Multi-mode AIGC cold-chain logistics path optimization decision-making method and system

Abstract

The invention relates to a multi-mode AIGC cold-chain logistics path optimization decision-making method and a multi-mode AIGC cold-chain logistics path optimization decision-making system. The method comprises the following steps: inputting position data, temperature data, road condition data and weather data in a cold-chain logistics system into an AIGC model of a stack auto-encoder for feature extraction and reconstruction error calculation to obtain a feature index set; performing space-time interval division and similarity matching on the cold chain distribution path to obtain an interval reference path library; inputting a hierarchical control model for stage processing to obtain an initial control parameter; an objective function including temperature constraint, time window constraint and vehicle dynamics constraint is constructed for solving, and a path optimization scheme is obtained; performing road condition random disturbance and temperature fluctuation influence analysis to obtain a path risk scoring index; and performing feedback correction and iterative optimization on the initial control parameter to obtain a target control parameter. According to the invention, the temperature control precision is ensured, and the distribution efficiency and the system reliability are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cold chain logistics path optimization, and in particular to a multi-modal AIGC cold chain logistics path optimization decision method and system. Background Art

[0002] With the rapid development of the cold chain logistics industry, the problem of distribution route optimization is becoming increasingly complex. Traditional path planning methods are difficult to take into account multiple goals such as temperature control, timeliness and transportation safety at the same time. Especially in the complex and changeable actual distribution environment, uncertain factors such as road conditions, weather effects and temperature fluctuations have brought huge challenges to cold chain distribution. Existing similarity matching methods are prone to introduce pseudo-similar paths during path planning, resulting in low prediction accuracy, and it is difficult to effectively deal with the time delay of the wire control steering system and the nonlinear characteristics of the vehicle.

[0003] The current cold chain logistics distribution system generally has the problem of insufficient utilization of multi-source heterogeneous data, and it is difficult to effectively integrate multimodal information such as location, temperature, road conditions and weather. At the same time, the existing control methods often adopt a single control strategy and cannot dynamically adjust the path characteristics under different working conditions, resulting in unsatisfactory control effects. Traditional optimization algorithms have low computational efficiency when dealing with multiple constraints and are difficult to meet real-time requirements. In terms of path risk assessment, existing methods lack systematic analysis of random disturbances in road conditions and temperature fluctuations, and cannot accurately predict and respond to potential risks. Especially in extreme weather or sudden changes in road conditions, the system's robustness and adaptability are insufficient, which can easily lead to problems such as delivery delays or abnormal temperature control. These problems seriously restrict the improvement of cold chain logistics distribution efficiency. Summary of the invention

[0004] The main purpose of the present invention is to provide a multimodal AIGC cold chain logistics path optimization decision-making method and system, which effectively improves the distribution efficiency and system reliability while ensuring the temperature control accuracy.

[0005] To achieve the above object, the present invention provides a multi-modal AIGC cold chain logistics path optimization decision method, comprising the following steps: The location data, temperature data, road condition data and weather data in the cold chain logistics system are input into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature index set; According to the characteristic indicator set, the cold chain distribution path is divided into time and space intervals and similarity matched to obtain an interval reference path library; The interval reference path library is input into a hierarchical control model, and steady-state path planning, dynamic compensation and steering control are hierarchically processed to obtain initial control parameters; Based on the initial control parameters, an objective function including temperature constraints, time window constraints and vehicle dynamics constraints is constructed, and a path optimization solution is obtained by solving the objective function through obstacle term transformation; Performing random disturbance of road conditions and temperature fluctuation impact analysis on the path optimization scheme to obtain a path risk score index; Feedback correction and iterative optimization are performed on the initial control parameters according to the path risk scoring index to obtain target control parameters.

[0006] The present invention also provides a multi-modal AIGC cold chain logistics path optimization decision system, comprising: The feature extraction module is used to input the location data, temperature data, road condition data and weather data in the cold chain logistics system into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature index set; A matching module, used to perform time-space interval division and similarity matching on the cold chain distribution path according to the characteristic indicator set to obtain an interval reference path library; A hierarchical module, used for inputting the interval reference path library into a hierarchical control model, hierarchically processing steady-state path planning, dynamic compensation and steering control, and obtaining initial control parameters; A solution module, used for constructing an objective function including temperature constraints, time window constraints and vehicle dynamics constraints based on the initial control parameters, solving the objective function through obstacle term transformation, and obtaining a path optimization solution; An analysis module, used to analyze the impact of random disturbances in road conditions and temperature fluctuations on the path optimization scheme to obtain a path risk score index; The optimization module is used to perform feedback correction and iterative optimization on the initial control parameters according to the path risk scoring index to obtain target control parameters.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention realizes the deep fusion and feature extraction of multimodal data through the AIGC model of the stacked autoencoder, thereby improving the efficiency of data utilization; adopts a similarity matching method based on time-space interval division to effectively avoid the interference of pseudo-similar engines; innovatively introduces a hierarchical control model to realize the coordinated optimization of steady-state path planning, dynamic compensation and steering control; through obstacle term transformation and progressive solution strategy, the solution efficiency of multi-constraint optimization problems is significantly improved; based on the analysis of the impact of random disturbances in road conditions and temperature fluctuations, a complete risk assessment mechanism is established; feedback correction and iterative optimization mechanisms are adopted to realize dynamic optimization and precise adjustment of control parameters. This method effectively improves the distribution efficiency and system reliability while ensuring the accuracy of temperature control. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the steps of a multi-modal AIGC cold chain logistics path optimization decision method in one embodiment of the present invention; Figure 2 It is a structural block diagram of a multi-modal AIGC cold chain logistics path optimization decision system in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 , this embodiment provides a multi-modal AIGC cold chain logistics path optimization decision method, including the following steps: S1, input the location data, temperature data, road condition data and weather data in the cold chain logistics system into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature index set; Among them, data collection and preprocessing are performed. In the cold chain logistics system, location data, temperature data, road condition data and weather data are key input elements. To ensure the effective combination of these multimodal data, they are normalized to the maximum and minimum values ​​to unify the value range of each data and avoid analysis deviations caused by different dimensions. Time series alignment is performed to ensure that data from different sources can correspond in the time dimension to form a standardized multimodal input matrix. The standardized multimodal input matrix is ​​input into the first layer encoder of the stacked autoencoder. In this process, 128 3×3 convolution kernels and a maximum pooling layer with a step size of 2 are used for feature dimensionality reduction. The convolution operation helps to extract local features, while the pooling layer effectively reduces the feature dimension while maintaining important information. After this processing, the first layer feature map is generated, and then the LeakyReLU activation function is used to enhance the nonlinear features and mine the potential features in the data, making the subsequent analysis more accurate. The first layer feature map is input into the second layer encoder. In this layer, 256 3×3 convolution kernels and a maximum pooling layer with the same stride of 2 are applied for feature compression. The LeakyReLU activation function and batch normalization are used to improve the performance of features and the stability of training, and the second-layer feature map is obtained. After being processed by two layers of encoders, the dimension of the data is effectively compressed, and the expressiveness of the features is significantly enhanced. The second-layer feature map is input into the generator network of the AIGC model. The design of the generator network contains 8 residual blocks, each of which is composed of two layers of 512 3×3 convolution kernels, and a self-attention mechanism is introduced between the residual blocks to enhance the model's ability to focus on important features. The generated feature vector is input into the first-layer decoder. In this layer, 256 3×3 deconvolution kernels and an upsampling layer with a stride of 2 are used for feature reconstruction. Through the deconvolution operation, the features are gradually restored to a higher dimension, and residual connections are made with the second-layer feature map. The residual connection helps to retain more information, making the reconstruction process more stable and effective. After being processed by the ReLU activation function, the first-layer reconstructed features are obtained. The first layer of reconstructed features are input to the second layer decoder. 128 3×3 deconvolution kernels and an upsampling layer with the same step size of 2 are applied to restore the features, and residual connections are made with the first layer feature maps. Finally, after being processed by the ReLU activation function, the reconstructed data matrix is ​​obtained. The weighted combination of the L1 norm and L2 norm between the reconstructed data matrix and the standardized multimodal input matrix is ​​calculated to obtain the feature reconstruction error. The evaluation of the reconstruction loss directly reflects the performance of the model and the effectiveness of the generated feature vector. Based on this reconstruction error, the generated feature vector is subjected to principal component analysis and denoising to optimize the quality of the feature index set.

[0014] S2, divide the cold chain distribution path into time and space intervals and perform similarity matching according to the characteristic indicator set to obtain the interval reference path library; Specifically, the feature indicator set is segmented in two dimensions of time window and space grid. The feature indicator set is converted into a sequence of spatiotemporal feature blocks, each of which has a clear division in time and space, so as to reflect the spatiotemporal characteristics of the distribution path. Based on the sequence of spatiotemporal feature blocks, the spatiotemporal correlation matrix of the distribution path is constructed. By calculating the path distance and analyzing the time overlap rate, a set of path segmentation points is obtained. These segmentation points are key nodes of the distribution path and help to identify the boundaries of different path segments. The set of path segmentation points is density calculated and locally clustered to obtain the initial interval boundary value. Cluster analysis is performed on the data to determine the boundaries of each path segment. The cold chain distribution path is segmented according to the initial interval boundary value, and the feature mean and variance of each path segment are calculated to obtain the interval feature vector. The interval feature vector contains the main information of each path segment and provides basic data for subsequent similarity matching. The similarity of the interval feature vector is calculated. Based on the matching of distance threshold and time window overlap, these indicators can be used to identify which paths are similar in time and space to form a similar path group. In order to refine the analysis, similar path groups are grouped according to spatiotemporal coverage, and each group of data is statistically analyzed to obtain an interval path template, which reflects the general characteristics of the path under specific conditions. The interval path template is feature extracted and standardized to ensure the availability and consistency of the template in different application scenarios. Through feature extraction, important features in the path template can be mined, thereby providing more effective information in subsequent decision-making. Standardization ensures data consistency, allowing path templates from different sources to be compared and analyzed under the same framework. Based on the processed reference path feature set, multi-level index construction and data organization are performed to finally form an interval reference path library. The interval reference path library contains analyzed and processed path features, and can provide a fast retrieval function in practical applications, thereby supporting path optimization decisions for cold chain logistics.

[0015] S3, inputting the interval reference path library into the hierarchical control model, performing hierarchical processing on steady-state path planning, dynamic compensation and steering control, and obtaining initial control parameters; It should be noted that the interval reference path library is input into the steady-state control layer of the hierarchical control model. The path features are decomposed by the steady-state path planner, and the trajectory smoothing is implemented to eliminate the sharp changes and irregularities in the path, so as to obtain the steady-state reference trajectory. By analyzing and optimizing the path features, the generated reference trajectory is ensured to be feasible and safe in practical applications. The steady-state reference trajectory is input into the variable steering ratio control unit of the hierarchical control model. The steady-state planning control quantity is obtained by calculating the front and rear wheel steering angle ratio and analyzing it in combination with the road adhesion constraint. This analysis not only considers the dynamic characteristics of the vehicle, but also fully considers the influence of the road conditions on the vehicle driving to ensure the rationality and effectiveness of the steady-state planning control quantity. The steady-state planning control quantity is input into the dynamic compensation layer of the hierarchical control model. The vertical load change is compensated and calculated by the segmented affine tire model to ensure the stability and maneuverability of the vehicle under different load conditions, and the dynamic compensation control quantity is obtained. The dynamic compensation control quantity is input into the adaptive secondary adjustment unit. In this unit, the compensation correction control quantity is obtained by compensating for the road adhesion conditions and correcting the yaw rate. The compensation correction control quantity is input into the tilt steering control layer of the hierarchical control model. The steering control quantity is obtained by calculating the yaw angle tracking error and analyzing the lateral stability. This analysis ensures the stability and control accuracy of the vehicle during driving by evaluating the steering state and driving trajectory of the vehicle to adapt to various complex road conditions. After completing the calculation of the above control quantities, the steady-state planning control quantity, dynamic compensation control quantity and steering control quantity are input into the inter-layer coordinator of the hierarchical control model. The inter-layer coordinator performs coordination processing according to the weight distribution of each control quantity to achieve an effective balance between different control objectives. The core of the coordination processing is to optimize the overall control strategy to ensure that the vehicle can be controlled in the best way under various dynamic conditions, and finally obtain the initial control parameters.

[0016] S4, constructing an objective function including temperature constraints, time window constraints and vehicle dynamics constraints based on the initial control parameters, solving it through obstacle term transformation, and obtaining a path optimization solution; Specifically, the temperature dynamic response calculation is performed on the initial control parameters, and the temperature change of the refrigerated box is modeled through the temperature transfer model to obtain the temperature constraint function. The temperature constraint function is converted into a logarithmic obstacle term, and the upper and lower limits of the temperature are mathematically described by the extrusion function to obtain the temperature obstacle term. By setting the obstacle term, the temperature is ensured to fluctuate within a safe range to avoid exceeding the requirements of the refrigeration conditions. Based on the requirements of the delivery time window, the time constraints of the initial control parameters are constructed. The time penalty function is used to quantify the delay and advance conditions to obtain the time window constraint function to ensure that the logistics distribution is carried out within the specified time range to avoid the decline in customer satisfaction or cargo loss caused by delays. The time window constraint function is converted into an exponential obstacle term, and the boundary of the time window is softened by smoothing to form a time obstacle term. Softening can improve the flexibility of the optimization process and make it easier to adjust the path when approaching the boundary. The vehicle dynamics model is calculated for the initial control parameters. The vehicle motion is described in detail by the tire mechanics model and the center of mass dynamics model to obtain the dynamic constraint function. This function can reflect the dynamic characteristics of the vehicle during actual driving, including speed, acceleration, and steering characteristics. Through the calculation of this model, the actual handling ability and movement restrictions of the vehicle are taken into account during the path optimization process to avoid unrealistic path planning. The temperature barrier term, time barrier term and dynamic constraint function are weighted and combined to form a combined objective function. The objective function comprehensively considers the influence of temperature control, time constraints and vehicle dynamics, and hierarchically processes each constraint through priority sorting to ensure that the influence of each factor can be reasonably balanced during the optimization process. The combined objective function is optimized and calculated in real time. The constraint boundary is shrunk by the progressive barrier tightening method, so that the system gradually approaches the optimal solution. The optimization process takes into account the control parameters in the current state and continuously adjusts them according to historical data and real-time feedback to ensure the effectiveness of path selection in a dynamic environment. The path nodes are reconstructed and connected according to the optimal control sequence, and finally the path optimization solution is obtained.

[0017] S5, analyzing the impact of random disturbance of road conditions and temperature fluctuation on the path optimization scheme to obtain a path risk score index; Among them, the probability distribution modeling of road condition changes is carried out according to the route optimization scheme. The core of the modeling process is to determine the possibility of different road condition changes, and generate a road condition disturbance distribution matrix based on this, which effectively describes the types and frequencies of road condition changes that may occur in different time periods. Based on the road condition disturbance distribution matrix, a road condition disturbance sequence is generated to reflect various road condition changes that may be encountered in the actual distribution process, such as traffic congestion, road construction and weather changes. The vehicle dynamic response calculation of the route optimization scheme is carried out based on the road condition disturbance sequence, taking into account the driving performance of the vehicle under different road conditions, and analyzing how the dynamic performance of the vehicle changes under these disturbances. At the same time, the fluctuation characteristics of the temperature control system are analyzed to evaluate the fluctuation range of the temperature control system in the cold chain logistics under the influence of road condition changes. The temperature fluctuation range is input into the Markov jump model, and the transfer characteristics of the temperature fluctuation are calculated by the state transition probability. The Markov model can help understand the probability distribution of temperature changes under different states and their evolution over time. The calculated temperature response curve shows the temperature change trend in the cold chain logistics process under different road condition disturbances and external environmental changes. On this basis, the temperature response curve is segmented and integrated to quantify the risk of temperature exceeding the limit. It helps to identify whether the temperature will exceed the safety limit under different circumstances, and finally obtains a temperature risk index, providing a quantitative analysis of the temperature control risk in cold chain logistics. A multi-factor coupling risk model is constructed based on the road disturbance sequence and the temperature risk index. Through this model, the impact of road condition changes and temperature fluctuations on the logistics process is comprehensively considered. In order to analyze the propagation characteristics of risks, the Bayesian network is used to calculate the risk propagation characteristics. The Bayesian network can dynamically describe and analyze the interaction between different factors through the dependency relationship between nodes to form a risk propagation matrix. The risk propagation matrix is ​​subjected to characteristic decomposition to extract the risk characteristic vector. Based on the risk characteristic vector, a multi-dimensional comprehensive evaluation calculation can comprehensively consider the influence of different factors and finally obtain the path risk score index.

[0018] S6, feedback correction and iterative optimization are performed on the initial control parameters according to the path risk score indicators to obtain the target control parameters.

[0019] Specifically, a risk adjustment matrix is ​​generated according to the path risk score index to reflect the degree of influence of different risk factors on the initial control parameters. By analyzing the risk score index, the parameters that need to be adjusted in actual operation are determined to ensure the safety and efficiency of the path. According to the risk adjustment matrix, the steady-state planning control quantity in the initial control parameters is corrected and calculated, and the steady-state control quantity is adjusted to make it more suitable for the current risk situation, and the steady-state parameter correction value is obtained. By combining the risk adjustment matrix with the steady-state planning control quantity, the adjustment range and direction required for the steady-state control quantity under different risk conditions can be effectively identified. Similarly, based on the risk adjustment matrix, the dynamic compensation control quantity in the initial control parameters is compensated and corrected to obtain the dynamic parameter correction value, ensuring that in the dynamic compensation control process, the influence of the external environment and road conditions on the vehicle movement can be fully considered, and then the dynamic compensation strategy is adjusted so that the vehicle can still maintain good controllability and stability in different scenarios. According to the risk adjustment matrix, the steering control quantity in the initial control parameters is corrected for deviation to obtain the steering parameter correction value. The goal of the correction is to reduce the safety hazards caused by inaccurate steering control, ensure the steering stability of the vehicle during driving, and avoid accidents. The steady-state parameter correction value, dynamic parameter correction value and steering parameter correction value are input into the inter-layer coordinator for weight distribution. The weights of each correction parameter are reasonably distributed to ensure that each factor can be fully valued and reasonably adjusted in the process of forming the final control parameters. By coordinating the generation of correction parameters, an effective balance between different control quantities is achieved, forming a control scheme that comprehensively considers all factors. Constraint condition verification and iterative calculation are performed on the coordinated correction parameters. Verify whether the coordinated correction parameters meet the predetermined constraints, such as temperature range, time window and vehicle dynamics restrictions, to ensure that the optimized control parameters can still meet all operating requirements in actual applications and avoid potential risks caused by parameter non-compliance. The target control parameters are determined by convergence judgment, that is, after multiple iterative calculations, if the target control parameters tend to be stable within a certain range, it is considered that the optimization goal has been achieved.

[0020] In one example, the location data, temperature data, road condition data and weather data in the cold chain logistics system are input into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature indicator set, including: collecting the location data, temperature data, road condition data and weather data in the cold chain logistics system, and performing maximum and minimum value normalization and time alignment processing on the location data, temperature data, road condition data and weather data to obtain a standardized multimodal input matrix; inputting the standardized multimodal input matrix into the first layer encoder of the stacked autoencoder, performing feature dimensionality reduction through 128 3×3 convolution kernels and a maximum pooling layer with a step size of 2, and processing through the LeakyReLU activation function to obtain a first layer feature map; inputting the first layer feature map into the second layer encoder, performing feature compression through 256 3×3 convolution kernels and a maximum pooling layer with a step size of 2, and processing through the LeakyReLU activation function and batch normalization to obtain a second layer feature map; The figure is input into the generator network of the AIGC model. The generator network contains 8 residual blocks, each residual block consists of two layers of 512 3×3 convolution kernels, and a self-attention mechanism is added between the residual blocks to obtain a generated feature vector; the generated feature vector is input into the first-layer decoder, and the feature is reconstructed through 256 3×3 deconvolution kernels and an upsampling layer with a step size of 2, and a residual connection is made with the second-layer feature map, and after processing with the ReLU activation function, the first-layer reconstructed feature is obtained; the first-layer reconstructed feature is input into the second-layer decoder, and the feature is restored through 128 3×3 deconvolution kernels and an upsampling layer with a step size of 2, and a residual connection is made with the first-layer feature map, and after processing with the ReLU activation function, the reconstructed data matrix is ​​obtained; the weighted combination reconstruction loss of the L1 norm and the L2 norm is calculated for the reconstructed data matrix and the standardized multimodal input matrix to obtain the feature reconstruction error, and the generated feature vector is subjected to principal component analysis and denoising according to the feature reconstruction error to obtain a feature indicator set.

[0021] In this example, relevant information is collected from multiple data sources, including location data, temperature data, traffic data, and weather data. These data are multimodal and have different properties and formats. In subsequent processing, maximum and minimum value normalization and time series alignment are performed to ensure that all types of data are analyzed at the same scale. Maximum and minimum value normalization refers to linearly converting data to the [0,1] interval. The specific formula is: ; in, Represents the original data, and are the minimum and maximum values ​​of the data, respectively. After normalization, the units and ranges of various types of data are unified, which is convenient for subsequent analysis and processing. Timing alignment ensures that the timestamps of various types of data are consistent so that they can be compared and analyzed within the same time window. The standardized multimodal input matrix is ​​input into the first layer encoder of the stacked autoencoder. The structure of the encoder includes 128 3×3 convolution kernels, combined with a maximum pooling layer with a step size of 2 for feature dimensionality reduction. The calculation formula for the convolution operation is: ; in, is a pixel value of the output feature map, is the input data, is the convolution kernel, and are the height and width of the convolution kernel respectively. After convolution, the feature map is processed by the LeakyReLU activation function. The formula of LeakyReLU is: in, The value is usually 0.01, which means that there is a small slope for negative values ​​to prevent the "death" of neurons. The first layer feature map is input to the second layer encoder. The second layer encoder performs feature compression through 256 3×3 convolution kernels and a maximum pooling layer with a stride of 2. The process also includes the application of the LeakyReLU activation function and batch normalization. The formula for batch normalization is: ; in, and are the mean and variance of the current batch, respectively. is a small constant used to prevent division by zero errors. The second layer feature map is input into the generator network of the AIGC model. The generator network consists of 8 residual blocks, each of which contains two layers of 512 3×3 convolution kernels. In the construction of the residual block, the self-attention mechanism is introduced to improve the network's ability to focus on features. The residual block inputs The output obtained by convolution Add together to form a new output: ; This structure enables the network to retain input information during learning, which helps prevent gradient disappearance. The generator network outputs the generated feature vector. The generated feature vector is input to the first layer decoder for feature reconstruction, and the first layer decoder is processed by 256 3×3 deconvolution kernels and an upsampling layer with a stride of 2. The deconvolution operation is regarded as the inverse operation of the input feature map. Its calculation formula is similar to that of convolution, but the weights use the inverse convolution kernel. The reconstructed features are residually connected with the second layer feature map and processed by the ReLU activation function to obtain the first layer of reconstructed features. The first layer of reconstructed features are input to the second layer decoder for feature restoration. This layer is processed by 128 3×3 deconvolution kernels and an upsampling layer with a stride of 2, and residually connected with the first layer feature map, and then through the ReLU activation function to obtain the reconstructed data matrix. The reconstructed data matrix is ​​compared with the standardized multimodal input matrix, and the reconstruction loss of the weighted combination is calculated using the L1 norm and L2 norm. The calculation formula for the reconstruction loss is: ; in, is the normalized multimodal input matrix, is the reconstructed data matrix, and is the weight coefficient. By calculating the reconstruction loss, the feature reconstruction error is obtained, and the principal component analysis and noise reduction processing are performed on the generated feature vector according to this error to extract effective features and finally obtain the feature index set.

[0022] In one example, the cold chain distribution path is divided into time and space intervals and similarity matched according to the feature indicator set to obtain an interval reference path library, including: performing two-dimensional segmentation of the feature indicator set in time window and space grid to obtain a time and space feature block sequence, and constructing a time and space correlation matrix of the distribution path based on the time and space feature block sequence, and obtaining a set of path segmentation points by calculating the path distance and the time overlap rate; performing density calculation and local aggregation on the set of path segmentation points to obtain an initial interval boundary value, and segmenting the cold chain distribution path according to the initial interval boundary value, and calculating the characteristic mean and variance of each path segment to obtain an interval feature vector; performing similarity calculation on the interval feature vector, matching based on the distance threshold and the time window overlap to obtain similar path groups, and grouping the similar path groups according to the time and space coverage, performing statistical analysis on each group of data to obtain an interval path template; performing feature extraction and standardization on the interval path template to obtain a reference path feature set, and performing multi-level index construction and data organization based on the reference path feature set to obtain an interval reference path library.

[0023] In this example, the feature indicator set is segmented in two dimensions: time window and space grid. The data is segmented by time and space to form a sequence of spatiotemporal feature blocks. The size of the time window is set according to business needs. For example, it is set to 30 minutes, while the space grid divides the distribution area into small grids. Assume that each grid is 1 km x 1 km. The two-dimensional segmentation method enables the data to reflect changes in time and space, which is convenient for subsequent analysis and calculation. For example, the time window is set to , the spatial grid is , then the characteristic index set is represented as a matrix , where each row corresponds to a time window and a spatial grid, expressed as: ; in, Indicates time window and the Feature data in a spatial grid, is the number of time windows, is the number of spatial grids. Through this structure, a spatiotemporal association matrix is ​​constructed to associate the features in the spatiotemporal feature block sequence, reflecting the relationship between the features. The path segmentation point set is obtained by calculating the path distance and the time overlap rate. The path distance calculation uses the Euclidean distance formula: ; in, and are the coordinates of each point in the path. The time overlap rate is obtained by calculating the degree of overlap between different time windows. The overlap rate is expressed as: ; Through these two calculation methods, we get the path segmentation point set , contains key segmentation points to help with subsequent analysis. Density calculation and local aggregation are performed on the path segmentation point set to determine the initial interval boundary value. Density calculation uses Gaussian kernel density estimation, expressed as: ; in, is the Gaussian kernel function, is the bandwidth parameter, is each sample point. The density function can help identify dense areas in the data set and determine the initial interval boundary value. According to the initial interval boundary value, the cold chain distribution path is segmented and the characteristic mean and variance of each segment are calculated. The characteristic mean and variance of each segment are combined into a new interval feature vector, which is expressed as: ; The similarity of interval feature vectors is calculated and matched based on the distance threshold and time window overlap. The similarity is calculated by cosine similarity, and the formula is: ; in, and are two feature vectors respectively. Based on this similarity, similar path groups are obtained, and these path groups are grouped according to the temporal and spatial coverage. Each group of data is statistically analyzed to obtain the interval path template. The reference path feature set is formed by feature extraction and standardization of the interval path template. The standardization process uses the Z-score standardization method: ; in, is the original data, is the mean, is the standard deviation. The final reference path feature set forms a structured data set that can be used for subsequent path optimization and decision support. For example, suppose that in the feature indicator set, there are the following data: ; After the dual-dimensional segmentation processing of time window and space grid, a sequence of spatiotemporal feature blocks is formed, and in the path analysis, the path segmentation point set is obtained as , where each segmentation point represents an important state change or feature transition. Based on these segmentation points, the spatiotemporal correlation matrix of the distribution path is effectively constructed to provide data support for subsequent path optimization.

[0024] In one example, an interval reference path library is input into a hierarchical control model, and steady-state path planning, dynamic compensation and steering control are hierarchically processed to obtain initial control parameters, including: inputting the interval reference path library into a steady-state control layer of the hierarchical control model, performing path feature decomposition and trajectory smoothing processing through a steady-state path planner to obtain a steady-state reference trajectory, and inputting the steady-state reference trajectory into a variable steering ratio control unit of the hierarchical control model to obtain a steady-state planning control amount through front and rear wheel steering angle ratio calculation and road adhesion constraint analysis; inputting the steady-state planning control amount into a dynamic compensation layer of the hierarchical control model, and performing stepwise processing through stepwise processing. The segment affine tire model performs compensation calculation on the vertical load change to obtain the dynamic compensation control quantity, and the dynamic compensation control quantity is input into the adaptive secondary adjustment unit of the hierarchical control model, and the compensation correction control quantity is obtained through road adhesion condition compensation and yaw rate correction; the compensation correction control quantity is input into the tilt steering control layer of the hierarchical control model, and the steering control quantity is obtained through yaw angle tracking error calculation and lateral stability analysis; the steady-state planning control quantity, dynamic compensation control quantity and steering control quantity are input into the inter-layer coordinator of the hierarchical control model, and coordinated according to the weight distribution of each control quantity to obtain the initial control parameters.

[0025] In this example, the interval reference path library is input into the steady-state control layer of the hierarchical control model. The path feature decomposition and trajectory smoothing are performed by the steady-state path planner to eliminate noise and unnecessary fluctuations in the path. The path feature decomposition uses the wavelet transform method to represent the original path as multiple components of different frequencies. The formula is expressed as: ; in, is the original path data, are the wavelet coefficients, is the wavelet basis function. In this way, important path features are extracted and then the trajectory is smoothed, for example using a moving average filter: ; in, is the smoothed path, is the sliding window size. After processing, the steady-state reference trajectory is obtained. The steady-state reference trajectory is input into the variable steering ratio control unit of the hierarchical control model. The steering angle ratio of the front and rear wheels is calculated to ensure that the vehicle maintains the best path tracking ability when turning. The steering angle is calculated using the following formula: ; in, is the front wheel steering angle, is the wheelbase, is the turning radius. At the same time, the road adhesion constraint analysis is performed to ensure that the vehicle can be effectively controlled under various road conditions. The calculation of road adhesion is achieved by combining the friction coefficient and the vertical load, expressed as: ; in, is the lateral force, is the friction coefficient, is the vertical load. The steady-state planning control quantity is obtained through calculation. The steady-state planning control quantity is input into the dynamic compensation layer of the hierarchical control model. The vertical load change is compensated by the segmented affine tire model to cope with the decrease in handling performance caused by the load change. The expression of the affine model is expressed as: ; in, is the lateral force, is the lateral stiffness, is the lateral slip angle, is the compensation coefficient related to the load. Through the above compensation calculation, the dynamic compensation control amount is obtained. The dynamic compensation control amount is input into the adaptive secondary adjustment unit of the hierarchical control model. The dynamic compensation control amount is optimized through road adhesion condition compensation and yaw rate correction. The yaw rate correction is achieved by the following formula: ; in, is the yaw rate, is the vehicle speed. After adjustment, the compensation correction control quantity is obtained. The compensation correction control quantity is input into the tilt steering control layer of the hierarchical control model. At this stage, the yaw angle tracking error is calculated and the lateral stability analysis is performed to ensure the stability of the vehicle in a dynamic environment. The yaw angle tracking error is calculated by the difference between the expected value and the actual value, and the formula is: ; in, is the yaw angle error, is the desired yaw angle, is the actual yaw angle. Based on this error, the final steering control quantity is determined to ensure that the vehicle can accurately track the predetermined trajectory. The steady-state planning control quantity, dynamic compensation control quantity and steering control quantity are input into the inter-layer coordinator of the hierarchical control model. The weight distribution of each control quantity is coordinated to comprehensively consider the influence of various factors on the control performance. Assume that the weights are , and , then use the following formula to get the initial control parameters : ; in, , and They are steady-state planning control, dynamic compensation control and steering control. Through weighted combination, the initial control parameters are obtained, which can effectively cope with various dynamic changes in cold chain logistics.

[0026] In one example, an objective function including temperature constraints, time window constraints and vehicle dynamics constraints is constructed based on the initial control parameters, and is solved by obstacle term transformation to obtain a path optimization solution, including: calculating the temperature dynamic response of the initial control parameters, modeling the temperature change of the refrigerated box through a temperature transfer model to obtain a temperature constraint function, and converting the temperature constraint function into a logarithmic obstacle term, mathematically describing the upper and lower limits of the temperature through an extrusion function to obtain a temperature obstacle term; constructing time constraints for the initial control parameters based on the delivery time window requirements, quantifying delays and advances through a time penalty function to obtain a time window constraint function, and converting the time window constraint function into a logarithmic obstacle term. The function is converted into an exponential obstacle term, and the time window boundary is softened by smoothing to obtain the time obstacle term; the vehicle dynamics model is calculated for the initial control parameters, and the vehicle motion is described by the tire mechanics model and the center of mass dynamics model to obtain the dynamic constraint function, and the temperature obstacle term, time obstacle term and dynamic constraint function are weightedly combined, and each constraint is hierarchically processed by priority sorting to obtain the combined objective function; the combined objective function is optimized and calculated in real time iteratively, and the constraint boundary is shrunk by the progressive barrier tightening method to obtain the optimal control sequence, and the path nodes are reconstructed and connected according to the optimal control sequence to obtain the path optimization solution.

[0027] In this example, the temperature dynamic response calculation is performed on the initial control parameters, and the temperature change of the refrigerated box is described by the temperature transfer model. The temperature transfer model is expressed by the one-dimensional heat conduction equation as follows: ; In this equation, Represents the temperature inside the refrigerator. is the thermal diffusivity, is the position coordinate, is time. By solving this equation, we can obtain the temperature distribution at different times and spatial positions, and establish a temperature constraint function, which describes the constraints on temperature changes over time. The obtained temperature constraint function is expressed as: ; in, and These are the maximum and minimum temperature limits specified in cold chain logistics. The temperature constraint function is converted into a logarithmic barrier term to form a temperature barrier term to ensure that the temperature limit is reasonably applied. The form of the logarithmic barrier term is: ; The constraints of the upper and lower temperature limits are further described mathematically through the extrusion function, and the temperature barrier term is defined. The extrusion function is usually: ; Through this step, a quantified temperature barrier term is obtained. The time constraints of the initial control parameters are constructed based on the delivery time window requirements. In this process, the delay and advance conditions are quantified by the time penalty function, which is formalized as: ; in, and are the start and end time of the delivery respectively. The penalty function can effectively quantify the impact caused by time delay or advance, and obtain the time window constraint function. In order to make the time constraint adapt to the needs of the optimization algorithm, the time window constraint function is converted into an exponential obstacle term in the form of: ; in, is an adjustment factor that can control the penalty intensity. In order to optimize the boundary of the processing time window, smoothing is used to soften the boundary of the time window to ensure that no sudden changes in constraints are generated during the optimization process. The vehicle dynamics model is calculated for the initial control parameters, and the vehicle motion is described by the tire mechanics model and the center of mass dynamics model. The tire mechanics model is expressed as: ; in, is the lateral force, is the lateral stiffness, is the lateral slip angle. The dynamic description of the vehicle's center of mass is expressed using Newton's second law: ; in, is the vehicle mass, is the acceleration, is the total external force on the vehicle. Combining the above models, we get a constraint function that includes all dynamic characteristics. We weightedly combine the temperature barrier term, time barrier term and dynamic constraint function to construct a comprehensive objective function. The form of this combined function is: ; in, , and They are the weight coefficients of temperature, time and dynamic constraints, is a function describing the dynamic constraints. Each constraint is hierarchically processed by priority sorting to ensure that each constraint can reasonably reflect its importance during the optimization process. The combined objective function is optimized and calculated in real time. The progressive barrier tightening method is used to shrink the constraint boundary to ensure that the set limit will not be exceeded during the continuous optimization process. This process is achieved through the following iterative formula: ; in, It is The solution of the iteration, is the step length, is the gradient of the objective function. The optimal control sequence obtained by this method is used to reconstruct and connect the path nodes to obtain the final path optimization solution.

[0028] In one example, a path optimization scheme is analyzed for the impact of random disturbances in road conditions and temperature fluctuations to obtain a path risk score index, including: based on the path optimization scheme, a probability distribution model is performed on the road condition changes to obtain a road condition disturbance distribution matrix, and a road condition disturbance sequence is generated based on the road condition disturbance distribution matrix; based on the road condition disturbance sequence, a vehicle dynamic response calculation is performed on the path optimization scheme, and a fluctuation characteristic analysis is performed on the temperature control system to obtain a temperature fluctuation range; the temperature fluctuation range is input into a Markov jump model, the temperature fluctuation transfer characteristic is calculated by the state transition probability to obtain a temperature response curve, and the temperature response curve is segmented and integrated to quantify the temperature over-limit risk to obtain a temperature risk index; a multi-factor coupling risk model is constructed based on the road condition disturbance sequence and the temperature risk index, and the risk propagation characteristic is calculated by a Bayesian network to obtain a risk propagation matrix; a characteristic decomposition operation is performed on the risk propagation matrix to obtain a risk characteristic vector, and a multi-dimensional comprehensive evaluation calculation is performed based on the risk characteristic vector to obtain a path risk score index.

[0029] In this example, the probability distribution modeling of road condition changes is performed to obtain the relevant road condition disturbance distribution matrix. This process involves the analysis of historical road condition data and the use of statistical methods to establish a probability model, using models such as normal distribution or gamma distribution. For example, suppose that the travel time of a delivery route is observed to conform to the normal distribution through historical data, and its probability density function is expressed as: ; in, is the travel time, is the average travel time, is the standard deviation. In this way, a road condition disturbance distribution matrix is ​​generated, in which each element represents the probability distribution of driving time under different road conditions. This matrix not only includes the time factor, but also combines the road condition type (such as congestion, smooth traffic, etc.) to form a multi-dimensional disturbance distribution matrix. Based on the road condition disturbance distribution matrix, a road condition disturbance sequence is generated to reflect the possible changes in road conditions in different time periods or under different road conditions. The generation method includes using Monte Carlo simulation to extract sample data from the road condition disturbance distribution by random sampling. For example, the disturbance sequence is generated by the following formula: ; in, Indicates at time The disturbance of time, is the number of samples generated. Based on the generated road disturbance sequence, the vehicle dynamic response calculation is performed. This calculation involves the vehicle motion model, considering the performance of the vehicle under different road conditions. The vehicle dynamics model is expressed as: ; in, is the total force acting on the vehicle, is the vehicle mass, is the acceleration, is the friction force. This model is used to calculate the acceleration changes of the vehicle under different road conditions, and then analyze the fluctuation characteristics of the temperature control system. Especially in cold chain logistics, the stability of the temperature control system is closely related to the acceleration and deceleration changes of the vehicle during driving. The temperature variation range is obtained through experiments or simulations. For example, by observing the changes in the temperature of the refrigerated box at different speeds, the temperature fluctuation range is obtained: ; in, and Respectively represent the minimum and maximum values ​​of temperature fluctuation. The temperature fluctuation range is input into the Markov jump model, which can calculate the transfer characteristics of temperature fluctuation through the state transition probability. The Markov jump model is described by the state transfer matrix. Each element in Indicates from the state Transfer to state The state transition of temperature is described as: ; By calculating the state transition of temperature, a temperature response curve is obtained, which shows the change of temperature at different time points. The temperature response curve is segmented and integrated to quantify the risk of temperature exceeding the limit. The quantification of temperature exceeding the limit risk is expressed by the following formula: ; in, is the indicator function of temperature exceeding the limit. The final temperature risk index is It can effectively reflect the risk of temperature control in the cold chain logistics process. On this basis, a multi-factor coupling risk model is constructed, combining the road disturbance sequence and temperature risk indicators to comprehensively evaluate the path risk. Bayesian network provides an effective calculation method. By establishing the conditional probability relationship between nodes, the Bayesian network is expressed as: ; in, Represents the set of all influencing factors, pa yes The parent node set of . The risk propagation characteristics are calculated through the Bayesian network to obtain the risk propagation matrix. The elements in the matrix represent the degree of mutual influence between different risk sources. The risk propagation matrix is ​​subjected to eigendecomposition to extract the main risk characteristics. This process is implemented using singular value decomposition to obtain the risk feature vector. The risk feature vector after eigendecomposition is expressed as: ; in, and are the left and right singular vectors, respectively, is a singular value matrix. Based on the risk characteristic vector, a multi-dimensional comprehensive evaluation calculation is performed to obtain the path risk score index. The comprehensive evaluation model is expressed as: ; in, For each risk factor, is the corresponding risk value. The final path risk score index can provide a scientific basis for subsequent path decision-making.

[0030] In one example, feedback correction and iterative optimization are performed on the initial control parameters according to the path risk scoring index to obtain the target control parameters, including: generating a risk adjustment matrix according to the path risk scoring index; correcting and calculating the steady-state planning control quantity in the initial control parameters according to the risk adjustment matrix to obtain the steady-state parameter correction value; compensating and correcting the dynamic compensation control quantity in the initial control parameters based on the risk adjustment matrix to obtain the dynamic parameter correction value; performing deviation correction on the steering control quantity in the initial control parameters according to the risk adjustment matrix to obtain the steering parameter correction value; inputting the steady-state parameter correction value, the dynamic parameter correction value and the steering parameter correction value into the inter-layer coordinator for weight distribution to obtain the coordinated correction parameter; performing constraint condition checking and iterative calculation on the coordinated correction parameter, and determining the target control parameter through convergence judgment.

[0031] In this example, a risk adjustment matrix is ​​generated based on the path risk score indicators to reflect the impact of different risk factors on system performance. When generating the risk adjustment matrix, the path risk score indicators are standardized to convert risk factors of different dimensions into a unified measure. Through standardization, a risk adjustment matrix is ​​constructed. , whose elements represent the influence of each risk score on the adjustment of control parameters. The matrix is ​​expressed as: ; in, Indicates risk indicators Control parameters The influence coefficient of . According to the risk adjustment matrix, the initial control parameters are corrected. For the correction calculation of the steady-state planning control quantity, the formula is expressed as: ; in, is the corrected steady-state control quantity, is the initial control quantity, is the corresponding risk adjustment matrix element, is the standardized risk score. Through this step, the steady-state parameter correction value is obtained. Based on the same principle, a similar formula is used for the compensation correction of the dynamic compensation control amount: ; in, represents the risk-adjusted dynamic compensation control amount, It represents the influence coefficient of dynamic compensation in the risk adjustment matrix. Through this process, the dynamic parameter correction value is obtained. For the deviation correction of the steering control amount, a similar method is used, and the formula is: ; in, Indicates the corrected steering control amount, is the adjustment coefficient related to steering control. After calculation, the steady-state parameter correction value, dynamic parameter correction value and steering parameter correction value are obtained respectively. The above correction values ​​are input into the inter-layer coordinator for weight distribution. When allocating weights, the weights are set according to the specific needs of the system and historical data. Assume that the weights of steady-state, dynamic and steering control quantities are set as , and , the coordination correction parameters are expressed as: ; In this formula, represents the final coordinated correction parameter, which comprehensively considers the influence of different control quantities by weighted summation. Constraint check and iterative calculation are performed on the coordinated correction parameter to ensure that the obtained control parameter can meet the system operation constraints in practical applications. Assume that the system constraints are expressed by a function It means that the correction parameters are valid if and only if the value of this function is within the acceptable range. The constraints are defined as: ; According to this condition, adjust the parameters. If the condition is not met, adjust , which is realized through iterative calculation. The steps of iterative calculation are expressed as: ; in, For the The calibration parameters for the iterations, is the learning rate, is the gradient of the constraint function. Through continuous iteration, the target control parameter that satisfies the constraint condition is gradually approached. The target control parameter is determined by convergence judgment, and the parameter change is checked to see if it is lower than the preset threshold. For example, when When , the algorithm is considered to have converged. is the set minimum value. For example, suppose that in a cold chain transportation, the initial steady-state planning control amount is 100, the dynamic compensation control amount is 50, and the steering control amount is 30. After risk scoring, the risk adjustment matrix is ​​as follows: ; Assume that the standardized risk score is , , . Using the previous formula, correction is performed, and the corrected control amount is finally obtained as: 54; Similarly, the correction values ​​of dynamic compensation control and steering control are calculated. Through the weight distribution of the coordinator, the coordination correction parameters are obtained and constraint checks and iterations are performed until convergence.

[0032] Reference Figure 2 , this embodiment provides a multi-modal AIGC cold chain logistics path optimization decision system, including: Feature extraction module 1, used to input the location data, temperature data, road condition data and weather data in the cold chain logistics system into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature index set; Matching module 2 is used to divide the cold chain distribution path into time and space intervals and perform similarity matching according to the characteristic index set to obtain an interval reference path library; The hierarchical module 3 is used to input the interval reference path library into the hierarchical control model, perform hierarchical processing on the steady-state path planning, dynamic compensation and steering control, and obtain the initial control parameters; A solution module 4 is used to construct an objective function including temperature constraints, time window constraints and vehicle dynamics constraints based on initial control parameters, and solve the objective function through obstacle term transformation to obtain a path optimization solution; Analysis module 5, used to analyze the impact of random disturbance of road conditions and temperature fluctuation on the path optimization plan to obtain a path risk score index; The optimization module 6 is used to perform feedback correction and iterative optimization on the initial control parameters according to the path risk scoring index to obtain the target control parameters.

[0033] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0034] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0035] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0036] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0037] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0038] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, system, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, system, article or method. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, system, article or method including the element.

[0039] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A multi-modal AIGC cold chain logistics path optimization decision method, characterized in that: The following steps are involved: The location data, temperature data, road condition data and weather data in the cold chain logistics system are input into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature index set; According to the characteristic indicator set, the cold chain distribution path is divided into time and space intervals and similarity matching is performed to obtain an interval reference path library; The interval reference path library is input into a hierarchical control model, and steady-state path planning, dynamic compensation and steering control are hierarchically processed to obtain initial control parameters; Based on the initial control parameters, an objective function including temperature constraints, time window constraints and vehicle dynamics constraints is constructed, and a path optimization solution is obtained by solving the objective function through obstacle term transformation; Performing random disturbance of road conditions and temperature fluctuation impact analysis on the path optimization scheme to obtain a path risk score index; Feedback correction and iterative optimization are performed on the initial control parameters according to the path risk scoring index to obtain target control parameters.

2. The multimodal AIGC cold chain logistics path optimization decision method according to claim 1 is characterized in that: The location data, temperature data, road condition data and weather data in the cold chain logistics system are input into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature index set, including: Collecting location data, temperature data, road condition data and weather data in the cold chain logistics system, and performing maximum and minimum value normalization and time series alignment processing on the location data, temperature data, road condition data and weather data to obtain a standardized multimodal input matrix; Input the standardized multimodal input matrix into the first layer encoder of the stacked autoencoder, perform feature dimension reduction through 128 3×3 convolution kernels and a maximum pooling layer with a step size of 2, and process it through a LeakyReLU activation function to obtain a first layer feature map; The first-layer feature map is input into the second-layer encoder, and feature compression is performed through 256 3×3 convolution kernels and a maximum pooling layer with a stride of 2. After LeakyReLU activation function and batch normalization processing, the second-layer feature map is obtained. The second layer feature map is input into the generator network of the AIGC model. The generator network contains 8 residual blocks, each of which consists of two layers of 512 3×3 convolution kernels, and a self-attention mechanism is added between the residual blocks to obtain a generated feature vector. The generated feature vector is input into the first layer decoder, and the feature is reconstructed through 256 3×3 deconvolution kernels and an upsampling layer with a step size of 2, and a residual connection is performed with the second layer feature map, and a ReLU activation function is processed to obtain the first layer reconstructed feature; The first layer of reconstructed features are input into the second layer decoder, and the features are restored through 128 3×3 deconvolution kernels and an upsampling layer with a step size of 2. The features are residually connected with the first layer feature map and processed by the ReLU activation function to obtain the reconstructed data matrix. A weighted combined reconstruction loss of the L1 norm and the L2 norm is calculated for the reconstructed data matrix and the standardized multimodal input matrix to obtain a feature reconstruction error, and principal component analysis and noise reduction processing are performed on the generated feature vector according to the feature reconstruction error to obtain a feature index set.

3. The multimodal AIGC cold chain logistics path optimization decision method according to claim 2 is characterized in that: The step of performing time-space interval division and similarity matching on the cold chain distribution path according to the characteristic indicator set to obtain an interval reference path library includes: The characteristic indicator set is segmented in two dimensions, i.e., time window and space grid, to obtain a spatiotemporal characteristic block sequence, and a spatiotemporal correlation matrix of the delivery path is constructed based on the spatiotemporal characteristic block sequence, and a set of path segmentation points is obtained by calculating the path distance and the time overlap rate; Density calculation and local aggregation are performed on the path segmentation point set to obtain an initial interval boundary value, and the cold chain distribution path is segmented according to the initial interval boundary value, and the characteristic mean and variance of each path segment are calculated to obtain an interval characteristic vector; Calculate the similarity of the interval feature vectors, match them based on the distance threshold and the time window overlap to obtain similar path groups, group the similar path groups according to the spatiotemporal coverage, perform statistical analysis on each group of data, and obtain an interval path template; The interval path template is subjected to feature extraction and standardization processing to obtain a reference path feature set, and multi-level index construction and data organization are performed based on the reference path feature set to obtain an interval reference path library.

4. The multimodal AIGC cold chain logistics path optimization decision method according to claim 3 is characterized in that: The interval reference path library is input into the hierarchical control model, and steady-state path planning, dynamic compensation and steering control are processed hierarchically to obtain initial control parameters, including: The interval reference path library is input into the steady-state control layer of the hierarchical control model, and the path feature decomposition and trajectory smoothing processing are performed by the steady-state path planner to obtain a steady-state reference trajectory, and the steady-state reference trajectory is input into the variable steering ratio control unit of the hierarchical control model, and the steady-state planning control amount is obtained by front and rear wheel steering angle ratio calculation and road adhesion constraint analysis; The steady-state planning control amount is input into the dynamic compensation layer of the hierarchical control model, and the vertical load change is compensated and calculated by a piecewise affine tire model to obtain a dynamic compensation control amount, and the dynamic compensation control amount is input into the adaptive secondary adjustment unit of the hierarchical control model to obtain a compensation correction control amount by road adhesion condition compensation and yaw rate correction; The compensation correction control amount is input into the tilt steering control layer of the hierarchical control model, and the steering control amount is obtained through yaw angle tracking error calculation and lateral stability analysis; The steady-state planning control quantity, the dynamic compensation control quantity and the steering control quantity are input into the inter-layer coordinator of the hierarchical control model, and coordinated according to the weight distribution of each control quantity to obtain initial control parameters.

5. The multimodal AIGC cold chain logistics path optimization decision method according to claim 4 is characterized in that: The objective function including temperature constraint, time window constraint and vehicle dynamics constraint is constructed based on the initial control parameters, and is solved by obstacle term transformation to obtain a path optimization solution, including: Performing temperature dynamic response calculation on the initial control parameters, modeling the temperature change of the refrigerated box through a temperature transfer model to obtain a temperature constraint function, performing logarithmic obstacle term conversion on the temperature constraint function, mathematically describing the upper and lower limits of temperature through an extrusion function to obtain a temperature obstacle term; Based on the delivery time window requirement, the initial control parameters are time-constrained, the delay and advance are quantified by the time penalty function to obtain the time window constraint function, and the time window constraint function is converted into an exponential barrier term, and the time window boundary is softened by smoothing to obtain the time barrier term; Performing vehicle dynamics model calculation on the initial control parameters, describing the vehicle motion through tire mechanics model and center of mass dynamics model, obtaining a dynamics constraint function, performing weighted combination on the temperature barrier term, the time barrier term and the dynamics constraint function, performing hierarchical processing on each constraint through priority sorting, and obtaining a combined objective function; The combined objective function is optimized and calculated iteratively in real time, the constraint boundary is shrunk by a progressive barrier tightening method to obtain an optimal control sequence, and the path nodes are reconstructed and connected according to the optimal control sequence to obtain a path optimization solution.

6. The multimodal AIGC cold chain logistics path optimization decision method according to claim 5 is characterized in that: The path optimization scheme is subjected to random disturbance of road conditions and temperature fluctuation impact analysis to obtain a path risk score index, including: According to the path optimization scheme, a probability distribution model is performed on the road condition change to obtain a road condition disturbance distribution matrix, and a road condition disturbance sequence is generated based on the road condition disturbance distribution matrix; Based on the road condition disturbance sequence, the vehicle dynamic response calculation is performed on the path optimization scheme, and the fluctuation characteristic analysis is performed on the temperature control system to obtain the temperature fluctuation range; The temperature fluctuation range is input into the Markov jump model, the temperature fluctuation transfer characteristics are calculated by the state transition probability, a temperature response curve is obtained, and the temperature response curve is piecewise integrated to quantify the temperature over-limit risk and obtain a temperature risk index; A multi-factor coupling risk model is constructed according to the road condition disturbance sequence and the temperature risk index, and the risk propagation characteristics are calculated through a Bayesian network to obtain a risk propagation matrix; Performing a characteristic decomposition operation on the risk propagation matrix to obtain a risk characteristic vector, and performing a multi-dimensional comprehensive evaluation calculation based on the risk characteristic vector to obtain a path risk scoring index.

7. The multimodal AIGC cold chain logistics path optimization decision method according to claim 6 is characterized in that: The step of performing feedback correction and iterative optimization on the initial control parameters according to the path risk scoring index to obtain target control parameters includes: generating a risk adjustment matrix according to the pathway risk scoring indicators; Performing correction calculation on the steady-state planning control amount in the initial control parameter according to the risk adjustment matrix to obtain a steady-state parameter correction value; Based on the risk adjustment matrix, a dynamic compensation control amount in the initial control parameter is compensated and corrected to obtain a dynamic parameter correction value; Performing deviation correction on the steering control amount in the initial control parameter according to the risk adjustment matrix to obtain a steering parameter correction value; Inputting the steady-state parameter correction value, the dynamic parameter correction value and the steering parameter correction value into an inter-layer coordinator for weight distribution to obtain a coordination correction parameter; The coordination correction parameters are subjected to constraint condition checking and iterative calculation, and target control parameters are determined through convergence judgment.

8. A multi-modal AIGC cold chain logistics path optimization decision system, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the system comprises: The feature extraction module is used to input the location data, temperature data, road condition data and weather data in the cold chain logistics system into the AIGC model of the stacked autoencoder for feature extraction and reconstruction error calculation to obtain a feature index set; A matching module, used to perform time-space interval division and similarity matching on the cold chain distribution path according to the characteristic indicator set to obtain an interval reference path library; A hierarchical module, used for inputting the interval reference path library into a hierarchical control model, hierarchically processing steady-state path planning, dynamic compensation and steering control, and obtaining initial control parameters; A solution module, used for constructing an objective function including temperature constraints, time window constraints and vehicle dynamics constraints based on the initial control parameters, solving the objective function through obstacle term transformation, and obtaining a path optimization solution; An analysis module, used to analyze the impact of random disturbances in road conditions and temperature fluctuations on the path optimization scheme to obtain a path risk score index; The optimization module is used to perform feedback correction and iterative optimization on the initial control parameters according to the path risk scoring index to obtain target control parameters.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images