Product sales prediction method and system based on BP neural network

Abstract

The invention relates to a product sales prediction method and system based on a BP neural network, and relates to the field of product sales prediction, and the method comprises the steps: constructing a product sales network graph; obtaining a plurality of historical product sales records, constructing a plurality of historical product sales network samples, and training and obtaining a product sales prediction model in combination with a double-layer loss function; receiving a to-be-predicted target product, extracting a plurality of predetermined sales points and a plurality of predetermined sales channels, obtaining predetermined sales strategies for the plurality of predetermined sales points at the same time, and establishing a predetermined product sales graph; and based on the predetermined product sales graph, using the product sales prediction model to perform sales prediction on the target product, and obtaining the predicted sales of the target product. According to the method and the device, the technical problem of low product sales volume prediction precision in a complex sales network in the prior art is solved, and the technical effect of improving the product sales volume prediction accuracy in the complex sales network is achieved by constructing the product sales network graph and performing decomposition prediction on the predetermined sales link of the target product.

Description

Technical Field

[0001] The present invention relates to the field of product sales forecasting, and in particular to a product sales forecasting method and system based on BP neural network. Background Art

[0002] In the current market environment, enterprises are often faced with diversified sales channels and complex sales networks. In such a complex sales network, accurately predicting product sales is crucial for enterprises to formulate sales strategies and inventory management. Traditional product sales forecasting is usually based on time series analysis or regression analysis, which can achieve good results in simple sales networks. However, when faced with complex sales networks, these methods are difficult to capture the mutual influence between sales points and the differences in sales strategies, resulting in a large deviation between the prediction results and the actual sales. In addition, existing sales forecasting methods usually model the entire sales network as a whole, ignoring the particularity of different sales links, and cannot fully utilize the structural information of the sales links, resulting in a further decrease in prediction accuracy. Therefore, the existing technology has low accuracy in product sales forecasting in complex sales networks. Summary of the invention

[0003] The present invention aims at solving the technical problem of low accuracy of product sales prediction in complex sales networks in the prior art, and provides a product sales prediction method and system based on BP neural network to solve the problem.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: In a first aspect, the present invention provides a product sales prediction method based on a BP neural network, comprising: constructing a product sales network graph, wherein the product sales network graph is a directed acyclic graph, wherein nodes in the directed acyclic graph represent sales points, and edges in the directed acyclic graph represent sales channels, and each sales point has multiple preset sales strategies; obtaining multiple historical product sales records, and based on the product sales network graph, constructing multiple historical product sales network samples, and training and acquiring a product sales prediction model in combination with a double-layer loss function, wherein the product sales prediction model includes multiple node sales prediction models, each node sales prediction model is composed of a basic prediction layer and a BP neural network layer, and the basic prediction layer is composed of multiple prediction models; receiving a target product to be predicted, obtaining a predetermined sales link for the target product, decomposing the predetermined sales link, extracting multiple predetermined sales points and multiple predetermined sales channels, and simultaneously obtaining predetermined sales strategies for multiple predetermined sales points, and forming a predetermined product sales graph; based on the predetermined product sales graph, using the product sales prediction model, predicting the sales of the target product, and obtaining the predicted sales of the target product.

[0005] In a second aspect, the present invention provides a product sales forecasting system based on a BP neural network, comprising: a sales graph construction module, used to construct a product sales network graph, the product sales network graph is a directed acyclic graph, the nodes in the directed acyclic graph represent sales points, the edges in the directed acyclic graph represent sales channels, and each sales point has multiple preset sales strategies; a model training module, used to obtain multiple historical product sales records, and based on the product sales network graph, construct multiple historical product sales network samples, and train and obtain a product sales forecasting model in combination with a double-layer loss function, wherein the product sales forecasting model includes multiple node sales forecasting models, each node sales forecasting model is composed of a basic forecasting layer and a BP neural network layer, and the basic forecasting layer is composed of multiple prediction models; a target product analysis module, used to receive a target product to be predicted, obtain a predetermined sales link for the target product, decompose the predetermined sales link, extract multiple predetermined sales points and multiple predetermined sales channels, and simultaneously obtain predetermined sales strategies for multiple predetermined sales points to form a predetermined product sales graph; a sales forecasting module, used to predict the sales of the target product based on the predetermined product sales graph and using the product sales forecasting model to obtain the predicted sales of the target product.

[0006] In a third aspect, the present application provides an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; wherein the processor is used to execute the product sales prediction method based on the BP neural network provided in the present application.

[0007] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, and the computer program is used to execute the product sales prediction method based on BP neural network provided in the present application.

[0008] The beneficial effects of the present invention are: Construct a product sales network diagram, in which nodes represent sales points and edges represent sales channels. Each sales point has multiple preset sales strategies, so as to clearly represent the structure of a complex sales network; obtain multiple historical product sales records, construct multiple historical product sales network samples based on the product sales network diagram, and train and obtain a product sales prediction model in combination with a double-layer loss function, wherein the product sales prediction model includes multiple node sales prediction models, each node sales prediction model is composed of a basic prediction layer and a BP neural network layer, and the basic prediction layer is composed of multiple prediction models. By introducing the node sales prediction model and the double-layer loss function, the influence of sales strategies and sales channels of different sales points is fully considered to improve the accuracy of product sales prediction; receive the target product to be predicted, obtain the predetermined sales link for the target product, decompose the predetermined sales link, extract multiple predetermined sales points and multiple predetermined sales channels, and simultaneously obtain the predetermined sales strategies for multiple predetermined sales points, form a predetermined product sales map, make full use of the structural information of the sales link, and improve the prediction accuracy; based on the predetermined product sales map, use the product sales prediction model to predict the sales of the target product, obtain the predicted sales of the target product, and obtain a more accurate target product sales prediction result by using the predetermined product sales map and the product sales prediction model in a targeted manner. By constructing a product sales network diagram, introducing a node sales prediction model and a double-layer loss function, and decomposing and analyzing the predetermined sales links, the problem of low product sales prediction accuracy in complex sales networks is effectively solved, and the prediction accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A schematic diagram of the process of the product sales forecasting method based on BP neural network provided by the present invention; Figure 2 A schematic diagram of the structure of a product sales forecasting system based on a BP neural network provided by the present invention; Figure 3 A schematic diagram of the structure of an electronic device provided by the present invention; Figure 4 A schematic diagram of the structure of a computer-readable storage medium provided by the present invention.

[0010] In the accompanying drawings, the components represented by the reference numerals are as follows: Sales map construction module 11, model training module 12, target product analysis module 13, sales forecasting module 14, electronic device 200, memory 210, processor 220, first computer program 211, computer readable storage medium 300, second computer program 311. DETAILED DESCRIPTION

[0011] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0012] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0013] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention.

[0014] Embodiment 1: like Figure 1 As shown, the embodiment of the present invention provides a product sales prediction method based on BP neural network, including: S100: Construct a product sales network graph, wherein the product sales network graph is a directed acyclic graph, wherein nodes in the directed acyclic graph represent sales points, edges in the directed acyclic graph represent sales channels, and each sales point has multiple preset sales strategies.

[0015] Specifically, in the process of constructing the product sales network graph, the nodes of the graph, namely the sales points, are first determined. Each sales point is a key link in the product sales network, representing a sales subject in the product sales process, which can be different types of sales subjects such as manufacturers, wholesalers, retailers, and online platforms. By abstracting these sales subjects as nodes in the graph, the product sales network is structured. Next, the edges connecting the sales points, namely the sales channels, are determined. The sales channel represents the path and method of the product flowing from one sales point to another. Each edge has its specific direction, indicating the flow direction of the product. At the same time, each edge also has additional attributes to describe the characteristics of the sales channel, such as channel type, channel capacity, channel cost, channel time, channel priority, etc. By giving directions and attributes to the sales channels, the flow process of the product in the sales network can be more accurately portrayed. After that, for each sales point, multiple sales strategies are preset. Sales strategy refers to a series of measures and methods taken in the sales process to improve sales performance and achieve sales goals. Among them, the preset sales strategy is designed and selected according to different sales scenarios, target customers, market environment and other factors. For example, we can formulate differentiated pricing strategies, promotion strategies, and advertising strategies for different target customer groups; formulate off-season and peak-season strategies, holiday strategies, etc. for different sales periods; formulate regional specialty strategies and localization strategies for different sales areas. By presetting multiple sales strategies at each sales point, we can provide sufficient optimization space for subsequent sales optimization.

[0016] By constructing a product sales network diagram that includes sales points, sales channels, and preset sales strategies, the product's sales network structure and sales process are fully represented, providing a basis for subsequent sales forecasts, sales optimization, and other analyses.

[0017] S200: Obtain multiple historical product sales records, and based on the product sales network diagram, construct multiple historical product sales network samples, and train to obtain a product sales prediction model in combination with a double-layer loss function, wherein the product sales prediction model includes multiple node sales prediction models, each node sales prediction model is composed of a basic prediction layer and a BP neural network layer, and the basic prediction layer is composed of multiple prediction models.

[0018] Specifically, in order to train the product sales prediction model, it is necessary to obtain sufficient historical sales data. By collecting and organizing the company's past product sales records, multiple historical product sales records containing information such as sales time, sales location, sales quantity, and sales amount are obtained, providing a real sample basis for model training. After obtaining multiple historical product sales records, these records are associated with the previously constructed product sales network diagram to construct multiple historical product sales network samples. Each historical product sales network sample is the actual sales situation in the sales network in a specific historical period. Then, based on the constructed multiple historical product sales network samples, combined with the double-layer loss function, the product sales prediction model is trained.

[0019] The product sales prediction model consists of multiple node sales prediction models, each of which is responsible for predicting the sales of the corresponding node in the sales network. Each node sales prediction model consists of a basic prediction layer and a BP neural network layer. The basic prediction layer consists of multiple prediction models, which can be time series prediction models, regression models, random forests, support vector machines, etc., to capture the basic trends and laws of node sales. The output of the basic prediction layer will be used as the input of the BP neural network layer. The BP neural network layer receives the output of the basic prediction layer and further optimizes and corrects the sales prediction of a single sales point through nonlinear transformation and weight learning, thereby learning the correlation influence and high-level features between sales points and improving the accuracy of sales prediction for a single sales point. In the model training process, a double-layer loss function is used as the optimization target. The first-layer loss function focuses on the prediction performance of the basic prediction layer. By comparing the difference between the output of the basic prediction model and the actual sales, the parameters of the basic prediction model are optimized. The second-layer loss function focuses on the prediction performance of the BP neural network layer. By comparing the difference between the output of the BP neural network layer and the actual sales, the weight parameters of the BP neural network layer are optimized to improve the accuracy of sales prediction for a single sales point.

[0020] By training a product sales prediction model with excellent performance, we can make full use of historical product sales records and product sales network diagrams to accurately predict product sales at each sales point in the future, providing support for sales forecasting.

[0021] S300: receiving a target product to be predicted, obtaining a predetermined sales link for the target product, decomposing the predetermined sales link, extracting multiple predetermined sales points and multiple predetermined sales channels, and simultaneously obtaining predetermined sales strategies for multiple predetermined sales points, and forming a sales map for the predetermined product.

[0022] Specifically, when it is necessary to predict the sales volume of a target product, firstly, relevant information of the target product to be predicted is received, including product type, specification model, target market, etc. According to the characteristics and sales requirements of the target product, the predetermined sales link for the target product is obtained. The predetermined sales link represents a series of sales points and sales channels that the target product passes through from production to final sales, reflecting the expected circulation path of the target product in the sales network. After obtaining the predetermined sales link, it is decomposed to extract multiple predetermined sales points and multiple predetermined sales channels that constitute the predetermined sales link. The predetermined sales point refers to each sales node involved in the predetermined sales link, such as manufacturers, regional general distributors, provincial distributors, prefecture-level city distributors, retailers, etc. The predetermined sales channel represents the sales channel connecting adjacent predetermined sales points, such as wholesale channels, retail channels, online channels, offline channels, etc.

[0023] While extracting the scheduled sales points and scheduled sales channels, obtain the scheduled sales strategy for each scheduled sales point. The scheduled sales strategy refers to the sales strategy set for each scheduled sales point when predicting sales, including pricing strategy, promotion strategy, advertising strategy, regional characteristics strategy, etc. These scheduled sales strategies are pre-formulated based on the characteristics of the target product, market environment, target customers and other factors, and are used to guide and optimize the sales activities of the target product at each scheduled sales point. After obtaining the scheduled sales points, scheduled sales channels and scheduled sales strategies, a scheduled product sales map is constructed. The scheduled product sales map is a specific sales network structure constructed for the target product, which contains information such as the scheduled sales link, scheduled sales points, scheduled sales channels and corresponding scheduled sales strategies of the target product. The scheduled product sales map represents the expected sales path and sales strategy of the target product in the sales network, and provides necessary input information for subsequent sales forecasts.

[0024] By constructing a predetermined product sales map including predetermined sales points, predetermined sales channels and predetermined sales strategies for the target product to be predicted, the characteristics and sales requirements of the target product are fully considered, laying the foundation for the subsequent use of the product sales forecasting model to predict the sales of the target product.

[0025] S400: Based on the predetermined product sales graph, using the product sales forecasting model, forecasting the sales of the target product, and obtaining the forecast sales of the target product.

[0026] Specifically, first, the sales graph of the scheduled product is input into the product sales forecasting model, and the product sales forecasting model forecasts the sales of each scheduled sales point in the scheduled product sales graph. For each scheduled sales point, the corresponding node sales forecasting model predicts the expected sales of the sales point according to the basic forecasting layer and the BP neural network layer. In the forecasting process, the basic forecasting layer of the node sales forecasting model first uses multiple basic forecasting models, such as time series forecasting model, regression model, random forest, support vector machine, etc., to make a preliminary forecast of the sales of the scheduled sales point. Then, the BP neural network layer receives the output of the basic forecasting layer, further corrects and optimizes the preliminary forecast results, and obtains the final predicted sales of the scheduled sales point. By forecasting the sales of each scheduled sales point in the scheduled product sales graph in turn, a set of predicted sales data of the scheduled sales points is obtained. By summarizing these predicted sales data, the overall predicted sales of the target product in the entire scheduled sales chain, that is, the predicted sales of the target product, can be obtained. The predicted sales of the target product comprehensively considers the expected sales of the target product at each scheduled sales point, as well as the correlation influence and sales network structure between the scheduled sales points, and reflects the overall sales expectation of the target product under a given scheduled sales chain and sales strategy.

[0027] Furthermore, the embodiment of the present application also includes: S210: traverse multiple historical product sales records to obtain a first historical product sales record; S220: Based on the product sales network diagram, the first historical product sales record is decomposed to obtain a plurality of first historical sales links; S230: Based on the product sales network diagram, perform secondary decomposition on the plurality of first historical sales links, and decompose each first historical sales link into a plurality of first historical sales points and a plurality of first historical sales channels; S240: Obtaining a first historical sales strategy and first historical sales data for each first historical sales point; S250: for each first historical sales link, starting from the bottom first historical sales point, combining the corresponding first historical sales strategy, the first historical sales data and the associated first historical sales channel, recursively constructing a first historical sales point sample upward, and aggregating multiple first historical sales point samples to form a first historical product sales network sample; S260: Repeat the process until multiple historical product sales records are traversed to obtain multiple historical product sales network samples.

[0028] In a preferred embodiment, first, multiple historical product sales records are traversed, and each historical product sales record is obtained one by one as a first historical product sales record. Then, the first historical product sales record is disassembled once based on the product sales network diagram to obtain multiple first historical sales links. The first historical sales link represents the complete flow path of the product from production to sales in the first historical product sales record, including the various sales points and sales channels passed through. Next, multiple first historical sales links are secondary disassembled based on the product sales network diagram, and each first historical sales link is disassembled into multiple first historical sales points and multiple first historical sales channels. The first historical sales point represents each sales node involved in the first historical sales link, such as a manufacturer, a regional general distributor, a provincial distributor, a prefecture-level city distributor, a retailer, etc. The first historical sales channel represents the sales channel connecting adjacent first historical sales points, such as a wholesale channel, a retail channel, an online channel, an offline channel, etc. Subsequently, the first historical sales strategy and the first historical sales data of each first historical sales point are obtained. The first historical sales strategy refers to the sales strategy actually adopted by each first historical sales point in the historical product sales record, including pricing strategy, promotion strategy, advertising strategy, regional feature strategy, etc. The first historical sales volume data refers to the number of products actually sold by each first historical sales point in the historical product sales record.

[0029] Then, for each first historical sales link, a first historical sales point sample is constructed. Specifically, first, the underlying sales point of the first historical sales link is determined, that is, the sales point closest to the end consumer in the link, such as a retailer. Taking the bottom sales point as the starting point, obtain the corresponding first historical sales strategy and first historical sales data as the basis for constructing the first historical sales point sample; then, in the product sales network diagram, find the sales channel directly connected to the bottom sales point, and along the sales channel, obtain the upper-level sales point connected to it, such as the prefecture-level city dealer, and obtain the first historical sales strategy and first historical sales data of the upper-level sales point; then, take the first historical sales data of the bottom sales point as the label, and take its first historical sales strategy, the first historical sales strategy of the upper-level sales point and the associated sales channel information as the feature, indicating the sales situation of the bottom sales point under the given upper-level sales point strategy and sales channel; continue to recursively upward on the product sales network diagram, take the upper-level sales point just now as the new starting point, and repeat the above adjacent sampling and sample construction process until reaching the top-level sales point of the first historical sales link, such as the manufacturer; in this recursive process, each sales point will be associated with the sales label of its lower-level sales point to form a multi-level first historical sales point sample. By performing layer-by-layer recursive fitting on the product sales network diagram, all first historical sales point samples on the complete first historical sales link are obtained. Each sample not only contains the sales strategy and sales data of the sales point itself, but also associates the sales information and sales channel characteristics of its subordinate sales points, reflecting the hierarchical structure and dependency relationship in the sales network. Then, all the first historical sales point samples on the first historical sales link are summarized to form a complete first historical product sales network sample. This sample covers each sales point on the first historical sales link, as well as the association relationship and multi-level influence in the product sales network diagram, providing rich information. After that, it is repeated until all historical product sales records are traversed, and multiple historical product sales network samples are obtained to provide support for training product sales prediction models.

[0030] By utilizing the product sales network diagram, we conducted multi-level decomposition and sample construction of historical product sales records, and obtained multiple historical product sales network samples, which not only included the historical sales data and sales strategies of each sales point, but also reflected the topological relationship and multi-level correlation influence between the sales points, providing rich and targeted data support for training high-quality product sales prediction models.

[0031] Furthermore, the embodiment of the present application also includes: S270: The double-layer loss function includes a first-layer loss function and a second-layer loss function; S271: The loss function of the first layer is: ; in, is the loss function value of the first layer, is the weight coefficient of the variance term, which is used to control the importance of the variance in the first layer loss function. is the weight coefficient of the mean error term, which is used to control the importance of the mean error in the first layer loss function. , represents a set of n predicted output values ​​based on the prediction layer, represents the variance of n predicted output values, Indicates the actual sales value. represents the mean of n predicted output values, Represents the absolute error between the mean of n predicted output values ​​and the actual sales value; S272: The second layer loss function is: ; in, is the loss function value of the second layer, represents the predicted output value of the BP neural network, represents the mean square error, is the weight coefficient of the regularization term, which is used to control the importance of regularization in the second layer loss function. Represents the weight parameter set of the BP neural network, Represents the L2 norm squared of the weight parameter.

[0032] In a preferred embodiment, a double-layer loss function is introduced, which includes a first-layer loss function and a second-layer loss function. The purpose of the double-layer loss function is to consider the prediction performance of the basic prediction layer and the BP neural network layer when training the product sales prediction model to obtain a more accurate and stable prediction result.

[0033] The first-layer loss function focuses on the prediction performance of the basic prediction layer, aiming to minimize the prediction error of the basic prediction model. The first-layer loss function is , consists of two parts, namely the variance term and the mean error term. represents the variance of the n predicted output values ​​output by the base prediction layer, where Represents the set of n predicted output values ​​of the basic prediction layer. The variance term measures the discreteness of the prediction results of the basic prediction model. A larger variance means the instability of the prediction results. Mean error term Represents the mean of the n predicted output values ​​output by the basic prediction layer The actual sales value The mean error term measures the overall deviation of the prediction results of the basic prediction model. In the first layer loss function, the variance term and the mean error term are multiplied by the weight coefficients and , which is used to control the importance of the variance term and the mean error term in the loss function. and The value of can balance the contribution of variance and mean error to the loss function, thereby optimizing the prediction performance of the basic prediction layer.

[0034] The second layer loss function focuses on the prediction performance of the BP neural network layer, aiming to minimize the prediction error of the BP neural network. The second layer loss function is , consists of two parts, namely the mean square error term and the regularization term. Represents the predicted output value of the BP neural network The actual sales value The mean square error term measures the difference between the prediction result of the BP neural network and the true value. The regularization term Represents the weight parameter of BP neural network The L2 norm squared is multiplied by the regularization coefficient λ. The regularization term is used to control the complexity of the BP neural network model and prevent the model from overfitting. By adding the regularization term, the model is encouraged to learn simpler and more generalizable parameters, thereby improving the robustness of the model. In the second-layer loss function, the weights of the mean square error term and the regularization term are controlled by λ. By adjusting the value of λ, the impact of the prediction error and model complexity on the loss function can be balanced, thereby optimizing the prediction performance of the BP neural network layer.

[0035] The double-layer loss function takes into account the prediction performance of the basic prediction layer and the BP neural network layer through the coordinated optimization of the first-layer loss function and the second-layer loss function. The first-layer loss function improves the prediction stability and accuracy of the basic prediction model by minimizing the variance and mean error; the second-layer loss function improves the prediction accuracy and generalization ability of the BP neural network by minimizing the mean square error and controlling the model complexity. The introduction of the double-layer loss function helps to train a product sales prediction model with better performance, thereby providing more reliable and accurate sales prediction results.

[0036] Furthermore, the embodiment of the present application also includes: S500: extracting multiple predetermined sales points from the predetermined product sales graph, and extracting multiple predetermined sales channels associated with the multiple predetermined sales points, combining the multiple node sales prediction models, obtaining the predicted sales of the multiple nodes, and recursively summarizing to form the predicted sales of the target product; S600: extracting the expected sales volume of the target product, and when the predicted sales volume of the target product is less than the expected sales volume of the target product, optimizing the sales graph of the predetermined product until the predicted sales volume of the target product is greater than or equal to the expected sales volume of the target product.

[0037] In a preferred embodiment, first, multiple scheduled sales points are extracted from the scheduled product sales map. Scheduled sales points refer to various sales nodes when predicting the sales volume of the target product, such as manufacturers, regional general distributors, provincial distributors, prefecture-level city distributors, retailers, etc. Then, multiple scheduled sales channels associated with multiple scheduled sales points are extracted. Scheduled sales channels refer to sales channels connecting different scheduled sales points, such as wholesale channels, retail channels, online channels, offline channels, etc. Next, combined with multiple node sales prediction models, the sales volume of each scheduled sales point is predicted, and the historical sales data, sales strategies, associated sales channels and other factors of the sales point are comprehensively considered to obtain multiple node predicted sales volume. After obtaining the node predicted sales volume of each scheduled sales point, these node predicted sales volume are accumulated layer by layer by recursive aggregation to form the predicted total sales volume of the target product on the entire scheduled product sales map, that is, the predicted sales volume of the target product. The process of recursive aggregation fully considers the hierarchical relationship and quantity transfer between the scheduled sales points, and the obtained predicted sales volume of the target product can reflect the expected sales of the entire sales network.

[0038] Subsequently, the expected sales volume of the target product is extracted. The expected sales volume of the target product is a pre-set target product sales volume value based on production plans, market demand forecasts, sales targets and other factors, representing the ideal sales volume level that is expected to be achieved. Then, the predicted sales volume of the target product is compared with the expected sales volume of the target product. If the predicted sales volume of the target product is less than the expected sales volume of the target product, it means that the current scheduled product sales map and sales strategy may not meet the sales expectations of the enterprise and need to be optimized and adjusted. At this time, the sales map of the scheduled product is optimized. While keeping the structure of the scheduled product sales map unchanged, the predicted sales volume of each scheduled sales point is improved by adjusting the sales strategy of each scheduled sales point, optimizing resource allocation, strengthening sales channels and other methods, thereby improving the predicted total sales volume of the entire sales network. The optimization process is carried out through multiple rounds of iterations. After each round of optimization, the predicted sales volume of the target product is recalculated and compared with the expected sales volume of the target product. If the predicted sales volume of the optimized target product is greater than or equal to the expected sales volume of the target product, it means that the optimization has achieved the expected effect and the optimization process can be stopped; otherwise, the next round of optimization is continued until the conditions are met.

[0039] By dynamically optimizing the sales chart of scheduled products based on the predicted sales of target products and further combining actual needs and expectations, the predicted results can be closer to the company's sales targets, which can help to adjust sales strategies in a timely manner, reasonably allocate resources, maximize product market performance, and achieve sales targets.

[0040] Furthermore, the embodiment of the present application also includes: S610: Obtain node sales bases of multiple predetermined sales points, and calculate the unit sales volume of each predetermined sales point, where the unit sales volume is equal to the node predicted sales volume divided by the node sales base; S620: Select target predetermined sales points for optimization in descending order according to the unit sales volume; S630: For a target predetermined sales point, obtain multiple preset sales strategies, apply the multiple preset sales strategies to the target predetermined sales point in sequence, and re-execute the node sales forecast to obtain multiple node candidate forecast sales; S640: Selecting a predetermined sales strategy with the largest predicted sales volume among multiple candidate node predicted sales volumes as the optimal sales strategy for the target predetermined sales point; S650: Based on the optimal sales strategy of the target predetermined sales point, updating the predetermined product sales map, and recalculating the predicted sales volume of the target product; S660: Iterative execution: when the predicted sales volume of the target product is greater than or equal to the expected sales volume of the target product, stop optimizing the sales graph of the scheduled product and obtain an updated sales graph of the scheduled product.

[0041] In a preferred embodiment, when optimizing the sales graph of a predetermined product, first, the node sales cardinality of multiple predetermined sales points is obtained. The node sales cardinality refers to the benchmark sales volume of each predetermined sales point when predicting sales volume, which is determined based on factors such as the historical sales data, market size, and customer groups of the sales point. Then, the unit sales volume of each predetermined sales point is calculated. The unit sales volume is equal to the node predicted sales volume of the sales point divided by its node sales cardinality, which represents the predicted sales volume of each cardinality unit of the sales point on average. The higher the unit sales volume, the higher the sales efficiency of the sales point, and the more sales volume can be brought by each cardinality unit on average. Subsequently, according to the calculated unit sales volume, the multiple predetermined sales points are sorted in order from small to large. A predetermined sales point with a small unit sales volume indicates that its sales efficiency is relatively low and there is room for optimization. According to the order of unit sales volume from small to large, the predetermined sales points are selected in turn as the optimization target, that is, the target predetermined sales point. In this way, the predetermined sales points with low sales efficiency can be optimized first to increase their sales volume.

[0042] Then, for the selected target predetermined sales point, multiple preset sales strategies are obtained. The preset sales strategy refers to a series of sales plans preset for the sales point, such as price strategy, promotion strategy, advertising strategy, etc. Then, these preset sales strategies are applied to the target predetermined sales point in turn, and the node sales forecasting model is re-executed to obtain multiple node candidate predicted sales. Each node candidate predicted sales corresponds to a preset sales strategy, which represents the expected sales of the target predetermined sales point under the strategy. Then, from the multiple node candidate predicted sales, the one with the largest predicted sales is selected, and the corresponding preset sales strategy is used as the optimal sales strategy for the target predetermined sales point. This means that among all the preset sales strategies, this strategy can maximize the expected sales of the target predetermined sales point, and is therefore regarded as the optimal strategy. Subsequently, the sales strategy of the target predetermined sales point is updated to the optimal sales strategy, and the strategies of the sales channels and other sales points associated with it are adjusted accordingly. The updated sales graph of the predetermined product can reflect the optimized sales strategy configuration. Then, based on the updated sales graph of the predetermined product, the predicted sales of the target product is recalculated to obtain the optimized forecast result. Repeat the execution to iteratively optimize the sales graph of the predetermined product. Each round of iteration will select a target scheduled sales point with a small unit sales volume, apply the optimal sales strategy to it, and update the scheduled product sales map and the target product predicted sales volume. When the optimized target product predicted sales volume is greater than or equal to the target product expected sales volume, it means that the optimization has achieved the expected effect and the iterative optimization process is stopped. At this point, the updated scheduled product sales map is obtained, in which the sales strategy configuration and predicted sales volume results are optimized.

[0043] By optimizing the weak links in the sales graph of the scheduled products in a targeted manner and adjusting the sales strategy, the sales efficiency and predicted sales of individual sales points can be improved, thereby driving the sales of the entire sales network. At the same time, through iterative optimization, the expected sales of the target products are constantly approached, making the optimization results more in line with actual needs.

[0044] Furthermore, the embodiment of the present application also includes: S661: Setting a preset node number threshold and a sales volume difference threshold; S662: When the number of predetermined sales points corresponding to the preset node number threshold is optimized in the order of small to large unit sales volume, a first sales volume difference between the predicted sales volume of the target product at this time and the expected sales volume of the target product is calculated; S663: If the first sales volume difference is greater than the sales volume difference threshold, performing sales strategy optimization on all remaining predetermined sales points simultaneously; S664: Update the scheduled product sales graph based on the optimization result, recalculate the target product predicted sales volume as the node's optimal predicted sales volume, and obtain the updated scheduled product sales graph.

[0045] In a preferred embodiment, a phased optimization mechanism is introduced by setting a preset node number threshold and a sales volume difference threshold to improve optimization efficiency.

[0046] First, set the preset node number threshold and sales volume difference threshold. The preset node number threshold indicates the upper limit of the number of scheduled sales points that need to be optimized during the optimization process. When the number of optimized scheduled sales points reaches this threshold, a phased evaluation is triggered. The sales volume difference threshold indicates the maximum acceptable difference between the predicted sales volume of the target product and the expected sales volume of the target product during the phased evaluation. If the actual difference exceeds this threshold, it means that the current optimization effect is not ideal and further optimization measures are needed. When optimizing the scheduled sales points in order from small to large according to the unit sales volume, after optimizing the number of scheduled sales points corresponding to the preset node number threshold, calculate the difference between the predicted sales volume of the target product and the expected sales volume of the target product at this time, that is, the first sales volume difference. This difference reflects the gap between the predicted sales volume and the expected sales volume of the entire sales network after optimizing a certain number of scheduled sales points.

[0047] Subsequently, the first sales difference is compared with the sales difference threshold. If the first sales difference is greater than the sales difference threshold, it means that the current optimization effect is not ideal, and only optimizing the number of scheduled sales points corresponding to the preset node number threshold cannot meet the expected sales requirements. At this time, the sales strategy optimization is performed on all the remaining scheduled sales points at the same time. In other words, instead of optimizing one by one in the order of unit sales volume, all the remaining scheduled sales points are optimized in parallel, and different sales strategy combinations are tried in order to obtain a larger optimization space and better optimization effect. After the sales strategy optimization is performed on all the remaining scheduled sales points at the same time, an optimization result is obtained. Based on the optimization result, the sales graph of the scheduled product is updated, and the sales strategies of all the scheduled sales points are updated to the optimized strategies. Then, the predicted sales of the target product are recalculated to obtain the optimized prediction result, which is called the node full-optimal predicted sales. The node full-optimal predicted sales reflects the expected sales level of the entire sales network after all the scheduled sales points are fully optimized. At the same time, the updated sales graph of the scheduled product is obtained, which contains the optimized sales strategy configurations of all the scheduled sales points.

[0048] By introducing preset node number thresholds and sales volume difference thresholds, a phased optimization strategy is provided. In the early stage of optimization, the scheduled sales points with low sales efficiency are optimized first in the order of unit sales volume, so as to obtain greater optimization effects at a lower optimization cost. After optimization to a certain extent, if the expected sales volume requirements are still not met, all remaining scheduled sales points are optimized simultaneously, and a wider range of strategy combinations are tried in order to obtain better results. This phased optimization mechanism can strike a balance between optimization efficiency and optimization effect, which can not only quickly improve the overall performance of the sales network, but also thoroughly optimize weak links when necessary, so as to ultimately make the predicted sales volume as close to or as close to the expected sales volume as possible.

[0049] Furthermore, the embodiment of the present application also includes: S665: If the node's all-predicted sales volume is less than the target product's expected sales volume, calculate a second sales volume difference between the node's all-predicted sales volume and the target product's expected sales volume; S666: Obtain an expanded scheduled sales node according to the second sales volume difference, generate an expanded scheduled sales link according to the expanded scheduled sales node, and update the scheduled product sales graph.

[0050] In a preferred embodiment, when the fully predicted sales volume of nodes still cannot reach the expected sales volume of the target product, the sales network is further expanded by expanding the predetermined sales nodes and predetermined sales links in order to increase the predicted sales volume and meet the expected sales volume requirements.

[0051] First, compare the node's optimal predicted sales volume with the target product's expected sales volume. If the node's optimal predicted sales volume is still less than the target product's expected sales volume, it means that the expected sales volume target cannot be achieved by optimizing the existing scheduled sales points and sales strategies alone. At this time, calculate the difference between the node's optimal predicted sales volume and the target product's expected sales volume, i.e., the second sales volume difference. The second sales volume difference indicates that, on the basis of the existing sales network and strategy optimization, additional sales volume is required to make up for the gap between the predicted sales volume and the expected sales volume. Then, based on the second sales volume difference, determine the number and type of scheduled sales nodes that need to be expanded, i.e., expand the scheduled sales nodes. Expanding scheduled sales nodes refers to newly added sales points outside the existing scheduled product sales map, which are used to expand the sales network and improve sales coverage and sales capabilities. The selection of expanded scheduled sales nodes is based on market research, customer analysis, regional characteristics and other factors, and selects sales points with greater sales potential and stronger regional influence. After determining the expanded scheduled sales nodes, generate corresponding expanded scheduled sales links based on these nodes. Expanding scheduled sales links refers to connecting the expanded scheduled sales nodes with existing scheduled sales points and sales channels to form new sales paths and sales relationships. When generating the expanded reservation sales link, the geographical location, transportation conditions, cooperative relationship and other factors between the newly added nodes and the existing nodes are considered to ensure that the expanded sales network can operate efficiently and smoothly. After that, the expanded reservation sales nodes and expanded reservation sales links are integrated into the original reservation product sales map to form an updated and expanded reservation product sales map.

[0052] When the existing sales network and strategy optimization reaches a bottleneck and cannot meet the expected sales requirements, continuous improvement and dynamic adaptation can be achieved by introducing new strategic sales nodes and sales links, so that the sales network can be flexibly adjusted according to actual needs and market changes, and continuously optimized and improved to better support the achievement of sales targets.

[0053] The product sales prediction method based on BP neural network provided by the embodiment of the present invention has at least the following technical effects: Construct a product sales network diagram, which is a directed acyclic graph. The nodes in the directed acyclic graph represent sales points, and the edges in the directed acyclic graph represent sales channels. Each sales point has multiple preset sales strategies. By clearly representing the relationship between sales points and sales channels, it lays the foundation for subsequent prediction model training and sales forecasting. Obtain multiple historical product sales records, and based on the product sales network diagram, construct multiple historical product sales network samples. Combined with the double-layer loss function, train and obtain the product sales forecasting model, where the product sales forecasting model includes multiple node sales forecasting models. Each node sales forecasting model consists of a basic forecasting layer and a BP neural network layer. The basic forecasting layer consists of multiple forecasting models, which provide support for the sales forecasting of the target product. Receive the target product to be predicted, obtain the predetermined sales link for the target product, decompose the predetermined sales link, extract multiple predetermined sales points and multiple predetermined sales channels, and obtain the predetermined sales strategies for multiple predetermined sales points at the same time. Build a predetermined product sales diagram, associate the sales information of the target product with the product sales network diagram, and prepare for sales forecasting. Based on the predetermined product sales graph, the product sales forecasting model is used to forecast the sales of the target product, obtain the predicted sales of the target product, and achieve accurate forecasting of product sales in complex sales networks.

[0054] Embodiment 2: like Figure 2 As shown, based on the same inventive concept as the product sales prediction method based on the BP neural network provided in the first embodiment, the embodiment of the present invention also provides a product sales prediction system based on the BP neural network, including: A sales graph construction module 11 is used to construct a product sales network graph, wherein the product sales network graph is a directed acyclic graph, wherein nodes in the directed acyclic graph represent sales points, edges in the directed acyclic graph represent sales channels, and each sales point has multiple preset sales strategies; The model training module 12 is used to obtain multiple historical product sales records, build multiple historical product sales network samples based on the product sales network diagram, and train and obtain a product sales prediction model in combination with a double-layer loss function, wherein the product sales prediction model includes multiple node sales prediction models, each of which is composed of a basic prediction layer and a BP neural network layer, and the basic prediction layer is composed of multiple prediction models; The target product analysis module 13 is used to receive the target product to be predicted, obtain the predetermined sales link for the target product, decompose the predetermined sales link, extract multiple predetermined sales points and multiple predetermined sales channels, and simultaneously obtain the predetermined sales strategies for the multiple predetermined sales points to form a predetermined product sales map; The sales volume forecasting module 14 is used to forecast the sales volume of the target product based on the predetermined product sales graph and using the product sales volume forecasting model to obtain the forecasted sales volume of the target product.

[0055] Furthermore, the model training module 12 includes the following execution steps: Traverse multiple historical product sales records and obtain the first historical product sales record; Based on the product sales network diagram, the first historical product sales record is disassembled to obtain a plurality of first historical sales links; Based on the product sales network diagram, a plurality of first historical sales links are secondary disassembled, and each first historical sales link is disassembled into a plurality of first historical sales points and a plurality of first historical sales channels; Obtain the first historical sales strategy and the first historical sales data for each first historical sales point; For each first historical sales link, starting from the bottom first historical sales point, combining the corresponding first historical sales strategy, the first historical sales data and the associated first historical sales channel, recursively construct the first historical sales point sample upward, and aggregate multiple first historical sales point samples to form the first historical product sales network sample; The execution is repeated until multiple historical product sales records are traversed to obtain multiple historical product sales network samples.

[0056] Furthermore, the model training module 12 also includes the following execution steps: The double-layer loss function includes a first-layer loss function and a second-layer loss function; The loss function of the first layer is: ; in, is the loss function value of the first layer, is the weight coefficient of the variance term, which is used to control the importance of the variance in the first layer loss function. is the weight coefficient of the mean error term, which is used to control the importance of the mean error in the first layer loss function. , represents a set of n predicted output values ​​based on the prediction layer, represents the variance of n predicted output values, Indicates the actual sales value. represents the mean of n predicted output values, Represents the absolute error between the mean of n predicted output values ​​and the actual sales value; The loss function of the second layer is: ; in, is the loss function value of the second layer, represents the predicted output value of the BP neural network, represents the mean square error, is the weight coefficient of the regularization term, which is used to control the importance of regularization in the second layer loss function. Represents the weight parameter set of the BP neural network, Represents the L2 norm squared of the weight parameter.

[0057] Furthermore, the embodiment of the present application also includes a sales optimization module, which includes the following execution steps: Extracting multiple predetermined sales points from the predetermined product sales graph, and extracting multiple predetermined sales channels associated with the multiple predetermined sales points, combining the multiple node sales prediction models, obtaining the multiple node predicted sales, and recursively summarizing to form the target product predicted sales; The expected sales volume of the target product is extracted, and when the predicted sales volume of the target product is less than the expected sales volume of the target product, the sales graph of the predetermined product is optimized until the predicted sales volume of the target product is greater than or equal to the expected sales volume of the target product.

[0058] Furthermore, the sales optimization module also includes the following execution steps: Obtaining node sales bases of multiple predetermined sales points, and calculating the unit sales volume of each predetermined sales point, where the unit sales volume is equal to the node forecast sales volume divided by the node sales base; According to the unit sales volume, select the target predetermined sales points for optimization in order from small to large; For a target predetermined sales point, a plurality of preset sales strategies are obtained, the plurality of preset sales strategies are applied to the target predetermined sales point in sequence, and the node sales forecast is re-executed to obtain a plurality of node candidate forecast sales; Selecting a predetermined sales strategy with the largest predicted sales volume among multiple node candidate predicted sales volumes as the optimal sales strategy for the target predetermined sales point; Based on the optimal sales strategy of the target predetermined sales point, updating the predetermined product sales map and recalculating the predicted sales volume of the target product; Iterative execution, when the predicted sales volume of the target product is greater than or equal to the expected sales volume of the target product, stop optimizing the sales graph of the scheduled product and obtain an updated sales graph of the scheduled product.

[0059] Furthermore, the sales optimization module also includes the following execution steps: Set preset node number threshold and sales volume difference threshold; When the number of predetermined sales points corresponding to the preset node number threshold is optimized in the order of small to large unit sales volume, a first sales volume difference between the predicted sales volume of the target product at this time and the expected sales volume of the target product is calculated; If the first sales volume difference is greater than the sales volume difference threshold, performing sales strategy optimization on all remaining predetermined sales points simultaneously; The scheduled product sales graph is updated based on the optimization result, the predicted sales volume of the target product is recalculated as the node's optimal predicted sales volume, and an updated scheduled product sales graph is obtained.

[0060] Furthermore, the sales optimization module also includes the following execution steps: If the node's fully predicted sales volume is less than the target product's expected sales volume, calculate a second sales volume difference between the node's fully predicted sales volume and the target product's expected sales volume; An expanded scheduled sales node is obtained according to the second sales volume difference, and an expanded scheduled sales link is generated according to the expanded scheduled sales node, and the scheduled product sales map is updated.

[0061] Embodiment three: See also Figure 3 , Figure 3 Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an electronic device 200 provided by an embodiment of the present invention includes a memory 210, a processor 220, and a first computer program 211 stored in the memory 210 and executable on the processor 220. When the processor 220 executes the first computer program 211, a product sales prediction method based on a BP neural network is implemented.

[0062] Embodiment 4: See also Figure 4 , Figure 4 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. Figure 4 As shown, this embodiment provides a computer-readable storage medium 300 on which a second computer program 311 is stored. When the second computer program 311 is executed by a processor, a product sales prediction method based on a BP neural network is implemented.

[0063] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0064] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0065] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or multiple boxes.

[0066] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction system, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0067] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0068] Although preferred embodiments of the present invention have been described, additional changes and modifications may occur to these embodiments once those skilled in the art understand the basic inventive concepts.

[0069] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the present invention and its equivalent technology, the present invention is also intended to include these changes and variations.

Claims

1. A product sales forecasting method based on BP neural network, characterized in that: include: Constructing a product sales network graph, wherein the product sales network graph is a directed acyclic graph, the nodes in the directed acyclic graph represent sales points, the edges in the directed acyclic graph represent sales channels, and each sales point has multiple preset sales strategies; Acquire multiple historical product sales records, construct multiple historical product sales network samples based on the product sales network diagram, and train and acquire a product sales prediction model in combination with a double-layer loss function, wherein the product sales prediction model includes multiple node sales prediction models, each node sales prediction model is composed of a basic prediction layer and a BP neural network layer, and the basic prediction layer is composed of multiple prediction models; Receiving a target product to be predicted, obtaining a predetermined sales link for the target product, decomposing the predetermined sales link, extracting a plurality of predetermined sales points and a plurality of predetermined sales channels, and simultaneously obtaining a predetermined sales strategy for the plurality of predetermined sales points, and forming a sales map for the predetermined product; Based on the predetermined product sales graph, the product sales forecasting model is used to forecast the sales of the target product to obtain the forecast sales of the target product.

2. The product sales prediction method based on BP neural network according to claim 1 is characterized in that: Acquire multiple historical product sales records, and construct multiple historical product sales network samples based on the product sales network diagram, including: Traverse multiple historical product sales records and obtain the first historical product sales record; Based on the product sales network diagram, the first historical product sales record is disassembled to obtain a plurality of first historical sales links; Based on the product sales network diagram, a plurality of first historical sales links are secondary disassembled, and each first historical sales link is disassembled into a plurality of first historical sales points and a plurality of first historical sales channels; Obtain the first historical sales strategy and the first historical sales data for each first historical sales point; For each first historical sales link, starting from the bottom first historical sales point, combining the corresponding first historical sales strategy, the first historical sales data and the associated first historical sales channel, recursively construct the first historical sales point sample upward, and aggregate multiple first historical sales point samples to form the first historical product sales network sample; The execution is repeated until multiple historical product sales records are traversed to obtain multiple historical product sales network samples.

3. The product sales forecasting method based on BP neural network according to claim 2 is characterized in that: include: The double-layer loss function includes a first-layer loss function and a second-layer loss function; The loss function of the first layer is: ； in, is the loss function value of the first layer, is the weight coefficient of the variance term, which is used to control the importance of the variance in the first layer loss function. is the weight coefficient of the mean error term, which is used to control the importance of the mean error in the first layer loss function. , represents a set of n predicted output values ​​based on the prediction layer, represents the variance of n predicted output values, Indicates the actual sales value. represents the mean of n predicted output values, Represents the absolute error between the mean of n predicted output values ​​and the actual sales value; The loss function of the second layer is: ； in, is the loss function value of the second layer, represents the predicted output value of the BP neural network, represents the mean square error, is the weight coefficient of the regularization term, which is used to control the importance of regularization in the second layer loss function. Represents the weight parameter set of the BP neural network, Represents the L2 norm squared of the weight parameter.

4. The product sales prediction method based on BP neural network according to claim 1 is characterized in that: The method further comprises: Extracting multiple predetermined sales points from the predetermined product sales graph, and extracting multiple predetermined sales channels associated with the multiple predetermined sales points, combining the multiple node sales prediction models, obtaining multiple node predicted sales, and recursively summarizing to form a target product predicted sales; The expected sales volume of the target product is extracted, and when the predicted sales volume of the target product is less than the expected sales volume of the target product, the sales graph of the predetermined product is optimized until the predicted sales volume of the target product is greater than or equal to the expected sales volume of the target product.

5. The product sales forecasting method based on BP neural network according to claim 4 is characterized in that: Optimizing the sales graph of the predetermined product includes: Obtaining node sales bases of multiple predetermined sales points, and calculating the unit sales volume of each predetermined sales point, where the unit sales volume is equal to the node forecast sales volume divided by the node sales base; According to the unit sales volume, select the target predetermined sales points for optimization in order from small to large; For a target predetermined sales point, a plurality of preset sales strategies are obtained, the plurality of preset sales strategies are applied to the target predetermined sales point in sequence, and the node sales forecast is re-executed to obtain a plurality of node candidate forecast sales; Selecting a predetermined sales strategy with the largest predicted sales volume among multiple node candidate predicted sales volumes as the optimal sales strategy for the target predetermined sales point; Based on the optimal sales strategy of the target predetermined sales point, updating the predetermined product sales map and recalculating the predicted sales volume of the target product; Iterative execution, when the predicted sales volume of the target product is greater than or equal to the expected sales volume of the target product, stop optimizing the sales graph of the scheduled product and obtain an updated sales graph of the scheduled product.

6. The product sales forecasting method based on BP neural network according to claim 5 is characterized in that: The method further comprises: Set preset node number threshold and sales volume difference threshold; When the number of predetermined sales points corresponding to the preset node number threshold is optimized in the order of small to large unit sales volume, a first sales volume difference between the predicted sales volume of the target product at this time and the expected sales volume of the target product is calculated; If the first sales volume difference is greater than the sales volume difference threshold, performing sales strategy optimization on all remaining predetermined sales points simultaneously; The scheduled product sales graph is updated based on the optimization result, the predicted sales volume of the target product is recalculated as the node's optimal predicted sales volume, and an updated scheduled product sales graph is obtained.

7. The product sales forecasting method based on BP neural network according to claim 6 is characterized in that: The method further comprises: If the node's fully predicted sales volume is less than the target product's expected sales volume, calculate a second sales volume difference between the node's fully predicted sales volume and the target product's expected sales volume; An expanded scheduled sales node is obtained according to the second sales volume difference, and an expanded scheduled sales link is generated according to the expanded scheduled sales node, and the scheduled product sales map is updated.

8. The product sales forecasting system based on BP neural network is characterized by: The method for predicting product sales based on a BP neural network according to any one of claims 1 to 7 comprises: A sales graph construction module, the sales graph construction module is used to construct a product sales network graph, the product sales network graph is a directed acyclic graph, the nodes in the directed acyclic graph represent sales points, the edges in the directed acyclic graph represent sales channels, and each sales point has multiple preset sales strategies; A model training module, the model training module is used to obtain multiple historical product sales records, build multiple historical product sales network samples based on the product sales network diagram, and train to obtain a product sales prediction model in combination with a double-layer loss function, wherein the product sales prediction model includes multiple node sales prediction models, each node sales prediction model is composed of a basic prediction layer and a BP neural network layer, and the basic prediction layer is composed of multiple prediction models; A target product analysis module, the target product analysis module is used to receive a target product to be predicted, obtain a predetermined sales link for the target product, decompose the predetermined sales link, extract multiple predetermined sales points and multiple predetermined sales channels, and simultaneously obtain a predetermined sales strategy for multiple predetermined sales points, and form a predetermined product sales map; The sales volume forecasting module is used to forecast the sales volume of the target product based on the predetermined product sales graph and using the product sales volume forecasting model to obtain the forecasted sales volume of the target product.

9. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program, thereby implementing the product sales prediction method based on BP neural network as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium, characterized in that: The storage medium stores a computer software program, and when the computer software program is executed by the processor, the product sales prediction method based on the BP neural network as described in any one of claims 1 to 7 is implemented.

Images