Intelligent warehouse location planning optimization method based on hybrid particle swarm algorithm

Abstract

The application discloses an intelligent warehouse location planning optimization method based on a hybrid particle swarm algorithm, which comprises the following steps: step 1, establishing a warehouse space model based on Manhattan distance and designing an overall fitness function; step 2, establishing an in-out warehouse frequency prediction model based on a long short-term memory network; step 3, performing random disturbance and stretching processing on the prediction result of the in-out warehouse frequency prediction model to expand the frequency distribution range; step 4, based on the particle swarm algorithm, realizing dynamic search and convergence balance of the location optimization by adaptively adjusting the inertia weight; step 5, based on the genetic algorithm, introducing crossover and mutation operations in the particle updating process to enhance the diversity of the location distribution; and step 6, based on the simulated annealing algorithm, introducing a temperature parameter and a poor solution acceptance mechanism to jump out of a local optimal solution and improve the global location optimization capability. The application effectively reduces the material flow path length and time consumption and significantly improves the operation efficiency and management capability of the intelligent warehouse system.

Description

Technical Field

[0001] This invention belongs to the field of intelligent warehouse location planning, and relates to an intelligent warehouse location planning optimization method based on a hybrid particle swarm optimization algorithm. Background Technology

[0002] Automated production processes based on warehousing and logistics planning are crucial for the development of various industries. Against this backdrop, intelligent warehouse location planning and optimization has received widespread attention. Its focus lies in warehouse space modeling, inbound / outbound frequency prediction, and intelligent location planning. The purpose of warehouse space modeling is to establish quantitative standards for location planning; inbound / outbound frequency prediction aims to forecast future inbound / outbound data; and intelligent location planning aims to rationally allocate warehouse locations and improve warehousing efficiency.

[0003] In the research process of intelligent warehouse location planning, the following problems were found: (1) Existing methods often concentrate goods with high entry and exit frequency in locations closer to the entrance and exit, which can easily lead to high-frequency goods piling up in local areas, resulting in uneven warehouse space utilization and congestion and path conflicts; (2) Traditional location planning models are mostly based on static data for one-time optimization, lacking the ability to predict future demand changes; (3) They are prone to getting trapped in local optima, especially in high-dimensional spaces or under complex constraints, making it difficult to guarantee search efficiency and solution quality. Summary of the Invention

[0004] To address the aforementioned technical problems in existing technologies, this invention proposes an intelligent warehouse location planning and optimization method based on a hybrid particle swarm optimization algorithm, the specific technical solution of which is as follows:

[0005] A smart warehouse location planning and optimization method based on a hybrid particle swarm optimization algorithm includes:

[0006] Step 1: Establish a warehouse spatial model based on Manhattan distance and design the overall fitness function;

[0007] Step 2: Establish an inbound / outbound frequency prediction model based on an LSTM (Long Short-Term Memory) network;

[0008] Step 3: Randomly perturb and stretch the prediction results of the inbound / outbound frequency prediction model to expand the frequency distribution range;

[0009] Step 4: Based on the particle swarm optimization algorithm, the dynamic search and convergence balance of the storage location optimization is achieved by adaptively adjusting the inertia weight;

[0010] Step 5: Based on the genetic algorithm, crossover and mutation operations are introduced in the particle update process to enhance the diversity of reservoir distribution;

[0011] Step 6: Based on the simulated annealing algorithm, by introducing temperature parameters and a poor solution acceptance mechanism, we can escape local optima and improve the global storage location optimization capability.

[0012] Furthermore, in step 1, the three-dimensional coordinates of the storage space are modeled in a grid, the Manhattan distance from the outbound and inbound locations to each storage location is calculated, and a height penalty term is introduced to quantify the material flow cost within the warehouse.

[0013] The formula for calculating the Manhattan distance is:

[0014] D = |x - x0| + |y - y0| + |h - h0|

[0015] Where (x0,y0,h0) is the location of the outbound or inbound port, (x,y,h) is the three-dimensional coordinate of the target storage location, and D is the Manhattan distance between the outbound port and the target storage location;

[0016] The warehouse space model incorporates a height penalty term when calculating the Manhattan distance:

[0017] D′=D+β·h,

[0018] Where β is the height penalty coefficient, and h is the height of the storage location;

[0019] A frequency dispersion penalty term is also introduced, calculated as follows:

[0020]

[0021] Where λ is the penalty coefficient, used to control the influence weight of the penalty term in the overall fitness function, N is the number of SKUs in inventory, α is the sensitivity coefficient of frequency difference, controlling the convergence speed of the penalty term, |f i -f j | represents the difference in outbound frequency between SKUi and SKUj, and D(i,j) is the Manhattan distance between SKUi and SKUj. When the outbound frequencies of SKUi and SKUj are similar, the penalty term is larger, prompting the algorithm to distribute them. When the outbound frequencies of SKUi and SKUj differ significantly, the penalty term tends to 0, and SKUs with large frequency differences are relatively concentrated. When the storage distance D(i,j) increases, the penalty term decreases, encouraging SKUs with similar frequencies to be distributed further apart in space. The penalty term increases rapidly as the frequency difference and distance decrease.

[0022] Furthermore, in step 1, the overall fitness function is:

[0023]

[0024] Among them, f out,i and f in,iD represents the inbound and outbound frequency of SKUi. out,i and D in,i w represents the Manhattan distance from the SKU to the entrance / exit. i and v i Here, h represents the weight and volume of the SKU, and h is the floor height of the SKU in the warehouse space.

[0025] Furthermore, step 2 specifically includes:

[0026] First, in the data preprocessing stage, multidimensional historical frequency data is used to normalize the historical frequency data. The normalization formula is as follows:

[0027]

[0028] Among them, f' i,j and g' i,j These are the normalized inbound and outbound frequencies, f i,j and g i,j The original frequencies of the j-th SKU at the i-th time step are min(f) and j, respectively. j ) and max(f j ) are the minimum and maximum outbound frequencies of the j-th inventory unit SKU in all time steps; after normalization, all frequency data are compressed into the [0,1] interval;

[0029] Next, the historical frequency data is divided into time windows to construct a sliding time series dataset:

[0030]

[0031] Among them, X (k) It is the input sequence for the k-th time window, with dimensions τ×N, y (k) It is the prediction frequency for subsequent moments in the corresponding time window;

[0032] A multi-layer LSTM structure is used to establish an inbound / outbound prediction model. The hidden layers extract the temporal features of frequency changes, and the output layer uses linear activation layers to output the predicted inbound / outbound frequency of SKUs at future time points. The predicted future frequency distribution... for:

[0033]

[0034] Among them, W out For the last layer of weights, b out For bias, h t The hidden state at the current time step;

[0035] The model parameters are optimized by adjusting the adaptive learning rate and using the mean squared error loss function, as shown in the formula:

[0036]

[0037] in, To predict the frequency, f i This is the actual frequency;

[0038] The Adam adaptive learning rate algorithm was used to adjust the parameters of the inbound / outbound prediction model.

[0039] Furthermore, step 3 specifically includes:

[0040] First, random perturbations are introduced by superimposing Gaussian noise on the prediction results to simulate the possible fluctuations and uncertainties in future frequencies, thereby avoiding overly regular frequency distributions.

[0041] Subsequently, a stretching factor is introduced to expand the frequency distribution range, and some frequency values ​​are appropriately enlarged or reduced to increase the difference and diversity of the frequency distribution.

[0042] After combining random perturbation with stretching, the future frequency distribution is as follows:

[0043]

[0044] Where 1+∈ is the random fluctuation coefficient and S is the stretching coefficient.

[0045] Furthermore, in step 4, the particle swarm optimization algorithm continuously adjusts the position and velocity of particles by simulating their search behavior in the solution space to seek the optimal storage location distribution. The particle velocity update formula and position update formula are as follows:

[0046] v i (t+1)=w(t)·v i (t)+c1·r1·(p best,i -x i (t))+c2·r2·(g best -x i (t)),

[0047] x i (t+1)=x i (t)+v i (t+1),

[0048] Where, x i (t) represents the storage location distribution of the i-th particle in the t-th iteration, v i (t) represents the velocity of the i-th particle, p best,i For the optimal solution in the history of the i-th particle, g bestt is the global optimal solution in the history of the population, w(t) is the inertia weight, c1 and c2 are learning factors representing the acceleration coefficient, and r1 and r2 are random numbers in the range [0,1][0,1][0,1].

[0049] Furthermore, the particle velocity update employs a probability mapping mechanism, mapping the particle velocity to probabilities using the Sigmoid function:

[0050]

[0051] Among them, P i,j Let v be the probability of particle i jumping in dimension j. i,j (t) represents the velocity of the i-th particle in the j-th dimension.

[0052] Furthermore, in step 4, an adaptive inertia weight mechanism is introduced, and the inertia weight formula is as follows:

[0053]

[0054] Among them, w max The initial inertia weight is set to 0.90 to enhance exploration capabilities. min The final inertia weight is set to 0.40 to enhance convergence capability; t is the current iteration number, and T is the maximum iteration number. In the early stage of iteration, the inertia weight is increased to enhance global exploration, thereby conducting a wider search in the storage space and preventing getting trapped in local optima. In the later stage of iteration, the inertia weight is gradually reduced to speed up the algorithm's convergence, thereby finely adjusting the storage location distribution and ensuring that the storage and retrieval efficiency of the inventory unit SKU is optimized.

[0055] Furthermore, in step 5, a genetic algorithm (GA) is integrated into the particle swarm optimization algorithm. During particle update, two sets of inventory unit SKU arrangements are periodically and randomly selected for crossover operations. By exchanging some SKU positions, new inventory location combinations are generated to broaden the search space. Simultaneously, a random mutation mechanism is introduced to randomly exchange the SKU arrangements in the inventory locations. The genetic algorithm formula is as follows:

[0056]

[0057] Where S1 and S2 are the SKU arrangements of two particles, S is the current SKU arrangement, and p c The crossover rate controls the frequency of crossover operations, ranging from 0.7 to 0.9, p m The mutation rate controls the frequency of mutation operations, typically set to 0.1–0.2. t represents the generation interval for genetic operations, typically set to 5–20.

[0058] Furthermore, in step 6, the dynamic adjustment of the temperature parameter specifically involves:

[0059] The temperature T(t) of generation t is:

[0060] T(t) = T0·α t ,

[0061] Where T0 is the initial temperature, α is the cooling coefficient, which is 0.9≤α≤0.99, and t is the current iteration number. When the initial temperature is higher, the simulated annealing algorithm is more likely to accept inferior solutions, increasing the probability of escaping local optima. As the number of iterations increases, the temperature gradually decreases, the probability of accepting inferior solutions decreases, and the algorithm converges to the global optimum.

[0062] Furthermore, the inferior solution acceptance mechanism is specifically as follows:

[0063] The acceptance probability of a suboptimal solution is based on the Metropolis criterion:

[0064]

[0065] Where P is the probability of accepting the inferior solution, ΔE = E new -E current Let ΔE be the fitness difference between the new solution and the current solution. When ΔE < 0, it means that the new solution is better than the current solution, so the new solution must be accepted. When ΔE > 0, it means that the new solution is worse than the current solution, so the worse solution is accepted with probability P. When the temperature is high, P is large, and it is easier to accept the worse solution and escape the local optimum. When the temperature is low, P is small, and it gradually converges and stabilizes near the global optimum.

[0066] The beneficial effects of this invention include: by combining particle swarm optimization with genetic algorithm and simulated annealing algorithm, the distribution of storage locations is optimized, significantly improving storage efficiency and material turnover speed.

[0067] First, this method designs a warehouse space model based on Manhattan distance. By gridding the three-dimensional coordinates of the storage space, the Manhattan distance from the warehouse and the receiving location to each storage location is calculated, thereby quantifying the material circulation cost in the warehouse.

[0068] Then, this method introduces an inbound / outbound frequency prediction model based on an LSTM (Long Short-Term Memory) network. By performing time series modeling on historical inbound / outbound frequency data, the distribution of SKU inbound / outbound frequencies over a future period is predicted. Random perturbations and stretching coefficients are added to the prediction results to broaden the frequency distribution range, prevent repetition of predicted frequencies, and thus enhance the model's generalization ability.

[0069] Finally, this method employs an improved particle swarm optimization (PSO) algorithm for warehouse location planning. The PSO algorithm achieves a dynamic balance between search and convergence capabilities by adaptively adjusting inertia weights; the genetic algorithm introduces crossover and mutation operations during particle updates to enhance population diversity and prevent local optima; and the simulated annealing algorithm improves global search capabilities by introducing temperature parameters and a suboptimal solution acceptance mechanism to escape local optima. The overall fitness function comprehensively considers factors such as inbound / outbound frequency, Manhattan distance between warehouse locations, cargo weight, and volume during optimization, while also introducing a frequency dispersion penalty term to prevent high-frequency SKUs from being concentrated in one location, optimizing the overall frequency distribution and further improving warehousing efficiency.

[0070] This method can visualize the optimized warehouse layout, making the storage locations more rational and placing high-frequency SKUs closer to the entrances and exits. This effectively reduces the length of material flow paths and time consumption, and significantly improves the operating efficiency and management capabilities of the intelligent warehousing system. Attached Figure Description

[0071] Figure 1 This is a flowchart of the intelligent warehouse location planning and optimization method according to an embodiment of the present invention;

[0072] Figure 2 This is a Boolean matrix diagram showing the location and cargo characteristics of an embodiment of the present invention.

[0073] Figure 3 This is a schematic diagram of the LSTM inbound / outbound frequency prediction model according to an embodiment of the present invention.

[0074] Figure 4 This is a comparison diagram of data frequency expansion before and after in an embodiment of the present invention;

[0075] Figure 5 This is a diagram showing the optimization results of intelligent warehouse location planning in an embodiment of the present invention. Detailed Implementation

[0076] To make the objectives, technical solutions, and technical effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0077] like Figure 1 As shown in the figure, this invention discloses an intelligent warehouse location planning and optimization method based on a hybrid particle swarm optimization algorithm, including:

[0078] Step 1: Establish a warehouse space model based on Manhattan distance and design the overall fitness function.

[0079] Specifically, a warehouse space model based on Manhattan distance was designed. First, the three-dimensional coordinates of the warehouse space were modeled as a grid, dividing the warehouse area into regular three-dimensional grid cells composed of length, width, and height. The Manhattan distance from the warehouse entry and exit locations to each storage location was calculated, and a height penalty term was introduced to quantify the material flow cost within the warehouse.

[0080] The formula for calculating the Manhattan distance is:

[0081] D = |x - x0| + |y - y0| + |h - h0|

[0082] Where (x0, y0, h0) represents the location of the outbound or inbound entrance, (x, y, h) represents the three-dimensional coordinates of the target storage location, and D represents the Manhattan distance between the entrance / exit and the target storage location. This distance measurement method can accurately reflect the number of grids that materials actually need to cross during storage and retrieval in a regular warehouse grid, thereby effectively quantifying the flow path and handling costs of materials within the warehouse.

[0083] Because material retrieval at high-rise racks is difficult and time-consuming in real-world warehousing environments, the model considers the multi-layered racking structure of the warehouse when calculating the Manhattan distance, incorporating the height dimension into the distance calculation and introducing a height penalty term.

[0084] D′=D+β·h,

[0085] Where β is the height penalty coefficient and h is the height of the storage location.

[0086] Meanwhile, to prevent high-frequency SKUs from accumulating in local areas, this embodiment of the invention introduces a frequency dispersion penalty term, calculated as follows:

[0087]

[0088] Where λ is the penalty coefficient, used to control the influence weight of the penalty term in the fitness function, N is the number of SKUs, α is the sensitivity coefficient of frequency difference, controlling the convergence speed of the penalty term, |f i -f j | represents the difference in outbound frequency between SKUi and SKUj, and D(i,j) is the Manhattan distance between SKUi and SKUj. When the outbound and inbound frequencies of SKUi and SKUj are close, i.e., |f i -f j The smaller the value, the larger the penalty term, prompting the algorithm to distribute the data. When the inbound and outbound frequencies of SKUi and SKUj differ significantly, i.e., |f i -f jWhen the distance between storage locations D(i,j) is large, the penalty term tends to be 0, allowing SKUs with large frequency differences to be relatively concentrated. As the storage location distance D(i,j) increases, the penalty term decreases, encouraging SKUs with similar frequencies to be distributed further apart spatially. The penalty term increases rapidly as the frequency difference and distance decrease, effectively preventing the local accumulation of SKUs with similar frequencies and promoting a more balanced distribution of SKUs within the storage location space.

[0089] Design the overall fitness function in warehouse optimization, wherein the overall fitness function is:

[0090]

[0091] Among them, f out,i and f in,i D represents the inbound and outbound frequency of SKUi. out,i and D in,i w represents the Manhattan distance from the SKU to the entrance / exit. i and v i Let represent the weight and volume of the SKU, h be the floor height of the SKU in the warehouse space, and Penalty be a frequency dispersion penalty term to prevent the accumulation of high-frequency SKUs, optimize the overall frequency distribution, and further improve warehouse efficiency. This function serves as an evaluation index for measuring the quality of SKU location distribution. Figure 2 The diagram shown is a Boolean matrix representing the characteristics of the cargo location and cargo.

[0092] Step 2: As Figure 3 As shown, an inbound / outbound frequency prediction model based on an LSTM (Long Short-Term Memory) network is established.

[0093] By performing time series modeling on historical inbound and outbound frequency data, we can predict the trend of SKU inbound and outbound frequency changes in the future.

[0094] In the data preprocessing stage, the multidimensional historical frequency data is normalized to eliminate numerical differences between different SKU frequencies and prevent gradient explosion or vanishing problems during model training. The minimum-maximum normalization formula is as follows:

[0095]

[0096] Among them, f' i,j and g' i,j These are the normalized inbound and outbound frequencies, f i,j and g i,j The original frequencies of the j-th SKU at the i-th time step are min(f) and j, respectively. j ) and max(f j) represent the minimum and maximum outbound frequencies of the j-th SKU across all time steps. After normalization, all frequency data are compressed to the [0,1] interval to prevent gradient explosion or vanishing problems caused by excessive numerical differences during model training.

[0097] Next, the historical frequency data is divided into time windows to construct a sliding time series dataset:

[0098] X (k) =[f' k ,f' k+1 ,…,f' k+τ-1 ],

[0099] y (k) =f' k+τ ,

[0100] Among them, X (k) It is the input sequence for the k-th time window, with dimensions τ×N, y (k) It is the prediction frequency for subsequent times within the corresponding time window.

[0101] The model learns the time dependence and long-term trend of historical frequency changes using an LSTM network. It employs a multi-layer LSTM structure, where hidden layers extract the temporal features of frequency changes, and the output layer uses linear activation layers to output predicted future SKU inbound / outbound frequencies. The predicted future frequency distribution is then analyzed. for:

[0102]

[0103] Among them, W out For the last layer of weights, b out For bias, h t The hidden state at the current time step.

[0104] The model parameters are optimized by adjusting the adaptive learning rate and using the mean squared error (MSE) loss function, as shown in the formula:

[0105]

[0106] in, To predict the frequency, f i The values ​​represent the true frequencies. Optimization employs the Adam adaptive learning rate algorithm for parameter adjustment to prevent getting trapped in local optima, ensuring the model effectively captures frequency fluctuations during the fitting process and improving prediction accuracy.

[0107] Step 3: Randomly perturb and stretch the model prediction results to expand the frequency distribution range.

[0108] Specifically, after completing the prediction of inbound and outbound frequencies, in order to prevent the future frequency distribution generated by the model from being too smooth or having too high frequency repetition, the prediction results are further subjected to random perturbation and stretching to increase the difference and diversity of the frequency distribution.

[0109] First, random perturbations are introduced by superimposing Gaussian noise onto the prediction results to simulate the potential fluctuations and uncertainties in future frequencies, thus preventing the frequency distribution from becoming overly regular. Then, a stretching coefficient is introduced to expand the frequency distribution range, appropriately amplifying or shrinking some frequency values ​​to increase the variability and diversity of the frequency distribution. After combining the random perturbation and stretching treatment, the future frequency distribution is as follows:

[0110]

[0111] Where 1+∈ is the random fluctuation coefficient and S is the stretching coefficient.

[0112] like Figure 4 As shown, this is a comparison before and after data frequency expansion.

[0113] Step 4: Based on the adaptive particle swarm optimization algorithm, the dynamic search and convergence balance of the storage location optimization is achieved by adaptively adjusting the inertia weight.

[0114] Specifically, the particle swarm optimization algorithm continuously adjusts the position and velocity of particles by simulating their search behavior in the solution space to seek the optimal storage location distribution. The particle velocity update formula and position update formula are as follows:

[0115] v i (t+1)=w(t)·v i (t)+c1·r1·(p best,i -x i (t))+c2·r2·(g best -x i (t)),

[0116] x i (t+1)=x i (t)+v i (t+1),

[0117] Where, x i (t) represents the storage location distribution of the i-th particle in the t-th iteration, v i (t) represents the velocity of the i-th particle, p best,i For the optimal solution in the history of the i-th particle, g best t is the global optimal solution in the history of the population, w(t) is the inertia weight, c1 and c2 are learning factors representing the acceleration coefficient, and r1 and r2 are random numbers in the range [0,1][0,1][0,1].

[0118] An adaptive inertia weight mechanism was designed, enabling particles to dynamically adjust their exploration and convergence capabilities during the search process. The inertia weight formula is as follows:

[0119]

[0120] Among them, w max This is the initial inertia weight, typically set to 0.90, to enhance exploration capabilities. min The final inertia weight is typically set to 0.40 to enhance convergence. t represents the current iteration number, and T represents the maximum iteration number. In the early stages of iteration, a larger inertia weight is used to enhance global exploration, thereby conducting a broader search in the storage space and preventing getting trapped in local optima. In the later stages of iteration, the inertia weight is gradually reduced to accelerate the algorithm's convergence speed, thus finely adjusting the storage location distribution and ensuring optimized SKU access efficiency.

[0121] Furthermore, the particle velocity update employs a probability mapping mechanism, mapping the particle velocity to probabilities via the Sigmoid function:

[0122]

[0123] Among them, P i,j Let v be the probability of particle i jumping in dimension j. i,j (t) represents the velocity of the i-th particle in the j-th dimension, to introduce a moderate degree of randomness, prevent the search process from being too rigid, and help to escape local optima.

[0124] Step 5: Based on the genetic algorithm, crossover and mutation operations are introduced in the particle update process to enhance the diversity of storage location distribution and prevent local optima.

[0125] Specifically, to further enhance the diversity and flexibility of storage location optimization, this invention integrates a genetic algorithm (GA) with the particle swarm optimization (PSO) algorithm. The GA introduces crossover and mutation operations during storage location optimization. Specifically, during particle update, two sets of SKU arrangements are periodically and randomly selected for crossover. By exchanging some SKU positions, new storage location combinations are generated to broaden the search space. Simultaneously, the algorithm introduces a random mutation mechanism, randomly exchanging the SKU arrangements in storage locations with a certain probability. This mutation operation breaks the limitation of fixed solutions, introduces new storage location layout possibilities, and thus prevents the population from getting trapped in local optima during the search process. The GA formula is:

[0126]

[0127] Where S1 and S2 are the SKU arrangements of two particles, S is the current SKU arrangement, and p cThe crossover rate controls the frequency of crossover operations, typically set to 0.7–0.9. m The mutation rate controls the frequency of mutation operations, typically set to 0.1–0.2. t represents the generation interval for genetic operations, typically set to 5–20. Crossover and mutation operations effectively break local optima, enhance the diversity of the solution space, maintain a high solution space exploration capability during SKU location optimization, and further improve SKU access efficiency and overall distribution balance.

[0128] Step 6: Based on the simulated annealing algorithm, by introducing temperature parameters, we can escape local optima, improve the global storage location optimization capability, and break the frequency accumulation effect.

[0129] Specifically, simulated annealing is introduced into a hybrid optimization framework combining particle swarm optimization and genetic algorithms. By introducing a temperature parameter and a suboptimal solution acceptance mechanism, the algorithm's ability to escape local optima is improved, preventing high-frequency SKUs from accumulating near the inlet and outlet, and promoting a more balanced distribution of SKUs in space. The core formula of the simulated annealing algorithm is the dynamic adjustment of the temperature parameter, where the temperature T(t) in generation t is:

[0130] T(t) = T0·α t ,

[0131] Where T0 is the initial temperature, α is the cooling coefficient (usually 0.9 ≤ α ≤ 0.99), and t is the current iteration number. A higher initial temperature makes the algorithm more likely to accept inferior solutions, increasing the probability of escaping local optima. As the number of iterations increases and the temperature gradually decreases, the probability of accepting inferior solutions decreases, eventually converging to the global optimum. The acceptance probability of inferior solutions is based on the Metropolis criterion:

[0132]

[0133] Where P is the probability of accepting the inferior solution, ΔE = E new -E current Let T(t) be the fitness difference between the new solution and the current solution. When the new solution is better than the current solution (ΔE < 0), the new solution is always accepted; when the new solution is worse than the current solution (ΔE > 0), the worse solution is accepted with probability P. At higher temperatures, P is larger, making it easier to accept worse solutions and escape local optima. At lower temperatures, P is smaller, leading to gradual convergence and stabilization near the global optimum.

[0134] Finally, the optimized warehouse layout will be visualized, such as... Figure 5 As shown in the results, the warehouse location planning and optimization results indicate that the method of the present invention has achieved good results in establishing warehouse models, predicting future inventory data, and intelligently allocating warehouse locations in intelligent warehouse location optimization.

[0135] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the implementation process of the present invention has been described in detail above, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent warehouse location planning optimization method based on a hybrid particle swarm algorithm, characterized in that, The application relates to a warehouse space optimization method based on Manhattan distance and a warehouse space optimization device. Step 1: a warehouse space model based on Manhattan distance is established, and an overall fitness function is designed; Step 2: a warehouse-in and warehouse-out frequency prediction model based on an LSTM long short-term memory network is established; Step 3: the prediction result of the warehouse-in and warehouse-out frequency prediction model is subjected to random disturbance and stretching treatment, and the frequency distribution range is expanded; Step 4: based on a particle swarm algorithm, inertia weight is adaptively adjusted to realize dynamic search and convergence balance of the warehouse location optimization; Step 5: based on a genetic algorithm, crossover and mutation operations are introduced in the particle updating process to enhance the diversity of the warehouse location distribution; Step 6: based on a simulated annealing algorithm, a temperature parameter and a poor solution accepting mechanism are introduced to jump out of a local optimal solution and improve the global warehouse location optimization capability; In step 4, the particle swarm algorithm simulates the search behavior of particles in the solution space, and continuously updates and adjusts the position and speed of the particles to seek an optimal warehouse location distribution, wherein a speed updating formula and a position updating formula of the particles are respectively as follows: ， ， wherein, is the bin position distribution of the i-th particle at the i-th iteration, is the velocity of the i-th particle, is the best solution of the i-th particle in history, is the global best solution of the population in history, is the inertia weight, , is the learning factor that represents the acceleration coefficient, , is a random number ranging in [0, 1]; The speed updating of the particles adopts a probability mapping mechanism, and the speed of the particles is mapped into probability through a Sigmoid function: ， where Pij(t) is the probability of a particle i jumping in dimension j at time t, Vj(i) is the velocity of the ith particle in the jth dimension, and N is the total number of particles. i,j i,j (t) is the velocity of the ith particle in the jth dimension at time t.​ In step 4, an adaptive inertia weight mechanism is introduced, and an inertia weight formula is as follows: ， wherein, is an initial inertia weight, set to 0.90, enhancing the ability to explore, is a final inertia weight, set to 0.40, enhancing the ability to converge; is the current iteration number, is the maximum iteration number; in the early stage of iteration, the inertia weight is increased to enhance the global exploration, so as to search more widely in the space of the storage location and prevent falling into local optimum; in the later stage of iteration, the inertia weight is gradually reduced, so that the convergence speed of the algorithm is accelerated, thereby the distribution of the storage location is finely adjusted, and the access efficiency of the inventory unit SKU is optimized.

2. The intelligent warehousing location planning optimization method of claim 1, wherein, In step 1, the three-dimensional coordinates of the warehouse space are subjected to grid modeling, the Manhattan distances from the warehouse-in and warehouse-out positions to each warehouse location are calculated, and a height penalty term is introduced to quantize the transfer cost of materials in the warehouse; The Manhattan distance calculation formula is as follows: ， wherein, is a position of a loading or unloading port, is a three-dimensional coordinate of a target storage location, is a Manhattan distance between the loading or unloading port and the target storage location; When the warehouse space model calculates the Manhattan distance, the height penalty term is introduced: ， wherein, is a highly penalizing coefficient, is the height of the bin location; A frequency dispersion penalty term is also introduced, and a calculation formula is as follows: ， wherein, is a penalty coefficient, used to control the influence weight of the penalty term in the overall fitness function, is the number of inventory units of SKU i, is a sensitivity coefficient of frequency difference, controlling the convergence speed of the penalty term, is the difference of the frequency of SKU i and SKU j, is the Manhattan distance between SKU i and SKU j; when the frequency of SKU i and SKU j is close, the penalty term value is large, prompting the algorithm to store them separately; when the frequency difference of SKU i and SKU j is large, the penalty term tends to 0, and the SKU with large frequency difference is relatively concentrated. When the distance between the storage locations increases, the penalty term decreases, encouraging the SKU with similar frequency to be distributed farther apart in space; the penalty term increases rapidly with the decrease of the frequency difference and the distance.

3. The intelligent warehousing location planning optimization method of claim 2, wherein, In step 1, the overall fitness function is as follows: ， wherein, and is the frequency of the SKUi to enter or exit the warehouse, and is the Manhattan distance of the SKU to the entrance or exit, and is the weight and volume of the SKU, is the height of the SKU in the warehouse space.

4. The intelligent warehousing location planning optimization method of claim 1, wherein, The step 2 specifically includes: Firstly, in the data preprocessing stage, multi-dimensional historical frequency data are used to normalize the historical frequency data, and a normalization formula is as follows: ， in, and These are the normalized inbound and outbound frequencies, and These represent the original frequencies of the j-th SKU at the i-th time step. and Let be the minimum and maximum outbound frequencies of the j-th inventory unit SKU across all time steps; after normalization, all frequency data are compressed to . Within the range; Then, the historical frequency data are divided into time windows to construct a sliding time series data set: ， wherein, is the input sequence of the th time window, with dimension , is the predicted frequency corresponding to the subsequent time instant of the time window. A multi-layer LSTM structure is used to establish the warehouse in-out prediction model, wherein the hidden layer is used to extract the time characteristics of frequency changes, the output layer uses a linear activation layer to output the future time SKU in-out frequency prediction value, and the predicted future frequency distribution is: ， wherein, is the last layer weight, is the bias, is the hidden state of the current time step; Through adaptive learning rate adjustment and mean square error loss function optimization model parameter, a formula is as follows: ， wherein, is the predicted frequency, is the true frequency; The optimization of the warehouse-in and warehouse-out prediction model adopts an Adam adaptive learning rate algorithm to adjust the parameters.

5. The intelligent warehousing location planning optimization method of claim 1, wherein, The step 3 specifically includes: Firstly, random disturbance is introduced by superimposing Gaussian noise on the prediction result to simulate the volatility and uncertainty of the future frequency, so that the frequency distribution is prevented from being too regularized; Then, a stretching coefficient is introduced to expand the frequency distribution range, and part of the frequency values are moderately enlarged or reduced to increase the difference and diversity of the frequency distribution; After the random disturbance and stretching treatment are combined, the future frequency distribution is as follows: ， wherein is a random fluctuation coefficient, is a stretching coefficient.

6. The intelligent warehousing location planning optimization method of claim 1, wherein, In step 5, the genetic algorithm GA is fused on the basis of the particle swarm algorithm, in the particle updating process, two groups of SKU arrangements are regularly selected to perform the crossover operation, new warehouse location combinations are generated by exchanging part of the SKU positions, and the search space is widened; meanwhile, a random variation mechanism is introduced to randomly exchange the arrangement of the SKUs on the warehouse locations, and a genetic algorithm formula is as follows: ， where S1 and S2 are the SKU arrangement of two particles, S is the current SKU arrangement, is the crossover rate, which controls the frequency of the crossover operation, and is usually taken as 0.7~0.9, is the mutation rate, which controls the frequency of the mutation operation, and is usually taken as 0.1~0.2, and t is the interval number of the genetic operation, and is usually taken as 5~20.

7. The intelligent warehousing location planning optimization method of claim 1, wherein, In step 6, the dynamic adjustment of the temperature parameter is as follows: The temperature T(t) of the tth generation is as follows: ， wherein, T0 is the initial temperature, is the cooling coefficient, taken as t is the current iteration number; when the initial temperature is high, the simulated annealing algorithm is more likely to accept a poor solution, increasing the probability of jumping out of a local optimal solution. As the iteration number increases, the temperature gradually decreases, the probability of accepting a poor solution decreases, and the global optimal solution is converged.

8. The intelligent warehousing location planning optimization method of claim 7, wherein, The poor solution accepting mechanism is as follows: The acceptance probability of a poor solution depends on the Metropolis criterion: ， where P is the probability of accepting a worse solution, is the fitness difference between the new solution and the current solution, and if the new solution is better than the current solution, then the new solution is accepted; if the new solution is worse than the current solution, then the new solution is accepted with a probability P; P is larger when the temperature is higher, and it is easier to accept a worse solution and jump out of a local optimum; P is smaller when the temperature is lower, and it is easier to converge gradually and stabilize near the global optimum.

Images