Supply chain demand prediction system and method based on artificial intelligence

Abstract

The invention provides a supply chain demand prediction system and method based on artificial intelligence, and belongs to the technical field of data prediction. The supply chain demand prediction system based on artificial intelligence comprises a demand prediction module used for obtaining first demand data of a target supply chain by using a demand prediction model based on influence factor data corresponding to the target supply chain; the data analysis module is used for judging whether to output a parameter optimization demand of the demand prediction model or not according to the second demand data and the first demand data of the target supply chain; and the parameter optimization module is used for responding to the parameter optimization demand, based on the second demand data and the influence factor data corresponding to the second demand data, the optimization model is used for obtaining an optimization parameter corresponding to the demand prediction model, and the optimization parameter is used for indicating the first equipment to retrain the demand prediction model based on the optimization parameter. According to the invention, the accuracy and reliability of supply chain demand prediction can be improved by dynamically adjusting the demand prediction model.

Description

Technical Field

[0001] The present disclosure relates to the technical field of data prediction, and in particular, to a supply chain demand prediction system and method based on artificial intelligence. Background Art

[0002] In the context of today's global economic integration, supply chain management has become a key factor for enterprises to enhance their competitiveness. As the core link of supply chain management, the accuracy of supply chain demand prediction is directly related to many aspects such as an enterprise's production plan, inventory management, logistics distribution, and customer service level.

[0003] Traditional supply chain demand prediction methods mainly rely on the statistical analysis of historical data based on prediction algorithms. However, for supply chain demand data, it is greatly affected by various factors, such as seasons, price changes, industry development, etc. Therefore, the actual prediction effect of existing supply chain demand prediction models has a large gap with the training effect during model training, and the prediction accuracy is poor.

[0004] Therefore, an accurate and reliable supply chain demand prediction system is needed. Summary of the Invention

[0005] Embodiments of the present disclosure provide a supply chain demand prediction system and method based on artificial intelligence to solve the problem that the actual prediction of existing demand prediction models is not accurate enough due to changes in influencing factors.

[0006] A demand prediction module, configured to obtain first demand data of a target supply chain by using a demand prediction model based on influencing factor data corresponding to the target supply chain; A data analysis module, configured to determine whether to output a parameter optimization requirement for the demand prediction model according to second demand data and the first demand data of the target supply chain; the second demand data is the actual demand data of the target supply chain; A parameter optimization module, configured to, in response to the parameter optimization requirement, obtain optimized parameters corresponding to the demand prediction model by using an optimization model based on the second demand data and the influencing factor data corresponding to the second demand data, and send the optimized parameters to a first device; A first device, configured to retrain the demand prediction model based on the optimized parameters, and instruct the demand prediction module to perform supply chain demand prediction based on the demand prediction model after retraining.

[0007] In an exemplary embodiment of the present disclosure, the data analysis module is specifically configured to output a parameter optimization requirement for the demand prediction model in response to a deviation between the first demand data and the second demand data being greater than a preset deviation threshold.

[0008] In an exemplary embodiment of the present disclosure, the optimization model is an ant colony model; before obtaining the optimization parameters of the demand prediction model using the optimization model based on the second demand data and the influencing factor data corresponding to the second demand data, the parameter optimization module is further configured to: Determine the solution space of the optimization model based on the training parameters and structural parameters corresponding to the demand prediction model, and determine the hyperparameters of the optimization model based on the second demand data and the influencing factor data corresponding to the second demand data.

[0009] In an exemplary embodiment of the present disclosure, the hyperparameters include a pheromone heuristic factor and an expected heuristic factor; The parameter optimization module is specifically configured to: In response to the existence of data in the target score with a score greater than the first score, reduce the reference value of the pheromone heuristic factor based on the first step size to obtain the pheromone heuristic factor, and increase the reference value of the expected heuristic factor based on the second step size to obtain the expected heuristic factor; The target score is obtained by using the random forest algorithm to score the influencing factor data corresponding to the second demand data based on the second demand data.

[0010] In an exemplary embodiment of the present disclosure, the hyperparameters include a pheromone heuristic factor and an expected heuristic factor; The parameter optimization module is specifically configured to: In response to the data scores in the target score being all less than the second score, reduce the reference value of the expected heuristic factor based on the third step size to obtain the expected heuristic factor, and increase the reference value of the pheromone heuristic factor based on the fourth step size to obtain the pheromone heuristic factor; The target score is obtained by using the random forest algorithm to score the influencing factor data corresponding to the second demand data based on the second demand data.

[0011] In an exemplary embodiment of the present disclosure, the parameter optimization module is further configured to: Determine the step sizes for adjusting the pheromone heuristic factor and the expected heuristic factor according to the dispersion degree of the target score.

[0012] In an exemplary embodiment of the present disclosure, the data analysis module is further configured to: Use the second demand data as the dependent variable and the first influencing factor data as the independent variable to obtain the data scores of each first influencing factor data using the random forest model; the first influencing factor data is the data that affects the supply demand of the target supply chain; Screen the first influencing factor data based on the data scores of each first influencing factor data to obtain the influencing factor data corresponding to the target supply chain.

[0013] In an exemplary embodiment of the present disclosure, the data analysis module is further configured to: Screen the first influencing factor data according to a predetermined analysis period to obtain the influencing factor data corresponding to the target supply chain; the first influencing factor data is the data that affects the supply demand of the target supply chain. Determine the data characteristics of the influencing factor data corresponding to the target supply chain; the data characteristics are used to instruct the demand forecasting module to select a corresponding demand forecasting model.

[0014] In an exemplary embodiment of the present disclosure, the demand forecasting models include a long short-term memory network model and a multi-layer perceptron model. The demand forecasting module is specifically configured to: In response to the proportion of non-linear characteristics in the influencing factor data corresponding to the target supply chain being greater than a preset proportion, use the long short-term memory network model based on the influencing factor data corresponding to the target supply chain to obtain the first demand data of the target supply chain. In response to the proportion of non-linear characteristics in the influencing factor data corresponding to the target supply chain being less than or equal to the preset proportion, use the long short-term memory network model based on the influencing factor data corresponding to the target supply chain to obtain the first forecast data; and use the multi-layer perceptron model based on the influencing factor data corresponding to the target supply chain to obtain the second forecast data; perform weighted calculation on the first forecast data and the second forecast data to obtain the first demand data of the target supply chain.

[0015] In a second aspect, an embodiment of the present disclosure provides an artificial intelligence-based supply chain demand forecasting method, including: using a demand forecasting model based on the influencing factor data corresponding to the target supply chain to obtain the first demand data of the target supply chain. Determine whether to output the parameter optimization requirement of the demand forecasting model according to the second demand data and the first demand data of the target supply chain; the second demand data is the actual demand data of the target supply chain. In response to the parameter optimization requirement, use the optimization model based on the second demand data and the influencing factor data corresponding to the second demand data to obtain the optimized parameters corresponding to the demand forecasting model. Retrain the demand forecasting model based on the optimized parameters to perform supply chain demand forecasting based on the retrained demand forecasting model.

[0016] The beneficial effects of an artificial intelligence-based supply chain demand forecasting system and method provided by an embodiment of the present disclosure are as follows: The present disclosure proposes a strategy for dynamically adjusting the demand forecasting model, which can update the demand forecasting model according to the deviation between the demand data predicted by the demand forecasting model and the actual demand data, so as to ensure the self-adaptability and robustness of the demand forecasting model and improve the accuracy and precision of supply chain demand forecasting. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following will briefly introduce the accompanying drawings required for describing the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present disclosure. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a schematic structural diagram of a supply chain demand forecasting system based on artificial intelligence provided by an embodiment of the present disclosure; Figure 2 It is a schematic flowchart of a supply chain demand forecasting method based on artificial intelligence provided by an embodiment of the present disclosure. Specific embodiments

[0019] To enable those skilled in the art to better understand this solution, the following will clearly describe the technical solutions in the embodiments of this solution in conjunction with the accompanying drawings in the embodiments of this solution. Obviously, the described embodiments are part of the embodiments of this solution, rather than all of the embodiments. Based on the embodiments in this solution, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this solution.

[0020] The terms "including" and any other variations in the description and claims of this solution and the above accompanying drawings mean "including but not limited to", intending to cover non-exclusive inclusion and not limited to the examples listed in the text. In addition, terms such as "first" and "second" are used to distinguish different objects, rather than to describe a specific order.

[0021] The following will describe the implementation of the present disclosure in detail in conjunction with specific accompanying drawings: Figure 1 It is a schematic structural diagram of a supply chain demand forecasting system based on artificial intelligence provided by an embodiment of the present disclosure. Refer to Figure 1 The supply chain demand forecasting system based on artificial intelligence includes: A demand forecasting module 10, configured to obtain first demand data of the target supply chain by using a demand forecasting model based on influence factor data corresponding to the target supply chain; A data analysis module 11, configured to determine whether to output a parameter optimization requirement for the demand forecasting model according to second demand data and the first demand data of the target supply chain; the second demand data is the actual demand data of the target supply chain; A parameter optimization module 12, configured to, in response to the parameter optimization requirement, obtain optimized parameters corresponding to the demand forecasting model by using an optimization model based on the second demand data and influence factor data corresponding to the second demand data, and send the optimized parameters to the first device 13; The first device 13 is configured to retrain the demand prediction model based on the optimized parameters and instruct the demand prediction module 10 to perform supply chain demand prediction based on the retrained demand prediction model.

[0022] In this embodiment, the target supply chain is a specific supply chain targeted by the present disclosure. For example, it can be a cold chain transportation supply chain, a raw material procurement supply chain, a product sales supply chain, or the overall supply chain of a certain industry. The influencing factor data is data related to the target supply chain demand. For example, it can be historical sales data, commodity price data, etc. The influencing factor data can be determined according to expert experience or prior knowledge, or can be screened by a classification model.

[0023] The demand prediction model is a mathematical model constructed based on artificial intelligence algorithms, and its basic model can be a neural network model, a linear regression model, or a decision tree model, etc.; the demand prediction model is trained with the demand data corresponding to the historical target supply chain and the influencing factor data corresponding to the historical target supply chain.

[0024] The first demand data is the predicted data of the target supply chain demand. For example, it can be the predicted product sales quantity, the predicted quantity of products to be stocked, the predicted quantity of parts to be purchased, or the predicted product production efficiency, etc.

[0025] The second demand data is the actual demand data of the target supply chain. For example, it can be the actual product sales quantity, the actual product cost data, the actual data of purchased parts, or the actual product production efficiency, etc.

[0026] The parameter optimization requirement can essentially be a piece of information or an instruction signal, and its triggering condition can be the deviation between the second demand data and the first demand of the target supply chain. Specifically, the data analysis module 11 is specifically configured to output the parameter optimization requirement of the demand prediction model in response to the deviation between the first demand data and the second demand data being greater than a preset deviation threshold.

[0027] Wherein, the deviation between the first demand data and the second demand data can be calculated by calculating the mean square error, mean absolute error, or mean absolute percentage error between the two data to obtain the deviation between the first demand data and the second demand data. The t-test can also be used to test the deviation between the first demand data and the second demand data, or more simply, the ratio of the difference to the true value can be calculated as the deviation. It should be noted that the preset deviation thresholds corresponding to different deviation calculation methods should be different.

[0028] When the data analysis module 11 calculates that the deviation between the first demand data and the second demand data is large, that is, greater than the corresponding preset deviation threshold, the current demand prediction model parameters are unreasonable, and thus a parameter optimization requirement for the demand prediction model is output.

[0029] The parameter optimization module 12 is a functional module for optimizing the parameters of the demand prediction model in the present disclosure. When the parameter optimization module 12 receives the parameter optimization requirement, it can input the actual demand data of the target supply chain and its corresponding influencing factor data into the optimization model to obtain the optimized parameters, and send the optimized parameters to the first device 13. Then, the first device 13 retrains the demand prediction model. After the retraining is completed, the first device 13 instructs the demand prediction module 10 to perform prediction using the retrained demand prediction model based on the influencing factor data corresponding to the target supply chain. The first device 13 can essentially be a computer or a training system.

[0030] Among them, the optimization model is a model for optimizing the parameters of the demand prediction model, such as an ant colony algorithm model, a genetic algorithm model, a particle swarm algorithm model, etc. The optimized parameters can essentially be the parameters of the demand prediction model. For example, if the demand prediction model is a Long - Short Term Memory Network (LSTM) model, the optimized parameters can be the learning rate, the number of neurons in the hidden layer, the number of layers, or the input sequence length of the LSTM model. The first device 13 can be a server, a computer, etc., which is used to receive the optimized parameters and retrain the demand prediction model based on these optimized parameters.

[0031] For example, an electronics manufacturing enterprise wants to predict the market demand for a certain model of smartphone in the next month using the above - mentioned artificial - intelligence - based supply chain demand prediction system.

[0032] The demand prediction module 10: receives the influencing factor data related to the supply chain of this model of smartphone, including the historical sales data of the past year, the selling prices of the same - type mobile phones of competitors in the current market, the recent market research data, etc. The demand prediction model uses the LSTM model. The demand prediction model can input the collected influencing factor data into the LSTM model. After the calculation and analysis of the model, the demand prediction data for this model of smartphone in the next month, that is, the first demand data, is obtained. Suppose the prediction result is a demand of 10,000 units next month.

[0033] Data analysis module 11: After one month, obtain the actual sales data (the second demand data) of the smart phone last month. Assume that 12,000 units were actually sold last month. Compare the first demand data (i.e., 10,000 units) with the second demand data (i.e., 12,000 units), and calculate the deviation between the two. If the deviation exceeds a pre-set threshold (e.g., 15%), it is determined that the parameters of the demand prediction model need to be optimized, and the parameter optimization requirement is output.

[0034] Parameter optimization module 12: When receiving the parameter optimization requirement, input the influencing factor data corresponding to the actual sales data last month (the second demand data) (such as the market situation last month, etc.) into the optimization model (assumed to be an ant colony algorithm model). The ant colony algorithm model searches for the optimal parameter combination in the solution space to obtain the optimized parameters for the LSTM model. Send these optimized parameters to the server (the first device 13) that trains the LSTM model. The server retrains the LSTM model according to the received optimized parameters, expecting to obtain more accurate results in the next prediction.

[0035] It should be noted that the action of the foregoing data analysis module 11 performing the analysis after one month is only an embodiment and not a limitation on time. Specifically, if the data predicted by the demand prediction model is 12 hours, the time for the data analysis module 11 to perform the action should be 12 hours later. It can be understood that the time for the data analysis module 11 to perform the action of outputting the parameter optimization requirement of the demand prediction model is at least after the prediction time of the demand prediction model.

[0036] In an alternative embodiment, the system can perform the retraining operation of the demand prediction model based on the optimized parameters according to a preset cycle. Before performing the next retraining operation, the demand prediction module 10 performs the supply chain demand prediction based on the demand prediction model after this retraining.

[0037] From the above, it can be concluded that the present disclosure proposes a strategy for dynamically adjusting the demand prediction model, which can update the demand prediction model according to the deviation between the demand data predicted by the demand prediction model and the actual demand data, thereby ensuring the self-adaptability and robustness of the demand prediction model and improving the accuracy and precision of the supply chain demand prediction.

[0038] In an embodiment of the present disclosure, the optimization model is an ant colony model; before obtaining the optimized parameters corresponding to the demand prediction model by using the optimization model based on the second demand data and the influencing factor data corresponding to the second demand data, the parameter optimization module 12 is further configured to: Determine the solution space of the optimization model based on the training parameters and structural parameters corresponding to the demand prediction model, and determine the hyperparameters of the optimization model based on the second demand data and the influencing factor data corresponding to the second demand data.

[0039] In this embodiment, the training parameters corresponding to the demand prediction model may be the learning rate, the number of iterations, etc., and the structural parameters may be the number of layers of the neural network model, the number of neurons in each layer, etc. The solution space of the optimization model contains possible parameter combinations, and the ant colony model will search for the optimal parameter combination within this solution space to optimize the demand prediction model.

[0040] The solution space includes dimensions and ranges. The dimension of the solution space is consistent with the sum of the number of training parameters and structural parameters of the demand prediction model. For example, if the optimization parameters to be solved are set as the learning rate, the number of neurons in the hidden layer, and the number of layers of the demand prediction model, then the dimension of the solution space is 3. The range of the solution space can be set according to the range of parameter settings of the demand prediction model when solving similar problems, or according to the default range of the demand prediction model. For example, the range of the learning rate can be set to 0.001 - 0.1. It should be noted that the range of the learning rate is only limited to the limitation and value in the learning rate dimension.

[0041] The hyperparameters of the optimization model include the pheromone evaporation coefficient.

[0042] The pheromone evaporation coefficient can be determined according to the stability value of the second demand data. For example, in response to the fluctuation value of the second demand data being less than the first fluctuation value, the reference value of the pheromone evaporation coefficient is reduced based on the first evaporation step size; In response to the fluctuation value of the second demand data being less than the second fluctuation value, the reference value of the pheromone evaporation coefficient is increased based on the second evaporation step size.

[0043] The fluctuation value can be determined by calculating the variance or standard deviation of the second demand data. The fluctuation value can be equal to the variance or standard deviation of the second demand data. The larger the variance, the greater the fluctuation of the data and the more dispersed the data; the smaller the variance, the more stable the data. If the volatility of the second demand data is large, that is, greater than the second fluctuation value, the value of the pheromone evaporation coefficient can be increased to make the pheromone evaporate faster, so that the algorithm can adapt to the change of data faster and avoid ants relying too much on past search experience and falling into local optimality. On the contrary, if the volatility of the second demand data is small, that is, less than the first fluctuation value, the pheromone evaporation coefficient can be set smaller to allow more pheromone to accumulate, so that ants can better use historical information for searching. The first fluctuation value and the second fluctuation value can be set based on experience, and the first evaporation step size and the second evaporation step size can be obtained from the data during the experiment. The reference value of the pheromone evaporation coefficient is the default value of the ant colony algorithm model.

[0044] As can be seen from the above, the present disclosure determines the solution space of the ant colony model based on the training parameters and structural parameters of the demand prediction model, ensuring that the search scope covers all possible combinations of key parameters, thereby finding a better parameter configuration. The present disclosure dynamically adjusts the pheromone evaporation coefficient according to the fluctuation of the second demand data, enabling the ant colony model to better adapt to the changes in the supply chain demand data at different stages, obtaining more accurate demand prediction model parameters, and thus achieving the effect of improving the accuracy and reliability of supply chain demand prediction.

[0045] Among the hyperparameters of the ant colony algorithm, in addition to the pheromone evaporation coefficient, there are also the pheromone heuristic factor and the expectation heuristic factor. The determination process of the pheromone heuristic factor and the expectation heuristic factor can be executed by the parameter optimization module 12. Specifically, in response to the existence of data in the target score with a score greater than the first score, the reference value of the pheromone heuristic factor is reduced based on the first step size to obtain the pheromone heuristic factor, and the reference value of the expectation heuristic factor is increased based on the second step size to obtain the expectation heuristic factor. The target score is based on the second demand data and is obtained by scoring the influencing factor data corresponding to the second demand data using the random forest algorithm.

[0046] In this embodiment, the target score is obtained using the random forest algorithm. Specifically, the second demand data can be used as the output, that is, the dependent variable, and the influencing factor data corresponding to the second demand data can be used as the input, that is, the independent variable, and input into the random forest algorithm. The random forest algorithm scores according to the importance of each influencing factor to the second demand data (i.e., the actual demand data of the target supply chain) to obtain the target score.

[0047] The pheromone heuristic factor represents the degree of emphasis of ants on the pheromone concentration when choosing a path, which helps to guide the ant colony to search along the paths that have been proven to be better in the past. In the supply chain demand prediction scenario, for example, if it is found in a previous search that when certain parameters of the demand prediction model are set to specific values, the demand prediction accuracy of the demand prediction model for a certain type of product is relatively high, then the pheromone concentration on the path corresponding to this parameter combination will be relatively high.

[0048] The expected heuristic factor reflects the degree of expectation of ants for the target state when choosing a path, that is, the degree of emphasis on the direct heuristic information from the current node to the target node, and can guide ants to search in a direction that seems closer to the target. In this embodiment, if the data with a higher target score is the price change trend, proving that the price change trend has a greater impact on the supply chain demand, the price change trend can be set as the direction of expected exploration or expected solution, which is equivalent to the direct heuristic information from the current node to the target node. When the expected heuristic factor is large, ants will pay more attention to this direct heuristic information and be more inclined to choose paths (i.e., price change trends) that seem more likely to be close to the optimal parameter combination for search, thereby accelerating the search effect and finding a parameter combination that can make the demand prediction model perform better.

[0049] Among them, the principle of adjusting the pheromone heuristic factor and the expected heuristic factor according to the target score is as follows: when there is data in the target score whose score is greater than the first score, it indicates that some influencing factors have a more significant impact on the demand. In this regard, the pheromone heuristic factor should be reduced and the expected heuristic factor should be increased to highlight the key factors, so that the ant colony model pays more attention to the expected heuristic information, which conforms to the logic that when there are obvious important influencing factors, these factors should be used preferentially to guide the search. Because the key factors may contain information that plays a decisive role in demand prediction, strengthening the dependence on the expected heuristic information helps to quickly locate better parameter combinations that match these factors.

[0050] In this embodiment, the first step length and the second step length can be determined based on the data in the experimental process, or can be set based on the reference range of the ant colony algorithm model itself.

[0051] In an embodiment of the present disclosure, the parameter optimization module 12 is further specifically configured to: in response to the data scores in the target score being all less than the second score, reduce the reference value of the expected heuristic factor based on the third step length to obtain the expected heuristic factor, and increase the reference value of the pheromone heuristic factor based on the fourth step length to obtain the pheromone heuristic factor; the reference values of the pheromone heuristic factor and the expected heuristic factor can be the default values of the ant colony model.

[0052] Similar to the aforementioned logic, when the data scores of the impact data corresponding to all the second demand data are relatively low, the expected heuristic factor is decreased and the pheromone heuristic factor is increased. This is because there is no obvious dominant factor in the data at this time. For example, the scores of factors such as season, price, and product life cycle are all small, and there is no prominent dominant factor. Then, the accumulation and dissemination of pheromone can search within a wider solution space, avoiding the ant colony model from falling into a local optimal solution due to over-reliance on less reliable heuristic information. For example, it is found that the change of season has a certain impact on the supply chain demand, causing the ant colony to move closer to the seasonal impact factor, resulting in the neglect of the impact of price and product life cycle on the supply chain demand.

[0053] Among them, the third step length and the fourth step length can be determined based on the data in the experimental process, or can be set based on the reference range of the ant colony algorithm model itself.

[0054] It can be concluded from the above that the present disclosure scores the impact factors of the target supply chain through the random forest algorithm, and dynamically adjusts the pheromone heuristic factor and the expected heuristic factor of the ant colony algorithm according to the scoring results, so that the present disclosure can adjust the search strategy according to the characteristics of the data to adapt to the demand prediction tasks under different data scenarios. The present disclosure dynamically adjusts the hyperparameters of the ant colony algorithm, enabling the ant colony model to better adapt to the characteristics of the supply chain impact data, improving the efficiency and accuracy of the parameter optimization process of the supply chain demand prediction model. Specifically, when there are key factors or dominant factors in the supply chain demand impact data, the ant colony model can more quickly locate the parameter combination of a better demand prediction model; when there is no obvious dominant factor in the supply chain demand impact data, the ant colony model can search within a wider space, thereby finding the parameter configuration of the demand prediction model that is more suitable for the current supply chain impact factors, improving the prediction performance of the demand prediction model, so as to improve the accuracy and reliability of the supply chain demand prediction.

[0055] In an embodiment of the present disclosure, the parameter optimization module 12 is further configured to: Determine the step lengths for adjusting the pheromone heuristic factor and the expected heuristic factor according to the dispersion degree of the target scores.

[0056] In this embodiment, the dispersion degree can be obtained by calculating the coefficient of variation of the target scores. The larger the coefficient of variation, the greater the dispersion degree, and the coefficient of variation can be used as the dispersion degree. The step lengths for adjusting the pheromone heuristic factor and the expected heuristic factor can be the step lengths for increasing the aforementioned two parameters, or can be the step lengths for decreasing the aforementioned two parameters. In this embodiment, the step lengths for adjusting the pheromone heuristic factor and the expected heuristic factor can include the first step length, the second step length, the third step length, and the fourth step length.

[0057] Specifically, both the first step length and the fourth step length can be determined by the first formula, and the first formula can be: , where represents the first step size or the fourth step size, represents the minimum step size, represents the maximum step size, represents the coefficient of variation, that is, the degree of dispersion, represents the minimum value of the coefficient of variation, represents the maximum value of the coefficient of variation, represents the first adjustment coefficient. The minimum step size, the maximum step size, the minimum value of the coefficient of variation, and the maximum value of the coefficient of variation can be calculated according to expert experience or data during the experiment.

[0058] The second step size and the third step size can be determined by the second formula, and the second formula can be: , where represents the second step size or the third step size, represents the first adjustment coefficient.

[0059] The reason why the first step size and the fourth step size can use the same formula is that they are both adjustments to the reference value of the pheromone heuristic factor. The reason why the second step size and the third step size can use the same formula is that they are both adjustments to the reference value of the expected heuristic factor. Since the adjustment degrees of the pheromone heuristic factor and the expected heuristic factor are different, the same formula cannot be adopted. and are the parameters for controlling the sizes of the two step sizes, and are obtained according to the actual application scenario and data fitting.

[0060] By analyzing the first formula and the second formula, it can be concluded that: When the coefficient of variation is large, it indicates that the data has a large degree of dispersion. At this time, a larger step size should be adopted to quickly search the solution space; when the coefficient of variation is small, it indicates that the data has a small degree of dispersion, and a smaller step size should be adopted for fine adjustment.

[0061] Because when the degree of dispersion of the target score is large, it indicates that there are some key factors in the influencing factor data that have a greater influence on the supply chain demand. The parameter optimization module 12 selects a larger adjustment step size, which can make the algorithm more inclined to adjust in the direction related to the key factors when searching the solution space. On the contrary, when the degree of dispersion of the target score is small, it means that the influence of each influencing factor on the demand is relatively average, and there are no particularly prominent key factors. To avoid the algorithm deviating from the optimal solution due to too large an adjustment amplitude, the parameter optimization module 12 can select a smaller adjustment step size and gradually fine-tune the pheromone heuristic factor and the expected heuristic factor, so that the algorithm can find a suitable parameter combination during the search process in a relatively stable solution space.

[0062] For example, in an electronic product supply chain, after scoring the data of influencing factors such as competitors' prices and product functional characteristics through the random forest algorithm, it is found that the dispersion degree of the target scores is relatively large, and the score of the market trend factor is much higher than other factors. Then, when the parameter optimization module 12 adjusts the pheromone heuristic factor and the expected heuristic factor, a relatively large adjustment step size is adopted to enable the ant colony algorithm to more quickly focus on the parameter search related to the market trend, so as to optimize the electronic product demand prediction model and make it more accurately reflect the impact of the market trend on demand.

[0063] Through the foregoing description, the hyperparameters of the ant colony model can be determined. During the iterative optimization process, the ant colony algorithm should determine the optimization effect and direction according to the fitness function. In this embodiment, the fitness function can be set as the mean absolute error, the mean square error, or the weighted average error, etc. The termination condition of the iteration can be that the difference between consecutive fitness function values is less than the preset difference or the number of iterations reaches the preset number, such as 3 - 6. The remaining parameters can be set according to experience or the default values of the ant colony model.

[0064] It can be concluded from the above that the present disclosure determines the step size for adjusting the pheromone heuristic factor and the expected heuristic factor based on the dispersion degree of the target scores, enabling the ant colony algorithm to adapt to various changes when dealing with the diversity and uncertainty of supply chain demand prediction. In the actual operation of the supply chain, the market demand changes rapidly. By adjusting the pheromone heuristic factor and the expected heuristic factor, the present disclosure enables the ant colony algorithm to accurately balance the relationship between exploration and exploitation during the iterative optimization process. For example, when a new demand signal appears in the market, the present disclosure can explore new demands with the adjusted factors and has a greater chance of finding the parameter combination that can most accurately reflect the market demand, thereby improving the prediction accuracy of the supply chain demand prediction model.

[0065] In an embodiment of the present disclosure, the data analysis module 11 is further configured to: Using the random forest model with the second demand data as the dependent variable and the first influencing factor data as the independent variable, obtain the data scores of each first influencing factor data; the first influencing factor data is the data that affects the supply and demand of the target supply chain; Based on the data scores of each first influencing factor data, screen the first influencing factor data to obtain the influencing factor data corresponding to the target supply chain.

[0066] In this embodiment, the target score is determined based on a random forest model. Specifically, the second demand data, i.e., the actual demand data of the supply chain, can be used as the dependent variable, and the first influencing factor data can be input into the random forest model as the independent variable, so as to obtain the scores of each data in the first influencing factor data. The first influencing factor data can be data that is preliminarily screened by those skilled in the art and considered to be related to the target supply chain.

[0067] After obtaining the data scores of each first influencing factor data, they can be compared with a preset score threshold. In response to the data score being higher than or equal to the preset score, the corresponding data is used as the influencing factor data corresponding to the target supply chain.

[0068] The data analysis module 11 can use the data in the first influencing factor data whose data scores are greater than the preset score as the influencing factor data corresponding to the target supply chain.

[0069] From the above, it can be concluded that the present disclosure can score each influencing factor based on the relationship between the input independent variable (the first influencing factor data) and the dependent variable (the second demand data) through the random forest model, providing data support for data screening and the aforementioned parameter adjustment.

[0070] In an embodiment of the present disclosure, the data analysis module 11 is further configured to: Screen the first influencing factor data according to a predetermined analysis period to obtain the influencing factor data corresponding to the target supply chain; the first influencing factor data is data that affects the supply and demand of the target supply chain; Determine the data characteristics of the influencing factor data corresponding to the target supply chain; The data characteristics are used to instruct the demand forecasting module 10 to select a corresponding demand forecasting model.

[0071] In this embodiment, the demand of the supply chain is dynamically affected by various factors, and the influence of each factor also changes in different time periods. Therefore, various influencing factors can be screened according to a predetermined analysis period (such as monthly, quarterly, etc.), which helps to timely capture the dynamic changes in the influence of factors such as market environment and seasonal changes on the supply chain demand, and ensure that the screened influencing factors can reflect the current actual situation. The screening process and screening conditions are the same as those for screening the first influencing factor data by the aforementioned random forest model.

[0072] From the above, it can be concluded that the present disclosure enables the data analysis module 11 to capture the influence of dynamic factors such as market environment and seasonal changes on the supply chain demand by regularly screening the influencing factors, which helps the demand forecasting module 10 to select the most appropriate demand forecasting model for prediction according to the latest influencing factor data, thereby improving the accuracy of supply chain demand forecasting.

[0073] In one embodiment of the present disclosure, the demand forecasting model includes a long short-term memory network model and a multi-layer perceptron model; The demand forecasting module 10 is specifically configured to: In response to the proportion of non-linear features in the influencing factor data corresponding to the target supply chain being greater than a preset proportion, based on the influencing factor data corresponding to the target supply chain, use the long short-term memory network model to obtain the first demand data of the target supply chain; In response to the proportion of non-linear features in the influencing factor data corresponding to the target supply chain being less than or equal to the preset proportion, based on the influencing factor data corresponding to the target supply chain, use the long short-term memory network model to obtain the first forecast data; and based on the influencing factor data corresponding to the target supply chain, use the multi-layer perceptron model to obtain the second forecast data; perform weighted calculation on the first forecast data and the second forecast data to obtain the first demand data of the target supply chain.

[0074] In this embodiment, linear fitting and non-linear fitting can be performed on the influencing factor data corresponding to the target supply chain, and the corresponding goodness of fit can be obtained, such as the linear fitting error and the non-linear fitting error .

[0075] Then, the proportion of non-linear features can be calculated through the third formula, and the third formula can be: , where is the proportion of linear features.

[0076] In this embodiment, considering that the long short-term memory network model includes structures such as memory units, making it have advantages in processing non-linear data. Compared with the prediction of non-linear data, its prediction effect on linear data is relatively poor, while the multi-layer perceptron model is relatively direct and effective in processing linear data. Therefore, when the influencing factor data has linear features, the influencing factor data can be input into the LSTM model and the multi-layer perceptron model respectively, and the weighted calculation is performed on the prediction data obtained by both to determine the first demand data of the target supply chain. The weights of the weighted calculation can be evenly distributed, or determined according to the proportion of non-linear features in the data. For example, the weights are adjusted according to the proportion of non-linear features, and the proportion of non-linear features is positively correlated with the allocated weights corresponding to the long short-term memory network model.

[0077] When the proportion of non-linear features is low, increase the weight of the multi-layer perceptron model and decrease the weight of the LSTM model; conversely, when the proportion of non-linear features is high, increase the weight of the LSTM model. The adjustment step size of the weight can be set based on experience.

[0078] As can be seen from the above, the present disclosure adjusts the output according to whether the influencing factors exhibit linear or non-linear characteristics, enabling the present disclosure to meet the demand forecasting requirements of the supply chain in different scenarios and improving the accuracy and reliability of supply chain demand forecasting.

[0079] A supply chain demand forecasting system based on artificial intelligence corresponding to the above embodiment Figure 2 is a schematic flowchart of a supply chain demand forecasting method based on artificial intelligence provided by an embodiment of the present disclosure. For the sake of illustration, only parts related to the embodiments of the present disclosure are shown. The supply chain demand forecasting method based on artificial intelligence in this embodiment can be executed by the supply chain demand forecasting system in any of the above embodiments. Refer to Figure 2 and the supply chain demand forecasting method based on artificial intelligence includes: S101: Using a demand forecasting model based on the influencing factor data corresponding to the target supply chain to obtain the first demand data of the target supply chain.

[0080] S102: Judging whether to output the parameter optimization requirement of the demand forecasting model according to the second demand data and the first demand data of the target supply chain; the second demand data is the actual demand data of the target supply chain.

[0081] S103: In response to the parameter optimization requirement, based on the second demand data and the influencing factor data corresponding to the second demand data, use an optimization model to obtain the optimized parameters corresponding to the demand forecasting model.

[0082] S104: Retrain the demand forecasting model based on the optimized parameters to perform supply chain demand forecasting based on the retrained demand forecasting model.

[0083] As can be seen from the above, the present disclosure proposes a strategy for dynamically adjusting the demand forecasting model, which can update the demand forecasting model according to the deviation between the demand data predicted by the demand forecasting model and the actual demand data, thereby ensuring the self-adaptability and robustness of the demand forecasting model and improving the accuracy and precision of supply chain demand forecasting.

[0084] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them; although the present disclosure has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. An artificial intelligence-based supply chain demand forecasting system, characterized in that, Including: A demand forecasting module, configured to use a demand forecasting model based on the influencing factor data corresponding to the target supply chain to obtain the first demand data of the target supply chain; A data analysis module, configured to determine whether to output a parameter optimization requirement of the demand forecasting model according to the second demand data and the first demand data of the target supply chain; The second demand data is the actual demand data of the target supply chain; A parameter optimization module, configured to, in response to the parameter optimization requirement, based on the second demand data and the influencing factor data corresponding to the second demand data, use an optimization model to obtain the optimized parameters corresponding to the demand forecasting model, and send the optimized parameters to a first device; The first device is configured to retrain the demand forecasting model based on the optimized parameters, and instruct the demand forecasting module to perform supply chain demand forecasting based on the retrained demand forecasting model.

2. The supply chain demand forecasting system based on artificial intelligence according to claim 1, wherein Specifically, the data analysis module is configured to output the parameter optimization requirement of the demand forecasting model in response to that the deviation between the first demand data and the second demand data is greater than a preset deviation threshold.

3. The supply chain demand forecasting system based on artificial intelligence according to claim 1, wherein, The optimization model is an ant colony model; before using the optimization model to obtain the optimized parameters corresponding to the demand forecasting model based on the second demand data and the influencing factor data corresponding to the second demand data, the parameter optimization module is further configured to: Determine the solution space of the optimization model based on the training parameters and structure parameters corresponding to the demand forecasting model, and determine the hyperparameters of the optimization model based on the second demand data and the influencing factor data corresponding to the second demand data.

4. The supply chain demand forecasting system based on artificial intelligence according to claim 3, wherein The hyperparameters include a pheromone heuristic factor and an expected heuristic factor; Specifically, the parameter optimization module is configured to: In response to that there is data in the target score whose score is greater than the first score, reduce the reference value of the pheromone heuristic factor based on the first step size to obtain the pheromone heuristic factor, and increase the reference value of the expected heuristic factor based on the second step size to obtain the expected heuristic factor; The target score is obtained by using the random forest algorithm to score the influencing factor data corresponding to the second demand data based on the second demand data.

5. The supply chain demand forecasting system based on artificial intelligence according to claim 3, wherein The hyperparameters include a pheromone heuristic factor and an expected heuristic factor; Specifically, the parameter optimization module is configured to: In response to that all the data scores in the target score are less than the second score, reduce the reference value of the expected heuristic factor based on the third step size to obtain the expected heuristic factor, and increase the reference value of the pheromone heuristic factor based on the fourth step size to obtain the pheromone heuristic factor; The target score is obtained by using the random forest algorithm to score the influencing factor data corresponding to the second demand data based on the second demand data.

6. The artificial intelligence-based supply chain demand forecasting system according to claim 4 or 5, characterized in that The parameter optimization module is further configured to: Determine the step sizes for adjusting the pheromone heuristic factor and the expected heuristic factor according to the dispersion degree of the target score.

7. The supply chain demand forecasting system based on artificial intelligence according to claim 1, characterized in that, The data analysis module is further configured to: Use the second demand data as the dependent variable and the first influencing factor data as the independent variable, and use the random forest model to obtain the data scores of each first influencing factor data; the first influencing factor data is the data that affects the supply demand of the target supply chain. Screen the first influencing factor data based on the data scores of the first influencing factor data to obtain the influencing factor data corresponding to the target supply chain.

8. The supply chain demand forecasting system based on artificial intelligence according to claim 1, characterized in that, The data analysis module is further configured to: Screen the first influencing factor data according to a predetermined analysis period to obtain the influencing factor data corresponding to the target supply chain; the first influencing factor data is data that affects the supply demand of the target supply chain; Determine the data characteristics of the influencing factor data corresponding to the target supply chain; the data characteristics are used to instruct the demand forecasting module to select a corresponding demand forecasting model.

9. The artificial intelligence-based supply chain demand forecasting system according to claim 8, wherein The demand forecasting models include a long short-term memory network model and a multi-layer perceptron model; The demand forecasting module is specifically configured to: In response to the proportion of non-linear features in the influencing factor data corresponding to the target supply chain being greater than a preset proportion, use the long short-term memory network model based on the influencing factor data corresponding to the target supply chain to obtain the first demand data of the target supply chain; In response to the proportion of non-linear features in the influencing factor data corresponding to the target supply chain being less than or equal to the preset proportion, use the long short-term memory network model based on the influencing factor data corresponding to the target supply chain to obtain the first forecast data; and use the multi-layer perceptron model based on the influencing factor data corresponding to the target supply chain to obtain the second forecast data; perform weighted calculation on the first forecast data and the second forecast data to obtain the first demand data of the target supply chain.

10. A supply chain demand forecasting method based on artificial intelligence, characterized in that, It includes: Use the demand forecasting model based on the influencing factor data corresponding to the target supply chain to obtain the first demand data of the target supply chain; Judge whether to output the parameter optimization requirement of the demand forecasting model according to the second demand data of the target supply chain and the first demand data; the second demand data is the actual demand data of the target supply chain; In response to the parameter optimization requirement, use the optimization model based on the second demand data and the influencing factor data corresponding to the second demand data to obtain the optimized parameters corresponding to the demand forecasting model; Retrain the demand forecasting model based on the optimized parameters to perform supply chain demand forecasting based on the retrained demand forecasting model.

Images