Logistics supply chain management system

Abstract

The invention relates to the field of logistics supply, and discloses a logistics supply chain management system which comprises a data access module, the data access module imports multi-source data into the system and transmits the data to a data storage and management module, the data access module is connected with the data storage and management module, and the data storage and management module is connected with the data access module. The data storage and management module is connected with a data fusion and feature weighting module, and the data fusion and feature weighting module extracts data in the data storage and management module and predicts the time node demand after multiple features in cooperation with a set space-time fusion prediction model; and performing prediction calculation on the demand quantity at different time points through the space-time fusion prediction model. According to the method, the data fusion and feature weighting module is combined with the space-time fusion prediction model, multi-source data is integrated, sudden demands and long-tail commodity fluctuation can be effectively dealt with, and the demand prediction accuracy is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics supply, in particular to a logistics supply chain management system. BACKGROUND

[0002] The logistics supply chain management system belongs to the important field of logistics and supply chain management, computer information processing and data analysis technology. The system integrates resources such as logistics, information flow and capital flow, aiming to realize the efficient operation of the entire chain from suppliers to consumers. Demand forecasting, as a key link, directly affects the scientificity and rationality of decisions such as inventory management, procurement planning and distribution arrangement.

[0003] At present, most logistics supply chain management systems mainly rely on traditional statistical models such as linear regression, time series analysis, or rely on artificial experience for judgment in demand forecasting. Traditional statistical models establish fixed mathematical relationships based on historical data, and predict future demand by analyzing past sales data, market trends and other factors. Artificial experience relies on the industry knowledge and market perception accumulated by management personnel in long-term work to make subjective judgments on demand changes. These two methods can make certain predictions of demand for regular goods in a relatively stable market environment.

[0004] However, these existing demand forecasting methods have significant defects. On the one hand, in the face of sudden demand such as large-scale promotion activities on e-commerce platforms and sudden changes in consumer habits, traditional statistical models lack a dynamic response mechanism to irregular events and cannot timely capture the dramatic changes in the market, resulting in a serious disconnection between predicted results and actual demand. On the other hand, for long-tail goods, their demand often shows low frequency and instability, and traditional models are difficult to dig out the complex influencing factors behind the demand fluctuations of such goods, and artificial experience is also difficult to accurately grasp their change rules, ultimately leading to low demand forecasting accuracy. Such low-accuracy demand forecasting can cause inventory accumulation or stockout problems, seriously affecting the operational efficiency and economic benefits of the logistics supply chain management system. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a logistics supply chain management system, which solves the problem of low prediction accuracy caused by the inability to respond to sudden demand and grasp the demand fluctuation rules of long-tail goods in the existing demand forecasting methods relying on traditional statistical models and artificial experience, thereby affecting the operational efficiency and economic benefits of the logistics supply chain management system.

[0006] In order to achieve the above object, the present application is realized by the following technical scheme: a logistics supply chain management system, comprising a data access module, the data access module imports multi-source data into the system, and transmits the data to a data storage and management module, the data access module is connected with the data storage and management module, the data storage and management module is responsible for receiving the data and storing the data, the data storage and management module is connected with a data fusion and feature weighting module, the data fusion and feature weighting module extracts the data in the data storage and management module, and cooperates with a set of space-time fusion prediction model to predict the demand quantity of the time node after multi-feature, the space-time fusion prediction model is used for predicting and calculating the demand quantity of different time points, the data fusion and feature weighting module is connected with a probability prediction and risk quantification module, the probability prediction and risk quantification module extracts the data in the data storage and management module, and cooperates with a set of calculation formula to calculate the loss risk of actual sales deviating from prediction, the data fusion and feature weighting module and the probability prediction and risk quantification module are both connected with a dynamic parameter optimization and feedback module, the dynamic parameter optimization and feedback module optimizes the formula weight in the data fusion and feature weighting module and the probability prediction and risk quantification module according to the subsequent data feedback and cooperates with the optimization formula, the dynamic parameter optimization and feedback module is connected with a supply chain planning module, the supply chain planning module sets different planning schemes according to the calculation structure of the data fusion and feature weighting module and the probability prediction and risk quantification module, the supply chain planning module is connected with a logistics scheduling module, and the logistics scheduling module sets different scheduling schemes according to the calculation structure of the data fusion and feature weighting module and the probability prediction and risk quantification module.

[0007] Preferably, the multi-source data in the data access module includes weather API, competitor price database and logistics GPS.

[0008] Preferably, the space-time fusion prediction model in the data fusion and feature weighting module is specifically:

[0009]

[0010] Wherein: represents the predicted demand quantity of time point t; f LSTM represents the mapping function of long short-term memory network; y t-1 represents the real demand quantity of historical time point; x t represents the external feature vector of time point t; w i represents the weight of the i th external feature; g i represents the nonlinear conversion function of the i th external feature; k represents the total number of external features.

[0011] Preferably, the calculation formula in the probability prediction and risk quantification module is specifically:

[0012]

[0013] Wherein: represents the loss function value corresponding to the quantile T; y t represents the actual demand at time point t, represents the predicted demand at time point t corresponding to quantile T; τ represents the target quantile; T represents the total length of the time window.

[0014] Preferably, the optimization formula in the dynamic parameter optimization and feedback module is specifically:

[0015]

[0016] Wherein: θ represents the model parameter set; represents the historical data set up to time t; represents the newly added real-time data; represents the data posterior probability distribution of parameter θ under the condition; α represents proportional to the symbol, indicating that the distributions on both sides of the equation are proportional.

[0017] Preferably, the different planning schemes in the supply chain planning module include:

[0018] When the predicted demand is high: emergency replenishment expansion: increase safety stock based on P90 prediction value; emergency capacity start: link with foundry or supplier, allocate 30%-50% emergency orders;

[0019] When the predicted demand is normal: periodic rolling replenishment: develop regular replenishment plan according to P50 value, purchase in batches (weekly / monthly dimension); capacity balance: 80% regular production line load, 20% reserved flexible capacity;

[0020] When the predicted demand is low: inventory control: reduce replenishment quantity according to P10 benchmark (only maintain 50% safety stock); tail inventory cleaning: start promotion or allocation to low demand area to avoid overstock.

[0021] Preferably, the different scheduling schemes in the logistics scheduling module include:

[0022] When the predicted demand is high: multi-warehouse cooperative direct delivery: start cross-regional warehouse direct delivery to terminal network, bypass central warehouse, priority transportation: high-value goods prefer air transportation / special line, guarantee time limit;

[0023] When the predicted demand is normal: economic path planning: optimize vehicle load rate, select low-cost route, standard time limit guarantee: distribute according to T+1 or T+2 mode;

[0024] Low demand forecast: intensive distribution: combine multiple orders, reduce distribution frequency, flexible capacity: use small vehicles or crowd-sourcing logistics to reduce costs.

[0025] The present application provides a logistics supply chain management system. Has the following beneficial effects:

[0026] 1、The data fusion and feature weighting module in the application combines the space-time fusion prediction model, integrates multi-source data, can effectively deal with sudden demand and long-tail commodity fluctuation, and significantly improves the demand prediction accuracy.

[0027] 2、The probability prediction and risk quantification module in the application accurately assesses the loss risk of actual sales deviating from prediction through scientific calculation formula, and provides data support for supply chain decision-making.

[0028] 3、The dynamic parameter optimization and feedback module in the application optimizes model parameters according to real-time data, optimizes the algorithm in the data fusion and feature weighting module and the probability prediction and risk quantification module according to subsequent feedback information, realizes self-adaptive effect, and guarantees the long-term effectiveness of prediction and decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 The system flowchart of the present application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the specification of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0031] Embodiment:

[0032] Please refer to the drawings Figure 1The embodiment of the present application provides a logistics supply chain management system, which comprises a data access module, the data access module imports multi-source data into the system, the multi-source data comprises weather API, competitive product price database and logistics GPS, and the data is transmitted to a data storage and management module, the data access module is connected with the data storage and management module, the data storage and management module is responsible for receiving the data and storing the data, the data storage and management module is connected with a data fusion and feature weighting module, the data fusion and feature weighting module extracts the data in the data storage and management module, and cooperates with a set space-time fusion prediction model to predict the demand quantity of a time node after multi-feature, the demand quantity of different time points is predicted and calculated through the space-time fusion prediction model, and the space-time fusion prediction model is specifically:

[0033]

[0034] Wherein: The predicted demand quantity of the time point t is represented by f LSTM The mapping function of the long short-term memory network (LSTM) is represented by y t-1 The real demand quantity of the historical time point (the known sales data of the previous week) is represented by x t The external feature vector of the time point t (including the promotion strength, the competitive product price fluctuation index and the weather temperature fusion feature) is represented by w i The weight of the i-th external feature is represented by g i The nonlinear conversion function of the i-th external feature is represented by k, and k represents the total number of external features.

[0035] The data fusion and feature weighting module is connected with a probability prediction and risk quantification module, the probability prediction and risk quantification module extracts the data in the data storage and management module, and cooperates with a set calculation formula to calculate the loss risk of the actual sales quantity deviating from the prediction, and the calculation formula is specifically:

[0036]

[0037] Wherein: The loss function value corresponding to the quantile T is represented by y t The actual demand quantity of the time point t (the real observed daily sales quantity) is represented by y The predicted demand quantity of the time point t corresponding to the quantile T is represented by τ, τ represents the target quantile, and T represents the total length of the time window (the length period of prediction, including the daily sales quantity of the future 30 days);

[0038] The data fusion and feature weighting module and the probability prediction and risk quantification module are connected with a dynamic parameter optimization and feedback module, the dynamic parameter optimization and feedback module is according to subsequent data feedback, and the formula weight in the data fusion and feature weighting module and the probability prediction and risk quantification module is optimized by cooperating with the optimization formula, and the optimization formula is specifically:

[0039]

[0040] Wherein: θ represents a model parameter set (including LSTM network weight, feature weight w i ); Represent the historical data set up to time t; Represent the newly added real-time data (the latest sales data collected on the same day, weather change information); Represent the data The posterior probability distribution of parameter θ under the condition;∝ represents proportional to the symbol, indicating that the distribution on both sides of the equation is proportional;

[0041] The dynamic parameter optimization and feedback module is connected with a supply chain planning module, the supply chain planning module is respectively provided with different planning schemes according to the calculation structure of the data fusion and feature weighting module and the probability prediction and risk quantification module, and the planning scheme includes:

[0042] When the predicted demand is high: emergency replenishment expansion: increase safety stock based on P90 predicted value;Emergency capacity start: link to foundry or supplier, distribute 30%-50% emergency order;

[0043] When the predicted demand is normal: periodic rolling replenishment: make regular replenishment plan according to P50 value, and purchase in batches (week / month dimension);Capacity balance: 80% regular production line load, 20% reserved flexible capacity;

[0044] When the predicted demand is low: inventory control: reduce the replenishment amount according to P10 benchmark (only maintain 50% safety stock);Tail clearance: start promotion or allocation to low demand area to avoid overstock;

[0045] The supply chain planning module is connected with a logistics scheduling module, the logistics scheduling module is respectively provided with different scheduling schemes according to the calculation structure of the data fusion and feature weighting module and the probability prediction and risk quantification module, and the scheduling scheme includes:

[0046] When the predicted demand is high: multi-warehouse cooperative direct delivery: start cross-regional warehouse direct delivery to terminal network, bypass the central warehouse, priority transportation: high-value goods preferentially use air transportation / special line to ensure time limit;

[0047] When the demand is normal: Economic route planning: optimize the vehicle load rate, choose low-cost routes (such as night road transportation), standard time guarantee: T+1 or T+2 mode distribution;

[0048] When the demand is low: intensive distribution: combine multiple orders to reduce distribution frequency (once a week), flexible capacity: use small vehicles or crowd-sourced logistics to reduce costs.

[0049] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A logistics supply chain management system, characterized in that, The system includes a data access module that imports multi-source data into the system and transmits this data to a data storage and management module. The data access module is connected to the data storage and management module, which is responsible for receiving and storing this data. The data storage and management module is also connected to a data fusion and feature weighting module. This module extracts data from the data storage and management module and, in conjunction with a pre-defined spatiotemporal fusion prediction model, predicts the demand at different time points based on multiple features. The spatiotemporal fusion prediction model is used to calculate and predict the demand at different time points. Finally, the data fusion and feature weighting module is connected to a probability prediction and risk quantification module, which extracts data from the data storage and management module. The system calculates the loss risk due to deviations between actual and predicted sales using set formulas. Both the data fusion and feature weighting module and the probability prediction and risk quantification module are connected to a dynamic parameter optimization and feedback module. This module optimizes the formula weights in these modules based on subsequent data feedback and the optimization formulas. A supply chain planning module is also connected to this module, which sets different planning schemes based on the calculation structures of the data fusion and feature weighting module and the probability prediction and risk quantification module. Finally, a logistics scheduling module is connected to this module, which sets different scheduling schemes based on the calculation structures of the data fusion and feature weighting module and the probability prediction and risk quantification module.

2. The logistics supply chain management system according to claim 1, characterized in that, The data access module includes multi-source data such as weather API, competitor price database, and logistics GPS.

3. A logistics supply chain management system according to claim 1, characterized in that, The spatiotemporal fusion prediction model in the data fusion and feature weighting module is specifically as follows: in: The projected demand at time point t; f LSTM The mapping function representing the Long Short-Term Memory network; y t-1 The actual demand at a historical point in time; x t The external feature vector representing time point t; w i The weight representing the i-th external feature; g i The nonlinear transformation function represents the i-th external feature; k represents the total number of external features.

4. A logistics supply chain management system according to claim 1, characterized in that, The specific calculation formula in the probability prediction and risk quantification module is as follows: in: The value of the loss function corresponding to the quantile T; y t This represents the actual demand at time point t. τ represents the predicted demand corresponding to the quantile T at time point t; τ represents the target quantile; and T represents the total length of the time window.

5. A logistics supply chain management system according to claim 1, characterized in that, The optimization formula in the dynamic parameter optimization and feedback module is as follows: Where: θ represents the set of model parameters; This represents the set of historical data up to time t. Represents newly added real-time data; Representative data Given the condition, the posterior probability distribution of parameter θ; ∝ represents proportionality, indicating that the distributions on both sides of the equation are proportional.

6. A logistics supply chain management system according to claim 1, characterized in that, The different planning schemes in the supply chain planning module include: When demand is predicted to be high: Emergency replenishment and capacity expansion: Adjust safety stock upward based on P90 forecast value; Emergency production capacity activation: Coordinate with contract manufacturers or suppliers to allocate 30% to 50% of emergency orders; When demand is expected to be normal: Periodic rolling replenishment: Develop a regular replenishment plan based on the P50 value and purchase in batches (weekly / monthly); Capacity balance: 80% of regular production line load, 20% reserved flexible capacity; When demand is predicted to be low: Inventory management: Reduce replenishment based on P10 (maintain only 50% safety stock); Clearing out leftover stock: Initiate promotions or transfer to low-demand areas to avoid stockpiling.

7. A logistics supply chain management system according to claim 1, characterized in that, The different scheduling schemes in the logistics scheduling module include: When demand is expected to be high: Multi-warehouse collaborative direct delivery: Initiate direct delivery from cross-regional warehouses to terminal outlets, bypassing central warehouses; Priority transportation: High-value goods are given priority for air freight / dedicated lines to ensure timeliness. When demand is expected to be normal: Economic route planning: Optimize vehicle load factor, select low-cost routes, and ensure standard delivery time: Deliver according to T+1 or T+2 mode. When demand is predicted to be low: consolidated delivery: merge multiple orders to reduce delivery frequency; flexible transportation capacity: switch to smaller vehicles or crowdsourced logistics to reduce costs.

Images