A logistics demand prediction and transport capacity elastic deployment method based on big data

Abstract

This invention discloses a method for logistics demand forecasting and flexible capacity allocation based on big data, relating to the field of big data processing. The method includes: establishing a high-quality data foundation through multi-source data collection and standardized processing; extracting core features using a hybrid model to achieve accurate demand forecasting; simultaneously integrating real-time capacity data to construct a dynamic evaluation system; generating an initial matching scheme based on the forecast results using a multi-objective optimization algorithm; and dynamically adjusting the scheme through real-time feedback; finally, evaluating the execution effect from four dimensions: timeliness, cost, efficiency, and service; and reverse-optimizing the forecasting model and allocation algorithm to form a closed-loop system of "data-driven - intelligent decision-making - dynamic control - feedback iteration". The advantages of this invention are: it uses full-link data-driven processing as its core, achieving accurate demand forecasting through multi-source data standardized processing and hybrid modeling; relying on IoT dynamic capacity management and multi-objective optimization algorithms to achieve flexible allocation; and efficiently balancing the multi-dimensional objectives of logistics operations.

Description

Technical Field

[0001] This invention relates to the field of big data processing, and in particular to a method for predicting logistics demand and flexibly allocating transportation capacity based on big data. Background Technology

[0002] With the explosive growth of e-commerce, deepening globalization, and consumers' increasing demands for timeliness, logistics demand exhibits high-frequency fluctuations and regional imbalances. Traditional static capacity planning models are ill-equipped to handle sudden order peaks, seasonal fluctuations, or supply-demand imbalances caused by unforeseen events. Demand forecasting, by integrating time series analysis, machine learning, and big data technologies, quantifies the correlation between historical data, market trends, and external variables to dynamically predict freight volume and reduce the "bullwhip effect." Meanwhile, flexible capacity allocation relies on digital platforms, the sharing economy model, and intelligent algorithms to achieve real-time optimization of resources such as vehicles, warehousing, and manpower, seeking a balance between cost and service levels.

[0003] Current logistics demand forecasting and capacity allocation methods struggle to achieve end-to-end collaborative optimization capabilities. At the data level, silos are prevalent, with insufficient integration of internal and external data sources. Preprocessing lacks standardized procedures and quality verification, leading to biased decision-making. Forecasting models often employ single algorithms, failing to capture non-linear demand fluctuations and long-term dependencies, or failing to accurately fit linear trends. Furthermore, they lack dynamic error correction mechanisms, resulting in insufficient forecast accuracy and refinement. Capacity management relies on fixed contracts or manual scheduling, leading to delayed status updates, insufficient integration of idle capacity, poor multimodal transport coordination, and reactive adjustments in response to emergencies. Optimization algorithms often focus on single objectives, neglecting multi-dimensional balance, lacking dynamic adjustment and global rebalancing capabilities, and employing a single evaluation dimension without a closed-loop iteration mechanism. This results in high empty-load rates, significant cost waste, and difficulty adapting to complex and ever-changing logistics operation scenarios. Summary of the Invention

[0004] To improve existing methods, this paper proposes a big data-based approach for logistics demand forecasting and flexible capacity allocation. This approach is driven by end-to-end data, achieves accurate demand forecasting through standardized processing of multi-source data and hybrid modeling, achieves flexible allocation by relying on IoT dynamic capacity management and multi-objective optimization algorithms, and combines four-dimensional evaluation and closed-loop iteration mechanisms to efficiently balance the multi-dimensional objectives of logistics operations.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for logistics demand forecasting and flexible capacity allocation based on big data includes: Collect multi-source logistics-related data across the entire domain, clean and denoise the collected data, standardize and transform it, and fill in missing values. Establish a data quality verification mechanism and build a unified data foundation. Temporal and correlation features are extracted from the preprocessed unified data to generate a set of derived features. Redundant terms are eliminated through feature importance evaluation, core features are retained, and standardized feature vectors are formed. A hybrid modeling approach combining deep learning and traditional statistical models is adopted. The core feature vector is input, an adaptive adjustment mechanism is established, the prediction error is corrected in real time, and the logistics demand prediction results are output by region, type, and time-sensitivity. Based on the Internet of Things, the transportation capacity status information of the entire domain is collected, the transportation capacity is classified and graded, and the fulfillment capability is evaluated in combination with historical data. A dynamic database is built to update the transportation capacity status information, classification and grading results and historical data in real time. Guided by demand forecasting results, a multi-objective optimization algorithm is designed, prioritizing objectives while taking into account multiple constraints. Through preliminary matching in different regions and global adjustment, an initial capacity matching scheme is generated. Based on the initial capacity matching plan, the plan is optimized by combining real-time feedback data to form an executable allocation plan. A dynamic adjustment mechanism is established to capture abnormal situations in real time and make dynamic adjustments through redundant calls and cross-regional scheduling. An evaluation system is established based on four dimensions: timeliness, cost, transportation efficiency, and service quality. The results are quantitatively compared with the preset goals to form an evaluation result. Based on the evaluation results, the logistics demand forecasting model and capacity allocation algorithm are iteratively optimized, parameters and constraints are adjusted, and a closed-loop mechanism is constructed.

[0006] Preferably, the process of collecting multi-source logistics-related data across the entire domain, cleaning and denoising the collected data, standardizing and transforming it, filling in missing values, establishing a data quality verification mechanism, and constructing a unified data foundation specifically includes: Collect multi-source logistics-related data across the entire domain, including logistics order data, cargo attribute data, transportation trajectory data, warehouse turnover data, regional economic data, meteorological data, and traffic condition data; The collected heterogeneous data is cleaned and denoised, and outliers, duplicates and missing data with missing rates exceeding a preset threshold are removed. Data of different formats and dimensions are converted into a unified standard format through data mapping rules, and classification imputation method is used to complete data with low missing rates. Establish a data quality verification mechanism to verify the integrity, consistency, and accuracy of preprocessed data.

[0007] Preferably, the step of extracting time-series and correlation features from the preprocessed unified data, generating a derived feature set, eliminating redundant terms through feature importance evaluation, retaining core features, and forming a standardized feature vector specifically includes: Feature extraction is performed on the preprocessed unified data to obtain time-series features of order generation time, transportation cycle, seasonal changes, and holiday distribution, and to generate the order volume fluctuation coefficient, peak period distribution, and timeliness demand level derived features within the time window. Analyze the correlation characteristics between cargo attributes and transportation modes, the correlation characteristics between regional economic activity and logistics demand, and the impact characteristics of meteorological and traffic factors on transportation timeliness to construct a multi-dimensional feature set; Redundant features are eliminated by feature importance assessment algorithm, and core features that have a significant impact on logistics demand forecasting are retained to form a standardized feature vector.

[0008] Preferably, the hybrid modeling approach combining deep learning and traditional statistical models, which inputs core feature vectors, establishes an adaptive adjustment mechanism, corrects prediction errors in real time, and outputs logistics demand prediction results by region, type, and time-sensitivity, specifically includes: A hybrid modeling approach combining deep learning models and traditional statistical models is adopted, with the generated core feature vectors as input; The deep learning model selects a time-series prediction network to capture the nonlinear fluctuations and long-term dependencies of demand, while a traditional statistical model fits the linear trends and periodic characteristics of demand. Establish a model adaptive adjustment mechanism to monitor prediction errors in real time. When the error exceeds the preset threshold, automatically optimize model parameters, update feature weights, and dynamically correct the prediction results by combining real-time data. Output logistics demand prediction results for the future preset time period by region, cargo type, and timeliness level.

[0009] Preferably, the step of collecting full-domain transportation capacity status information based on the Internet of Things, classifying and grading transportation capacity, evaluating performance capabilities in conjunction with historical data, and constructing a dynamic database to update transportation capacity status information, classification and grading results, and historical data in real time specifically includes: The system collects real-time status information of its own transportation capacity, cooperative transportation capacity, and idle social transportation capacity through IoT devices, including vehicle location, load status, vehicle condition level, driver qualifications, and available time window. Classify and grade transportation resources, evaluate the fulfillment rate, failure rate and cost control capability of each transportation capacity based on historical operating data, and classify transportation capacity levels according to indicators such as transportation capacity, timeliness guarantee level and service quality rating. Build a dynamic capacity database to update capacity status information, rating results, and historical operational data in real time.

[0010] Preferably, the step of designing a multi-objective optimization algorithm based on demand forecasting results, setting priority objectives and taking into account multiple constraints, and generating an initial capacity matching scheme through regional preliminary matching and global adjustment specifically includes: Guided by logistics demand forecasting results and combined with dynamic capacity database information, a multi-objective optimization algorithm is designed to set capacity matching priority objectives, including maximizing the timeliness achievement rate, minimizing transportation costs, maximizing capacity utilization, and minimizing empty-running rate. During the algorithm's operation, constraints are set, including capacity and load limits, transportation route limits, timeliness requirements, and vehicle condition suitability requirements. By breaking down the supply and demand matching task using dynamic programming, preliminary matching is first achieved by region, and then global capacity adjustment is carried out for cross-regional and high-time-efficiency demands to generate an initial capacity matching plan.

[0011] Preferably, the step of optimizing the initial capacity matching scheme based on real-time feedback data to form an executable allocation scheme, establishing a dynamic adjustment mechanism, capturing abnormal situations in real time, and dynamically adjusting through redundant calls and cross-regional scheduling specifically includes: Based on the initial capacity matching plan, the plan is optimized by combining real-time feedback data to generate an executable capacity flexible allocation plan, which clarifies the transportation tasks, route planning, loading time, unloading time and connection nodes of each capacity. Establish a dynamic adjustment mechanism for allocation plans to capture sudden changes in demand, transportation capacity failures, and abnormal road conditions in real time. When an anomaly occurs, the plan adjustment process is triggered, and an adjusted allocation plan is generated through methods such as calling up redundant capacity reserves, cross-regional capacity scheduling, and replanning transportation routes.

[0012] Preferably, the evaluation system established from four dimensions—timeliness, cost, transport capacity efficiency, and service quality—quantifies and compares the execution results with the preset targets to form the evaluation results, specifically including: Establish a multi-dimensional evaluation indicator system, setting evaluation indicators from four dimensions: timeliness guarantee, cost control, transportation efficiency, and service quality; The timeliness guarantee indicators include on-time order fulfillment rate and timeliness deviation rate; the cost control indicators include unit cargo transportation cost and empty driving cost ratio; the capacity efficiency indicators include capacity utilization rate and vehicle turnover rate; and the service quality indicators include cargo damage rate and customer satisfaction. By comparing the actual execution results with the predicted targets and preset standards, a quantitative evaluation of the allocation effect is completed, and an evaluation result is formed.

[0013] Preferably, the iterative optimization of the logistics demand forecasting model and capacity allocation algorithm based on the evaluation results, adjusting parameters and constraints, and constructing a closed-loop mechanism specifically includes: Based on the evaluation results, the logistics demand forecasting model and capacity allocation algorithm are iteratively optimized. For the prediction model, feature weights are adjusted and model parameters are optimized by combining actual needs and deviation data, and the data collection dimensions are expanded to improve prediction accuracy; For the allocation algorithm, the objective function weights are optimized, the constraint conditions are adjusted, and historical optimization experience and anomaly handling cases are incorporated to improve the algorithm's adaptability and robustness. A closed-loop mechanism is formed, consisting of data collection, predictive modeling, capacity allocation, effect evaluation, and iterative optimization.

[0014] Compared with the prior art, the advantages of the present invention are: By collecting multi-source data across the entire domain and performing standardized preprocessing, coupled with a quality verification mechanism, a reliable data foundation is built, laying the groundwork for accurate decision-making. A hybrid modeling approach, merging deep and shallow models and incorporating feature optimization techniques, enables refined demand forecasting by region, type, and timeframe, taking into account both linear and nonlinear patterns, significantly improving forecast accuracy. A dynamic capacity database is built based on the Internet of Things (IoT), enabling capacity classification, grading, and real-time management. Combined with multi-objective optimization algorithms and dynamic adjustment mechanisms, it achieves efficient capacity adaptation through regional matching and global allocation, while also quickly responding to anomalies, ensuring flexibility and stability in allocation. A four-dimensional evaluation system and closed-loop iteration mechanism are established to continuously optimize model and algorithm parameters, effectively balancing timeliness, cost, efficiency, and service quality, significantly improving the intelligence, refinement, and overall competitiveness of logistics operations. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method proposed in this invention; Figure 2 This is a schematic diagram of the logistics data collection and processing proposed in this invention; Figure 3 This is a schematic diagram illustrating the construction of the logistics demand characteristic engineering proposed in this invention; Figure 4 This is a schematic diagram illustrating the construction of the dynamic adaptive hybrid prediction model proposed in this invention; Figure 5 This is a schematic diagram of the capacity resource perception and status assessment proposed in this invention; Figure 6 This is a schematic diagram illustrating the design and deployment of the capacity optimization algorithm proposed in this invention; Figure 7 This is a schematic diagram illustrating the generation of the flexible capacity allocation scheme proposed in this invention; Figure 8 This is a schematic diagram illustrating the evaluation and optimization of the blending effect proposed in this invention; Figure 9 This is a schematic diagram illustrating the iterative optimization of the prediction model and allocation algorithm proposed in this invention. Detailed Implementation

[0016] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0017] See Figure 1 As shown, a method for logistics demand forecasting and flexible capacity allocation based on big data includes: Step 1: Collect multi-source logistics-related data across the entire domain, clean and denoise the collected data, standardize and transform it, and complete missing values. Establish a data quality verification mechanism and build a unified data foundation. Step 2: Extract time-series and correlation features from the preprocessed unified data, generate a derived feature set, eliminate redundant items through feature importance evaluation, retain core features, and form a standardized feature vector; Step 3: Using a hybrid modeling approach that combines deep learning with traditional statistical models, the core feature vector is input, an adaptive adjustment mechanism is established to correct prediction errors in real time, and the logistics demand prediction results are output by region, type, and timeliness. Step 4: Based on the Internet of Things, collect the overall transportation capacity status information, classify and grade the transportation capacity, combine historical data to evaluate the performance capability, and build a dynamic database to update the transportation capacity status information, classification and grading results and historical data in real time; Step 5: Based on the demand forecast results, design a multi-objective optimization algorithm, set priority objectives and take into account multiple constraints, and generate an initial capacity matching scheme through preliminary matching in different regions and global adjustment. Step Six: Based on the initial capacity matching plan, optimize the plan by combining real-time feedback data to form an executable allocation plan, establish a dynamic adjustment mechanism, capture abnormal situations in real time, and make dynamic adjustments through redundant calls and cross-regional scheduling. Step 7: Establish an evaluation system based on four dimensions: timeliness, cost, transportation efficiency, and service quality; quantitatively compare the execution results with the preset goals to form evaluation results. Step 8: Based on the evaluation results, iteratively optimize the logistics demand forecasting model and capacity allocation algorithm, adjust parameters and constraints, and build a closed-loop mechanism.

[0018] See Figure 2 As shown, the process involves collecting multi-source logistics data across the entire domain, cleaning and denoising the collected data, standardizing and transforming it, completing missing values, establishing a data quality verification mechanism, and constructing a unified data foundation. Specifically, this includes: Collect multi-source logistics-related data across the entire domain, including logistics order data, cargo attribute data, transportation trajectory data, warehouse turnover data, regional economic data, meteorological data, and traffic condition data; The collected heterogeneous data is cleaned and denoised, and outliers, duplicates and missing data with missing rates exceeding a preset threshold are removed. Data of different formats and dimensions are converted into a unified standard format through data mapping rules, and classification imputation method is used to complete data with low missing rates. Establish a data quality verification mechanism to verify the integrity, consistency, and accuracy of preprocessed data.

[0019] Specifically, statistical analysis is used to identify outliers. For numerical data such as weight and volume, data deviating from the mean by three times the standard deviation is removed. For data such as order time and trajectory coordinates, data exceeding a reasonable range is removed. Duplicate values ​​are retained based on the "latest timestamp" principle, and the rest are deleted. Data corresponding to fields with a missing rate exceeding 5% is directly removed. Data standardization is implemented, and a unified data format specification is established. Through a pre-set data mapping table, data fields from different platforms and formats are uniformly mapped to standard fields. Data with a low missing rate (≤5%) is completed using a classification imputation method. Numerical fields are filled with the median of data from the same category and region, textual fields are filled with the mode, and time-based fields are filled according to business logic calculations. A three-dimensional data quality verification mechanism is established. Integrity verification ensures that core fields are not missing. Consistency verification verifies that the address code matches the actual address and that the order weight and volume conform to the corresponding cargo density range. Accuracy verification compares offline documents with system data through sampling, with the error rate controlled within 1%. Data that passes verification is entered into a unified data base.

[0020] See Figure 3 As shown, time-series and correlation features are extracted from the preprocessed unified data to generate a derived feature set. Redundant terms are eliminated through feature importance evaluation, and core features are retained to form a standardized feature vector. Specifically, this includes: Feature extraction is performed on the preprocessed unified data to obtain time-series features of order generation time, transportation cycle, seasonal changes, and holiday distribution, and to generate the order volume fluctuation coefficient, peak period distribution, and timeliness demand level derived features within the time window. Analyze the correlation characteristics between cargo attributes and transportation modes, the correlation characteristics between regional economic activity and logistics demand, and the impact characteristics of meteorological and traffic factors on transportation timeliness to construct a multi-dimensional feature set; Redundant features are eliminated by feature importance assessment algorithm, and core features that have a significant impact on logistics demand forecasting are retained to form a standardized feature vector.

[0021] Specifically, the system deeply mines and derives time-series features, employing a sliding time window method to split data, dividing it into multi-granularity time windows by hour, day, week, and month to accurately capture the time-series patterns of logistics demand. For order time-series features, it extracts basic time-series fields such as order generation timestamp, estimated fulfillment cycle, and actual transportation time, calculating the month-on-month growth rate and year-on-year fluctuation coefficient of order volume within each time window, and marking peak, trough, and duration periods for orders. Combining Gregorian calendar holidays, traditional lunar calendar festivals, e-commerce promotional events, and regional festivals, it generates a "holiday weight factor" to quantify the impact of different nodes on demand increments. Based on historical data mining of seasonal variation patterns, such as the time-series characteristics of upgraded fresh food order timeliness demand in summer and the surge in cold chain transportation demand in winter, it derives a "seasonal demand adaptation coefficient" to improve the time-series feature set. Redundant features were eliminated by combining feature contribution evaluation with cross-validation: the explanatory power of each feature on logistics demand fluctuations was statistically analyzed, and the top 80% of features were ranked and selected. After multiple rounds of cross-validation, redundant items with multicollinearity and information overlap were eliminated, and core features such as time-series fluctuation coefficient, holiday weight factor, cargo-transportation mode matching degree, regional demand potential index, and environmental impact correction factor were retained. The selected features were standardized, the feature value range was unified, and they were classified and sorted according to "time-series features - related features". Each core feature was assigned a corresponding weight, and a standardized feature vector with unified dimensions and clear logic was constructed.

[0022] See Figure 4 As shown, a hybrid modeling approach combining deep learning and traditional statistical models is adopted. Core feature vectors are input, an adaptive adjustment mechanism is established to correct prediction errors in real time, and the output logistics demand prediction results by region, type, and timeframe specifically include: A hybrid modeling approach combining deep learning models and traditional statistical models is adopted, with the generated core feature vectors as input; The deep learning model selects a time-series prediction network to capture the nonlinear fluctuations and long-term dependencies of demand, while a traditional statistical model fits the linear trends and periodic characteristics of demand. Establish a model adaptive adjustment mechanism to monitor prediction errors in real time. When the error exceeds the preset threshold, automatically optimize model parameters, update feature weights, and dynamically correct the prediction results by combining real-time data. Output logistics demand prediction results for the future preset time period by region, cargo type, and timeliness level.

[0023] Specifically, the core modules and hierarchical connections of the model are as follows: The core comprises three main modules: First, the LSTM nonlinear feature capture module, with three hidden layers. The first layer is the input mapping layer, which converts feature vectors into hidden states recognizable by the model; the second layer is the temporal memory layer, which retains long-term demand patterns and forgets invalid short-term fluctuations through a gating mechanism; the third layer is the output mapping layer, which outputs the nonlinear demand prediction component. Second, the ARIMA linear trend fitting module, containing an autoregressive layer, a difference layer, and a moving average layer. The difference layer eliminates data non-stationarity, the autoregressive layer fits the linear correlation of historical demand, and the moving average layer smooths random fluctuations, outputting the linear demand prediction component. Third, the adaptive fusion and adjustment module, bidirectionally connected to the first two modules, is responsible for fusing the results of the two components and dynamically optimizing the model. The hierarchical connections adopt a "feature input → parallel operation of two modules → component fusion → result output" chain. The fusion module achieves coupling of the two components through dynamic weight allocation, with the weights adjusted in real time according to the prediction error. Training data is taken from generated standardized feature vectors and divided into training, validation, and test sets in a 7:2:1 ratio. Initialization parameters: LSTM module hidden layer neurons are set to 128, initial learning rate is 0.01, batch size is 32, number of iterations is 100, and dropout coefficient is 0.2; ARIMA module autoregression order is set to 3, difference order to 1, and moving average order to 3. Training steps: First, train the two modules separately. LSTM uses the training set to iteratively update weights, and validates every 10 epochs using the validation set. If the validation error increases for 3 consecutive epochs, an early stopping mechanism is triggered. ARIMA calibrates linear parameters and optimizes the residual sum of squares using the training set. Then, the fusion module is trained jointly, and the overall accuracy is validated using the test set. The output weights of the two modules are dynamically adjusted until the prediction error is below a preset threshold.

[0024] Scenario adaptation is achieved through module fine-tuning. For fresh food cold chain scenarios, the LSTM module's ability to capture short-term weather and time-sensitive demand characteristics is enhanced, and the sensitivity of the memory layer to temperature factors is increased. For cross-regional bulk cargo scenarios, the linear trend fitting capability of the ARIMA module is optimized to adapt to demand fluctuations driven by long-term economic data. Feature vectors are directly accessed, and the input dimensions are organized according to "time-series features + correlation features" to output three-dimensional demand prediction results by region, cargo type, and time-sensitive time, along with the prediction confidence of each dimension. The output data is directly connected to the capacity optimization algorithm.

[0025] See Figure 5 As shown, based on the Internet of Things (IoT) to collect full-domain transportation capacity status information, transportation capacity is classified and graded, and historical data is combined to assess fulfillment capabilities. A dynamic database is constructed to update transportation capacity status information, classification and grading results, and historical data in real time. Specifically, this includes: The system collects real-time status information of its own transportation capacity, cooperative transportation capacity, and idle social transportation capacity through IoT devices, including vehicle location, load status, vehicle condition level, driver qualifications, and available time window. Classify and grade transportation resources, evaluate the fulfillment rate, failure rate and cost control capability of each transportation capacity based on historical operating data, and classify transportation capacity levels according to indicators such as transportation capacity, timeliness guarantee level and service quality rating. Build a dynamic capacity database to update capacity status information, rating results, and historical operational data in real time.

[0026] Specifically, the transportation capacity is classified and graded for management and status assessment. First, it is classified according to transportation capacity and service type into four major categories: heavy cargo transportation capacity, cold chain transportation capacity, general cargo transportation capacity, and same-city delivery capacity. Each category is further subdivided into subcategories according to vehicle specifications. The grading adopts a percentage-based quantitative assessment, with core indicators including transportation capacity, timeliness guarantee level, service quality, and operational stability.

[0027] See Figure 6 As shown, guided by demand forecasting results, a multi-objective optimization algorithm is designed, setting priority objectives and considering multiple constraints. Through preliminary regional matching and global adjustment, an initial capacity matching scheme is generated, specifically including: Guided by logistics demand forecasting results and combined with dynamic capacity database information, a multi-objective optimization algorithm is designed to set capacity matching priority objectives, including maximizing the timeliness achievement rate, minimizing transportation costs, maximizing capacity utilization, and minimizing empty-running rate. During the algorithm's operation, constraints are set, including capacity and load limits, transportation route limits, timeliness requirements, and vehicle condition suitability requirements. By breaking down the supply and demand matching task using dynamic programming, preliminary matching is first achieved by region, and then global capacity adjustment is carried out for cross-regional and high-time-efficiency demands to generate an initial capacity matching plan.

[0028] Specifically, a multi-objective optimization algorithm architecture is adopted, defining four core objectives and dynamic priority rules: maximizing the timeliness of delivery, prioritizing high-frequency, time-sensitive orders such as fresh produce and emergency supplies; minimizing transportation costs, reducing unit cargo transportation and empty-run costs through route optimization and grouping, with priority applied to general bulk cargo orders; maximizing capacity utilization, prioritizing the allocation of orders from the same region and route to idle capacity to reduce wasted capacity; and minimizing empty-run rate, planning return orders based on return routes to achieve a closed loop of "outbound fulfillment + return cargo delivery." The multi-objective optimization objective function is:

[0029] in, To optimize the overall score across multiple objectives, , , , The weighting coefficients for timeliness compliance rate, cost control, capacity utilization rate, and empty-run rate. To ensure the timeliness of compliance, This represents the actual total transportation cost. To preset a maximum transportation cost, To improve capacity utilization, Empty running rate of transport capacity; Before the algorithm runs, all constraints are reviewed to form feasible control rules: Regarding load constraints, the total weight and volume of goods are strictly matched with the vehicle's rated load and cargo compartment volume, with a 10% safety margin reserved to prevent overloading; regarding route constraints, traffic data and policy restrictions are considered to avoid congested roads and temporary control areas, with cross-regional orders prioritizing highway + national road combinations; regarding timeliness constraints, loading, transportation, and unloading time nodes are scheduled in reverse order based on the order's timeliness level, with emergency buffer time reserved; regarding vehicle condition compatibility constraints, cold chain orders are matched with transport capacity equipped with refrigeration equipment and meeting operating standards, while large-item orders are assigned to dedicated heavy-duty vehicles; regarding compliance constraints, the operating qualifications of transport capacity and the professional scope of drivers are verified. The matching task is decomposed using dynamic programming and executed in two steps: The first step is preliminary matching by region. Matching units are divided according to administrative divisions and transportation radius. The demand orders in each unit are initially matched with the available transportation capacity in that region. Priority is given to matching orders and transportation capacity of the same type of goods and the same time level. For example, same-city delivery orders are matched with short-distance transportation capacity such as 4.2-meter box trucks, and cross-regional large-item orders are matched with semi-trailer transportation capacity, generating a preliminary matching list for the region. The second step is global transportation capacity adjustment. For cross-regional, high-time-sensitivity, and regionally insufficient transportation capacity, a global allocation mechanism is activated to call on suitable transportation capacity from surrounding areas. At the same time, idle high-quality transportation capacity in society is activated as a supplement. The location and availability of adjustable transportation capacity are queried in real time through the dynamic transportation capacity database. The transportation capacity scheduling cost and time loss are calculated to select the optimal adjustment plan.

[0030] See Figure 7 As shown, based on the initial capacity matching plan, the plan is optimized by combining real-time feedback data to form an executable allocation plan. A dynamic adjustment mechanism is established to capture abnormal situations in real time. Dynamic adjustments are made through redundant calls and cross-regional scheduling, specifically including: Based on the initial capacity matching plan, the plan is optimized by combining real-time feedback data to generate an executable capacity flexible allocation plan, which clarifies the transportation tasks, route planning, loading time, unloading time and connection nodes of each capacity. Establish a dynamic adjustment mechanism for allocation plans to capture sudden changes in demand, transportation capacity failures, and abnormal road conditions in real time. When an anomaly occurs, the plan adjustment process is triggered, and an adjusted allocation plan is generated through methods such as calling up redundant capacity reserves, cross-regional capacity scheduling, and replanning transportation routes.

[0031] Specifically, the system connects to a dynamic capacity database, a real-time road condition monitoring system, and an order management platform to supplement the data collected in the initial plan. This includes the current real-time location of each transport unit, congestion at nearby loading points, remaining working hours for drivers, and progress of cargo preparation. Based on this supplementary data, the plan details are optimized, and the precise task allocation for each transport unit is clarified: loading order is prioritized according to the principle of "loading nearby and following the correct route" to avoid cross-regional backtracking; the optimal transportation route is planned based on real-time road conditions, marking service areas, emergency avoidance points, and alternative routes; precise time nodes are set, including loading start and end times, transit time, unloading deadline, and buffer time between each stage. For multi-capacity collaborative transportation orders, the connection nodes and responsibility boundaries of each transport unit are clarified, generating a standardized and executable plan that includes a task list, route map, time nodes, and responsible parties. After triggering the adjustment process, accurately pinpoint the core cause of the anomaly and its scope of impact. For example, if there is a capacity failure, the type of failure, repair time, and whether it is replaceable must be clearly identified. Then, the emergency allocation strategy is activated, and corresponding measures are taken for different scenarios: when there is insufficient capacity, the redundant capacity reserve is called up; when there is cross-regional scheduling, the surrounding regional dispatch centers are coordinated to coordinate available capacity; when the route is blocked, the optimal alternative route is generated through route replanning algorithm; when the demand changes, the capacity is rematched and the original task allocation is adjusted. The adjusted plan must be quickly verified for compliance, and it is checked whether it meets the constraints of load, timeliness, qualifications, etc.

[0032] See Figure 8 As shown, an evaluation system is established from four dimensions: timeliness, cost, transport capacity efficiency, and service quality. The system quantitatively compares the execution results with the preset goals, and the evaluation results specifically include: Establish a multi-dimensional evaluation indicator system, setting evaluation indicators from four dimensions: timeliness guarantee, cost control, transportation efficiency, and service quality; The timeliness guarantee indicators include on-time order fulfillment rate and timeliness deviation rate; the cost control indicators include unit cargo transportation cost and empty driving cost ratio; the capacity efficiency indicators include capacity utilization rate and vehicle turnover rate; and the service quality indicators include cargo damage rate and customer satisfaction. By comparing the actual execution results with the predicted targets and preset standards, a quantitative evaluation of the allocation effect is completed, and an evaluation result is formed.

[0033] Specifically, based on the four dimensions of timeliness guarantee, cost control, transportation efficiency, and service quality, concrete indicators are broken down and statistical rules are clarified, while weights are dynamically allocated according to business scenarios. Timeliness guarantee dimension (weight 30%): On-time delivery rate is calculated as the percentage of orders whose actual delivery time is within the agreed time limit; timeliness deviation rate is calculated as the percentage of the difference between actual and estimated time, and statistics are also divided into same-city, cross-regional, and cold chain scenarios. Cost control dimension (weight 25%): Unit cargo transportation cost is calculated as the ratio of total cost including fuel, tolls, and labor costs to cargo weight / volume; empty run cost ratio is calculated as the ratio of cost generated by empty run mileage to total transportation cost, with a focus on monitoring cross-regional return empty run situations. Transportation efficiency dimension (weight 25%): Transportation capacity utilization rate is calculated as the ratio of actual operating time to available time; vehicle turnover rate is calculated as the number of order batches completed by a single vehicle per day, and is evaluated according to transportation capacity level. Service quality dimension (weight 20%): Cargo damage rate is calculated as the percentage of damaged / deteriorated goods valued according to cargo type; customer satisfaction is calculated by weighting online ratings after order completion.

[0034] See Figure 9 As shown, based on the evaluation results, the logistics demand forecasting model and capacity allocation algorithm are iteratively optimized, parameters and constraints are adjusted, and a closed-loop mechanism is constructed, specifically including: Based on the evaluation results, the logistics demand forecasting model and capacity allocation algorithm are iteratively optimized. For the prediction model, feature weights are adjusted and model parameters are optimized by combining actual needs and deviation data, and the data collection dimensions are expanded to improve prediction accuracy; For the allocation algorithm, the objective function weights are optimized, the constraint conditions are adjusted, and historical optimization experience and anomaly handling cases are incorporated to improve the algorithm's adaptability and robustness. A closed-loop mechanism is formed, consisting of data collection, predictive modeling, capacity allocation, effect evaluation, and iterative optimization.

[0035] Specifically, for the hybrid prediction model, optimizations were made in three aspects: First, feature weight iteration: based on the feature contribution analysis in the evaluation report, the weights of features that significantly affect prediction accuracy were increased, such as adding the weight of temperature factors in cold chain scenarios and increasing the weight of regional economic data in cross-regional scenarios, while eliminating redundant features; Second, model parameter optimization: the number of neurons in the hidden layer of the LSTM module and the dropout coefficient were adjusted, and the order setting of the ARIMA module was optimized to adapt to the fluctuation patterns of different scenario requirements; Third, model structure fine-tuning: if the prediction deviation of a certain type of scenario is consistently high, an attention mechanism module was added to strengthen the capture of key features; After optimization, the prediction error and confidence index before and after optimization were compared through dual verification with test sets and real-time data to ensure that the prediction accuracy was improved by more than 10%; For the multi-objective optimization algorithm, the focus is on optimizing the objective function, constraints, and matching logic: At the objective function level, the weight allocation is adjusted based on the evaluation results. For example, if the empty-run rate is too high, the weight of matching return orders is increased; if the cargo damage rate exceeds the standard, the weight of vehicle condition adaptation is increased. At the constraint level, the constraint rules for special scenarios are refined, such as supplementing the operating threshold constraints of refrigeration equipment for cold chain orders and strengthening the route compliance constraints for hazardous chemical transportation. At the matching logic level, historical optimization experience and anomaly handling cases are incorporated to optimize the priority of global capacity allocation, shorten the response time of cross-regional capacity call, and transform step-by-step dynamic adjustment cases into algorithm rules to improve the algorithm's adaptability to sudden scenarios. After optimization, the robustness of the algorithm is tested by simulating different business scenarios.

[0036] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0037] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for logistics demand forecasting and flexible capacity allocation based on big data, characterized in that, include: Collect multi-source logistics-related data across the entire domain, clean and denoise the collected data, standardize and transform it, and fill in missing values. Establish a data quality verification mechanism and build a unified data foundation. Temporal and correlation features are extracted from the preprocessed unified data to generate a set of derived features. Redundant terms are eliminated through feature importance evaluation, core features are retained, and standardized feature vectors are formed. A hybrid modeling approach combining deep learning and traditional statistical models is adopted. The core feature vector is input, an adaptive adjustment mechanism is established, the prediction error is corrected in real time, and the logistics demand prediction results are output by region, type, and time-sensitivity. Based on the Internet of Things, the transportation capacity status information of the entire domain is collected, the transportation capacity is classified and graded, and the fulfillment capability is evaluated in combination with historical data. A dynamic database is built to update the transportation capacity status information, classification and grading results and historical data in real time. Guided by demand forecasting results, a multi-objective optimization algorithm is designed, prioritizing objectives while taking into account multiple constraints. Through preliminary matching in different regions and global adjustment, an initial capacity matching scheme is generated. Based on the initial capacity matching plan, the plan is optimized by combining real-time feedback data to form an executable allocation plan. A dynamic adjustment mechanism is established to capture abnormal situations in real time and make dynamic adjustments through redundant calls and cross-regional scheduling. An evaluation system is established based on four dimensions: timeliness, cost, transportation efficiency, and service quality. The results are quantitatively compared with the preset goals to form an evaluation result. Based on the evaluation results, the logistics demand forecasting model and capacity allocation algorithm are iteratively optimized, parameters and constraints are adjusted, and a closed-loop mechanism is constructed.

2. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The process of collecting multi-source logistics-related data across the entire domain, cleaning and denoising the collected data, standardizing and transforming it, filling in missing values, establishing a data quality verification mechanism, and constructing a unified data foundation specifically includes: Collect multi-source logistics-related data across the entire domain, including logistics order data, cargo attribute data, transportation trajectory data, warehouse turnover data, regional economic data, meteorological data, and traffic condition data; The collected heterogeneous data is cleaned and denoised, and outliers, duplicates and missing data with missing rates exceeding a preset threshold are removed. Data of different formats and dimensions are converted into a unified standard format through data mapping rules, and classification imputation method is used to complete data with low missing rates. Establish a data quality verification mechanism to verify the integrity, consistency, and accuracy of preprocessed data.

3. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The process of extracting time-series and correlation features from the preprocessed unified data, generating a derived feature set, eliminating redundant terms through feature importance evaluation, retaining core features, and forming a standardized feature vector specifically includes: Feature extraction is performed on the preprocessed unified data to obtain time-series features of order generation time, transportation cycle, seasonal changes, and holiday distribution, and to generate the order volume fluctuation coefficient, peak period distribution, and timeliness demand level derived features within the time window. Analyze the correlation characteristics between cargo attributes and transportation modes, the correlation characteristics between regional economic activity and logistics demand, and the impact characteristics of meteorological and traffic factors on transportation timeliness to construct a multi-dimensional feature set; Redundant features are eliminated by feature importance assessment algorithm, and core features that have a significant impact on logistics demand forecasting are retained to form a standardized feature vector.

4. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The hybrid modeling approach, which combines deep learning with traditional statistical models, inputs core feature vectors, establishes an adaptive adjustment mechanism, corrects prediction errors in real time, and outputs logistics demand prediction results by region, type, and time-sensitivity. Specifically, this includes: A hybrid modeling approach combining deep learning models and traditional statistical models is adopted, with the generated core feature vectors as input; The deep learning model selects a time-series prediction network to capture the nonlinear fluctuations and long-term dependencies of demand, while a traditional statistical model fits the linear trends and periodic characteristics of demand. Establish a model adaptive adjustment mechanism to monitor prediction errors in real time. When the error exceeds the preset threshold, automatically optimize model parameters, update feature weights, and dynamically correct the prediction results by combining real-time data. Output logistics demand prediction results for the future preset time period by region, cargo type, and timeliness level.

5. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The process of collecting full-domain transportation capacity status information based on the Internet of Things, classifying and grading transportation capacity, evaluating performance capabilities by combining historical data, and constructing a dynamic database to update transportation capacity status information, classification and grading results, and historical data in real time specifically includes: The system collects real-time status information of its own transportation capacity, cooperative transportation capacity, and idle social transportation capacity through IoT devices, including vehicle location, load status, vehicle condition level, driver qualifications, and available time window. Classify and grade transportation resources, evaluate the fulfillment rate, failure rate and cost control capability of each transportation capacity based on historical operating data, and classify transportation capacity levels according to indicators such as transportation capacity, timeliness guarantee level and service quality rating. Build a dynamic capacity database to update capacity status information, rating results, and historical operating data in real time.

6. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The process of designing a multi-objective optimization algorithm based on demand forecasting results, setting priority objectives while considering multiple constraints, and generating an initial capacity matching scheme through regional preliminary matching and global adjustment specifically includes: Guided by logistics demand forecasting results and combined with dynamic capacity database information, a multi-objective optimization algorithm is designed to set capacity matching priority objectives, including maximizing the timeliness achievement rate, minimizing transportation costs, maximizing capacity utilization, and minimizing empty-running rate. During the algorithm's operation, constraints are set, including capacity and load limits, transportation route limits, timeliness requirements, and vehicle condition suitability requirements. By decomposing the supply and demand matching task using dynamic programming, preliminary matching is first achieved by region, and then global capacity adjustment is carried out for cross-regional and high-time-efficiency demands to generate an initial capacity matching plan.

7. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The process of optimizing the initial capacity matching scheme based on real-time feedback data to form an executable allocation scheme, establishing a dynamic adjustment mechanism, capturing abnormal situations in real time, and dynamically adjusting through redundant calls and cross-regional scheduling specifically includes: Based on the initial capacity matching plan, the plan is optimized by combining real-time feedback data to generate an executable capacity flexible allocation plan, which clarifies the transportation tasks, route planning, loading time, unloading time and connection nodes of each capacity. Establish a dynamic adjustment mechanism for allocation plans to capture sudden changes in demand, transportation capacity failures, and abnormal road conditions in real time. When an anomaly occurs, the plan adjustment process is triggered, and an adjusted allocation plan is generated through methods such as calling up redundant capacity reserves, cross-regional capacity scheduling, and replanning transportation routes.

8. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The evaluation system, established from four dimensions—timeliness, cost, transport efficiency, and service quality—quantifies and compares the execution results with the preset targets, forming the evaluation results, which specifically include: Establish a multi-dimensional evaluation indicator system, setting evaluation indicators from four dimensions: timeliness guarantee, cost control, transportation efficiency, and service quality; The timeliness guarantee indicators include on-time order fulfillment rate and timeliness deviation rate; the cost control indicators include unit cargo transportation cost and empty driving cost ratio; the capacity efficiency indicators include capacity utilization rate and vehicle turnover rate; and the service quality indicators include cargo damage rate and customer satisfaction. By comparing the actual execution results with the predicted targets and preset standards, a quantitative evaluation of the allocation effect is completed, and an evaluation result is formed.

9. The method for logistics demand forecasting and flexible capacity allocation based on big data according to claim 1, characterized in that, The iterative optimization of the logistics demand forecasting model and capacity allocation algorithm based on the evaluation results, adjusting parameters and constraints, and constructing a closed-loop mechanism specifically includes: Based on the evaluation results, the logistics demand forecasting model and capacity allocation algorithm are iteratively optimized. For the prediction model, feature weights are adjusted and model parameters are optimized by combining actual needs and deviation data, and the data collection dimensions are expanded to improve prediction accuracy. For the allocation algorithm, the objective function weights are optimized, the constraint settings are adjusted, and historical optimization experience and anomaly handling cases are incorporated to improve the algorithm's adaptability and robustness. A closed-loop mechanism is formed, consisting of data collection, predictive modeling, capacity allocation, effect evaluation, and iterative optimization.

Images