This refers to transportation costs, expressed in yuan, which are related to travel distance and fuel consumption. The cost is for refrigeration, expressed in yuan, and is related to the operating time and power of the refrigeration equipment. This is the cost of goods damage, expressed in yuan, calculated based on the damage rate and the unit price of the goods. This is the penalty cost for breach of the time window, expressed in yuan, incurred when the actual arrival time exceeds the customer's time window.

[0040] Step S32: Construct the second objective function The goal is to maximize customer satisfaction. Customer satisfaction is influenced by a combination of arrival time and product quality; high customer satisfaction occurs with on-time arrival and good product quality, while low customer satisfaction occurs with delayed arrival or deteriorating product quality. The calculation formula is as follows: ;in, Average customer satisfaction is a dimensionless value. Total number of customers; For the first The customer satisfaction score can be calculated using a fuzzy membership function based on arrival time and goods quality. Indicates to From 1 to Sum.

[0041] Step S33: Construct the third objective function The goal is to minimize carbon emissions. Carbon emissions from cold chain logistics mainly come from two sources: fuel consumption by vehicles and electricity consumption by refrigeration equipment. The calculation formula is as follows: ;in, Total carbon emissions, in kg; This is the fuel carbon emission coefficient, expressed in kg / L, representing the amount of carbon dioxide equivalent produced by burning one liter of fuel. The total number of vehicles; This represents the total number of road segments; Fuel consumption per unit distance when unloaded, in L / km; This represents the increase in fuel consumption per unit load per unit distance, expressed in L / (kg・km), reflecting the impact of load on fuel consumption. For vehicles The current load capacity, in kg; For the first The distance between the road segments, in km; As a decision variable, when the vehicle After the first The value is 1 if the road segment is valid, otherwise the value is 0. The carbon emission coefficient for cooling power consumption is expressed in kg / kWh and is related to the power generation structure of the power grid. For vehicles The total power consumption for cooling is expressed in kWh. Indicates to From 1 to Summation; Indicates to From 1 to Summation. In this formula This partially reflects the linear relationship between vehicle fuel consumption and load; the greater the load, the higher the fuel consumption.

[0042] Step S34: Randomly generated A feasible solution is used as the initial population. Each individual uses an integer code to represent the delivery route, where the sequence of integers in the code represents the order in which customer nodes are visited. Feasible solutions must satisfy basic constraints such as vehicle capacity, time window, and temperature. Population size... The choice of solution requires a trade-off between computational efficiency and solution quality. Too small a solution may result in insufficient search space coverage, while too large a solution will increase computation time.

[0043] Step S35 calculates three objective function values ​​for each individual in the population and performs fast non-dominated sorting to divide the individuals into different Pareto rank levels. Non-dominated sorting is a core concept in multi-objective optimization; if an individual... Not inferior to individuals in all goals And at least one objective is superior to the individual. Then it is called Dominate Individuals not dominated by any other individual are called non-dominated solutions, forming the first-level Pareto front. After removing individuals from the first level, the remaining non-dominated solutions form the second level, and so on, until all individuals are layered. The time complexity of the fast non-dominated sorting algorithm is O(n log n). ,in The number of objective functions. Population size.

[0044] Step S36: Calculate the adaptive weighted crowding distance for each individual within the same Pareto level. Crowding distance measures the density of solutions around an individual; individuals with large crowding distances are located in sparse regions of the solution space, and retaining these individuals helps maintain the diversity of the solution set. This invention employs an adaptive weighting mechanism to assign dynamic weights to different objectives, enabling the algorithm to automatically adjust the importance of each objective based on the current distribution of the solution set. The calculation formula is as follows: ;in, For the first The crowding distance between individuals is a dimensionless value; This represents the summation of three objective functions; For the first The adaptive weights of each objective are dimensionless values. For the first After sorting the target, the first The next neighbor of each individual One target value; For the first After sorting the target, the first The first neighbor of each individual One target value; For the current population The maximum value of each objective; For the current population The minimum value of each objective. Adaptive weights. The coefficient of variation of each target value can be dynamically calculated. Targets with a large coefficient of variation indicate that the solution set is not evenly distributed in the direction of the target, and should be given a larger weight to promote the expansion in that direction.

[0045] Step S37 employs a binary tournament selection operator to select parent individuals based on Pareto rank and crowding distance. In each round of the binary tournament selection, two individuals are randomly selected for comparison; the individual with the lower Pareto rank wins. If the ranks are the same, the individual with the greater crowding distance wins, and the winning individual enters the parent pool. Simulated binary crossover and polynomial mutation are performed on the selected parent individuals to generate offspring individuals. Simulated binary crossover simulates the effect of single-point crossover in binary encoding, controlling the similarity between offspring and parents through a distribution exponent parameter. Polynomial mutation perturbs the gene values ​​of individuals; the perturbation magnitude is controlled by the distribution exponent parameter, which helps increase population diversity and escape local optima.

[0046] Step S38 merges the parent population with the offspring population to form a population of size [missing information]. A temporary population was selected. Priority was given to Pareto rank, and within the same rank, those with greater crowding distance were prioritized from the temporary population. Individuals enter the next generation of the population. This elite preservation strategy ensures that superior individuals are not lost in the evolutionary process, while maintaining population diversity through crowding distance selection.

[0047] Step S39 determines whether the maximum number of iterations has been reached. If it has, output all individuals in the current population with a Pareto level of 1 as the Pareto optimal solution set; otherwise, return to step S35 to continue iteration. The maximum number of iterations needs to be determined based on the problem size and computational resources, and can be adjusted by observing the convergence of the objective function value.

[0048] Step S310 uses the TOPSIS method to select the compromise optimal solution from the Pareto optimal solution set as the delivery route plan. TOPSIS is a multi-attribute decision-making method, short for Approximation of Ideal Solution Ranking. This method calculates the distance between each candidate solution and the positive ideal solution and the negative ideal solution, and selects the solution with the highest relative proximity as the final plan. The positive ideal solution is a virtual solution where all objectives are optimal, and the negative ideal solution is a virtual solution where all objectives are worst. The greater the relative proximity, the closer the solution is to the positive ideal solution and the farther it is from the negative ideal solution, representing the best overall compromise plan. Decision-makers can also select other plans from the Pareto optimal solution set based on actual preferences.

[0049] Step S4 involves dynamic replanning. During actual delivery, traffic conditions, weather conditions, and vehicle status may change, and the original delivery route may no longer be optimal or even infeasible. This invention establishes a dynamic replanning mechanism to monitor the delivery execution status in real time. When the deviation from expectations exceeds a certain level, route replanning is automatically triggered to re-optimize the remaining undelivered nodes, enabling the system to adapt to dynamic environments.

[0050] Step S41: Collect the current vehicle location and cabin temperature from the on-board terminal every preset sampling period. External ambient temperature and vehicle speed data are collected. The sampling period needs to balance real-time monitoring and communication resource consumption; it is typically set to several minutes. Simultaneously, real-time traffic congestion index of the road segment ahead of the vehicle is obtained from the traffic data interface. The traffic congestion index reflects the degree of road congestion; the higher the value, the more congested the road.

[0051] Step S42: Calculate the normalized temperature deviation. Normalization maps temperature deviations to a range of 0 to 1, facilitating comprehensive comparisons with other indicators. The calculation formula is as follows: ;in, This represents the normalized temperature deviation, which is a dimensionless value. Current temperature of the carriage With the target temperature setpoint The absolute difference, in °C; The upper limit of the permissible temperature for goods, in °C; This is the lower limit of the permissible temperature for goods, expressed in °C. It refers to the temperature inside the cargo compartment when it is exactly equal to the target temperature. The value is 0 when the temperature deviates from the allowable range boundary. It is close to 0.5.

[0052] Step S43 calculates the estimated arrival time based on the current location and real-time traffic conditions. And calculate the normalized delay time. The estimated arrival time can be calculated based on the remaining distance and real-time traffic conditions. The calculation formula is as follows: ;in, The normalized delay time is a dimensionless value; To maximize the function, ensure the result is non-negative; Estimated arrival time, in minutes; The latest arrival time requested by the customer, in minutes; The allowable delay tolerance time, in minutes. This applies when the estimated arrival time is earlier than the latest arrival time. The value is 0 when the delay time reaches the tolerance time. The value is 1.

[0053] Step S44: Calculate the normalized road condition change index. This indicator reflects the degree of change in actual road conditions relative to the assumptions made during planning. The calculation formula is as follows: ;in, The normalized road condition change index is a dimensionless value. This represents the current actual traffic congestion index; The traffic congestion index preset during the planning stage; This represents the maximum traffic congestion index. When the actual congestion level matches the preset value... A value of 0 indicates that the road is more congested than expected. It is a positive value.

[0054] Step S45: Calculate the dynamic replanning trigger index value. This indicator comprehensively considers three factors: temperature deviation, time delay, and changes in road conditions, and obtains a single trigger criterion through weighted summation. The calculation formula is as follows: ;in, The replanning trigger indicator value is a dimensionless value; Temperature deviation weighting coefficient; This is the weighting coefficient for delay time; Road condition weighting coefficient; Normalized temperature deviation; Normalization caused delays; This is a normalized road condition change index. The weighting coefficients can be set according to the importance of each factor in actual business operations; the sum of all weights is recommended to be 1 for easier threshold setting. For temperature-sensitive goods, the weights can be appropriately increased. For orders with high time sensitivity, the quantity can be appropriately increased. .

[0055] Step S46: Determine the trigger indicator value Does it exceed the preset threshold? Furthermore, the duration exceeds the preset minimum number of cycles. Setting a minimum number of cycles is to avoid frequent replanning due to short-term fluctuations; replanning is only triggered when the abnormal situation persists. If the triggering condition is met, proceed to step S47; otherwise, return to step S41 to continue monitoring.

[0056] Step S47 marks the nodes that have completed delivery as fixed nodes and constructs subproblems for the remaining undelivered nodes. Fixed nodes do not participate in the re-optimization calculation. This incremental replanning strategy has less computation and faster response speed compared to global replanning, making it suitable for real-time scheduling scenarios.

[0057] Step S48 employs an adaptive large-scale neighborhood search algorithm to solve the subproblem. This algorithm is a metaheuristic that searches the solution space by alternately performing destruction and repair operations. From the set of destruction operators, a roulette wheel selection method is used to remove some nodes. This set includes random removal, worst-case removal, Shaw similarity removal, and path removal. From the set of repair operators, a roulette wheel selection method is used to reinsert the removed nodes. This set includes greedy insertion, regret insertion, and random insertion. A simulated annealing criterion is used to determine whether to accept a new solution. If the new solution is better than the current solution, it is accepted directly; if it is worse, it is accepted with a certain probability, which decreases as the number of iterations increases. The selection weights of each operator are dynamically adjusted based on their performance in historical iterations, with operators that perform well having a higher probability of being selected. After iterating to the maximum number of iterations or exceeding a preset time limit, a re-optimized path solution is output.

[0058] Step S49 updates the re-optimized route plan to the vehicle terminal, guiding the driver to follow the new route through the navigation interface. Simultaneously, the route change information is synchronized to the dispatch center, allowing dispatchers to monitor delivery progress.

[0059] Step S5 involves temperature prediction and refrigeration control. Traditional refrigeration control uses a feedback control method, adjusting the refrigeration power only when the temperature exceeds the set range, resulting in control lag. This invention employs a feedforward control strategy based on temperature prediction, utilizing a deep learning model to predict future temperature changes in the carriage and adjust the refrigeration power in advance, reducing temperature fluctuations and lowering the risk of cargo damage and energy consumption.

[0060] Step S51: Construct the input feature vector for the temperature prediction model Input feature vector Including history The data includes the internal temperature sequence of the carriage, the external ambient temperature sequence, the cooling power sequence, the vehicle speed sequence, the door opening / closing status sequence, and the current cargo load at each time step. Carriage temperature is affected by multiple factors: external ambient temperature influences internal temperature through heat conduction via the carriage walls; vehicle speed affects the intensity of air convection heat transfer; door opening / closing leads to the exchange of hot and cold air; cargo load affects heat capacity; and cooling power determines cooling capacity. Using these factors as input features helps improve prediction accuracy. (Historical time steps) The choice of ...

[0061] Step S52 inputs the feature vector Input a pre-trained long short-term memory neural network model, output the future The sequence of predicted temperatures for the carriage at each time step is used, and the first predicted value is taken as the predicted temperature. Long Short-Term Memory (LSTM) neural networks are a variant of recurrent neural networks (RNNs). They address the vanishing gradient problem of traditional RNNs by introducing a gating mechanism and excel at handling long-term dependencies in time-series data. The model needs to be trained offline using historical delivery data and then deployed to the system for online inference after training.

[0062] Step S53: Calculate the predicted temperature deviation The predicted temperature deviation is the core basis for adjusting cooling power. A positive deviation indicates that the predicted temperature is higher than the target temperature, requiring increased cooling; a negative deviation indicates that the predicted temperature is lower than the target temperature, allowing for reduced cooling power. The calculation formula is as follows: ;in, To predict temperature deviation, the unit is °C; Temperature is predicted, and the unit is °C. Set the target temperature value in °C.

[0063] Step S54: Calculate the environment and vehicle speed correction coefficients. The actual cooling effect of the refrigeration system is affected by the ambient temperature and vehicle speed. Higher ambient temperatures increase the heat load, while higher vehicle speeds enhance air convection heat transfer. A correction factor is used to compensate for these factors. The calculation formula is as follows: ;in, This is a correction factor, and it is a dimensionless value. is the environmental temperature influence coefficient, which is a dimensionless parameter; The current ambient temperature is expressed in °C. The ambient temperature is for reference only, and the unit is °C. The speed influence coefficient is a dimensionless parameter. Current vehicle speed, in km / h; This is a reference speed, expressed in km / h. The correction factor is 1 when both ambient temperature and vehicle speed are equal to the reference value. The correction factor is adjusted accordingly when the ambient temperature is higher than the reference value or the vehicle speed deviates from the reference value.

[0064] Step S55: Calculate the optimal cooling power output value. This invention employs a PID control strategy based on predicted temperature deviation. PID control is the most widely used control algorithm in industrial control, achieving precise control of the controlled variable through a combination of proportional, integral, and derivative components. The calculation formula is as follows: ;in, This represents the optimal cooling power output value, expressed in kW. Basic cooling capacity, measured in kW; The proportional control coefficient is a dimensionless parameter that determines the strength of the response to the current deviation. To predict temperature deviation, the unit is °C; is the integral control coefficient, a dimensionless parameter used to eliminate steady-state error; This is the time integral value of the temperature deviation, in °C·s; is the differential control coefficient, a dimensionless parameter used to suppress the trend of deviation change; The rate of change of temperature deviation is expressed in °C / s. These are the correction coefficients. The proportional term provides control action proportional to the deviation; the larger the deviation, the stronger the control action. The integral term accumulates the deviation, gradually increasing the control action until the deviation is eliminated, even if the deviation is small but persistent. The derivative term responds to the trend of deviation changes, providing control action in advance to reduce overshoot when the deviation changes rapidly. The three control coefficients can be tuned using engineering tuning methods or self-tuning algorithms.

[0065] Step S56 determines the optimal cooling power output value. The power setting is sent to the refrigeration unit controller, which adjusts the actual output power of the refrigeration unit. The refrigeration unit controller then adjusts the compressor speed or regulating valve opening based on the received power setting to achieve continuous power regulation.

[0066] Step S57 If the predicted temperature If the temperature exceeds the permissible range for the goods, a temperature warning will be sent to the dispatch center and the driver's terminal. The warning information includes the current temperature, the predicted temperature, and the estimated time of exceeding the limit, prompting relevant personnel to pay attention and take necessary measures, such as checking the operating status of refrigeration equipment or adjusting the delivery plan.

[0067] Step S6 is data archiving. After a delivery task is completed, the system records the actual delivery route, the entire temperature curve, the actual cargo damage, and energy consumption data. It calculates the key performance indicators (KPIs) for the delivery task and stores the data in the historical database. Historical data can be used to analyze delivery operation efficiency, assess service quality, and identify problem areas. Simultaneously, historical data also serves as a data source for calibrating the cargo damage prediction model parameters and iteratively training the temperature prediction neural network model. By continuously accumulating data and optimizing the model, system performance is continuously improved.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-objective path dynamic optimization method based on cold chain logistics, characterized in that, Includes the following steps: S1. Data Acquisition and Initialization: Collect data on distribution center location, customer node location, customer demand, customer time window, temperature-sensitive parameters of goods category, vehicle configuration parameters, real-time traffic network data, and meteorological data. Preprocess the collected data to generate a delivery task dataset. S2. Cargo Damage Prediction: Based on the temperature-sensitive parameters of cargo categories and real-time temperature data, a cargo damage prediction model that considers temperature deviation and temperature fluctuation is adopted to calculate the quality decay rate of various types of cargo and the predicted cargo damage rate when they arrive at each customer node. S3. Multi-objective optimization solution: With the optimization objectives of minimizing total delivery cost, maximizing customer satisfaction, and minimizing carbon emissions, and with constraints of vehicle capacity, time window, and temperature, an improved non-dominated sorting genetic algorithm based on adaptive weighted congestion distance is used to solve the problem, generate a Pareto optimal solution set, and select a delivery route plan from it to be sent to the vehicle terminal for execution. S4. Dynamic Replanning: During the delivery process, real-time data on vehicle location, compartment temperature, and road traffic status are collected to calculate the dynamic replanning trigger index value. When the trigger index value exceeds the preset threshold, an adaptive large-scale neighborhood search algorithm is used to perform incremental local re-optimization on the remaining undelivered nodes to update the delivery route plan. S5. Temperature Prediction and Cooling Control: Based on historical temperature data and real-time operating data, the system uses a long short-term memory neural network to predict the cabin temperature within a set time period in the future. The system calculates the optimal cooling power output value based on the predicted temperature deviation to achieve feedforward control of the cooling system. S6. Data Archiving: After the delivery task is completed, record the actual delivery route, the temperature curve throughout the process, the actual damage to goods and energy consumption data, and store them in the historical database for model parameter optimization.

2. The multi-objective path dynamic optimization method based on cold chain logistics according to claim 1, characterized in that, Step S2 includes the following sub-steps: S21. Obtain the category-sensitive parameters of the currently transported goods, including the activation energy of quality degradation. Quality degradation pre-exponential factor Optimal storage temperature Temperature deviation sensitivity coefficient Temperature fluctuation sensitivity coefficient ; S22. Obtain the current time from the vehicle temperature sensor. Real-time temperature of the carriage Calculate the standard deviation of temperature data within a set time window prior to the current moment. ; S23. Calculate the quality decay rate at the current moment. The calculation formula is: ;in, for The rate of quality degradation at any given time (1 / h); The quality degradation pre-exponential factor (1 / h); It is a natural exponential function; The activation energy for quality decay (J / mol); The universal gas constant is 8.314 (J / (mol·K)). for Real-time temperature (K) of the carriage at any given moment; This refers to the temperature deviation sensitivity coefficient. For real-time temperature and optimal storage temperature The absolute difference (K); The optimal storage temperature (K); This is the temperature fluctuation sensitivity coefficient; for Set the standard deviation (K) of the temperature data within the time window before the specified time. S24. Based on the estimated arrival time of each customer node in the delivery route plan. Integrate the quality decay rate over time to calculate the arrival time of the goods. Loss rate at each customer node The calculation formula is: ;in, For the goods to arrive at the Loss rate (%) at each customer node; It is a natural exponential function; Represents the time from the start time 0 to the arrival time. Definite integral operations; for The rate of quality degradation at any given time (1 / h); To reach the Time (h) for each customer node; S25. Output the predicted loss rate of each customer node to the multi-objective optimization module for calculating the objective function of loss cost.

3. The multi-objective path dynamic optimization method based on cold chain logistics according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31. Construct the first objective function With the goal of minimizing total delivery costs, the calculation formula is as follows: ;in, Total delivery cost (RMB); Transportation cost (yuan); Refrigeration cost (RMB); Cost of damaged goods (yuan); Cost of penalty for breach of contract within the time window (RMB); S32. Constructing the second objective function With the goal of maximizing customer satisfaction, the calculation formula is as follows: ;in, Average customer satisfaction; Total number of customers; For the first Customer satisfaction score; Indicates to From 1 to Summation; S33. Constructing the third objective function With the goal of minimizing carbon emissions, the calculation formula is as follows: ;in, Total carbon emissions (kg); Fuel carbon emission coefficient (kg / L); The total number of vehicles; This represents the total number of road segments; Fuel consumption per unit distance when unloaded (L / km); Fuel consumption increment per unit load per unit distance (L / (kg·km)); For vehicles Current load capacity (kg); For the first Distance of the road segment (km); As a decision variable, vehicle After the first The value is 1 if the road segment is valid, otherwise the value is 0. The carbon emission coefficient for refrigeration power consumption (kg / kWh); For vehicles Total power consumption for cooling (kWh); Indicates to From 1 to Summation; Indicates to From 1 to Summation; S34, Randomly generated A feasible solution is used as the initial population, and each individual uses an integer code to represent the delivery path; S35. Calculate three objective function values ​​for each individual in the population, and perform fast non-dominated sorting to divide the individuals into different Pareto rank levels. S36. Calculate the adaptive weighted crowding distance for each individual within the same Pareto level. The calculation formula is: ;in, For the first The crowded distance between individuals; This represents the summation of three objective functions; For the first Adaptive weights for each objective; For the first After sorting the target, the first The next neighbor of each individual One target value; For the first After sorting the target, the first The first neighbor of each individual One target value; For the current population The maximum value of each objective; For the current population The minimum value of each objective; S37. Use a binary tournament selection operator to select parent individuals based on Pareto rank and crowding distance, and perform simulated binary crossover and polynomial mutation to generate offspring individuals. S38. Merge the parent and offspring populations, selecting populations based on Pareto rank priority, and then prioritizing those with greater crowding distance within the same rank. Individuals enter the next generation of the population; S39. Determine whether the maximum number of iterations has been reached. If it has, output the Pareto optimal solution set; otherwise, return to step S35 to continue iterating. S310. Use the TOPSIS method to select the compromise optimal solution from the Pareto optimal solution set as the delivery route scheme.

4. The multi-objective path dynamic optimization method based on cold chain logistics according to claim 1, characterized in that, Step S4 includes the following sub-steps: S41. Collect the current vehicle location and passenger compartment temperature from the vehicle terminal every preset sampling period. The system collects data on external ambient temperature and vehicle speed, and obtains the real-time traffic congestion index of the road ahead from a traffic data interface. ; S42. Calculate the normalized temperature deviation. The calculation formula is: ;in, Normalized temperature deviation; Current temperature of the carriage With the target temperature setpoint The absolute difference (°C); The upper limit of the permissible temperature for goods (°C); The lower limit of the permissible temperature for goods (°C); S43. Calculate the estimated arrival time based on the current location and real-time traffic conditions. Calculate normalized delay time The calculation formula is: ;in, Normalization caused delays; This is a function to find the maximum value. Estimated arrival time (min); The latest arrival time (in minutes) requested by the customer; The allowable delay tolerance time (min); S44. Calculate the Normalized Road Condition Change Index The calculation formula is: ;in, This is a normalized road condition change index; This represents the current actual traffic congestion index; The traffic congestion index preset during the planning stage; This represents the maximum value of the traffic congestion index. S45. Calculate the dynamic replanning trigger index value. The calculation formula is: ;in, The replanning trigger indicator value; Temperature deviation weighting coefficient; This is the weighting coefficient for delay time; Road condition weighting coefficient; Normalized temperature deviation; Normalization caused delays; This is a normalized road condition change index; S46. Determine the trigger indicator value Does it exceed the preset threshold? If the duration exceeds the preset minimum number of cycles, then proceed to step S47; otherwise, return to step S41 to continue monitoring. S47. Mark the nodes that have completed delivery as fixed nodes, and construct subproblems for the remaining undelivered nodes; S48. Adaptive large-scale neighborhood search algorithm is used to solve the subproblem: select a destruction operator from the set of destruction operators in a roulette wheel manner to remove some nodes, select a repair operator from the set of repair operators in a roulette wheel manner to re-insert the removed nodes, determine whether to accept the new solution according to the simulated annealing criterion, and output the re-optimized path scheme after iterating until the termination condition is met. S49. Update the optimized route plan to the vehicle terminal to guide the driver to perform delivery according to the new route.

5. The multi-objective path dynamic optimization method based on cold chain logistics according to claim 1, characterized in that, Step S5 includes the following sub-steps: S51. Constructing the input feature vector for the temperature prediction model The input feature vector Including history The time step includes the internal temperature sequence of the carriage, the external ambient temperature sequence, the cooling power sequence, the vehicle speed sequence, the door opening and closing status sequence, and the current cargo load. S52, Input feature vector Input a pre-trained long short-term memory neural network model, output the future The predicted temperature of the carriage at each time step is taken as the predicted temperature. ; S53, Calculate the predicted temperature deviation The calculation formula is: ;in, To predict temperature deviation (°C); To predict the temperature (°C); Set the target temperature (°C); S54. Computational Environment and Vehicle Speed ​​Correction Coefficient The calculation formula is: ;in, This is a correction factor; The environmental temperature influence coefficient; The current ambient temperature (°C); For reference ambient temperature (°C); This is the coefficient affecting vehicle speed; Current vehicle speed (km / h); For reference vehicle speed (km / h); S55. Calculate the optimal cooling power output value. The calculation formula is: ;in, The optimal cooling power output value (kW); Based on the cooling capacity (kW); This is the proportional control coefficient; To predict temperature deviation (°C); These are integral control coefficients; The time integral value of the temperature deviation (°C·s); These are the differential control coefficients; The rate of change of temperature deviation (°C / s); This is a correction factor; S56, Optimal cooling power output value Send the signal to the refrigeration unit controller to adjust the actual output power of the refrigeration unit; S57, If the predicted temperature If the temperature exceeds the permissible range for the cargo, a temperature warning message will be sent to the dispatch center and the driver's terminal.

6. A multi-objective path dynamic optimization system based on cold chain logistics, used to implement the method according to any one of claims 1 to 5, characterized in that, include: The data acquisition module includes an on-board data acquisition unit and an external data interface unit. The on-board data acquisition unit is installed on the delivery vehicle and is electrically connected to a temperature sensor inside the vehicle compartment, an ambient temperature sensor outside the vehicle compartment, a GPS positioning module, a vehicle speed sensor, and a door status sensor, respectively, for collecting data on vehicle compartment temperature, ambient temperature, vehicle position, vehicle speed, and door status. The external data interface unit is communicatively connected to a traffic data server and a meteorological data server, respectively, for acquiring road network data, real-time traffic data, and meteorological data. The data processing module is communicatively connected to the data acquisition module and is used to perform missing value imputation, outlier removal, and format unification processing on the acquired raw data to generate a standardized delivery task dataset. The cargo damage prediction module is communicatively connected to the data processing module and includes a parameter storage unit, a temperature statistics unit, and an attenuation calculation unit; the parameter storage unit is used to store temperature-sensitive parameters for each category of cargo. The temperature statistics unit is communicatively connected to the data acquisition module and is used to calculate the standard deviation of the temperature data. The attenuation calculation unit is communicatively connected to the parameter storage unit and the temperature statistics unit, respectively, and is used to calculate the quality attenuation rate and the loss rate. The multi-objective optimization module, communicatively connected to the data processing module and the cargo loss prediction module, includes an objective function calculation unit, a non-dominated sorting unit, a crowding distance calculation unit, and a genetic operation unit. The objective function calculation unit calculates three objective function values: total delivery cost, customer satisfaction, and carbon emissions. The non-dominated sorting unit, communicatively connected to the objective function calculation unit, performs rapid non-dominated sorting of the population. The crowding distance calculation unit, also communicatively connected to the non-dominated sorting unit, calculates adaptive weighted crowding distance. The genetic operation unit, communicatively connected to the crowding distance calculation unit, performs selection, crossover, and mutation operations and outputs a Pareto optimal solution set. The dynamic replanning module, which is communicatively connected to the data acquisition module, includes an index calculation unit, a trigger judgment unit, and a local re-optimization unit; the index calculation unit is used to calculate the normalized temperature deviation, normalized delay time, normalized road condition change index, and replanning trigger index value. The triggering judgment unit is communicatively connected to the indicator calculation unit and is used to determine whether the triggering condition is met; the local re-optimization unit is communicatively connected to the triggering judgment unit and the multi-objective optimization module respectively and is used to perform incremental local re-optimization on the remaining undelivered nodes. The temperature prediction and control module, communicatively connected to the data acquisition module, includes a feature extraction unit, a neural network inference unit, and a power calculation unit. The feature extraction unit extracts input feature vectors from historical data. The neural network inference unit, communicatively connected to the feature extraction unit, incorporates a long short-term memory neural network model to output predicted temperature values. The power calculation unit, communicatively connected to the neural network inference unit, calculates the optimal cooling power output value. The power calculation unit, communicatively connected to the refrigeration unit controller, sends cooling power control commands. The human-computer interaction module is communicatively connected to the multi-objective optimization module and the dynamic replanning module, and includes a dispatcher management terminal and a driver mobile terminal. The dispatcher management terminal is used to display delivery route plans and real-time monitoring information; the driver mobile terminal is set in the cab of the delivery vehicle and is wirelessly connected to the dynamic replanning module to receive navigation routes and early warning information. The data storage module is communicatively connected to the data processing module, the cargo damage prediction module, and the temperature prediction and control module, respectively, and is used to store historical delivery data and model parameters.

7. The multi-objective path dynamic optimization system based on cold chain logistics according to claim 6, characterized in that, In the cargo damage prediction module: The parameter storage unit is communicatively connected to the data storage module and reads the activation energy of quality decay, the pre-exponential factor of quality decay, the optimal storage temperature, the temperature deviation sensitivity coefficient, and the temperature fluctuation sensitivity coefficient of each type of goods from the data storage module. The temperature statistics unit receives real-time temperature data from the vehicle-mounted data acquisition unit, calculates the temperature standard deviation according to a set time window, and outputs it to the attenuation calculation unit. The attenuation calculation unit calculates the quality attenuation rate and the damage rate based on the received temperature-sensitive parameters, real-time temperature, and temperature standard deviation, and outputs the damage rate to the objective function calculation unit of the multi-objective optimization module.

8. The multi-objective path dynamic optimization system based on cold chain logistics according to claim 6, characterized in that, In the multi-objective optimization module: The objective function calculation unit receives road network distance data and customer time window data from the data processing module, and receives cargo loss rate data from the cargo loss prediction module. It calculates the objective function value of total delivery cost, the objective function value of customer satisfaction, and the objective function value of carbon emissions, and outputs the three objective function values ​​to the non-dominated sorting unit. The non-dominated sorting unit performs fast non-dominated sorting of all individuals in the population according to three objective function values, and outputs the sorting results to the crowding distance calculation unit. The crowding distance calculation unit calculates the adaptive weighted crowding distance of each individual within the same Pareto level based on the sorting results, and outputs the calculation results to the genetic operation unit. The genetic operation unit performs selection operations based on Pareto level and crowding distance, performs crossover and mutation operations on the selected individuals to generate offspring population, and outputs the final Pareto optimal solution set to the human-computer interaction module.

9. The multi-objective path dynamic optimization system based on cold chain logistics according to claim 6, characterized in that, In the dynamic replanning module: The index calculation unit receives vehicle temperature and vehicle location data from the vehicle data acquisition unit, receives real-time traffic congestion index from the external data interface unit, calculates normalized temperature deviation, normalized delay time, and normalized road condition change index, and calculates replanning trigger index value according to preset weight coefficients, and outputs the trigger index value to the trigger judgment unit. The triggering judgment unit compares the received triggering index value with a preset threshold and determines whether the duration of exceeding the threshold exceeds a preset minimum number of cycles. When the triggering condition is met, it sends a replanning start signal to the local re-optimization unit. After receiving the replanning start signal, the local re-optimization unit obtains the current delivery route plan from the multi-objective optimization module, performs incremental local re-optimization on the remaining undelivered nodes using an adaptive large-scale neighborhood search algorithm, and sends the re-optimized route plan to the driver's mobile terminal through the human-machine interaction module.

10. The multi-objective path dynamic optimization system based on cold chain logistics according to claim 6, characterized in that, In the temperature prediction and control module: The feature extraction unit reads historical temperature data from the data storage module and receives real-time operating data from the vehicle data acquisition unit. It extracts input feature vectors including the compartment temperature sequence, ambient temperature sequence, cooling power sequence, vehicle speed sequence, door status sequence, and cargo load, and outputs the input feature vectors to the neural network inference unit. After receiving the input feature vector, the neural network inference unit performs inference calculations through the built-in long short-term memory neural network model and outputs the temperature prediction value to the power calculation unit. The power calculation unit calculates the predicted temperature deviation based on the received temperature prediction value and the target temperature set value, calculates the correction coefficient by combining the ambient temperature and vehicle speed data received from the vehicle data acquisition unit, calculates the optimal cooling power output value based on the predicted temperature deviation and the correction coefficient, and converts the optimal cooling power output value into a control command and sends it to the refrigeration controller. The power calculation unit is also connected to the human-machine interaction module. When the temperature prediction value exceeds the allowable range, it pushes temperature warning information to the dispatcher management terminal and the driver mobile terminal.

Images