Warehouse logistics management method based on logistics big data

Abstract

The invention belongs to the technical field of logistics management, and discloses a warehouse logistics management method based on logistics big data. Comprising the steps of collecting logistics multi-source heterogeneous data; preprocessing the logistics multi-source heterogeneous data to obtain preprocessed heterogeneous data; performing mode recognition on the preprocessed heterogeneous data by using an improved automaton algorithm to obtain M logistics route modes; based on M logistics route modes, an optimal logistics planning scheme is calculated by using an optimization algorithm, warehouse logistics management is guided by using the optimal logistics planning scheme, and the logistics response speed and the service quality are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of logistics management, and more specifically, to a warehousing logistics management method based on logistics big data. Background Art

[0002] Existing logistics management technologies have exposed many problems in the face of increasingly complex supply chain environments and massive amounts of data. First, traditional methods are difficult to effectively process and integrate heterogeneous data from different sources, resulting in the inability to fully explore the value of data and insufficient basis for decision-making. Second, existing route planning algorithms often only consider static factors and cannot respond to dynamic factors such as traffic conditions and weather changes in a timely manner, resulting in low actual transportation efficiency. Third, traditional inventory management strategies lack flexibility and are difficult to cope with rapid changes in market demand, and inventory backlogs or shortages often occur. In addition, existing logistics optimization methods usually only focus on a single goal, such as cost minimization, while ignoring multi-dimensional indicators such as timeliness and service quality, making it difficult to meet the comprehensive needs of modern logistics. In actual operations, logistics companies often face problems such as information islands, unreasonable resource scheduling, and poor emergency response capabilities, which seriously affect overall operational efficiency. More importantly, it is difficult to achieve global optimization of the entire supply chain, resulting in poor coordination between various links in the supply chain and weak risk resistance. These problems are particularly prominent in special circumstances such as e-commerce promotions and public emergencies.

[0003] In view of this, the present invention proposes a warehousing logistics management method based on logistics big data to solve the above problems. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a warehouse logistics management method based on logistics big data, characterized by comprising: S1, collecting logistics multi-source heterogeneous data;

[0005] S2, preprocessing the logistics multi-source heterogeneous data to obtain preprocessed heterogeneous data;

[0006] S3, using the improved automaton algorithm to perform pattern recognition on the preprocessed heterogeneous data to obtain M types of logistics route patterns;

[0007] S4. Based on M types of logistics route patterns, optimize and calculate to obtain the optimal logistics planning solution, and use the optimal logistics planning solution to guide warehouse logistics management.

[0008] Furthermore, the logistics multi-source heterogeneous data includes logistics transportation data, logistics order data, warehousing data, weather data, traffic data and equipment data.

[0009] Furthermore, the method of preprocessing the logistics multi-source heterogeneous data includes:

[0010] Process missing values ​​in logistics multi-source heterogeneous data, remove duplicate data in logistics multi-source heterogeneous data, process outliers in logistics multi-source heterogeneous data, and obtain preliminary standard heterogeneous data; merge data from different data sources of preliminary standard heterogeneous data to obtain preliminary fused data; encode non-numeric data in preliminary fused data into numeric data, and normalize / standardize the numeric data in preliminary fused data to obtain pre-processed heterogeneous data;

[0011] The method of performing spatial alignment includes:

[0012] Project the original preliminary standard heterogeneous data to the preliminary spatial dimension; construct a statistical grid on the preliminary spatial dimension, divide the entire data space into regular grid cells, calculate the number of data points in each grid cell, and obtain the data distribution of each grid cell by statistics; calculate the data density of each grid cell based on the number of data points in each grid cell, preset an upper density threshold, and regard the grid cells above the upper density threshold as dense data areas; preset a lower density threshold, and regard the grid cells below the lower density threshold as sparse data areas;

[0013] Define an initial partition granularity as the basic partition of the entire data space. For the grid cells that are dense data areas, divide them into smaller sub-units with a side length of Among them, ρ_min is a preset minimum density threshold, ρ is the data density of the grid unit as the dense data area; ε is a small constant; α0 is the preset initial adjustment parameter, and α1 is the linear term coefficient;

[0014] Merge adjacent grid cells that are sparse data areas to form several partition units;

[0015] Construct a hierarchical structure, the hierarchical structure includes the top layer, the middle layer and the bottom layer; the basic partition of the entire data space is the top layer, the sub-units divided by the dense data area are the middle layer, and the partition unit is the bottom layer;

[0016] According to the spatial coordinates of the data points, the layer to which they belong is determined. For the data points belonging to the middle layer, they are processed more finely; for the data points belonging to the bottom layer, they are processed roughly; and then all the layers are aligned according to the spatial dimensions.

[0017] Furthermore, the method of performing time alignment includes:

[0018] For each data source, extract the timestamp sequence from the preliminary standard heterogeneous data, and convert the extracted timestamp sequence into a unified time format to obtain a unified timestamp sequence; calculate the time interval between adjacent timestamps in the unified timestamp sequence, and calculate the distribution characteristics of the time interval, including the mean μ, variance σ, median me, lower quartile Q1, upper quartile Q3 and skewness sk; use the distribution characteristics to calculate the uniform distribution index U;

[0019]

[0020] Among them, w1, w2, w3 and w4 are weight coefficients, and γ is the adjustment parameter;

[0021] A uniformity threshold is preset. If the uniform distribution index U is greater than or equal to the uniformity threshold, the timestamps of the preliminary standard heterogeneous data are uniformly distributed. For uniformly distributed timestamps, linear interpolation is used for time alignment.

[0022] Define several time windows, set the starting time point of the time window, and number the time windows; that is, the time window number; traverse the unified timestamp sequence, for each timestamp t in the unified timestamp sequence, calculate the time window number b to which t belongs; count the number of timestamps in the time window corresponding to b; obtain the time window count sequence C = [c_1, c_2, c_3, ..., c_m], where m is the total number of time windows, and c_m represents the number of timestamps in the mth time window;

[0023] The timestamp density sequence D = [d_i] is calculated based on the time window count sequence C; d_i is the density of the i-th time window;

[0024] Among them, ban is the bandwidth parameter of the kernel function, norm is the normalization coefficient; j is another index of the time window; c_j is the number of timestamps in the j-th time window; w_j is the weight value of the j-th time window; is a regulation function based on the time window index j;

[0025] Where len(j) is the length of the jth time window, pos(j) is the start time or center time of the jth time window, base(j) is the base function, which represents the initial weight of the jth time window; α, β, and δ are adjustable parameters;

[0026] The timestamp density sequence D is plotted into a curve in chronological order, which is the timestamp density curve. The timestamp density curve is peak detected to obtain the local maximum point p, and the lower critical value x_min and the upper critical value x_max are defined. The local maximum point p is fixed and expanded to obtain the dense area [p-x_min, p-x_max].

[0027] For timestamps in dense areas, spline interpolation or Gaussian kernel interpolation is used for time alignment.

[0028] Furthermore, the method of performing pattern recognition on the pre-processed heterogeneous data includes:

[0029] An automaton model is constructed based on preprocessed heterogeneous data, and the preprocessed heterogeneous data is mapped to a preset grid space, that is, the dimension and size of the grid space are defined; the spatial coordinates in the preprocessed heterogeneous data are mapped to the grid index of the grid space; a set of grid states in the grid space is defined; the neighborhood radius of the grid is defined as the Moore neighborhood, and state transition rules are preset, which include vehicle state transition rules, obstacle state transition rules, weather state transition rules, and traffic state transition rules;

[0030] Initialize the initial state, time step and total number of iterations of the grid space; for each time step, traverse all grids, calculate the state of the grid in the next time step based on the state of the current grid, the state of the neighboring grids and the state transition rule, and update the state of all grids, repeating until the total number of iterations is reached;

[0031] Extract key features from the grid space, including clustering features, dynamic features and topological features; clustering features are the clustering degree and clustering shape of the grid; dynamic features are the changing trend and periodicity of the grid state; topological features are the connectivity and obstacle distribution of the grid space;

[0032] The extracted key features are input into the classification / clustering algorithm to divide the grid space into different pattern areas, each of which corresponds to a logistics route pattern.

[0033] Furthermore, the vehicle state transition rule is:

[0034] If the current grid is idle and there is a logistics vehicle state in the adjacent grid, the current grid is transferred to the corresponding logistics vehicle state with the dynamic probability PL;

[0035] PL = (1-a1-a2-a3) × P_bas + a1 × f(v) + a2 × g(w) + a3 × h(r); where a1, a2 and a3 are three state weight coefficients; f(v) is the vehicle type influence function, g(w) is the weather state influence function, and h(r) is the traffic state influence function;

[0036] Among them, v_max is the number of the largest vehicle type;

[0037] Among them, w_max is the number of the maximum weather state;

[0038] Among them, r′, v′ and w′ are all positive numbers less than 1 and greater than 0, and r max is the number of the maximum traffic state;

[0039] If the current grid is in a logistics vehicle state, the current logistics vehicle state is maintained with a preset retention probability, or transferred to the logistics vehicle state of an adjacent grid; if the current grid is in an obstacle state, the logistics vehicle state cannot be transferred to the grid.

[0040] Furthermore, the obstacle state transfer rule is:

[0041] If the current grid is in an idle state and there is an obstacle state in the adjacent grid, the current grid is transferred to the corresponding obstacle state with a preset transfer probability;

[0042] If the current grid is in an obstacle state, keep the current obstacle state, or transfer to an idle state with the elimination probability pc;

[0043] Among them, b1, b2, b3 and b4 are probability parameters, o It is the number of the obstacle type. If the current grid is in the logistics vehicle state or storage state, the obstacle state cannot be transferred to this grid.

[0044] The weather state transition rules are:

[0045] If the current grid is idle, the entire study area is divided into several meteorological areas according to geographical location in advance, and a spatiotemporal probability model of weather conditions is constructed for each meteorological area. In the spatiotemporal probability model, given the current time and geographical location, the spatiotemporal probability model of the corresponding meteorological area is queried, and the probabilities of various weather conditions are calculated based on historical data. The weather state with the highest probability is used as the state of the current grid transfer.

[0046] If the current grid is in a weather state, the current weather state is maintained with the weather maintenance probability p_w, or it is transferred to another weather state with the weather transition probability p′-w;

[0047] p′_w=(1-p_w)×P(w′|w); where P(w′|w) is the conditional probability of the current weather state number w transferring to the weather state number w′;

[0048] If the current grid is in a logistics vehicle state or a storage state, the weather state will not have a direct impact; if the current grid is in an obstacle state, the weather state will be transferred to an idle state.

[0049] Furthermore, the preset method of the traffic state transition rule includes:

[0050] The road network is modeled as a directed weighted graph G(V, E), where the nodes in the node set V represent intersections or road sections, and the edges and their weights in the edge set E represent the length and capacity of the road;

[0051] Define vehicle agent attributes and behavior strategies; vehicle agent attributes include: current location, destination, vehicle model and travel time; behavior strategies include path selection strategy and vehicle speed adjustment strategy; set the simulation time step dt, at each time step, all vehicle agents move in the directed weighted graph according to the behavior strategy, and after the vehicle reaches the destination, it is removed from the directed weighted graph, and new vehicle agents are continuously generated in the directed weighted graph according to the spatiotemporal probability model;

[0052] At each time step, the number of vehicles on each node is counted, and the number of vehicles on the node is mapped to the traffic state of the node; for each node, the transfer of its traffic state is affected by the traffic state of the adjacent nodes, and is implemented by a rule-based transfer model or a data learning-based transfer model; the directed weighted graph G(V, E) is spatially mapped with the grid space to obtain a complex; a state transfer probability matrix is ​​set for the complex, and the complex with the state transfer probability matrix set is the traffic state transfer rule.

[0053] Furthermore, the method of spatially mapping the directed weighted graph G(V, E) to the grid space includes:

[0054] Each grid in the grid space is associated with a node in a directed weighted graph, and the size of the grid is set to be equivalent to the actual scale of the node; if there is an edge between adjacent grids, the corresponding grids are also connected; the vehicle agent is placed in the grid corresponding to its current position, and the attributes of the vehicle agent are mapped to different state values ​​of the grid. The same grid can accommodate several vehicle agents; at each time step, all vehicle agents move in the grid space according to the behavioral strategy. The vehicle agent selects the next step position according to the connectivity relationship between the grids and follows the behavioral strategy. Similarly, after the vehicle agent reaches the destination, it is removed from the grid space; the number of vehicle agents in each grid is counted, and the number of vehicle agents is mapped to the traffic state value of the grid.

[0055] Furthermore, the calculation method of the optimal logistics planning solution includes:

[0056] Constructing the optimization objective function Wherein, V1, V2 and V3 are optimization weight coefficients; I and J are nodes in the directed weighted graph, (I, J) represents the edge between node I and node J, c_IJ is the unit transportation cost of edge (I, J), x_IJ is a 0-1 variable, when the vehicle passes through edge (I, J), x_IJ = 1, otherwise it is 0; t_IJ is the travel time of edge (I, J); h is the index of the order, SP_h is the actual delivery time of order h, QP_h is the expected delivery time of order h, and N1 is the number of orders;

[0057] Define optimization decision variables including vehicle scheduling, route selection, and warehouse scheduling; construct constraints including vehicle load, time window, driving distance, and warehouse capacity, and adjust the constraints according to the characteristics of M logistics route modes;

[0058] Permutation coding is used to encode the optimization decision variables into a permutation; according to M logistics route patterns, an initial molecule group is generated, and the function value FL′ of the optimization objective function of each molecule is calculated;

[0059] Mapping function values ​​to molecular masses Among them, ω and ω1 are positive weight parameters used to adjust the relative importance of the two items, Fmax is the maximum function value of all molecules in the current molecule group; PO is the penalty term of the molecule, reflecting the degree of its violation of the constraint conditions;

[0060] Calculate the mass difference ΔM between any two molecules s1 and s2; based on the mass difference, calculate the attractive force FY_s1s2 or repulsive force FC_s1s2 between the two molecules; for each molecule, add up the attractive or repulsive forces exerted on it by all other molecules to obtain the total force;

[0061] For each molecule, add up the gravitational or repulsive forces exerted on it by all other molecules to obtain the total force; calculate the acceleration of each molecule based on the total force; update the molecular velocity based on the acceleration; update the molecular position based on the molecular velocity; preset an upper mass threshold, mutate molecules with a molecular mass greater than the upper mass threshold, and split them to produce new molecules; perform mutations to increase / decrease the number of vehicles or change the route; the new molecule inherits the position and velocity of the molecule before the split;

[0062] A lower mass threshold is preset, and molecules with a mass less than the lower mass threshold are polymerized; the polymerization is performed as direct replacement or fusion; the position and speed of the new molecules obtained after polymerization are calculated based on the weighted average of the mass of the polymerized molecules;

[0063] Until the standard deviation of the function value of the optimization objective function of the molecular group is less than the preset convergence threshold, the updating, mutation, splitting and aggregation are terminated; the molecule with the largest molecular mass at this time is extracted as the optimal solution; the optimal solution is decoded into the actual logistics route planning plan, which is the optimal logistics planning plan; the optimal logistics planning plan includes vehicle scheduling, route selection and warehouse scheduling.

[0064] Technical effects and advantages of a warehousing logistics management method based on logistics big data of the present invention:

[0065] The present invention can accurately capture the complex dynamic characteristics and potential laws of logistics data, greatly improving the accuracy of logistics prediction and decision-making. This data-driven intelligent management method can not only significantly optimize warehouse layout and inventory strategy, but also realize dynamic optimization of transportation routes, thereby greatly reducing logistics costs and improving resource utilization efficiency. The identified logistics route pattern contains valuable experience in logistics operations, guides subsequent planning decisions, avoids repeated exploration, and improves efficiency; at the same time, it can achieve global optimization of the entire supply chain, improve the flexibility and risk resistance of the supply chain, and the adaptive optimization algorithm can quickly generate the optimal solution in a complex and changeable logistics environment, significantly improve the logistics response speed and service quality, and improve and supplement the original model to form a virtuous circle of continuous iterative optimization; secondly, when conducting logistics planning, a variety of influencing factors are comprehensively considered to obtain a more comprehensive and practical optimal solution; thirdly, the identified logistics route pattern is used as a priori knowledge for optimization, which greatly improves the optimization efficiency and can flexibly adapt to logistics needs of different scales and types. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A schematic diagram of a warehousing logistics management method based on logistics big data of the present invention;

[0067] Figure 2 It is a schematic diagram of a warehouse logistics management system based on logistics big data of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] Example 1

[0070] See also Figure 1 As shown, the warehousing logistics management method based on logistics big data described in this embodiment includes: S1, collecting multi-source heterogeneous logistics data;

[0071] S2, preprocessing the logistics multi-source heterogeneous data to obtain preprocessed heterogeneous data;

[0072] S3, using the improved automaton algorithm to perform pattern recognition on the preprocessed heterogeneous data, and obtain M types of logistics route patterns; M is an integer greater than 1;

[0073] S4. Based on M types of logistics route patterns, optimize the calculation to obtain the optimal logistics planning solution, and use the optimal logistics planning solution to guide warehouse logistics management.

[0074] Logistics multi-source heterogeneous data includes logistics transportation data, logistics order data, warehousing data, weather data, traffic data and equipment data; logistics transportation data includes vehicle GPS trajectory data and logistics vehicle loading data; logistics order data includes customer order information, order status, delivery information and receipt information; warehousing data includes warehouse inventory data, inbound and outbound records and cargo storage locations;

[0075] Weather data includes weather conditions, temperature and rainfall along the route; traffic data includes road traffic conditions, congestion and construction information; equipment data includes vehicle equipment status data and storage equipment status data; these heterogeneous data come from a wide range of sources, with different data formats and structures, and require data preprocessing and fusion to provide valuable input data for subsequent feature extraction, pattern recognition and route planning.

[0076] The methods for preprocessing logistics multi-source heterogeneous data include:

[0077] Process missing values ​​in multi-source heterogeneous logistics data. For missing values, delete, interpolate, or use statistical values ​​(such as mean, median, etc.) to fill them according to the characteristics of the data; remove duplicate data in multi-source heterogeneous logistics data. For duplicate data records, keep one and delete other duplicate records; process outliers in multi-source heterogeneous logistics data. Outliers may be caused by errors in data collection, transmission or storage. They need to be judged and processed according to business rules, such as deletion, replacement, etc.; obtain preliminary standard heterogeneous data.

[0078] The preliminary standard heterogeneous data from different data sources are merged. Specifically, the merged time dimension and the initial spatial dimension are defined. The time dimension selects the minimum time granularity, such as seconds, minutes or hours; the spatial dimension selects longitude and latitude coordinates or geocoding (such as administrative division codes); each data source in the preliminary standard heterogeneous data is time-aligned according to the time dimension, and each data source in the preliminary standard heterogeneous data is spatially aligned according to the spatial dimension; preliminary fused data is obtained; non-numeric data (such as text) in the preliminary fused data is encoded and converted into numerical data, and the numerical data in the preliminary fused data is normalized / standardized to eliminate the dimensionality effect, so as to obtain preprocessed heterogeneous data.

[0079] Ways to perform time alignment include:

[0080] For each data source, extract the timestamp sequence from the preliminary standard heterogeneous data, and convert the extracted timestamp sequence into a unified time format to obtain a unified timestamp sequence; calculate the time interval between adjacent timestamps in the unified timestamp sequence, and calculate the distribution characteristics of the time interval, including the mean μ, variance σ, median me, lower quartile Q1, upper quartile Q3 and skewness sk; use the distribution characteristics to calculate the uniform distribution index U;

[0081]

[0082] Among them, w1, w2, w3 and w4 are weight coefficients, and their numerical values ​​sum to 1, and γ is an adjustment parameter used to adjust the nonlinearity of the skewness term.

[0083] A uniformity threshold is preset. If the uniform distribution index U is greater than or equal to the uniformity threshold, the timestamps of the preliminary standard heterogeneous data are uniformly distributed; otherwise, they are non-uniformly distributed. For uniformly distributed timestamps, linear interpolation is used for time alignment.

[0084] Define several time windows, such as one time window per day, set the starting time point of the time window, such as 2023-05-01 00:00:00; and number the time windows; that is, the time window number;

[0085] Traverse the unified timestamp sequence, and for each timestamp t in the unified timestamp sequence, calculate the time window number b to which t belongs; count the number of timestamps in the time window corresponding to b; and obtain the time window counting sequence C = [c_1, c_2, c_3, ..., c_m], where m is the total number of time windows and c_m represents the number of timestamps in the mth time window.

[0086] The timestamp density sequence D = [d_i] is calculated based on the time window count sequence C; d_i is the density of the i-th time window;

[0087] Among them, ban is the bandwidth parameter of the kernel function, which controls the smoothness of the kernel function; norm is the normalization coefficient, which is used to normalize the density value to the interval [0, 1]; j is another index of the time window (not indexing the same time window as i); c_j is the number of timestamps in the jth time window; w_j is the weight value of the jth time window; is a regulation function based on the time window index j, which is used to dynamically adjust the contribution of each time window.

[0088] Wherein, len(j) is the length of the j-th time window, pos(j) is the start time or center time of the j-th time window, base(j) is the basic function, which indicates the initial weight of the j-th time window. It can be a constant function or other function forms, such as linear function, exponential function, etc.; α, β and δ are adjustable parameters used to control the influence of the time window length, position and basic weight on the adjustment function.

[0089] The timestamp density sequence D is plotted into a curve in chronological order, which is the timestamp density curve. The peak value of the timestamp density curve is detected to obtain the local maximum point p, and the lower critical value x_min and the upper critical value x_max are defined. The local maximum point p is fixed and expanded to obtain the dense area [p-x_min, p-x_max]. For the timestamps in the dense area, spline interpolation or Gaussian kernel interpolation is used for time alignment.

[0090] Ways to perform spatial alignment include:

[0091] Project the original preliminary standard heterogeneous data (such as GPS coordinates) to the preliminary spatial dimension, such as a projection method based on a buffer, a topological structure or a probability model; construct a statistical grid on the preliminary spatial dimension, divide the entire data space (process different data sources of the preliminary standard heterogeneous data separately) into regular grid cells, calculate the number of data points in each grid cell, and obtain the data distribution of each grid cell.

[0092] Based on the number of data points in each grid cell, the data density of each grid cell is calculated, and an upper density threshold is preset. The grid cells above the upper density threshold are regarded as dense data areas; the lower density threshold is preset, and the grid cells below the lower density threshold are regarded as sparse data areas.

[0093] Define an initial partition granularity (such as road level) as the basic partition of the entire data space. For the grid cells that are dense data areas, divide them into smaller sub-units (such as road section level or finer). The side length of the sub-unit is Among them, ρ_min is a preset minimum density threshold used to control the minimum side length of the subunit, ρ is the data density of the grid unit as the dense data area; ε is a small constant, the purpose is to avoid the situation where the formula has singular values ​​or numerical instability when the density ρ is close to 0, ε can be a very small positive constant, such as 1e-6; α0 is a preset initial adjustment parameter, which represents the basic value of the mapping intensity, α1 is the linear term coefficient, which represents the slope of the mapping intensity changing with density; the two parameters α0 and α1 control the mapping intensity between the density and the side length of the subunit in different density areas; the adjacent grid units as sparse data areas are merged to form several partition units.

[0094] Construct a hierarchical structure, which includes the top layer, the middle layer and the bottom layer; take the basic partition of the entire data space as the top layer, the sub-units divided by the dense data area as the middle layer, and the partition unit as the bottom layer; determine the layer to which the data point belongs according to its spatial coordinates, and perform more detailed processing on the data points in the middle layer, such as trajectory reconstruction and anomaly detection; perform rough processing on the data points in the bottom layer, such as simple statistics and clustering; then align all the layers according to the spatial dimension (directly map the GPS coordinates or geocodes to the spatial dimension).

[0095] Methods for pattern recognition of preprocessed heterogeneous data include:

[0096] An automaton model is constructed based on preprocessed heterogeneous data, and the preprocessed heterogeneous data is mapped to a preset grid space, that is, the dimension (two-dimensional or three-dimensional) and size (number of rows and columns, and number of layers if three-dimensional) of the grid space are defined; the spatial coordinates (such as longitude and latitude) in the preprocessed heterogeneous data are mapped to the grid index of the grid space (using linear interpolation, nearest neighbor interpolation and other methods to discretize the continuous spatial coordinates to the grid index).

[0097] Define the set of states of the grids in the grid space. Specifically, the idle state (0): indicates that there is no logistics activity in the grid at present;

[0098] Logistics vehicle status (1, 2, 3, ...): Different positive integers represent different types of logistics vehicles occupying the grid, such as 1 for a small truck, 2 for a medium truck, 3 for a large truck, etc.;

[0099] Storage status (101, 102, 103, ...): A positive integer plus 100 indicates that the grid is a storage facility. Different numbers can represent different types of storage facilities, such as 101 for a small warehouse, 102 for a medium warehouse, 103 for a large warehouse, etc.

[0100] Obstacle status (-1, -2, -3, ...): Different negative integers represent different types of obstacles occupying the grid, such as -1 for road construction, -2 for traffic accidents, and -3 for natural disasters.

[0101] Weather status (201, 202, 203, ...): 200 plus different positive integers represent different weather conditions, such as 201 for sunny days, 202 for cloudy days, 203 for rainy days, etc.;

[0102] Traffic status (301, 302, 303, ...): 300 plus different positive integers represent different traffic statuses, such as 301 may represent smooth traffic, 302 may represent light congestion, 303 may represent severe congestion, etc.

[0103] By taking into account multiple factors such as logistics vehicles, storage facilities, obstacles, weather and traffic, the automaton model can better simulate the real logistics environment; different states can be distinguished on the visual interface by different colors or patterns.

[0104] The state set can be adjusted according to specific needs, and certain states can be added or deleted; at the same time, different weights are assigned to each state to indicate their degree of influence on the logistics route, providing strong support for the identification of logistics route patterns.

[0105] The neighborhood radius of the grid is defined as Moore neighborhood (8 neighborhoods), and the state transition rules are preset. The state transition rules include vehicle state transition rules, obstacle state transition rules, weather state transition rules, and traffic state transition rules.

[0106] The vehicle state transition rules are:

[0107] If the current grid is in an idle state (0) and there is a logistics vehicle state in the adjacent grid, the current grid is transferred to the corresponding logistics vehicle state with a dynamic probability PL;

[0108] PL=(1-a1-a2-a3)×P_bas+a1×f(v)+a2×g(w)+a3×h(r);

[0109] Among them, a1, a2 and a3 are three state weight coefficients, which are used to balance the influence of the three factors; f(v) is the vehicle type influence function, g(w) is the weather state influence function, and h(r) is the traffic state influence function, which are specifically numbered, that is, the vehicle type is numbered as v, v=1 represents a small vehicle, v=2 represents a medium-sized vehicle, v=3 represents a large vehicle, etc., the weather state is numbered as w, w=1 represents a sunny day, w=2 represents a cloudy day, w=3 represents a rainy day, etc., and the traffic state is numbered as r, r=1 represents a smooth road condition, r=2 represents a light congestion, r=3 represents a severe congestion, etc.;

[0110] Wherein, v′ is a positive number less than 1 and greater than 0, and v_max is the number of the largest vehicle type;

[0111] Wherein, w′ is a positive number less than 1 and greater than 0, and w_max is the number of the maximum weather state;

[0112] Wherein, r′ is a positive number less than 1 and greater than 0, and r_max is the number of the maximum traffic state.

[0113] It should be noted that the larger the vehicle type number, the higher the probability value (it is easier to occupy an empty grid); the worse the weather conditions, the higher the probability value (the vehicle is more likely to occupy an empty grid); the more congested the traffic conditions, the lower the probability value (it is difficult for the vehicle to occupy an empty grid);

[0114] If the current grid is in a logistics vehicle state, the current logistics vehicle state will be maintained with a preset retention probability, or transferred to the logistics vehicle state of an adjacent grid (simulating vehicle movement); if the current grid is in an obstacle state, the logistics vehicle state cannot be transferred to the grid.

[0115] The obstacle state transition rules are:

[0116] If the current grid is in an idle state (0) and there is an obstacle state in the adjacent grid, the current grid is transferred to the corresponding obstacle state with a preset transfer probability; if the current grid is in an obstacle state, the current obstacle state is maintained, or it is transferred to an idle state with an elimination probability pc (simulating the duration and removal of obstacles).

[0117] Among them, b1, b2, b3 and b4 are probability parameters, b1 controls the influence of obstacle type on the persistence probability, b2 is the displacement parameter of obstacle type, b3 controls the influence of weather conditions on the persistence probability, and b4 controls the influence of traffic conditions on the persistence probability; o is the number of the obstacle type (the same as the setting method of vehicle type); it can well characterize the nonlinear trend of the obstacle persistence probability changing with the three influencing factors.

[0118] If the current grid is in a logistics vehicle state or a storage state, the obstacle state cannot be transferred to that grid (the obstacle will block the vehicle and storage facilities).

[0119] The weather state transition rules are:

[0120] If the current grid is in an idle state (0), then the corresponding weather data is queried according to the current time and the geographical location of the grid to determine which weather state to transfer to; specifically, the entire study area is divided into several meteorological areas according to the geographical location in advance, and for each meteorological area, a spatiotemporal probability model of the weather state is constructed, such as a Gaussian mixture model (GMM), a Markov model, etc.; in the spatiotemporal probability model, the transition of weather states follows certain probability distributions, and there are transition probabilities between different weather states; given the current time and geographical location, the spatiotemporal probability model of the corresponding meteorological area is queried, and the probabilities of various weather states are calculated based on historical data; the weather state with the highest probability is used as the transition state of the current grid.

[0121] If the current grid is in a weather state, the current weather state is maintained with the weather maintenance probability p_w (obtained from historical data statistics), or it is transferred to another weather state with the weather transition probability p′_w;

[0122] p′_w=(1-p_w)×P(w′|w); where P(w′|w) is the conditional probability of the current weather state number w transferring to the weather state number w′, which is given by the above model; if the current grid is in a logistics vehicle state or a warehousing state, the weather state will not have a direct impact and the grid will be transferred according to normal rules; if the current grid is in an obstacle state, the weather state will transfer to an idle state.

[0123] The preset methods of traffic state transition rules include:

[0124] The road network is modeled as a directed weighted graph G(V, E), where the nodes in the node set V represent intersections or road sections, and the edges and their weights in the edge set E represent the length and capacity of the road;

[0125] Define vehicle agent attributes and behavior strategies; vehicle agent attributes include: current location (node), destination, vehicle model and travel time; behavior strategies include path selection strategies (such as the shortest path) and vehicle speed adjustment strategies (such as adjusting vehicle speed according to congestion conditions).

[0126] Set the simulation time step dt. At each time step, all vehicle agents move in the directed weighted graph according to the behavioral strategy. After the vehicle reaches the destination, it is removed from the directed weighted graph. According to the spatiotemporal probability model, new vehicle agents are continuously generated in the directed weighted graph.

[0127] At each time step, the number of vehicles at each node (intersection / road section) is counted, and the number of vehicles at the node is mapped to the traffic status of the node (such as smooth, congested, etc.). The traffic status of the node can be set to a discrete value or a continuous value (congestion degree).

[0128] For each node, the transfer of its traffic state is affected by the traffic state of adjacent nodes, which is implemented by a rule-based transfer model or a data learning-based transfer model (such as the Markov model); the transfer model reflects the dynamic characteristics of traffic flow, such as spread, evacuation, and periodicity.

[0129] The directed weighted graph G(V, E) is spatially mapped to the grid space to obtain a complex. Each grid in the grid space is associated with a node in the directed weighted graph. The size of the grid is set to be equivalent to the actual scale of the node (scaled down), such as 100 meters × 100 meters; if there is an edge e∈E between adjacent grids, the corresponding grids are also connected; the vehicle agent is placed in the grid corresponding to its current position, and the attributes of the vehicle agent (such as the vehicle model) can be mapped to different state values ​​(discrete values) of the grid. The same grid can accommodate several vehicle agents to represent the congestion situation.

[0130] At each time step, all vehicle agents move in the grid space according to the behavioral strategy. The vehicle agents select the next step position according to the behavioral strategy through the connectivity between the grids. Similarly, after the vehicle agents reach the destination, they are removed from the grid space.

[0131] Count the number of vehicle agents in each grid, and map the number of vehicle agents to the traffic state value of the grid, such as 0: idle; 1-10: unobstructed (the specific value indicates the number of vehicles); 11-20: slow moving; greater than 20: congested; set a state transition probability matrix for the complex (the traffic state transition of the grid is affected by the traffic state of the neighboring grid; the transition probability in the state transition probability matrix is ​​obtained by data learning, and some empirical rules can also be set, such as the probability of transitioning from a congested state to a smooth state is low, and the probability of transitioning from a congested state to a higher congested state is high. The more congested states in the neighborhood, the higher the probability of transitioning to a congested state; the complex with the state transition probability matrix set is the traffic state transition rule.

[0132] The traffic simulation of intelligent agents is naturally embedded in the grid space. The movement of intelligent agents will affect the traffic status of the grid, and the traffic status of the grid will affect the behavioral strategy of the intelligent agents. The two interact and evolve dynamically. The traffic simulation method can more microscopically and realistically characterize the spatiotemporal evolution process of traffic flow, and organically couple it with other factors (such as logistics, weather, etc.) through the automaton model, providing strong support for the overall logistics route pattern recognition.

[0133] Initialize the initial state, time step and total number of iterations of the grid space; for each time step, traverse all grids, calculate the state of the grid in the next time step based on the state of the current grid, the state of the neighboring grids (grids in the Moore neighborhood) and the state transition rule, and update the state of all grids, repeating until the total number of iterations is reached.

[0134] Key features are extracted from the grid space, including clustering features, dynamic features and topological features; clustering features are the clustering degree of the grid (the sum of the state values ​​of all grids in the grid neighborhood) and the clustering shape (the shape compactness of the clustering area, that is, the equivalent circularity index); dynamic features are the changing trend and periodicity of the grid state; topological features are the connectivity of the grid space (the ratio of the number of nodes of the maximum connected component to the total number of nodes) and obstacle distribution (the proportion of grids in obstacle state to the total number of grids).

[0135] For each grid, calculate the temporal trend of its state, use linear regression and other methods to fit the trend line, and use the slope as the trend. For each grid, extract the time series data of its state. Use spectral analysis and other time series decomposition methods to decompose the periodic components, and calculate the goodness of fit between the periodic components and the original time series data as the periodicity. The key features can reflect important information such as the spatial distribution and temporal evolution of logistics routes.

[0136] The extracted key features are input into the classification / clustering algorithm (such as support vector machine, DBSCAN clustering, K-Means clustering, etc.), and the grid space is divided into different pattern areas, each of which corresponds to a logistics route pattern; the dynamic evolution process of logistics data in time and space is simulated, and potential aggregation, diffusion and other patterns in the data are automatically discovered, so as to identify different logistics route patterns; it has strong explanatory power and can intuitively display the spatiotemporal distribution characteristics of the pattern.

[0137] The calculation method of the optimal logistics planning solution includes:

[0138] Constructing the optimization objective function Among them, V1, V2 and V3 are optimization weight coefficients used to balance the importance of the three objectives; I and J are nodes in a directed weighted graph, (I, J) represents the edge between node I and node J, c_IJ is the unit transportation cost of edge (I, J), x_IJ is a 0-1 variable, x_IJ = 1 when the vehicle passes through edge (I, J), otherwise it is 0; t_IJ is the travel time of edge (I, J); h is the index of the order, SP_h is the actual delivery time of order h, QP_h is the expected delivery time of order h, and N1 is the number of orders; it is used to define the service level.

[0139] The optimization decision variables are defined to include vehicle scheduling plan (number and type of vehicles), route selection plan (selecting the road sections according to M logistics route modes) and warehouse scheduling plan (warehouse in and out order).

[0140] The constraints constructed include vehicle load, time window, driving distance and storage capacity, and the constraints are adjusted according to the characteristics of M logistics route modes; for example, the traffic conditions, congestion level and other characteristics of each logistics route mode are analyzed; for modes with severe congestion, the time window constraints can be appropriately relaxed; for modes with good road conditions, the time window constraints can be appropriately reduced; the time window constraints can also be dynamically adjusted according to factors such as weather conditions and event impacts.

[0141] Permutation coding is used to encode the optimization decision variables (vehicle scheduling, route selection, warehouse scheduling) into a permutation; according to M logistics route patterns, an initial molecular group (including several molecules) is generated, and the function value FL′ of the optimization objective function of each molecule is calculated; the method of generating the initial molecular group is:

[0142] Feature extraction is performed for each logistics route mode, including route length, number of nodes passed, average transportation time and estimated cost; an initial number of molecules N3 is set, usually N3 should be greater than M to ensure sufficient diversity, and a weight value is assigned to each logistics route mode according to the quality of each logistics route mode. The higher the weight value, the better the mode. For each logistics route mode, several initial solutions based on the logistics route mode are generated, where the initial solution is proportional to the weight value of the logistics route mode.

[0143] Mapping function values ​​to molecular masses Among them, ω and ω1 are positive weight parameters used to adjust the relative importance of the two items, Fmax is the maximum function value of all molecules in the current molecule group; PO is the penalty term of the molecule, reflecting the degree of its violation of the constraint conditions;

[0144] Calculate the mass difference ΔM between any two molecules s1 and s2; based on the mass difference, calculate the attractive force FY_s1s2 or repulsive force FC_s1s2 between the two molecules; for each molecule, add up the attractive or repulsive forces exerted on it by all other molecules to obtain the total force.

[0145] Where G′ is the gravitational constant, R_s1s2 is the distance between two molecules s1 and s2;

[0146] Among them, H′ is the repulsion constant, n2 is the repulsion coefficient that controls the repulsion strength;

[0147] For each molecule, add up the attractive or repulsive forces exerted on it by all other molecules to get the total force;

[0148] Based on the total force, calculate the acceleration of each molecule (total force divided by molecular mass); based on the acceleration, update the molecular velocity (obtained by the acceleration equation); based on the molecular velocity, update the molecular position (speed time multiplied by velocity plus original position).

[0149] An upper mass threshold is preset, and molecules with a molecular mass greater than the upper mass threshold are mutated and split to generate new molecules; the mutation is performed to increase / decrease the number of vehicles or change the route; the new molecules inherit the position and speed of the molecules before the split; a lower mass threshold is preset, and molecules with a mass less than the lower mass threshold are aggregated to avoid falling into a local optimum; the aggregation is performed as direct replacement or fusion; the position and speed of the new molecules obtained after aggregation are calculated based on the weighted average of the masses of the aggregated molecules.

[0150] Until the standard deviation of the function value of the optimization objective function of the molecular group is less than the preset convergence threshold, the updating, mutation, splitting and aggregation are terminated; the molecule with the largest molecular mass at this time is extracted as the optimal solution; the optimal solution is decoded into the actual logistics route planning plan, which is the optimal logistics planning plan; the optimal logistics planning plan includes vehicle scheduling, route selection (combined with M logistics route modes) and warehouse scheduling.

[0151] Specifically, vehicle scheduling, route selection, and warehouse scheduling are encoded as a permutation, and the steps for decoding the permutation are as follows: according to the first few bits of the encoded permutation, the scheduling information such as the number of vehicles and vehicle models is decoded; according to the middle part of the encoded permutation, the driving route of each vehicle is decoded in combination with the M logistics route modes obtained previously. Specifically, according to the encoding, the corresponding road sections are selected from the M logistics route modes and spliced ​​into a complete route; according to the last few bits of the encoded permutation, the warehouse scheduling information of the order and time of in-and-out storage and out-of-stock is decoded; the decoded scheme is checked to ensure that the constraints are met. If not, local repair strategies are adopted, including route segmentation, customer point exchange, customer point insertion, and vehicle type upgrade; route segmentation is to split a long route that violates the constraints into two or more short routes; customer point exchange is to exchange customer points between different routes of the same vehicle so that the load capacity meets the constraints; customer point insertion is to insert certain customer points into other routes that do not violate the constraints; vehicle type upgrade is to upgrade small vehicles to large vehicles to increase the load capacity.

[0152] M types of logistics route patterns provide prior knowledge for route selection. The patterns are used as initial solutions, or are utilized in encoding, calculation, boundary processing and other links to improve performance. At the same time, the optimization process may also discover new and better patterns to improve and supplement the original patterns.

[0153] This embodiment can accurately capture the complex dynamic characteristics and potential laws of logistics data, greatly improve the accuracy of logistics prediction and decision-making. This data-driven intelligent management method can not only significantly optimize the warehouse layout and inventory strategy, but also realize the dynamic optimization of transportation routes, thereby significantly reducing logistics costs and improving resource utilization efficiency. The identified logistics route pattern contains valuable experience in logistics operations, guides subsequent planning decisions, avoids repeated exploration, and improves efficiency; at the same time, it can achieve global optimization of the entire supply chain, improve the flexibility and risk resistance of the supply chain, and the adaptive optimization algorithm can quickly generate the optimal solution in a complex and changeable logistics environment, significantly improve the logistics response speed and service quality, and improve and supplement the original model to form a virtuous cycle of continuous iterative optimization; secondly, when conducting logistics planning, a variety of influencing factors are comprehensively considered to obtain a more comprehensive and practical optimal solution; thirdly, the identified logistics route pattern is used as a priori knowledge for optimization, which greatly improves the optimization efficiency and can flexibly adapt to logistics needs of different scales and types.

[0154] Example 2

[0155] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a warehousing logistics management system based on logistics big data, including: a data acquisition module for collecting multi-source heterogeneous logistics data;

[0156] The data processing module is used to preprocess the logistics multi-source heterogeneous data to obtain preprocessed heterogeneous data;

[0157] The pattern recognition module is used to perform pattern recognition on the preprocessed heterogeneous data using an improved automaton algorithm to obtain M types of logistics route patterns;

[0158] The dynamic planning module optimizes and calculates the optimal logistics planning scheme based on M logistics route modes, and uses the optimal logistics planning scheme to guide warehouse logistics management; each module is connected through wired and / or wireless means to achieve data transmission between modules.

[0159] Example 3

[0160] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the warehouse logistics management method based on logistics big data provided above is implemented.

[0161] Since the electronic device introduced in this embodiment is an electronic device used to implement a warehousing and logistics management method based on logistics big data in the embodiment of this application, based on the warehousing and logistics management method based on logistics big data introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the warehousing and logistics management method based on logistics big data in the embodiment of this application, it belongs to the scope of protection of this application.

[0162] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0163] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A warehousing logistics management method based on logistics big data, characterized in that: include: S1. Collecting heterogeneous data from multiple sources of logistics; S2, preprocessing the logistics multi-source heterogeneous data to obtain preprocessed heterogeneous data; S3, using the improved automaton algorithm to perform pattern recognition on the preprocessed heterogeneous data to obtain M types of logistics route patterns; S4. Based on M types of logistics route patterns, optimize the calculation to obtain the optimal logistics planning solution, and use the optimal logistics planning solution to guide warehouse logistics management.

2. A warehousing logistics management method based on logistics big data according to claim 1, characterized in that: The logistics multi-source heterogeneous data includes logistics transportation data, logistics order data, warehousing data, weather data, traffic data and equipment data.

3. A warehousing logistics management method based on logistics big data according to claim 2, characterized in that: The method of preprocessing logistics multi-source heterogeneous data includes: Process missing values ​​in logistics multi-source heterogeneous data, remove duplicate data in logistics multi-source heterogeneous data, process outliers in logistics multi-source heterogeneous data, and obtain preliminary standard heterogeneous data; merge data from different data sources of preliminary standard heterogeneous data to obtain preliminary fused data; encode non-numeric data in preliminary fused data into numeric data, and normalize / standardize the numeric data in preliminary fused data to obtain pre-processed heterogeneous data; The method of performing spatial alignment includes: Project the original preliminary standard heterogeneous data to the preliminary spatial dimension; construct a statistical grid on the preliminary spatial dimension, divide the entire data space into regular grid cells, calculate the number of data points in each grid cell, and obtain the data distribution of each grid cell by statistics; calculate the data density of each grid cell based on the number of data points in each grid cell, preset an upper density threshold, and regard the grid cells above the upper density threshold as dense data areas; preset a lower density threshold, and regard the grid cells below the lower density threshold as sparse data areas; Define an initial partition granularity as the basic partition of the entire data space. For the grid cells that are dense data areas, divide them into smaller sub-units with a side length of Among them, ρ_min is a preset minimum density threshold, ρ is the data density of the grid unit as the dense data area; ε is a small constant; α0 is the preset initial adjustment parameter, and α1 is the linear term coefficient; Merge adjacent grid cells that are sparse data areas to form several partition units; Construct a hierarchical structure, the hierarchical structure includes the top layer, the middle layer and the bottom layer; the basic partition of the entire data space is the top layer, the sub-units divided by the dense data area are the middle layer, and the partition unit is the bottom layer; According to the spatial coordinates of the data points, the layer to which they belong is determined. For the data points belonging to the middle layer, they are processed more finely; for the data points belonging to the bottom layer, they are processed roughly; and then all the layers are aligned according to the spatial dimensions.

4. A warehousing logistics management method based on logistics big data according to claim 3, characterized in that: The time alignment method includes: For each data source, extract the timestamp sequence from the preliminary standard heterogeneous data, and convert the extracted timestamp sequence into a unified time format to obtain a unified timestamp sequence; calculate the time interval between adjacent timestamps in the unified timestamp sequence, and calculate the distribution characteristics of the time interval, including the mean μ, variance σ, median me, lower quartile Q1, upper quartile Q3 and skewness sk; use the distribution characteristics to calculate the uniform distribution index U; Among them, w1, w2, w3 and w4 are weight coefficients, and γ is the adjustment parameter; A uniformity threshold is preset. If the uniform distribution index U is greater than or equal to the uniformity threshold, the timestamps of the preliminary standard heterogeneous data are uniformly distributed. For uniformly distributed timestamps, linear interpolation is used for time alignment. Define several time windows, set the starting time point of the time window, and number the time windows; that is, the time window number; traverse the unified timestamp sequence, for each timestamp t in the unified timestamp sequence, calculate the time window number b to which t belongs; count the number of timestamps in the time window corresponding to b; obtain the time window count sequence C = [c_1, c_2, c_3, ..., c_m], where m is the total number of time windows, and c_m represents the number of timestamps in the mth time window; The timestamp density sequence D = [d_i] is calculated based on the time window count sequence C; d_i is the density of the i-th time window; Among them, ban is the bandwidth parameter of the kernel function, norm is the normalization coefficient; j is another index of the time window; c_j is the number of timestamps in the j-th time window; w_j is the weight value of the j-th time window; is a regulation function based on the time window index j; Where len(j) is the length of the jth time window, pos(j) is the start time or center time of the jth time window, base(j) is the base function, which represents the initial weight of the jth time window; α, β, and δ are adjustable parameters; The timestamp density sequence D is plotted into a curve in chronological order, which is the timestamp density curve. The timestamp density curve is peak detected to obtain the local maximum point p, and the lower critical value x_min and the upper critical value x_max are defined. The local maximum point p is fixed and expanded to obtain the dense area [p-x_min, p-x_max]. For timestamps in dense areas, spline interpolation or Gaussian kernel interpolation is used for time alignment.

5. A warehousing logistics management method based on logistics big data according to claim 4, characterized in that: The method of performing pattern recognition on preprocessed heterogeneous data includes: An automaton model is constructed based on preprocessed heterogeneous data, and the preprocessed heterogeneous data is mapped to a preset grid space, that is, the dimension and size of the grid space are defined; the spatial coordinates in the preprocessed heterogeneous data are mapped to the grid index of the grid space; a set of grid states in the grid space is defined; the neighborhood radius of the grid is defined as the Moore neighborhood, and state transition rules are preset, which include vehicle state transition rules, obstacle state transition rules, weather state transition rules, and traffic state transition rules; Initialize the initial state, time step and total number of iterations of the grid space; for each time step, traverse all grids, calculate the state of the grid in the next time step based on the state of the current grid, the state of the neighboring grids and the state transition rule, and update the state of all grids, repeating until the total number of iterations is reached; Extract key features from the grid space, including clustering features, dynamic features and topological features; clustering features are the clustering degree and clustering shape of the grid; dynamic features are the changing trend and periodicity of the grid state; topological features are the connectivity and obstacle distribution of the grid space; The extracted key features are input into the classification / clustering algorithm to divide the grid space into different pattern areas, each of which corresponds to a logistics route pattern.

6. A warehousing logistics management method based on logistics big data according to claim 5, characterized in that: The vehicle state transition rule is: If the current grid is idle and there is a logistics vehicle state in the adjacent grid, the current grid is transferred to the corresponding logistics vehicle state with the dynamic probability PL; PL = (1-a1-a2-a3) × P_bas + a1 × f(v) + a2 × g(w) + a3 × h(r); where a1, a2 and a3 are three state weight coefficients; f(v) is the vehicle type influence function, g(w) is the weather state influence function, and h(r) is the traffic state influence function; Among them, v_max is the number of the largest vehicle type; Among them, w_max is the number of the maximum weather state; Among them, r′, v′ and w′ are all positive numbers less than 1 and greater than 0, and r max is the number of the maximum traffic state; If the current grid is in a logistics vehicle state, the current logistics vehicle state is maintained with a preset retention probability, or transferred to the logistics vehicle state of an adjacent grid; if the current grid is in an obstacle state, the logistics vehicle state cannot be transferred to the grid.

7. A warehousing logistics management method based on logistics big data according to claim 6, characterized in that: The obstacle state transfer rule is: If the current grid is in an idle state and there is an obstacle state in the adjacent grid, the current grid is transferred to the corresponding obstacle state with a preset transfer probability; If the current grid is in an obstacle state, keep the current obstacle state, or transfer to an idle state with the elimination probability pc; Among them, b1, b2, b3 and b4 are probability parameters, o It is the number of the obstacle type. If the current grid is in the logistics vehicle state or storage state, the obstacle state cannot be transferred to this grid. The weather state transition rules are: If the current grid is idle, the entire study area is divided into several meteorological areas according to geographical location in advance, and a spatiotemporal probability model of weather conditions is constructed for each meteorological area. In the spatiotemporal probability model, given the current time and geographical location, the spatiotemporal probability model of the corresponding meteorological area is queried, and the probabilities of various weather conditions are calculated based on historical data. The weather state with the highest probability is used as the state of the current grid transfer. If the current grid is in a weather state, the current weather state is maintained with the weather maintenance probability p_w, or it is transferred to another weather state with the weather transition probability p′_w; p′_w=(1-p_w)×P(w′|w); where P(w′|w) is the conditional probability of the current weather state number w transferring to the weather state number w′; If the current grid is in a logistics vehicle state or a storage state, the weather state will not have a direct impact; if the current grid is in an obstacle state, the weather state will be transferred to an idle state.

8. A warehousing logistics management method based on logistics big data according to claim 7, characterized in that: The preset methods of traffic state transition rules include: The road network is modeled as a directed weighted graph G(V, E), where the nodes in the node set V represent intersections or road sections, and the edges and their weights in the edge set E represent the length and capacity of the road; Define vehicle agent attributes and behavior strategies; vehicle agent attributes include: current location, destination, vehicle model and travel time; behavior strategies include path selection strategy and vehicle speed adjustment strategy; set the simulation time step dt, at each time step, all vehicle agents move in the directed weighted graph according to the behavior strategy, and after the vehicle reaches the destination, it is removed from the directed weighted graph, and new vehicle agents are continuously generated in the directed weighted graph according to the spatiotemporal probability model; At each time step, the number of vehicles on each node is counted, and the number of vehicles on the node is mapped to the traffic state of the node; for each node, the transfer of its traffic state is affected by the traffic state of the adjacent nodes, and is implemented by a rule-based transfer model or a data learning-based transfer model; the directed weighted graph G(V, E) is spatially mapped with the grid space to obtain a complex; a state transfer probability matrix is ​​set for the complex, and the complex with the state transfer probability matrix set is the traffic state transfer rule.

9. A warehousing logistics management method based on logistics big data according to claim 8, characterized in that: The method of spatially mapping the directed weighted graph G(V, E) to the grid space includes: Each grid in the grid space is associated with a node in a directed weighted graph, and the size of the grid is set to be equivalent to the actual scale of the node; if there is an edge between adjacent grids, the corresponding grids are also connected; the vehicle agent is placed in the grid corresponding to its current position, and the attributes of the vehicle agent are mapped to different state values ​​of the grid. The same grid can accommodate several vehicle agents; at each time step, all vehicle agents move in the grid space according to the behavioral strategy. The vehicle agent selects the next step position according to the connectivity relationship between the grids and follows the behavioral strategy. Similarly, after the vehicle agent reaches the destination, it is removed from the grid space; the number of vehicle agents in each grid is counted, and the number of vehicle agents is mapped to the traffic state value of the grid.

10. A warehousing logistics management method based on logistics big data according to claim 9, characterized in that: The calculation method of the optimal logistics planning solution includes: Constructing the optimization objective function Wherein, V1, V2 and V3 are optimization weight coefficients; I and J are nodes in the directed weighted graph, (I, J) represents the edge between node I and node J, c_IJ is the unit transportation cost of edge (I, J), x_IJ is a 0-1 variable, when the vehicle passes through edge (I, J), x_IJ = 1, otherwise it is 0; t_IJ is the travel time of edge (I, J); h is the index of the order, SP_h is the actual delivery time of order h, QP_h is the expected delivery time of order h, and N1 is the number of orders; Define optimization decision variables including vehicle scheduling, route selection, and warehouse scheduling; construct constraints including vehicle load, time window, driving distance, and warehouse capacity, and adjust the constraints according to the characteristics of M logistics route modes; Permutation coding is used to encode the optimization decision variables into a permutation; according to M logistics route patterns, an initial molecule group is generated, and the function value FL′ of the optimization objective function of each molecule is calculated; Mapping function values ​​to molecular masses Among them, ω and ω1 are positive weight parameters used to adjust the relative importance of the two items, Fmax is the maximum function value of all molecules in the current molecule group; PO is the penalty term of the molecule, reflecting the degree of its violation of the constraint conditions; Calculate the mass difference ΔM between any two molecules s1 and s2; based on the mass difference, calculate the attractive force FY_s1s2 or repulsive force FC_s1s2 between the two molecules; for each molecule, add up the attractive or repulsive forces exerted on it by all other molecules to obtain the total force; For each molecule, add up the gravitational or repulsive forces exerted on it by all other molecules to obtain the total force; calculate the acceleration of each molecule based on the total force; update the molecular velocity based on the acceleration; update the molecular position based on the molecular velocity; preset an upper mass threshold, mutate molecules with a molecular mass greater than the upper mass threshold, and split them to produce new molecules; perform mutations to increase / decrease the number of vehicles or change the route; the new molecule inherits the position and velocity of the molecule before the split; A lower mass threshold is preset, and molecules with a mass less than the lower mass threshold are polymerized; the polymerization is performed as direct replacement or fusion; the position and speed of the new molecules obtained after polymerization are calculated based on the weighted average of the mass of the polymerized molecules; Until the standard deviation of the function value of the optimization objective function of the molecular group is less than the preset convergence threshold, the updating, mutation, splitting and aggregation are terminated; the molecule with the largest molecular mass at this time is extracted as the optimal solution; the optimal solution is decoded into the actual logistics route planning plan, which is the optimal logistics planning plan; the optimal logistics planning plan includes vehicle scheduling, route selection and warehouse scheduling.

Images