Logistics transportation scheme evaluation method and system

Abstract

The invention discloses a logistics transportation scheme evaluation method and system, and relates to the technical field of logistics management, and the method comprises the steps: obtaining a training sample; extracting a multi-dimensional feature vector; inputting a deep learning model to obtain prediction cost; determining a difference value between the predicted cost and the real cost, and adjusting the deep learning model to obtain a trained deep learning model; obtaining a to-be-evaluated logistics transportation scheme; extracting key information; inputting the key information into the trained deep learning model to obtain the estimated cost of each logistics transportation scheme to be evaluated; and selecting an optimal scheme. According to the invention, the cost and benefit of the transportation scheme can be evaluated according to an objective and unified standard, the optimization of the transportation route and resource allocation is facilitated, the overall efficiency of logistics transportation is improved, and intelligent monitoring and management of the logistics transportation process are realized by introducing advanced technologies such as deep learning and NLP.

Description

Technical Field

[0001] The present application relates to the field of logistics management technology, and in particular to a logistics transportation plan evaluation method and system. Background Art

[0002] Logistics companies are cost-sensitive companies. When formulating logistics and transportation plans for various needs, they always aim to minimize costs. However, there are many factors that affect the cost of the plan, including route selection, vehicle selection, energy replenishment strategy, etc. Since there are many influencing factors, and different factors have different impacts on costs, when there are multiple different logistics and transportation plans for the same logistics and transportation demand, it is necessary to evaluate the costs of different logistics and transportation plans in order to select the plan with the lowest cost. However, there is currently no technology for evaluating logistics and transportation plans, and judgments still rely on manual experience. This method is highly subjective, and the judgment standards are not unified, and the labor cost is very high. Summary of the invention

[0003] The embodiments of the present application provide a method and system for evaluating a logistics transportation plan, which are used to solve the problems of serious subjectivity and inconsistent standards in manual judgment in the prior art.

[0004] On the one hand, an embodiment of the present application provides a method for evaluating a logistics transportation plan, comprising: Get training samples; Extract multi-dimensional feature vectors from training samples; The multi-dimensional feature vector is input into a deep learning model. The deep learning model includes a CNN (Convolutional Neural Network) layer, an RNN (Recurrent Neural Network) / Transformer layer and an output layer connected in sequence. The CNN layer extracts the local pattern of the multi-dimensional feature vector, the RNN / Transformer layer extracts the long-distance dependency in the multi-dimensional feature vector, and the output layer uses a fully connected layer to output the prediction cost after considering the local pattern and long-distance dependency. Determine the difference between the predicted cost and the true cost corresponding to the training sample, and adjust the deep learning model using the loss function according to the difference to minimize the difference, thereby obtaining the trained deep learning model; Obtain the logistics and transportation solutions to be evaluated; Use the LLM (Large Language Module) semantic understanding model to extract key information from the logistics and transportation plan to be evaluated; Input key information into the trained deep learning model to obtain the estimated cost corresponding to each logistics transportation plan to be evaluated; The logistics and transportation plan with the lowest estimated cost is selected as the optimal plan.

[0005] On the other hand, the embodiment of the present application further provides a logistics transportation plan evaluation system, including: A sample acquisition module is used to acquire training samples; Feature extraction module, used to extract multi-dimensional feature vectors from training samples; A model input module is used to input a multi-dimensional feature vector into a deep learning model. The deep learning model includes a CNN layer, an RNN / Transformer layer, and an output layer connected in sequence. The CNN layer extracts the local pattern of the multi-dimensional feature vector, the RNN / Transformer layer extracts the long-distance dependency in the multi-dimensional feature vector, and the output layer uses a fully connected layer to output the prediction cost after considering the local pattern and the long-distance dependency. The model adjustment module is used to determine the difference between the predicted cost and the true cost corresponding to the training sample, and adjust the deep learning model according to the difference using the loss function to minimize the difference, thereby obtaining the trained deep learning model; Solution acquisition module, used to obtain the logistics and transportation solution to be evaluated; The information extraction module is used to extract key information from the logistics and transportation scheme to be evaluated using the semantic understanding model of LLM; The cost estimation module is used to input key information into the trained deep learning model to obtain the estimated cost corresponding to each logistics transportation plan to be evaluated; The scheme selection module is used to select the logistics transportation scheme to be evaluated with the lowest estimated cost as the optimal scheme.

[0006] A logistics transportation plan evaluation method and system in this application has the following advantages: 1. Improve the efficiency and accuracy of logistics and transportation: Through the solution method of this application, logistics and transportation companies can more accurately evaluate the cost and benefits of transportation solutions based on in-depth analysis and intelligent prediction of multi-source heterogeneous data. This not only helps to optimize transportation routes and resource allocation, but also reduces empty driving rates and waiting time, significantly improving the overall efficiency of logistics and transportation.

[0007] 2. Reduce operating costs and risks: The method of this application realizes intelligent monitoring and management of the logistics and transportation process by introducing advanced technologies such as deep learning and NLP (Natural Language Processing). It can recommend the most economical transportation plan for enterprises, effectively reduce transportation costs, and improve the profitability of enterprises.

[0008] 3. Promote the intelligent transformation of the logistics industry: The method of this application will promote the transformation of the logistics industry towards intelligence and digitalization. By integrating multi-source data, building deep learning models, and applying NLP technology, logistics and transportation companies can build an efficient and intelligent transportation evaluation and management system. This intelligent transformation can not only enhance the core competitiveness of the company, but also promote the upgrading and development of the entire logistics industry, and promote the coordination and integration of upstream and downstream of the industrial chain.

[0009] 4. Promote the in-depth mining and utilization of data value: This application realizes the comprehensive integration and deep fusion of data by integrating multi-source heterogeneous data. This not only enriches the data dimension, but also improves the reliability and prediction accuracy of the data. On this basis, enterprises can further explore and utilize the value of data to discover new business opportunities and growth points. For example, through in-depth analysis of transportation data, enterprises can understand the changing patterns of transportation demand in different regions and different time periods, and provide a basis for formulating more accurate marketing strategies and capacity planning. At the same time, data-driven decision-making methods will also become one of the important trends in the future development of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0011] Figure 1 A flow chart of a logistics transportation plan evaluation method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0013] Figure 1 A flow chart of a logistics transportation scheme evaluation method provided in an embodiment of the present application. The present application embodiment provides a logistics transportation scheme evaluation method, including: S100: Obtain training samples.

[0014] For example, building a deep learning model requires a large amount of sample data as support. In the logistics and transportation scenario, waybill data, TBox (Telematics Box) positioning data, vehicle speed, rotation speed, remaining energy consumption and other data of the vehicle host can all be used as part of the sample data. Some of these data are structured data and unstructured data, so they need to be uniformly parsed into structured data. During the parsing process, message queue technologies such as Kafka or RabbitMQ (RabbitMessage Queue) are first used to transmit and cache data in real time, and then big data processing frameworks such as Hadoop or Spark are used to distribute and parse the data in a structured manner.

[0015] In the process of integrating data from multiple data sources, the quality monitoring module is also used to monitor the stability and data quality of the data source in real time, including completeness, consistency, and timeliness. Specifically, machine learning algorithms, such as anomaly detection algorithms, can be used to automatically identify and report potential data quality issues to ensure the quality of data entering subsequent processing links.

[0016] After being monitored by the quality monitoring module, the structured data will be cleaned using an automated data cleaning tool based on a rule engine. The tool can automatically process the data according to preset cleaning rules, such as removing null values, processing abnormal values, format standardization, etc. At the same time, it also supports user-defined cleaning rules to adapt to different business scenarios and data characteristics.

[0017] Specifically, during the data cleaning process, you can use the uniqueness constraints of hash tables or databases to detect and delete duplicate data; identify and correct or delete noise data by setting thresholds and anomaly detection algorithms, such as Z-Score, IQR (Interquartile Range), etc.; write data conversion scripts to convert data in different formats into a unified format, such as JSON (JavaScript Object Notation) and CSV (Comma-Separated Values), and standardize field naming and data types.

[0018] For missing or abnormal data, this application also uses a variety of data repair and filling technologies, such as statistical filling methods, including mean, median, mode filling, etc., machine learning-based prediction filling, including using existing data training models to predict missing values, etc. These methods effectively improve the integrity and availability of data.

[0019] S110, extracting multi-dimensional feature vectors from training samples.

[0020] Exemplarily, in feature extraction, multiple means are used to extract multi-dimensional feature information. In addition to traditional statistical features such as mean, variance, kurtosis, etc., text mining, such as TF-IDF (Term Frequency–Inverse Document Frequency), Word2Vec, and image processing, such as CNN to extract image features, are also introduced to extract valuable feature information from unstructured data. Then, regular expressions, text mining and other technologies are used to extract structured feature information from text data. One-hot encoding, or label encoding, is performed on categorical feature information, such as vehicle type, energy type, cargo type, and provinces and cities through which the route is passed. For numerical feature information, such as route distance, Z-Score standardization or Min-Max normalization is used to make the feature information in the same dimension and distribution range. The encoded categorical feature information and the standardized or normalized numerical feature information constitute a multi-dimensional feature vector.

[0021] S120, input the multi-dimensional feature vector into a deep learning model, the deep learning model includes a CNN layer, an RNN / Transformer layer and an output layer connected in sequence, the CNN layer extracts the local pattern of the multi-dimensional feature vector, the RNN / Transformer layer extracts the long-distance dependency in the multi-dimensional feature vector, and the output layer uses a fully connected layer to output the prediction cost after considering the local pattern and the long-distance dependency.

[0022] For example, the deep learning model is constructed using a deep learning framework such as TensorFlow or PyTorch. The model structure combines the advantages of CNN and RNN / Transformer, and is specifically composed as follows: Input layer: receives a multi-dimensional feature vector as input.

[0023] Embedding layer: Converts the categorical feature vector after one-hot encoding or label encoding into a dense vector for subsequent processing.

[0024] CNN layers: extract local patterns in the multi-dimensional feature vector, especially those related to spatial locations, such as different sections of a route.

[0025] RNN / Transformer layer: Processes sequence data in multi-dimensional feature vectors, such as chronological transportation records, and captures long-range dependencies, such as the impact of vehicle performance changes on prices in different provinces and cities.

[0026] Attention layer: Connected after the RNN / Transformer layer, it is used to enhance the deep learning model's ability to capture interesting information, such as congestion on a specific road section, to further highlight the impact of this interesting information on the prediction results.

[0027] Output layer: Use a fully connected layer to output the predicted cost.

[0028] S130, determining the difference between the predicted cost and the true cost corresponding to the training sample, and adjusting the deep learning model using a loss function according to the difference to minimize the difference, thereby obtaining a trained deep learning model.

[0029] For example, considering that different factors may have different effects on cost prediction, this application uses weighted mean square error (WMSE) as the loss function:

[0030] in, N is the sample size, y i is the true cost, y i ’ is the forecast cost, ω i It is a weight set according to business needs and is used to adjust the contribution of different samples to the total loss.

[0031] Furthermore, the training samples include supervised learning samples with labels and semi-supervised learning samples without labels. For supervised learning samples, the difference is determined using the true cost contained in the corresponding label. Specifically, the deep learning model iteratively adjusts internal parameters such as weights and biases in the neural network to minimize the difference between the predicted cost and the true cost, and then evaluates the accuracy of the weight setting based on performance indicators such as the accuracy, recall rate, and F1 score of the deep learning model.

[0032] For semi-supervised learning samples, the difference is determined using the corresponding pseudo-label or unsupervised learning method. Specifically, for unlabeled data in the training samples, pseudo-labels or unsupervised learning methods can be used to extract useful information to assist model training and improve the generalization ability of the model. Pseudo-labels are generated through the model's own predictions, while unsupervised learning methods can use clustering, autoencoders, etc. However, consistency constraints need to be introduced to encourage the model to maintain consistency in its predictions of unlabeled data under different disturbances.

[0033] The training process of the deep learning model in this application is as follows: Initialization parameters: Before training begins, the model's parameters need to be randomly initialized. The initial values ​​of these parameters have a certain impact on the final performance of the model, but usually not too much, because the optimization algorithm during training will gradually adjust these values.

[0034] Forward propagation: Given the input data, the model will perform forward propagation according to its structure and calculate the corresponding output.

[0035] Loop calculation loss: The loss function is used to calculate the difference between the model's predicted cost and the true cost. This loss value is used to guide the optimization direction of the model.

[0036] Back propagation: Based on the loss value, the gradient of each parameter in the model is calculated through the back propagation algorithm. The gradient represents the sensitivity of the loss function to each parameter, that is, the impact of parameter changes on the loss value.

[0037] Parameter update: Use optimization algorithms such as gradient descent, Adam, etc. to update the model parameters according to the calculated gradient. The goal of optimization is to minimize the loss function, that is, to make the model's predicted cost closer to the true cost.

[0038] Iterative training: Repeat the above process until certain stopping conditions are met, such as reaching a preset number of iterations or the performance on the validation set no longer improving.

[0039] Hyperparameter selection: For vehicle type and energy type selection, the key hyperparameters may include learning rate, batch size, convolution kernel size, number of RNN / Transformer layers and units, etc.

[0040] Dynamically adjust the learning rate: Use 0.01 as the initial value, and use the learning rate to find the linear or exponential increase in one epoch. Record the loss of each step, and then find the point where the loss drops the fastest as the starting learning rate.

[0041] Tuning method: Use Bayes' theorem to efficiently find the optimal hyperparameter combination in the search space. This method is usually faster and more accurate than grid search and random search to find the optimal hyperparameter combination.

[0042] L1 / L2 regularization is used to prevent overfitting, and Dropout randomly discards some neuron connections during training.

[0043] S140, obtaining a logistics transportation plan to be evaluated.

[0044] For example, the logistics transportation plan to be evaluated is input by the user, and the user can use text input. The input content does not need to strictly follow a certain format or field name requirement. It only needs to organize the language naturally and describe the relevant information of the logistics scenario in text.

[0045] S150, using the semantic understanding model of LLM to extract key information in the logistics transportation plan to be evaluated.

[0046] For example, the semantic understanding model can use the open source Qwen-7b large model to deeply analyze the text input by the user. LLM can accurately capture the contextual meaning in the language through a large amount of pre-trained text data, and convert the natural language instructions input by the user into structured query parameters, such as the origin ID, destination ID, cargo type code, time window, etc.

[0047] After obtaining the structured query parameters, it is also necessary to associate the structured query parameters with basic data tables such as vehicle type, energy type, route distance, and provinces and cities passed through to prepare a data set for cost estimation, which is called key information.

[0048] Furthermore, basic data tables such as the vehicle type price reference table and energy replenishment price calculation table are dynamically updated based on real-time data and user feedback.

[0049] S160, inputting key information into the trained deep learning model to obtain the estimated cost corresponding to each logistics transportation plan to be evaluated.

[0050] For example, the trained deep learning model already has the ability to estimate the cost and has met the requirements in terms of estimated performance, so after inputting key information into the deep learning model, the output estimated cost can be obtained.

[0051] S170, selecting the logistics transportation scheme to be evaluated with the lowest estimated cost as the optimal scheme.

[0052] For example, since logistics companies are cost-sensitive companies, under the assumption that all logistics and transportation plans to be evaluated can meet a logistics and transportation task, the plan with the lowest estimated cost will be selected as the optimal plan.

[0053] The present application also provides a logistics transportation plan evaluation system, which includes the following functional modules: A sample acquisition module is used to acquire training samples; Feature extraction module, used to extract multi-dimensional feature vectors from training samples; A model input module is used to input a multi-dimensional feature vector into a deep learning model. The deep learning model includes a CNN layer, an RNN / Transformer layer, and an output layer connected in sequence. The CNN layer extracts the local pattern of the multi-dimensional feature vector, the RNN / Transformer layer extracts the long-distance dependency in the multi-dimensional feature vector, and the output layer uses a fully connected layer to output the prediction cost after considering the local pattern and the long-distance dependency. The model adjustment module is used to determine the difference between the predicted cost and the true cost corresponding to the training sample, and adjust the deep learning model according to the difference using the loss function to minimize the difference, thereby obtaining the trained deep learning model; Solution acquisition module, used to obtain the logistics and transportation solution to be evaluated; The information extraction module is used to extract key information from the logistics and transportation scheme to be evaluated using the semantic understanding model of LLM; The cost estimation module is used to input key information into the trained deep learning model to obtain the estimated cost corresponding to each logistics transportation plan to be evaluated; The scheme selection module is used to select the logistics transportation scheme to be evaluated with the lowest estimated cost as the optimal scheme.

[0054] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0055] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A logistics transportation plan evaluation method, characterized in that: include: Get training samples; Extracting a multi-dimensional feature vector from the training sample; Inputting the multidimensional feature vector into a deep learning model, the deep learning model comprising a CNN layer, an RNN / Transformer layer and an output layer connected in sequence, the CNN layer extracting a local pattern of the multidimensional feature vector, the RNN / Transformer layer extracting a long-distance dependency in the multidimensional feature vector, and the output layer using a fully connected layer to output a prediction cost after considering the local pattern and the long-distance dependency; Determine the difference between the predicted cost and the true cost corresponding to the training sample, and adjust the deep learning model according to the difference using a loss function to minimize the difference, thereby obtaining the trained deep learning model; Obtain the logistics and transportation solutions to be evaluated; Using the semantic understanding model of LLM to extract key information from the logistics transportation plan to be evaluated; Inputting the key information into the trained deep learning model to obtain an estimated cost corresponding to each of the logistics transportation plans to be evaluated; The logistics transportation plan to be evaluated with the lowest estimated cost is selected as the optimal plan.

2. A logistics transportation scheme evaluation method according to claim 1, characterized in that: After obtaining the training samples, the training samples and the logistics transportation scheme to be evaluated are subjected to structured analysis to obtain corresponding structured data.

3. A logistics transportation scheme evaluation method according to claim 2, characterized in that: After acquiring the structured data, an anomaly detection algorithm is used to monitor the quality of the structured data.

4. A logistics transportation scheme evaluation method according to claim 1, characterized in that: After obtaining the training samples, the data in the training samples and the logistics transportation scheme to be evaluated are cleaned, repaired and filled.

5. A method for evaluating a logistics transportation plan according to claim 1, characterized in that: When extracting the multidimensional feature vector in the training sample, statistical features, text mining and image processing techniques are used to extract the feature information in the training sample, the categorical feature information is uniquely encoded or label encoded, and the numerical feature information is standardized or normalized to obtain the multidimensional feature vector.

6. A method for evaluating a logistics transportation plan according to claim 1, characterized in that: The deep learning model also has an attention layer after the RNN / Transformer layer, and the attention layer enhances the ability of the deep learning model to capture information of interest.

7. A method for evaluating a logistics transportation plan according to claim 1, characterized in that: The training samples include supervised learning samples with labels and semi-supervised learning samples without labels. For the supervised learning samples, the difference is determined using the true cost contained in the corresponding label. For the semi-supervised learning samples, the difference is determined using the corresponding pseudo label or using an unsupervised learning method.

8. A method for evaluating a logistics transportation plan according to claim 1, characterized in that: In the process of training the deep learning model, Bayes' theorem is used to find the optimal hyperparameter combination in the search space, and the deep learning model is trained using the optimal hyperparameter combination.

9. A system using a logistics transportation scheme evaluation method according to any one of claims 1 to 8, characterized in that: include: A sample acquisition module is used to acquire training samples; A feature extraction module, used to extract multi-dimensional feature vectors from the training samples; A model input module, used to input the multi-dimensional feature vector into a deep learning model, wherein the deep learning model includes a CNN layer, an RNN / Transformer layer and an output layer connected in sequence, wherein the CNN layer extracts a local pattern of the multi-dimensional feature vector, the RNN / Transformer layer extracts a long-distance dependency in the multi-dimensional feature vector, and the output layer uses a fully connected layer to output a prediction cost after considering the local pattern and the long-distance dependency; A model adjustment module, used to determine the difference between the predicted cost and the true cost corresponding to the training sample, and adjust the deep learning model according to the difference using a loss function to minimize the difference, thereby obtaining the trained deep learning model; Solution acquisition module, used to obtain the logistics and transportation solution to be evaluated; An information extraction module, used to extract key information from the logistics transportation scheme to be evaluated by using the semantic understanding model of LLM; A cost estimation module, used to input the key information into the trained deep learning model to obtain an estimated cost corresponding to each of the logistics transportation plans to be evaluated; The scheme selection module is used to select the logistics transportation scheme to be evaluated with the lowest estimated cost as the optimal scheme.

Images