Supply chain replenishment decision optimization method, system, medium, and computer device

Abstract

The application belongs to the technical field of data processing. A supply chain replenishment decision optimization method, system, medium and computer equipment are provided. Multi-source features are associated with hyperparameters of a machine learning model. The hyperparameters are mapped to different dimensions of a hyperparameter search space, and each dimension corresponds to a value range of a hyperparameter. The multi-source data features are input into a pre-trained machine learning model to obtain a demand prediction result of the target product in a future period. A decision result of whether to replenish is generated according to current inventory data and the demand prediction result. The machine learning model adopts a fruit fly optimization algorithm for hyperparameter optimization, and fruit fly individuals search for an optimal solution in the hyperparameter search space as the optimized hyperparameters. The application improves the sensitivity of different hyperparameters to different features and realizes higher-precision decision making.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a supply chain replenishment decision optimization method, system, medium, and computer equipment. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of IoT and big data technologies, supply chain data exhibits characteristics of multi-source, heterogeneity, and dynamism, encompassing structured and unstructured data such as historical orders, inventory, market, logistics, and supplier capacity. To improve data processing capabilities, in digital supply chain management, enterprises mostly rely on machine learning models to process massive amounts of supply chain business data, achieving dynamic demand forecasting and flexible inventory optimization.

[0004] Most existing machine learning models for decision-making are trained and optimized using pre-collected data. However, even small deviations in hyperparameters (learning rate, tree depth, time window, etc.) can cause demand forecasts to deviate significantly from actual demand fluctuations, leading to erroneous decisions. Existing solutions lack effective correlation with supply chain data when optimizing hyperparameters for machine learning models. When data exhibits seasonal fluctuations or sudden anomalies, they cannot dynamically adjust the hyperparameter search strategy based on data changes, resulting in the final optimized machine learning model failing to meet predetermined decision-making accuracy requirements. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a supply chain replenishment decision optimization method, system, medium, and computer equipment. It associates multi-source features of different dimensions with hyperparameters of machine learning models, and maps the associated hyperparameters to different dimensions of the hyperparameter search space for individual search in the fruit fly optimization algorithm. This improves the sensitivity of different hyperparameters to different features and achieves higher-precision decision-making.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Firstly, the present invention provides a method for optimizing supply chain replenishment decisions.

[0008] A supply chain replenishment decision optimization method includes the following process:

[0009] The historical order data, user review data, and social media related data of the target product are preprocessed, and feature extraction is performed on the preprocessed data to obtain multi-source features;

[0010] Multi-source features are associated with hyperparameters of a machine learning model, and the hyperparameters are mapped to different dimensions of the hyperparameter search space, with each dimension corresponding to a range of values ​​for a hyperparameter.

[0011] The multi-source data features are input into a pre-trained machine learning model to obtain the demand forecast of the target product in a certain future period. Based on the current inventory data and the demand forecast, a decision on whether to replenish the stock is generated. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters. Individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0012] Secondly, the present invention provides a supply chain replenishment decision optimization system.

[0013] A supply chain replenishment decision optimization system includes:

[0014] The feature extraction unit is configured to: preprocess the acquired historical order data, user review data, and social media related data of the target product, and extract features from the preprocessed data to obtain multi-source features;

[0015] The search space construction unit is configured to: associate hyperparameters of multiple sources with machine learning models, map the hyperparameters to different dimensions of the hyperparameter search space, and each dimension corresponds to a range of values ​​for a hyperparameter.

[0016] The decision generation unit is configured to: input the multi-source data features into a pre-trained machine learning model to obtain the demand forecast result of the target product in a certain period of the future, and generate a decision result on whether to replenish the stock based on the current inventory data and the demand forecast result. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters, and individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0017] Thirdly, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium;

[0018] A processor, adapted to execute computer programs;

[0019] A computer-readable storage medium storing a computer program that, when executed by the processor, implements the supply chain replenishment decision optimization method as described in the first aspect of the present invention.

[0020] Fourthly, the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed as described in the first aspect of the present invention for optimizing supply chain replenishment decisions.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. This invention innovatively proposes a supply chain replenishment decision optimization method, which associates hyperparameters of multi-source data with machine learning models and maps these hyperparameters to different dimensions of the hyperparameter search space. This allows for a more accurate capture of the intrinsic relationship between different hyperparameters and multi-source data features, greatly improving the sensitivity of different hyperparameters to different features. For example, in supply chain scenarios involving multi-source data such as market trends, supplier delivery times, and inventory levels, some hyperparameters may be more sensitive to changes in market trend data, while others may react more strongly to fluctuations in inventory data. By mapping hyperparameters to different dimensions, the algorithm can perceive these differences more precisely, thereby adjusting parameters more effectively during model training and improving model performance.

[0023] 2. This invention uses a chaotic mapping initialization and a Gaussian-Cauchy hybrid mutation strategy to achieve global wide-area coverage in the early stage of the search by utilizing the ergodicity of the Tent chaotic mapping, thus avoiding the distribution bias of traditional random initialization. In the later stage of the iteration, the heavy-tailed characteristic of Cauchy mutation is introduced to break through local extrema. Combined with a dynamic decay step size strategy, a smooth transition from coarse-grained exploration to fine-grained development is achieved.

[0024] 3. This invention utilizes real-time triggered differential operations based on population state. Dynamic Gaussian mutation is applied to low-fitness individuals, while chaotic perturbation is used for high-fitness individuals, achieving adaptive matching of parameters and the search phase. From the perspective of individual optimization, applying dynamic Gaussian mutation to low-fitness individuals injects them with diverse exploration energy. Dynamic Gaussian mutation means that the mutation intensity is adjusted in real-time according to the algorithm's running state, allowing low-fitness individuals to explore the search space more extensively and flexibly. They are no longer limited to fixed mutation patterns but dynamically change the magnitude and direction of mutation based on the individual's fitness and the overall search progress. This gives low-fitness individuals a greater chance to escape potentially localized optima, discover more promising solution spaces, thereby improving their fitness and providing more possibilities for the evolution of the entire population.

[0025] 4. This invention integrates a multi-criteria stopping strategy, incorporating maximum iteration count, early stopping mechanism, fitness change rate threshold, and population diversity monitoring. This approach reduces computation time while ensuring solution quality. The maximum iteration count sets a basic timeframe for the algorithm, ensuring sufficient opportunities for search and optimization, preventing the algorithm from failing to find a satisfactory solution due to insufficient iterations. The early stopping mechanism stops the algorithm promptly when the solution quality improvement is minimal or nonexistent after several consecutive iterations, preventing meaningless repetitive computations and preserving already obtained good solutions. The fitness change rate threshold allows the algorithm to sensitively capture the trend of solution quality changes. When the fitness change rate falls below the set threshold, it indicates that the improvement space for the solution is very limited; stopping iteration at this point avoids unnecessary computational resource consumption without affecting solution quality. Population diversity monitoring ensures comprehensiveness of the solution from a population perspective. By monitoring the degree of difference among individuals in the population, it ensures that the algorithm covers a sufficiently broad solution space during the search process, avoiding premature convergence to local optima, thus providing strong support for obtaining high-quality solutions.

[0026] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0027] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0028] Figure 1 A flowchart illustrating a supply chain replenishment decision optimization method provided as an exemplary embodiment of the present invention;

[0029] Figure 2 A schematic diagram of a supply chain replenishment decision optimization system provided as an exemplary embodiment of the present invention;

[0030] Figure 3 A schematic diagram of a computer device provided for an exemplary embodiment of the present invention. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0033] As described in the background section, existing machine learning models cannot dynamically adjust their search strategies based on data changes during optimization, resulting in the final optimized model's prediction accuracy failing to meet predetermined requirements. Furthermore, existing optimization processes suffer from other issues. For example, grid search / random search methods are computationally inefficient in high-dimensional hyperparameter spaces, making it difficult to meet the timeliness requirements of real-time supply chain decision-making; swarm intelligence optimization algorithms (such as PSO and GA) are prone to getting trapped in local optima in complex non-convex search spaces, all of which reduce the decision accuracy of the final machine learning model. Therefore, this implementation proposes a supply chain replenishment decision optimization method, such as... Figure 1 As shown, the process includes the following:

[0034] S101: Preprocess the historical order data, user review data, and social media related data (discussion opinions related to the target product, etc., which are authorized data) of the acquired target product, and extract features from the preprocessed data to obtain multi-source features;

[0035] S102: Associate the multi-source features with the hyperparameters of the machine learning model, and map the hyperparameters to different dimensions of the hyperparameter search space, with each dimension corresponding to a range of values ​​for a hyperparameter.

[0036] S103: Input the multi-source data features into a pre-trained machine learning model (e.g., a convolutional neural network model or a long short-term memory neural network model) to obtain the demand forecast result of the target product in a certain period of the future. Generate a decision result on whether to replenish the stock based on the current inventory data and the demand forecast result. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters. Individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0037] In S101 of this implementation, specifically, it includes:

[0038] S101-1: Obtain historical order data for the target product, clean the data, and remove missing values, duplicate values, and outliers. Missing values ​​are handled using mean imputation, median imputation, or time-series interpolation; duplicate values ​​are directly deleted; outliers are identified and corrected or deleted using statistical methods (such as the 3σ principle) or the isolated forest algorithm. The cleaned data is then normalized, mapping numerical data to the [0,1] or [-1,1] interval.

[0039] S101-2: Collect user evaluation data. First, perform text cleaning to remove special characters and stop words (such as "de", "le", "zai"), use lexical analysis in natural language processing technology to segment the text into individual words, and perform word - part tagging. Then, use sentiment analysis algorithms, such as dictionary - based sentiment analysis methods or deep - learning - based sentiment analysis models (such as the LSTM - Attention model), to judge the sentiment polarity (positive, negative, neutral) of user evaluations, and quantify the sentiment analysis results into numerical values. For example, positive is 1, neutral is 0, and negative is - 1.

[0040] S101-3: Collect data related to the target product from social media platforms, use web crawler technology to capture data, and perform data deduplication and format unification. For social media text data, also perform pre - processing operations such as cleaning, word segmentation, and word - part tagging; in addition, extract features such as topic tags, like counts, comment counts, and share counts from social media associated data, and fuse these features with the text data.

[0041] S101-4: For the pre - processed historical order data, use the principal component analysis (PCA) method to extract key features, retaining the main information while reducing the data dimension; for the text data of user evaluations and social media, use word embedding techniques (such as Word2Vec, BERT) to map words to a low - dimensional vector space to obtain the distributed representation of the text, and then use a convolutional neural network (CNN) or a recurrent neural network (RNN) to extract text features from the text vectors.

[0042] In S102 of this implementation method, specifically, it includes:

[0043] Determine the hyperparameters of the machine learning model, such as the maximum depth of the decision tree model, the number of trees in the random forest model, the learning rate of the neural network model, the number of neurons in the hidden layer, etc. Establish an association between the multi - source features and these hyperparameters, and analyze the influence of different features on the hyperparameter values. For example, if the historical order data fluctuates greatly, it may be necessary to adjust the learning rate and the number of neurons in the hidden layer of the neural network model to better fit the data. Map each hyperparameter to different dimensions of the hyperparameter search space, and set the value range for each dimension. The value range can be determined based on experience, historical experimental data, or the theoretical limitations of the model.

[0044] For continuous parameters, such as learning rates in the range (0.001, 0.1), floating-point encoding and logarithmic scaling are used. For discrete parameters (such as activation function types), enumeration is employed. The parameter range setting must balance prior knowledge with dynamic expansion mechanisms. For example, the default range for XGBoost tree depth is typically set to 3-10, but if the optimal value is found to be close to the upper limit in experiments, it is expanded to 3-20 to cover potentially better solutions. For the curse of dimensionality problem that is prone to occur in high-dimensional spaces, dimensionality reduction can be achieved through principal component analysis (PCA) or sensitivity analysis to remove low-influence parameters (such as certain regularization terms contributing less than 5% to a specific dataset).

[0045] Optionally, in other implementations, statistical methods such as Pearson correlation coefficient and Spearman rank correlation coefficient can be used to calculate the correlation between multi-source features and hyperparameters. For example, analyzing the correlation between historical order data fluctuations and the learning rate of a neural network model: if the correlation coefficient is close to 1 or -1, it indicates a strong linear correlation; if the correlation coefficient is close to 0, it indicates a weak linear correlation. Besides correlation analysis, one can also explore the causal relationship between multi-source features and hyperparameters. This can be done through experimental design, fixing other factors and changing the value of a certain feature, observing the changes in hyperparameters. For example, in neural network model training, keeping other data features constant and only changing the degree of fluctuation in historical order data, then observing the changes in hyperparameters such as the learning rate and the number of neurons in the hidden layer to determine whether a causal relationship exists.

[0046] Optionally, in other implementations, a dynamic adjustment strategy for hyperparameters is formulated based on the established associations. When multi-source features change, the hyperparameters are automatically adjusted according to the strategy. For example, if a sudden increase in fluctuations in historical order data is detected, the learning rate of the neural network model can be automatically adjusted to a smaller value based on the previously established associations to avoid overfitting during training. At the same time, the number of neurons in the hidden layer can be appropriately increased to improve the model's ability to fit complex data.

[0047] Optionally, in other implementations, the differences in multi-source features among different users or in different scenarios are considered, and personalized hyperparameters are set for each user or scenario. For example, for users with high purchase frequency and large order amounts, their historical order data is relatively stable, so a larger learning rate and fewer hidden layer neurons can be set for these users; while for users with low purchase frequency and large fluctuations in order amounts, a smaller learning rate and more hidden layer neurons are set to improve the degree of personalization.

[0048] In S103 of this implementation, the machine learning model uses the fruit fly optimization algorithm for hyperparameter optimization, specifically including:

[0049] Tent chaotic mapping is used to generate initial solutions to address the uneven distribution caused by random initialization. Leveraging the ergodicity and initial value sensitivity of chaotic systems, the population more evenly covers potentially high-value regions within the search space. Each fruit fly individual corresponds to a set of hyperparameter combinations (such as learning rate, number of hidden nodes, etc.), and its position vector is generated through a Tent chaotic sequence, expressed by the formula:

[0050] (1);

[0051] in, Represents the modulo operation. Representative generation Random numbers between The dynamic adjustment parameter represents the current iteration number. Equation (1) preserves the chaotic characteristics through modular arithmetic and introduces a random perturbation term to enhance the exploration capability. After the initial population is generated, the chaotic variables need to be linearly mapped to the actual domain of the hyperparameters.

[0052] To further balance global search and local exploration, this invention employs a linear decay strategy to dynamically adjust the step size. A larger dynamic adjustment parameter is set in the initial iteration phase, allowing individual fruit flies to explore extensively within the hyperparameter space and enhancing coverage of unknown regions. As the number of iterations increases, the dynamic adjustment parameter is gradually reduced (e.g., ...). ,in The maximum number of iterations, (This refers to the current iteration number), which allows the search to focus on high-potential areas, improving the fineness of parameter adjustment, fitness evaluation, and search mechanism.

[0053] The design of the fitness function is a crucial step connecting the algorithm's search mechanism with the optimization objective of the actual problem. Its core task is to quantify the performance metrics of the hyperparameter combination (such as model accuracy and loss value) into comparable odor concentration values, thereby guiding the fruit fly population towards a better solution. In single-objective optimization scenarios, reciprocal mapping is typically used. D i =1 / loss converts the loss value into a positional value based on its distance from the origin. This method enhances the discriminative power of low-loss regions through non-linear scaling. Substituting the odor concentration value into the odor concentration determination function yields the odor concentration value for each fruit fly individual.

[0054] (2);

[0055] in, represent and The mapping function between them.

[0056] For example, in neural network tuning, when the cross-entropy loss value decreases from 0.5 to 0.2, It will jump from 2 to 5, significantly amplifying the competitive advantage of outstanding individuals.

[0057] For multi-objective optimization problems (such as simultaneously optimizing model accuracy and inference speed), a weighted fusion method needs to be introduced:

[0058] (3);

[0059] Among them, the weighting coefficient 、 It needs to be dynamically adjusted according to task priority. Represents reasoning delay, This represents accuracy.

[0060] By combining global exploration in the olfactory phase with local exploitation in the visual phase, the algorithm's performance in complex optimization problems is improved. In the olfactory phase, individual fruit flies employ a Gaussian perturbation strategy for wide-area exploration; the position update formula for each individual fruit fly is:

[0061] X i t+1 =X i t +N(0,σ 2 )•R step (t)(4;

[0062] Among them, the Gaussian noise term N(0,σ) 2 Provide random perturbation, dynamic search step size R step (t)=R max -( R max -R min •t / T varies with the number of iterations Linear decay, with a larger initial step size to cover unknown areas, and a smaller step size in the later stages to avoid oscillations, X i t Indicates individual fruit flies i exist t The position at time +1, X i t+1 Indicates individual fruit flies i exist t Location at any given moment N (0, σ 2 () represents the Gaussian noise term, which is provided with a random perturbation. R max This represents the maximum search step size, specifying the maximum distance a fruit fly individual may move during the search process. R min This represents the minimum search step size, which limits the minimum distance a fruit fly can move.t Represents the current iteration number. T This represents the total number of iterations.

[0063] Once the visual phase begins, the algorithm focuses on the current best individual X. best In the given area, Cauchy mutation-enhanced local search is employed, and the location update formula is as follows:

[0064] X i t+1 = X best +Cauchy(0,1)• η ( t (5);

[0065] in, η ( t ) represents the intensity of variation. X best This represents the optimal position found in the fruit fly population up to the current iteration. Cauchy(0,1) This represents the Cauchy distribution.

[0066] The heavy-tailed nature of the Cauchy distribution allows for a wider range of local jumps; for example, the learning rate might abruptly change from 0.05 to 0.08 in a single iteration, overcoming the limitations of traditional Gaussian mutation. (Mutation intensity) η ( t )= η 0•e -t / T As the iteration exponent decays, larger offsets are allowed initially for exploration. X best Neighborhood, fine-tuning later. η 0 represents the initial value of the variation intensity.

[0067] The synergy of the two stages mentioned above in this invention is achieved through an adaptive switching mechanism: when the population fitness variance is lower than a set threshold θ (e.g., 0.1), the visual stage is triggered; if the optimal solution has not improved for three consecutive generations, the olfactory stage is restarted.

[0068] Population state determination and manipulation are the core mechanisms of the Drosophila optimization algorithm for dynamically balancing global exploration and local exploitation. By monitoring the deviation between individual fitness and the population average fitness in real time, it triggers differentiated mutation strategies to optimize search efficiency. After each iteration, the algorithm first calculates the odor concentration value of each individual fruit fly. smell i and population average concentration smell avg When the individual concentration is below the average (i.e. smell i < smell avgWhen the condition is met, it is identified as "clustering," indicating that the individual may be trapped in a local optimum, and a Gaussian mutation operation needs to be applied.

[0069] X i ′ = X i + N (0,(( smell max - smell min ) / (smell avg - smell i ))•0.5 e -t / T (6);

[0070] in, X i ′ It is the location of the mutated consequence fly individual. X i This represents the position of the fruit fly individual before the mutation. N It represents a random number that follows a Gaussian distribution.

[0071] The mutation strategy dynamically adjusts the variance of Gaussian noise to adaptively match the perturbation intensity of low-fit individuals with the current iteration stage and population state. The noise intensity is higher in the initial iterations to enhance the ability to escape local optima, and gradually decreases in later iterations to improve convergence accuracy. If the fitness of the mutated individuals... smell i ′ If the solution is better than the original, replace the original solution; otherwise, retain the original solution to avoid unnecessary perturbations.

[0072] For individual fruit flies with concentrations higher than the average ( smell i < smell avg If the condition is ), it is determined to be a "divergent phenomenon". A perturbation sequence is generated using Logistic chaotic mapping. The ergodicity of the chaotic variables is used to perform a refined search of the high-fitness region. The perturbation formula is:

[0073] X i ′ = X i +0.1(1-t / T) 2 •(c k -0.5)(7;

[0074] in, X i ′This represents the location of the individual fruit flies after the perturbation treatment. X i c represents the position of the individual fruit flies after perturbation. k Represents the chaos intensity coefficient, chaos intensity coefficient c k The perturbation amplitude decreases nonlinearly with the number of iterations. In the early stage, the perturbation amplitude is large to cover potential solutions in the neighborhood, and in the later stage, it gradually shrinks to focus on fine-tuning the optimal solution.

[0075] The fruit fly optimization algorithm of this invention achieves efficient convergence and resource management through a multi-criteria joint stopping strategy. Its core design comprises three parts: basic conditions, dynamic judgment, and intelligent monitoring. The basic termination condition sets the maximum number of iterations T as a hard termination line, and introduces an early stopping mechanism: if the optimal fitness is achieved for K consecutive generations (e.g., 20 generations)... Smell best If no improvement is achieved, the process terminates early to avoid invalid computations. Dynamic convergence is determined by monitoring the rate of change in fitness. δ With population fitness variance σ 2 To achieve intelligent stopping, when δ <10 -5 (i.e., the first set threshold) or σ 2 The algorithm is considered to have converged when the value is less than 0.01 (i.e., the second set threshold). This invention dynamically adjusts the number of iterations by integrating time (e.g., 24 hours) and computational budget (e.g., 105 evaluations): if convergence is rapid in the early stages, the number of iterations is extended proportionally, and vice versa to save resources.

[0076] In S103 of this implementation, the following steps are also included: obtaining current inventory data, comparing the demand forecast result with the current inventory, setting a replenishment threshold, and generating a replenishment decision when the demand forecast result is greater than the sum of the current inventory and the replenishment threshold; otherwise, no replenishment decision is generated. The replenishment threshold can be set according to factors such as the company's inventory strategy and product sales characteristics.

[0077] Figure 2 A supply chain replenishment decision optimization system is shown, including:

[0078] The feature extraction unit 201 is configured to: preprocess the acquired historical order data, user review data, and social media related data of the target product, and extract features from the preprocessed data to obtain multi-source features;

[0079] The search space construction unit 202 is configured to: associate hyperparameters of multiple sources with machine learning models, map the hyperparameters to different dimensions of the hyperparameter search space, and each dimension corresponds to a range of values ​​for a hyperparameter.

[0080] The decision generation unit 203 is configured to: input the multi-source data features into a pre-trained machine learning model to obtain the demand forecast result of the target product in a certain period of the future, and generate a decision result on whether to replenish the stock based on the current inventory data and the demand forecast result. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters, and individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0081] It is understood that the aforementioned units can be individually or entirely merged into one or more other units, or some of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effects of the embodiments of this application. The aforementioned units are based on logical functional division. In practical applications, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of this application, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0082] According to another embodiment of this application, the system described in this embodiment can be constructed by running a computer program (including program code) capable of performing the steps involved in the corresponding method of the present invention on a general-purpose computing device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, loaded into the aforementioned computing device through the computer-readable recording medium, and run therein.

[0083] Figure 3 A computer device is shown, which includes a processor 301, a communication interface 302, and a computer-readable storage medium 303. The processor 301, communication interface 302, and computer-readable storage medium 303 can be connected via a bus or other means.

[0084] The communication interface 302 is used to receive and send data. The computer-readable storage medium 303 can be stored in the memory of the electronic device. The computer-readable storage medium 303 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the computer-readable storage medium 303.

[0085] The processor 301 is the computing and control core of the electronic device. It is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions to achieve the corresponding method flow or corresponding function.

[0086] The processor 301 is configured to perform the following process:

[0087] The historical order data, user review data, and social media related data of the target product are preprocessed, and feature extraction is performed on the preprocessed data to obtain multi-source features;

[0088] Multi-source features are associated with hyperparameters of a machine learning model, and the hyperparameters are mapped to different dimensions of the hyperparameter search space, with each dimension corresponding to a range of values ​​for a hyperparameter.

[0089] The multi-source data features are input into a pre-trained machine learning model to obtain the demand forecast of the target product in a certain future period. Based on the current inventory data and the demand forecast, a decision on whether to replenish the stock is generated. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters. Individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0090] The historical order data, user review data, and social media related data of the target product are preprocessed, and feature extraction is performed on the preprocessed data to obtain multi-source features;

[0091] Multi-source features are associated with hyperparameters of a machine learning model, and the hyperparameters are mapped to different dimensions of the hyperparameter search space, with each dimension corresponding to a range of values ​​for a hyperparameter.

[0092] The multi-source data features are input into a pre-trained machine learning model to obtain the demand forecast of the target product in a certain future period. Based on the current inventory data and the demand forecast, a decision on whether to replenish the stock is generated. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters. Individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0093] This invention also provides a computer-readable storage medium, which is a memory device in an electronic device for storing programs and data. It is understood that the computer-readable storage medium here may include both built-in storage media in the electronic device and extended storage media supported by the electronic device. The computer-readable storage medium provides storage space for storing the processing system of the electronic device.

[0094] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM memory or unstable memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0095] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to perform the following process:

[0096] The historical order data, user review data, and social media related data of the target product are preprocessed, and feature extraction is performed on the preprocessed data to obtain multi-source features;

[0097] Multi-source features are associated with hyperparameters of a machine learning model, and the hyperparameters are mapped to different dimensions of the hyperparameter search space, with each dimension corresponding to a range of values ​​for a hyperparameter.

[0098] The multi-source data features are input into a pre-trained machine learning model to obtain the demand forecast of the target product in a certain future period. Based on the current inventory data and the demand forecast, a decision on whether to replenish the stock is generated. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters. Individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0099] The present invention also provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the following process:

[0100] The historical order data, user review data, and social media related data of the target product are preprocessed, and feature extraction is performed on the preprocessed data to obtain multi-source features;

[0101] Multi-source features are associated with hyperparameters of a machine learning model, and the hyperparameters are mapped to different dimensions of the hyperparameter search space, with each dimension corresponding to a range of values ​​for a hyperparameter.

[0102] The multi-source data features are input into a pre-trained machine learning model to obtain the demand forecast of the target product in a certain future period. Based on the current inventory data and the demand forecast, a decision on whether to replenish the stock is generated. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters. Individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters.

[0103] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0104] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, digital cable) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0105] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A supply chain replenishment decision optimization method, characterized in that, Includes the following processes: The historical order data, user review data, and social media related data of the target product are preprocessed, and feature extraction is performed on the preprocessed data to obtain multi-source features; Associating multi-source features with hyperparameters of a machine learning model, mapping the hyperparameters to different dimensions of the hyperparameter search space, with each dimension corresponding to a range of values ​​for a hyperparameter; wherein associating order time series features and product category distribution features with hyperparameters of the machine learning model includes: quantifying the correlation between different features and hyperparameters through mutual information statistics, and optimizing the construction of the hyperparameter search space based on the correlation; The multi-source data features are input into a pre-trained machine learning model to obtain the demand forecast result of the target product in a certain period of the future. Based on the current inventory data and the demand forecast result, a decision result on whether to replenish the stock is generated. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters. Individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters. In the initial search phase, Tent chaotic mapping is used to generate initial solutions. Each fruit fly individual corresponds to a set of hyperparameter combinations, and the position vector of the fruit fly individual is generated through Tent chaotic sequence. During the olfactory stage, fruit flies use a Gaussian perturbation strategy to explore a wide area; upon entering the visual stage, they focus on the area where the current best individual is located and use Cauchy mutation to enhance the local search. When the population fitness variance falls below a set threshold, the visual phase is triggered; if the optimal solution does not improve for three consecutive generations, the olfactory phase is restarted. After each iteration, if the odor concentration of a fruit fly individual is lower than the population average, it is considered an aggregation phenomenon, and a Gaussian mutation operation is applied to this fruit fly individual; if the odor concentration of a fruit fly individual is greater than or equal to the population average, it is considered a divergence phenomenon, and a perturbation is applied to this fruit fly individual.

2. The supply chain replenishment decision optimization method as described in claim 1, characterized in that, In the fruit fly optimization algorithm, the Tent chaotic map is used to generate the initial solution. Each fruit fly corresponds to a set of hyperparameter combinations. The location of each fruit fly individual is: ,in, Represents the modulo operation. Representative generation Random numbers between This represents a dynamically adjusted parameter related to the current iteration number.

3. The supply chain replenishment decision optimization method as described in claim 1, characterized in that, In the olfactory optimization phase of fruit flies, the position update formula for each individual fruit fly is: X i t+1 =X i t +N(0,σ 2 )·R step (t), where N(0,σ) 2 ) is a sequence of elements with a mean of 0 and a variance of σ. 2 Gaussian distribution, R step (t) represents the dynamic search step size, X i t X represents the position of the i-th fruit fly individual at the t-th iteration. i t+1 This indicates the position of the i-th fruit fly individual at the (t+1)-th iteration; Once the visual phase begins, the position update formula for each individual fruit fly is: X i t+1 =X best +Cauchy(0,1)·η(t), where η(t) is the variation intensity, X best This represents the optimal position found in the fruit fly population up to the current iteration, where Cauchy(0,1) represents the Cauchy distribution. The odor concentration value is used as the fitness of individual fruit flies. When the population fitness variance of the fruit fly population is lower than a set threshold, the visual stage is triggered; if the optimal solution is not improved after three consecutive generations, the olfactory stage is restarted.

4. The supply chain replenishment decision optimization method as described in claim 3, characterized in that, After each iteration, when the odor concentration value of a certain fruit fly individual is... i Below the population average concentration of smell avg When this is determined to be an aggregation phenomenon, a Gaussian mutation operation is applied to this individual fruit fly: X i ′ =X i +N (0,((smell max -smell min ) / (smell avg -smell i ))·0.5e -t / T ), where X i ′ It refers to the location of the mutated consequence fly individual, smell. max and smell min X represents the maximum and minimum odor concentrations in the fruit fly population, respectively. t is the current iteration number, T is the total number of iterations, and N represents a random number following a Gaussian distribution. i Represents the position of the original fruit fly individual; When the odor concentration value of a certain fruit fly individual is... i Greater than or equal to the population average concentration (smell) avg If the phenomenon is observed, it is determined to be a divergent phenomenon, and a disturbance is applied to this individual fruit fly: X i ′ =X i +0.1(1-t / T) 2 ·(c k -0.5), where X i ′ c represents the position of the individual fruit fly after the perturbation. k This represents the chaos intensity coefficient.

5. The supply chain replenishment decision optimization method as described in any one of claims 1-4, characterized in that, The termination conditions of the fruit fly optimization algorithm include: The fruit fly optimization algorithm terminates when the maximum number of iterations T is reached, if the optimal fitness Smell value is maintained for K consecutive generations. best Without improvement, the fruit fly optimization algorithm terminates. The population fitness change rate and population fitness variance are calculated. The fruit fly optimization algorithm terminates when the population fitness change rate is less than a first set threshold or the population fitness variance is less than a second set threshold.

6. The supply chain replenishment decision optimization method as described in any one of claims 1-4, characterized in that, A decision on whether to replenish stock is generated based on current inventory data and demand forecasts, including: Obtain current inventory data, compare the demand forecast with the current inventory, set a replenishment threshold, and generate a replenishment decision when the demand forecast is greater than the sum of the current inventory and the replenishment threshold; otherwise, do not generate a replenishment decision.

7. A supply chain replenishment decision optimization system, characterized in that, include: The feature extraction unit is configured to: preprocess the acquired historical order data, user review data, and social media related data of the target product, and extract features from the preprocessed data to obtain multi-source features; The search space construction unit is configured to: associate hyperparameters of multiple sources with machine learning models, map the hyperparameters to different dimensions of the hyperparameter search space, and each dimension corresponds to a range of values ​​for a hyperparameter. Associating order time series features and product category distribution features with hyperparameters of machine learning models includes: quantifying the correlation between different features and hyperparameters through mutual information statistics, and optimizing the construction of the hyperparameter search space based on the correlation. The decision generation unit is configured to: input the multi-source data features into a pre-trained machine learning model to obtain the demand forecast result of the target product in a certain period of the future, and generate a decision result on whether to replenish the stock based on the current inventory data and the demand forecast result. The machine learning model uses the fruit fly optimization algorithm to optimize hyperparameters, and individual fruit flies search for the optimal solution in the hyperparameter search space as the optimized hyperparameters. In the initial search phase, Tent chaotic mapping is used to generate initial solutions. Each fruit fly individual corresponds to a set of hyperparameter combinations, and the position vector of the fruit fly individual is generated through Tent chaotic sequence. During the olfactory stage, fruit flies use a Gaussian perturbation strategy to explore a wide area; upon entering the visual stage, they focus on the area where the current best individual is located and use Cauchy mutation to enhance the local search. When the population fitness variance falls below a set threshold, the visual phase is triggered; if the optimal solution does not improve for three consecutive generations, the olfactory phase is restarted. After each iteration, if the odor concentration of a fruit fly individual is lower than the population average, it is considered an aggregation phenomenon, and a Gaussian mutation operation is applied to this fruit fly individual; if the odor concentration of a fruit fly individual is greater than or equal to the population average, it is considered a divergence phenomenon, and a perturbation is applied to this fruit fly individual.

8. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program that, when executed by the processor, implements the supply chain replenishment decision optimization method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1 to 6.

Images