Multi-dimensional dynamic charging full life cycle management system and method

Abstract

The invention discloses a full-life-cycle management system and a full-life-cycle management method for multi-dimensional dynamic charging. The method comprises the steps that an RFID chip is embedded in a cargo, and identification information including the unique ID of the cargo, the category of the cargo and the purchase date is written in the RFID chip; scanning the RFID chip through a cargo binding terminal, and associating information of a customer to establish a cargo electronic ledger; when the goods need to be transported in a cross-land mode, starting point information and destination information of the goods are obtained, a path optimization algorithm is called, and waybill data containing the transportation distance are generated; constructing a corresponding feature vector based on a transportation distance, a time period coefficient, energy consumption data and a vehicle capacity reciprocal in the waybill data; inputting the feature vector into a machine learning model obtained through historical waybill data training, and outputting predicted transportation cost; and comparing the first product result with the predicted transportation cost, and determining the final transportation cost according to the comparison result, the first product result being the product result of the basic unit price and the transportation distance.

Description

Technical Field

[0001] This application relates to the field of software technology, and particularly to a full life cycle management system and method for multi-dimensional dynamic charging. Background Art

[0002] With the rapid development of logistics transportation and Internet of Things technology, asset management based on radio frequency identification (RFID) technology has been widely applied in multiple industries. Especially in the full life cycle management of consumables, goods and other textiles, embedding RFID chips to achieve automatic identification and information tracking has become an important means to improve management efficiency and accuracy.

[0003] Currently, the transportation costs of consumables and goods usually adopt fixed charging or a simple calculation method of distance multiplied by unit price, lacking comprehensive consideration of multi-dimensional factors in the transportation process, resulting in unreasonable charging and difficulty in reflecting the actual transportation costs. At the same time, path optimization and dynamic cost prediction are not fully utilized during transportation, affecting transportation efficiency and cost control.

[0004] Therefore, how to construct a full life cycle management method based on multi-dimensional data dynamic charging, which can accurately reflect the actual transportation costs and reasonably formulate transportation fees, has important application value and practical significance. Summary of the Invention

[0005] To solve the above technical problems, this application provides a full life cycle management system and method for multi-dimensional dynamic charging.

[0006] The technical solutions provided in this application are described below: In the first aspect of this application, a full life cycle management method for multi-dimensional dynamic charging is provided, and the method includes: Embedding an RFID chip in the goods, and writing identification information including the unique ID of the goods, the category of the goods, and the purchase date into the RFID chip; Scanning the RFID chip through a goods-binding terminal and associating the affiliated customer information to establish an electronic goods ledger; When the goods need to be transported across regions, obtaining the starting point information and destination information of the goods, and calling a path optimization algorithm to generate waybill data including the transportation distance; Constructing a corresponding feature vector based on the transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity in the waybill data; Inputting the feature vector into a machine learning model trained by historical waybill data, and outputting a predicted transportation cost; Comparing the first product result with the predicted transportation cost, and determining the final transportation fee according to the comparison result, where the first product result is the product result of the basic unit price and the transportation distance.

[0007] Optionally, it further includes: Collect and record the RFID signal values of the goods embedded with RFID chips continuously for multiple times during washing and use; Calculate the RSSI mean value and the fluctuation amplitude based on the continuously collected RSSI signal values; Compare the calculated RSSI mean value with a preset mean threshold, and compare the fluctuation amplitude with a preset fluctuation threshold; Determine the damaged state of the goods according to the comparison results.

[0008] Optionally, the determining the damaged state of the goods according to the comparison results includes: When the RSSI mean value is lower than the preset mean threshold and / or the fluctuation amplitude is higher than the preset fluctuation threshold, it is determined that the goods are in a damaged state.

[0009] Optionally, the time period coefficient includes a weighted time cost parameter based on historical time period electricity prices, road congestion levels, and free parking durations in the parking lot, and the weighted time cost parameter is used to reflect the economic efficiency of transportation tasks in different time periods.

[0010] Optionally, the construction of the machine learning model includes: Collect historical waybill data within a preset time window, and the historical waybill data includes the following fields: transportation distance, start and end time periods, electricity price fluctuation information, actual energy consumption, vehicle model, vehicle load factor, parking fees, toll standards, historical traffic congestion coefficients, and actual transportation costs; Perform data cleaning and normalization processing on the historical waybill data, and perform feature crossing, discretization, and high-order combination to generate a multi-dimensional and highly coupled transportation task feature vector; Use a feature selection method based on information gain and L1 regularization to determine the optimal feature subset in the transportation task feature vector; Train multiple regression models based on the optimal feature subset, and the multiple regression models include random forest regression, gradient boosting tree, support vector regression, and multi-layer perceptron neural network, and obtain the generalization performance indicators corresponding to each model through cross-validation; Use a weighted integration method to fuse the multiple regression models to obtain a fusion model.

[0011] Optionally, the performing data cleaning and normalization processing on the historical waybill data, and performing feature crossing, discretization, and high-order combination to generate a multi-dimensional and highly coupled transportation task feature vector includes: Perform data cleaning on the historical waybill data, and the data cleaning includes performing mean filling or category mode filling on missing values, and performing elimination processing on outliers based on the statistical distribution boundary; Compress the values of the cleaned numerical data to a unified dimension using the min-max normalization method or the Z-score standardization method; Perform feature engineering on the normalized numerical data, and the feature engineering includes: Perform feature cross-operation to combine multiple basic features into interactive features; perform feature discretization operation to divide continuous features into multiple discrete intervals; construct high-order combined features, and the high-order combined features at least include unit energy consumption, unit parking space cost, and driver delivery efficiency; Construct a transportation task feature vector based on the processing results to form a multi-dimensional and highly coupled data structure as the model input.

[0012] Optionally, the method for determining the optimal feature subset in the transportation task feature vector by using the feature selection method based on information gain and L1 regularization includes: Calculate the discrimination ability of each candidate feature relative to the transportation cost based on the information gain method, and accordingly determine the feature subset with an information gain value higher than the preset threshold, and the candidate features are determined in the task feature vector; Train a sparse regression model based on the L1 regularization method, and automatically compress some feature coefficients to zero by minimizing the loss function with an L1 penalty term; Construct the optimal feature subset from the features obtained after information gain and L1 regularization.

[0013] Optionally, the calculating the discrimination ability of each candidate feature relative to the transportation cost based on the information gain method, and accordingly determining the feature subset with an information gain value higher than the preset threshold, and the candidate features are determined in the task feature vector includes: Discretize the transportation cost into several categories; Calculate the joint entropy and conditional entropy between each candidate feature and each category of the discretized transportation cost; Calculate the information gain value of each candidate feature according to the joint entropy and conditional entropy; Set an information gain threshold, and determine the candidate features with an information gain value higher than the information gain threshold as the feature subset.

[0014] Optionally, the calculating the joint entropy and conditional entropy between each candidate feature and each category of the discretized transportation cost includes: Calculate the joint frequency between each value of the candidate feature and each category of the transportation cost, and calculate the joint probability distribution; Calculate the joint entropy H(X, Y)=−∑ x，y p(x, y)logp(x, y), where X is the candidate feature and Y is the transportation cost of the corresponding category; Calculate the transportation cost entropy H(Y) corresponding to the category: H(Y)=−∑ y p(y)logp(y) and the conditional entropy of the candidate feature H(Y∣X)=H(X,Y)−H(X); Calculate the information gain IG(Y∣X)=H(Y)−H(Y∣X) according to the entropy, where H(X)=−∑ x p(x)logp(x) is the entropy of the candidate feature; Among them, the joint probability distribution p(x,y) represents the probability that the candidate feature X takes the value x and the corresponding transportation cost Y takes the value y, p(x) represents the probability that the candidate feature X takes the value x, that is, p(x,y) is accumulated for all y; p(y) represents the probability that the transportation cost Y takes the value y; the joint entropy H(X,Y) is used to measure the uncertainty of the joint distribution of the candidate feature and the transportation cost; H(X) represents the uncertainty of the candidate feature, H(Y) represents the uncertainty of the transportation cost category itself; the conditional entropy H(Y|X) represents the remaining uncertainty of the transportation cost Y under the condition that the candidate feature X is known; the information gain IG(Y|X) is used to measure the explanatory ability of the feature X for the classification variable Y; if the candidate feature X is highly correlated with the transportation cost, then H(Y|X) is small and the information gain IG(Y∣X) is larger, indicating that this feature is more valuable.

[0015] The second aspect of this application provides a full life cycle management system for multi-dimensional dynamic billing, including: An RFID unit for embedding an RFID chip in the goods, and identification information including the unique ID of the goods, the category of the goods, and the purchase date is written in the RFID chip; A binding unit for scanning the RFID chip through a goods binding terminal and associating the affiliated customer information to establish an electronic goods ledger; A waybill data generation unit for obtaining the starting point information and destination information of the goods and calling a path optimization algorithm to generate waybill data including the transportation distance when the goods need to be transported across regions; A feature vector construction unit for constructing a corresponding feature vector based on the transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity in the waybill data; A machine learning unit for inputting the feature vector into a machine learning model trained by historical waybill data and outputting a predicted transportation cost; A result output unit for comparing the first product result with the predicted transportation cost and determining the final transportation cost according to the comparison result, where the first product result is the product result of the basic unit price and the transportation distance.

[0016] It can be seen from the above technical solutions that this application has the following advantages: The present invention realizes the digital management of the entire life cycle of goods by embedding RFID chips in the goods, can accurately collect and associate the unique identity information and customer information of the goods, and improves the intelligence and automation level of asset management.

[0017] In addition, the present invention combines the path optimization algorithm and the construction of multi-dimensional feature vectors, and uses a machine learning model to dynamically predict the transportation cost, which can more comprehensively and accurately reflect the actual cost changes during the transportation process, and avoids the limitations of the traditional single-distance billing method.

[0018] By comparing the traditional billing product result with the dynamically predicted cost, the final transportation cost is reasonably determined, which not only ensures the economic rationality of the transportation service, but also effectively promotes the optimal allocation of transportation resources and cost savings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a schematic flowchart of an embodiment of the full life cycle management method for multi-dimensional dynamic billing provided in the present application; Figure 2 It is a schematic flowchart of a specific embodiment of step S105 in the full life cycle management method for multi-dimensional dynamic billing provided in the present application; Figure 3 It is a schematic flowchart of another embodiment of the full life cycle management method for multi-dimensional dynamic billing provided in the present application; Figure 4 It is a schematic structural diagram of an embodiment of the full life cycle management system for multi-dimensional dynamic billing provided in the present application; Figure 5 It is a schematic structural diagram of an embodiment of the full life cycle management device for multi-dimensional dynamic billing provided in the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Please refer to Figure 1 , the present application first provides an embodiment of the full life cycle management method for multi-dimensional dynamic billing, and this embodiment includes: S101. Embed an RFID chip in the goods, and write identification information including the unique ID of the goods, the category of the goods, and the purchase date into the RFID chip; Embed an RFID chip in the goods. The unique identification information of the goods is pre-written in the RFID chip, including but not limited to key information such as the unique ID of the goods, the category of the goods, and the purchase date. The unique ID is used to distinguish different goods, the category of the goods is used to identify the type and specification of the goods, and the purchase date is used to track the usage time and life cycle management of the goods.

[0022] S102. Scan the RFID chip through the goods-binding terminal and associate the customer information to establish an electronic ledger for the goods; Scan the RFID chip in the goods through the goods-binding terminal device. After scanning, associate the read RFID chip information with the customer information to establish a corresponding electronic ledger for the goods. The electronic ledger includes but not limited to information such as the unique ID, category, purchase date, customer information, and current status of the goods, realizing real-time digital management of each piece of goods.

[0023] S103. When the goods need to be transported across regions, obtain the starting point information and destination information of the goods, and call the path optimization algorithm to generate waybill data including the transportation distance; When the goods need to be transported across regions, the system automatically obtains the starting point location information and destination location information of the goods. Based on this location information, call the path optimization algorithm for path planning to optimize the transportation route. The path optimization algorithm can consider factors such as distance, traffic conditions, and time windows, and finally generate waybill data including detailed transportation information such as transportation distance for subsequent billing and scheduling. The path optimization module adopts an improved heuristic search algorithm (such as the A* algorithm or Dijkstra algorithm), combined with real-time traffic information and historical congestion patterns for dynamic adjustment. Some implementations can also use machine learning models to predict the passage time of road sections to improve the accuracy of path planning.

[0024] Specifically, when the goods need to be transported across regions, in this embodiment, the system obtains the location and generates waybill data according to the following process: In the consumable-binding terminal or the warehouse management system, read the inventory or loading position identifier of the goods to be transported. This position identifier corresponds one-to-one with the coordinates in the pre-established geographical information database (geographical coordinate library) to obtain the starting point longitude and latitude coordinates; Read the address information of the destination hotel or laundry center from the user order or scheduling instruction, and convert it into the destination longitude and latitude coordinates through the geocoding service (Geocoding API).

[0025] The system automatically calls the third-party map server interface to obtain the real-time road network data from the starting point to the destination, including the lengths of each road segment, the legal passing speeds, the historical and real-time congestion coefficients; synchronously obtains the possible road condition restriction information for each segment (such as restricted travel intervals, truck prohibited travel periods, bridge load limits, etc.) and time window constraints (such as the allowed time periods for hotel pick-up / delivery).

[0026] In the built-in path planning module, the obtained road network is abstracted as a weighted directed graph: Nodes: Road network intersections or key coordinate points; Edges: Segment connections, and the weights of the edges are weighted and synthesized according to the following multi-dimensional indicators: Length of the road segment (km); Estimated travel time = length of the road segment ÷ legal speed × congestion coefficient; Toll cost; Estimated vehicle energy consumption = length of the road segment × unit energy consumption (kWh / km); Time window violation penalty (if the estimated arrival time exceeds the destination unloading time window, an additional weight is added).

[0027] Call the Dijkstra algorithm or a multi-objective optimization variant based on A* search to calculate the optimal path on this weighted graph, with the optimal basis being the minimum total weighted cost.

[0028] The optimal path obtained in step 3.2 is translated into a continuous road segment sequence, and the following are calculated: Total transportation distance (meters or kilometers) = sum of the lengths of each road segment; Estimated total travel time (minutes) = sum of the estimated travel times of each road segment; Estimated total energy consumption (kWh) and toll fees (yuan); In the background waybill generation module, create a form containing the following fields: OrderID: Unique order number; StartCoords: Starting point longitude and latitude; EndCoords: Destination longitude and latitude; RouteSegments: List of road segments and their attributes; TotalDistance: Total transportation distance; EstimatedTime: Estimated travel time; EstimatedEnergy: Estimated energy consumption; EstimatedTollFees: Estimated toll fees; TimeWindowConstraints: Unloading time window; And persist the waybill data in the scheduling database for use in the construction of feature vectors and dynamic billing in subsequent steps S104–S106.

[0029] Through the above implementation, the system can automatically generate waybill data containing accurate transportation distances and cost elements based on real-time road conditions and multi-dimensional weight constraints, providing reliable and quantifiable inputs for subsequent multi-dimensional dynamic billing.

[0030] S104. Construct corresponding feature vectors based on the transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity in the waybill data; Based on the key parameters in the above waybill data, including transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity, construct corresponding transportation task feature vectors. These feature vectors reflect the multi-dimensional attributes of transportation tasks, can accurately describe various factors affecting costs during transportation, and provide comprehensive input data for machine learning models.

[0031] This step constructs transportation task feature vectors based on multi-dimensional key parameters for the waybill data generated in step S103. The specific implementation includes the following: Extract the following key metrics from the waybill data: Transportation distance d (unit: kilometers), which is the total transportation distance generated by route planning; Time period coefficient t, which is a cost weighting factor used to reflect the transportation time period (such as peak hours, night, etc.); Energy consumption data e (unit: kWh), calculated based on the route length and vehicle energy consumption model; Reciprocal of vehicle production capacity c = 1 / vehicle maximum load, (unit: tons −1 ), which reflects the impact of the vehicle's unit load on costs.

[0032] Perform normalization on the extracted numerical features, scale each metric to an interval of a unified dimension (such as [0, 1]) to avoid training biases caused by scale differences. Common normalization methods include min-max normalization.

[0033] Arrange the normalized key parameters in a fixed order to form a multi-dimensional feature vector: X = [d′, t′, e′, c′], where d, t′, e′, c′ represent the normalized transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity, respectively.

[0034] To enhance the model's ability to express complex transportation scenarios, the basic features can be extended and combined, such as products, squares, and cross terms. This step can be dynamically determined whether to execute based on historical experience or an automatic feature selection algorithm. Convert the finally constructed feature vector X into a format suitable for the input of a machine learning model (such as a numerical array or a tensor), and pass it to the prediction model in step S105.

[0035] S105. Input the feature vector into a machine learning model trained with historical waybill data and output the predicted transportation cost; Input the constructed feature vector of the transportation task into a machine learning model pre-trained based on historical waybill data. Through learning a large amount of historical transportation data, this machine learning model can output the corresponding predicted transportation cost, realizing dynamic and accurate prediction of the cost of the current transportation task.

[0036] See Figure 2 , specifically, this step may include: S1051. Collect historical waybill data within a preset time window. The historical waybill data includes the following fields: transportation distance, start and end time periods, electricity price fluctuation information, actual energy consumption, vehicle model, vehicle load factor, parking fees, toll standards, historical traffic congestion coefficient, and actual transportation cost; The system collects historical waybill data from a preset time window (such as the past three months, the past six months). The historical waybill data includes multiple key fields, including but not limited to: Transportation distance (km), reflecting the path length of each transportation; Start and end time periods, indicating the start time and end time of the transportation task, used to calculate the time period correlation coefficient; Electricity price fluctuation information, reflecting the electricity price changes during the corresponding time period of the transportation process; Actual energy consumption (kWh), statistically obtained based on the vehicle driving mileage and the energy consumption model; Vehicle model, indicating the specific model and specifications of the vehicle used for transportation; Vehicle load factor, the ratio of the actual load to the maximum load of the vehicle; Parking fees, the parking costs generated during the transportation process; Toll standards, determined according to the toll standards involved in the driving route; Historical traffic congestion coefficient, reflecting the congestion status of the transportation route and time period; Actual transportation cost, the transportation fees actually incurred in the historical data.

[0037] This dataset provides a real sample basis for subsequent model training and evaluation.

[0038] S1052. Perform data cleaning and normalization on the historical waybill data, and conduct feature crossing, discretization, and high-order combination to generate a multi-dimensional and highly coupled transportation task feature vector; Perform data cleaning on the historical waybill data described in step S1051, specifically including: Perform data cleaning on the historical waybill data. The data cleaning includes performing mean filling or category mode filling on missing values, and performing elimination processing on outliers based on statistical distribution boundaries; using the min-max normalization method or the Z-score standardization method to compress the values of the cleaned numerical data to a unified dimension; performing feature engineering processing on the normalized numerical data. The feature engineering processing includes: Perform feature crossing operations to combine multiple basic features into interaction features; perform feature discretization operations to divide continuous features into multiple discrete intervals; construct high-order combination features, and the high-order combination features at least include unit energy consumption, unit parking space cost, and driver delivery efficiency; Construct a transportation task feature vector based on the processing results to form a multi-dimensional and highly coupled data structure as the model input.

[0039] Among them, feature crossing is to multiply or combine two or more basic features to generate crossing features, such as "transportation distance × traffic congestion coefficient"; Discretization processing is to divide intervals for continuous features (such as transportation distance, energy consumption) and convert them into categorical features; High-order combination is to perform non-linear combinations such as squaring and cubing on features to enhance the model's ability to capture complex relationships.

[0040] Through the above steps, generate a multi-dimensional and highly coupled transportation task feature vector to provide rich input information for the machine learning model.

[0041] S1053. Use a feature selection method based on information gain and L1 regularization to determine the optimal feature subset in the transportation task feature vector; Based on the above-generated transportation task feature vectors, feature selection techniques are adopted to screen out the feature subset that contributes the most to predicting transportation costs. Specifically, it includes: calculating the discrimination ability of each candidate feature relative to transportation costs based on the information gain method, and accordingly determining the feature subset with an information gain value higher than a preset threshold, where the candidate features are determined in the task feature vectors; training a sparse regression model based on the L1 regularization method, and automatically compressing some feature coefficients to zero by minimizing the loss function with an L1 penalty term; constructing the features obtained after information gain and L1 regularization into an optimal feature subset. The calculating the discrimination ability of each candidate feature relative to transportation costs based on the information gain method, and accordingly determining the feature subset with an information gain value higher than a preset threshold, where the candidate features are determined in the task feature vectors includes: Discretize the transportation costs into several categories; Calculate the joint entropy and conditional entropy between each candidate feature and the discretized transportation costs of each category; Calculate the information gain value of each candidate feature according to the joint entropy and conditional entropy; Set an information gain threshold, and determine the candidate features with an information gain value higher than the information gain threshold as the feature subset.

[0042] Among them, calculating the joint entropy and conditional entropy between each candidate feature and the discretized transportation costs of each category includes: Calculate the joint frequency between each value of the candidate feature and the transportation costs of each category, and calculate the joint probability distribution; Calculate the joint entropy H(X, Y)=−∑ x，y p(x, y)logp(x, y), where X is the candidate feature and Y is the transportation cost of the corresponding category; Calculate the entropy of the transportation cost of the corresponding category H(Y)=−∑ y p(y)logp(y) and the conditional entropy of the candidate feature H(Y|X)=H(X, Y)−H(X); Calculate the information gain IG(Y|X)=H(Y)−H(Y|X) according to the entropy, where H(X)=−∑ x p(x)logp(x) is the entropy of the candidate feature.

[0043] Among them, the joint probability distribution p(x, y) represents the probability that the candidate feature X takes the value x and the corresponding transportation cost Y takes the value y. p(x) represents the probability that the candidate feature X takes the value x, that is, the sum of p(x, y) for all y. p(y) represents the probability that the transportation cost Y takes the value y. The joint entropy H(X, Y) is used to measure the uncertainty of the joint distribution of the candidate feature and the transportation cost. H(X) represents the uncertainty of the candidate feature, and H(Y) represents the uncertainty of the transportation cost category itself. The conditional entropy H(Y|X) represents the remaining uncertainty of the transportation cost Y given the candidate feature X. The information gain IG(Y|X) is used to measure the explanatory power of the feature X for the classification variable Y. If the candidate feature X is highly correlated with the transportation cost, then H(Y|X) is small and the information gain IG(Y|X) is larger, indicating that this feature is more valuable.

[0044] In this step, the information gain method is used to calculate the correlation between each feature and the transportation cost, and the features with information gain values lower than the preset threshold are removed. Combined with the sparse characteristics of L1 regularization, an L1 penalty term is introduced in the regression model training to achieve feature coefficient sparsification and further screen out the significantly influential features.

[0045] The finally obtained feature subset not only ensures the prediction performance but also effectively reduces the model complexity and overfitting risk.

[0046] S1054. Based on the optimal feature subset, train multiple regression models respectively. The multiple regression models include random forest regression, gradient boosting tree, support vector regression, and multi-layer perceptron neural network, and obtain the generalization performance indicators corresponding to each model through cross-validation. Based on the optimal feature subset selected in step S1053, train multiple regression models respectively, specifically including: Random Forest Regression, which realizes integrated prediction by constructing multiple decision trees; Gradient Boosting Decision Tree, which iteratively optimizes the weak learner to improve the accuracy; Support Vector Regression, which realizes non-linear regression by using kernel function mapping; Multilayer Perceptron (MLP), which has a multi-hidden layer structure and fits complex non-linear relationships.

[0047] The generalization performance metrics of each model are evaluated using a cross - validation method (such as K - fold cross - validation), including the mean squared error (MSE), mean absolute error (MAE), etc., to ensure the stable performance of the model on unseen data.

[0048] S1055. Use a weighted ensemble method to fuse the multiple regression models to obtain a fused model.

[0049] To further improve the prediction performance, a weighted ensemble method can be used to fuse the multiple regression models obtained through training. The specific implementation includes dynamically allocating the weights of each model according to the performance metrics of cross - validation; the fused prediction result is the weighted average of the prediction values of each model; the weight allocation scheme is optimized through the validation set to maximize the accuracy and robustness of the overall model.

[0050] The above embodiments significantly improve the accuracy and reliability of transportation cost prediction through systematic data processing, scientific feature engineering, multi - model training and fusion, and are applicable to the full - life - cycle management system of multi - dimensional dynamic billing.

[0051] S106. Compare the first product result with the predicted transportation cost, and determine the final transportation fee according to the comparison result. The first product result is the product result of the base unit price and the transportation distance.

[0052] Compare the first product result (i.e., the product of the base unit price and the transportation distance) in the basic billing mode with the predicted transportation cost obtained in step S105. Determine the final transportation fee according to the comparison result. When the predicted transportation cost is higher than the basic billing result, the system can automatically adjust the billing strategy to reflect the actual transportation cost, ensuring reasonable and fair billing, while guaranteeing the quality of transportation services and cost - effectiveness.

[0053] After obtaining the predicted transportation cost output by the fused model in step S105, the system continues to perform the following operations to determine the final transportation fee: The system first calculates the "first product result" based on a fixed billing rule. This result is the product value of a preset base unit price and the transportation distance, and the expression is as follows: C_base = P_unit×D Where: C_base represents the basic billing amount, P_unit represents the preset base unit price, with the unit of yuan / km; D represents the transportation distance obtained in step S103, with the unit of km.

[0054] The integrated regression model (integrating random forest, GBDT, SVR, and MLP) trained in system call step S105 takes the feature vector corresponding to the current transportation task as input and outputs the predicted transportation cost, denoted as C_predict. This value represents the actual transportation cost expected to occur in the actual operating environment. The system calculates the difference between the two: ΔC = C_predict - C_base And set a dynamic adjustment threshold (for example, a floating range of ±10%). Based on the magnitude and direction of the difference, the following judgments are made: If ∣ΔC∣ ≤ δC_base, it means that the predicted cost is close to the basic billing, and the final transportation cost is set to C_base; If ΔC > δC_base, it indicates that the basic billing may underestimate the transportation cost, and the system adjusts the final transportation cost upward according to the predicted value; If ΔC < -δC_base, it means that the basic billing may overestimate the transportation cost, and the system can set a discount or refund strategy to improve customer satisfaction.

[0055] Based on the above judgment results, the system finally determines the transportation cost C_final.

[0056] After the final transportation cost is generated, the system automatically updates the "transportation cost" field in the electronic ledger of the goods, records the calculation method and source (basic billing / prediction model / hybrid strategy) of this cost in the waybill information, and includes this cost in the customer's bill, and supports functions such as online query and transparent price comparison display.

[0057] Refer to Figure 3 , this application also provides another embodiment of the full life cycle management method for multi-dimensional dynamic billing, and this embodiment includes: S301. Embed an RFID chip in the goods, and write identification information including the unique ID of the goods, the category of the goods, and the purchase date into the RFID chip; S302. Scan the RFID chip through the goods-bound terminal and associate the customer information to establish an electronic ledger of the goods; S303. When the goods need to be transported across regions, obtain the starting point information and destination information of the goods, and call the path optimization algorithm to generate waybill data including the transportation distance; S304. Based on the transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity in the waybill data, construct the corresponding feature vector; S305. Input the feature vector into the machine learning model trained with historical waybill data and output the predicted transportation cost; S306. Compare the first product result with the predicted transportation cost, and determine the final transportation fee according to the comparison result. The first product result is the product result of the base unit price and the transportation distance. In this embodiment, steps S301 to S306 are similar to the relevant steps in the foregoing embodiment, and will not be described in detail here.

[0058] S307. For the goods embedded with RFID chips, collect and record the RFID signal values continuously for multiple times during washing and use. During the washing and use of the goods, the system performs continuous reading operations on the RFID chips embedded in the goods through RFID reading devices (such as RFID antennas installed in the washing line channel, laundry basket entrance, delivery / collection points, etc.). During each reading process, the system collects and records the RSSI (Received Signal Strength Indicator) value.

[0059] The system performs RSSI collection operations at multiple time points to obtain a sufficient number of samples. For example, at the beginning, middle, and end of the washing cycle, and at the recycling point after user use, to ensure that the data covers the signal changes under different physical states.

[0060] For example, for each piece of goods, the system can collect 20 to 50 valid RSSI signal data continuously for subsequent fluctuation analysis.

[0061] S308. Based on the continuously collected RSSI signal values, calculate the RSSI mean value and the fluctuation amplitude. The system performs statistical processing on the RSSI signal sequence collected in step S307, including but not limited to calculating the RSSI mean value and the RSSI fluctuation amplitude (variance or range). The system can select one or a combination of them for calculation according to the deployment environment, for evaluating the physical consistency and signal conduction stability between the chip and the fabric structure.

[0062] S309. Compare the calculated RSSI mean value with a preset mean value threshold, and compare the fluctuation amplitude with a preset fluctuation threshold. The system compares the statistical values obtained in step S308 with the pre-set standard thresholds. The setting of the thresholds can be obtained based on the statistical analysis of a large amount of RFID data of goods in normal states, such as establishing a normal signal model through means such as cluster analysis, confidence interval, and 3σ rule.

[0063] S310. Determine the damaged state of the goods according to the comparison result.

[0064] When the average RSSI is lower than the preset average threshold and / or the fluctuation range is higher than the preset fluctuation threshold, it indicates that the overall chip signal is weakened, which may be caused by fabric occlusion, poor chip connection after immersion in water, or physical damage to the goods; it is determined that the goods are in a damaged state. The system updates the status of the goods in the electronic ledger to "pending re-inspection" or "suspected damage" and pushes it to the operation personnel's task list. The system can also set the damage level, such as "slight fluctuation", "moderate abnormality", "severe damage", etc., for classification processing.

[0065] In this embodiment, through non-contact and batch automated RFID signal analysis means, the identification of the hidden damaged state of goods during their service life is realized. Compared with the traditional manual visual inspection method, this method has the following advantages: It can identify structural fatigue that is difficult to detect by the naked eye; it does not rely on manual inspection and is suitable for large-scale consumable management scenarios; data is recorded for each read and can be used for subsequent loss analysis and liability determination; combined with prediction algorithms, it can detect loss trends in advance and optimize the recycling and replacement cycles.

[0066] Refer to Figure 4 , this application also provides an embodiment of a system, which includes: An RFID unit 401 for embedding an RFID chip in the goods, and identification information including the unique ID of the goods, the category of the goods, and the purchase date is written in the RFID chip; A binding unit 402 for scanning the RFID chip through a goods binding terminal and associating the affiliated customer information to establish an electronic ledger of the goods; A waybill data generation unit 403 for obtaining the starting point information and destination information of the goods and calling a path optimization algorithm to generate waybill data including the transportation distance when the goods need to be transported across regions; A feature vector construction unit 404 for constructing corresponding feature vectors based on the transportation distance, time period coefficient, energy consumption data, and the reciprocal of vehicle production capacity in the waybill data; A machine learning unit 405 for inputting the feature vectors into a machine learning model trained by historical waybill data and outputting a predicted transportation cost; A result output unit 406 for comparing the first product result with the predicted transportation cost and determining the final transportation cost according to the comparison result, where the first product result is the product result of the basic unit price and the transportation distance.

[0067] Optionally, it further includes a damage detection unit 407 for: Collecting and recording the RFID signal values of the goods embedded with RFID chips continuously multiple times during washing and use; Calculate the RSSI mean value and the fluctuation amplitude based on the RSSI signal values collected continuously for multiple times; Compare the calculated RSSI mean value with a preset mean threshold, and compare the fluctuation amplitude with a preset fluctuation threshold; Determine the damaged state of the goods according to the comparison results.

[0068] Optionally, it further includes a damage detection unit 407, which is used for: When the RSSI mean value is lower than the preset mean threshold and / or the fluctuation amplitude is higher than the preset fluctuation threshold, determine that the goods are in a damaged state.

[0069] Optionally, the time period coefficient includes a weighted time cost parameter based on historical time period electricity prices, road congestion levels, and parking lot free time durations, and the weighted time cost parameter is used to reflect the economic efficiency of transportation tasks in different time periods.

[0070] Optionally, the machine learning model is constructed in the following manner: Collect historical waybill data within a preset time window, and the historical waybill data includes the following fields: transportation distance, start and end time periods, electricity price fluctuation information, actual energy consumption, vehicle model, vehicle load factor, parking fees, toll standards, historical traffic congestion coefficients, and actual transportation costs; Perform data cleaning and normalization processing on the historical waybill data, and perform feature crossing, discretization, and high-order combination to generate a multi-dimensional and highly coupled transportation task feature vector; Use a feature selection method based on information gain and L1 regularization to determine the optimal feature subset in the transportation task feature vector; Based on the optimal feature subset, train multiple regression models respectively. The multiple regression models include random forest regression, gradient boosting tree, support vector regression, and multi-layer perceptron neural network, and obtain the generalization performance indicators corresponding to each model through cross-validation; Use a weighted integration method to fuse the multiple regression models to obtain a fusion model.

[0071] Please refer to Figure 5 , this application also provides a full life cycle management device for multi-dimensional dynamic billing, including: A processor 501, a memory 502, an input / output unit 503, and a bus 504; The processor 501 is connected to the memory 502, the input / output unit 503, and the bus 504; The memory 502 stores a program, and the processor 501 calls the program to execute any of the above methods.

[0072] This application also relates to a computer-readable storage medium, on which a program is stored. The program, when running on a computer, causes the computer to execute any of the above methods.

[0073] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0074] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can be in electrical, mechanical, or other forms.

[0075] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0076] In addition, in each embodiment of this application, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0077] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, read-only memory), random access memories (RAM, random access memory), magnetic disks, or optical discs.

Claims

1. A full life cycle management method for multi-dimensional dynamic charging, characterized in that The method includes: Embedding an RFID chip in the goods, and writing identification information including the unique ID of the goods, the category of the goods, and the purchase date into the RFID chip; Scanning the RFID chip through a goods binding terminal, and associating the affiliated customer information to establish an electronic ledger of the goods; When the goods need to be transported across regions, obtaining the starting point information and destination information of the goods, and invoking a path optimization algorithm to generate waybill data including the transportation distance; Constructing a corresponding feature vector based on the transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity in the waybill data; Inputting the feature vector into a machine learning model trained by historical waybill data, and outputting the predicted transportation cost; Comparing the first product result with the predicted transportation cost, and determining the final transportation fee according to the comparison result, where the first product result is the product result of the base unit price and the transportation distance.

2. The full life cycle management method for multi-dimensional dynamic charging according to claim 1, characterized in that It also includes: Collecting and recording the RFID signal values of the goods embedded with the RFID chip continuously for multiple times during washing and use; Calculating the RSSI mean value and the fluctuation amplitude based on the continuously collected RSSI signal values; Comparing the calculated RSSI mean value with a preset mean value threshold, and comparing the fluctuation amplitude with a preset fluctuation threshold; Determining the damaged state of the goods according to the comparison result.

3. The full life cycle management method for multi-dimensional dynamic charging according to claim 2, characterized in that The determining the damaged state of the goods according to the comparison result includes: When the RSSI mean value is lower than the preset mean value threshold and / or the fluctuation amplitude is higher than the preset fluctuation threshold, it is determined that the goods are in a damaged state.

4. The full life cycle management method for multi-dimensional dynamic charging according to claim 1, characterized in that The time period coefficient includes a weighted time cost parameter based on historical time period electricity prices, road congestion levels, and free parking durations, and the weighted time cost parameter is used to reflect the economic efficiency of transportation tasks in different time periods.

5. The full life cycle management method for multi-dimensional dynamic charging according to claim 1, characterized in that, The construction of the machine learning model includes: Collecting historical waybill data within a preset time window, where the historical waybill data includes the following fields: transportation distance, start and end time periods, electricity price fluctuation information, actual energy consumption, vehicle model, vehicle load factor, parking fees, toll standards, historical traffic congestion coefficients, and actual transportation costs; Performing data cleaning and normalization processing on the historical waybill data, and performing feature crossing, discretization, and high-order combination to generate a multi-dimensional and highly coupled transportation task feature vector; Using a feature selection method based on information gain and L1 regularization to determine the optimal feature subset in the transportation task feature vector; Training multiple regression models based on the optimal feature subset, where the multiple regression models include random forest regression, gradient boosting tree, support vector regression, and multi-layer perceptron neural network, and obtaining the generalization performance indicators corresponding to each model through cross-validation; Fusing the multiple regression models using a weighted integration method to obtain a fusion model.

6. The full life cycle management method for multi-dimensional dynamic charging according to claim 5, characterized in that The performing data cleaning and normalization processing on the historical waybill data, and performing feature crossing, discretization, and high-order combination to generate a multi-dimensional and highly coupled transportation task feature vector includes: Performing data cleaning on the historical waybill data, where the data cleaning includes performing mean filling or category mode filling on missing values, and performing elimination processing based on statistical distribution boundaries on outliers; Compress the values of the cleaned numerical data to a unified dimension using the min-max normalization method or the Z-score standardization method; Perform feature engineering on the normalized numerical data, and the feature engineering includes: Perform feature cross-operation to combine multiple basic features into interactive features; perform feature discretization operation to divide continuous features into multiple discrete intervals; construct high-order combined features, and the high-order combined features at least include unit energy consumption, unit parking space cost, and driver delivery efficiency; Construct a transportation task feature vector based on the processing results to form a multi-dimensional and highly coupled data structure as the model input.

7. The full life cycle management method for multi-dimensional dynamic charging according to claim 5, characterized in that The method of determining the optimal feature subset in the transportation task feature vector by using the feature selection method based on information gain and L1 regularization includes: Calculate the discrimination ability of each candidate feature relative to the transportation cost based on the information gain method, and accordingly determine the feature subset with an information gain value higher than the preset threshold, and the candidate features are determined in the task feature vector; Train a sparse regression model based on the L1 regularization method, and automatically compress some feature coefficients to zero by minimizing the loss function with an L1 penalty term; Construct the features obtained after information gain and L1 regularization into an optimal feature subset.

8. The full life cycle management method for multi-dimensional dynamic charging according to claim 7, characterized in that The calculating the discrimination ability of each candidate feature relative to the transportation cost based on the information gain method, and accordingly determining the feature subset with an information gain value higher than the preset threshold, and the candidate features are determined in the task feature vector includes: Discretize the transportation cost into several categories; Calculate the joint entropy and conditional entropy between each candidate feature and the discretized transportation cost categories; Calculate the information gain value of each candidate feature according to the joint entropy and conditional entropy; Set an information gain threshold, and determine the candidate features with an information gain value higher than the information gain threshold as the feature subset.

9. The full life cycle management method for multi-dimensional dynamic charging according to claim 7, characterized in that The calculating the joint entropy and conditional entropy between each candidate feature and the discretized transportation cost categories includes: Calculate the joint frequency of each value of the candidate feature and each category of transportation cost, and calculate the joint probability distribution; Calculate the joint entropy H(X, Y) = -∑ x，y p(x, y) log p(x, y), where X is the candidate feature and Y is the transportation cost of the corresponding category; Calculate the transportation cost entropy H(Y) corresponding to the category: H(Y)=−∑ y p(y)logp(y) and the candidate feature conditional entropy H(Y∣X)=H(X,Y)−H(X); Calculate the information gain IG(Y|X) = H(Y) - H(Y|X) according to entropy, where H(X) = -∑ x p(x) log p(x) is the entropy of the candidate feature; Among them, the joint probability distribution p(x, y) represents the probability that the candidate feature X takes the value x and the corresponding transportation cost Y takes the value y, p(x) represents the probability that the candidate feature X takes the value x, that is, sum p(x, y) for all y; p(y) represents the probability that the transportation cost Y takes the value y; the joint entropy H(X, Y) is used to measure the uncertainty of the joint distribution of the candidate feature and the transportation cost; H(X) represents the uncertainty of the candidate feature, and H(Y) represents the uncertainty of the transportation cost category itself; the conditional entropy H(Y | X) represents the remaining uncertainty of the transportation cost Y given the candidate feature X; the information gain IG(Y | X) is used to measure the explanatory ability of the feature X for the classification variable Y; if the candidate feature X is highly correlated with the transportation cost, then H(Y|X) is small and the information gain IG(Y∣X) is larger, indicating that the feature is more valuable.

10. The full life cycle management system for multi-dimensional dynamic charging, characterized in that, Including: An RFID unit for embedding an RFID chip in the goods, and identification information including the unique ID of the goods, the goods category, and the purchase date is written in the RFID chip; A binding unit for scanning the RFID chip through a goods binding terminal and associating the customer information to establish an electronic ledger for goods; A waybill data generation unit for obtaining the starting point information and destination information of goods and calling a path optimization algorithm to generate waybill data including the transportation distance when the goods need to be transported across regions; A feature vector construction unit for constructing corresponding feature vectors based on the transportation distance, time period coefficient, energy consumption data, and reciprocal of vehicle production capacity in the waybill data; A machine learning unit for inputting the feature vectors into a machine learning model trained with historical waybill data and outputting the predicted transportation cost; A result output unit for comparing the first product result with the predicted transportation cost and determining the final transportation cost according to the comparison result, where the first product result is the product result of the basic unit price and the transportation distance.

Images