Big data analysis-based sales prediction and strategy optimization method and related equipment thereof

Abstract

The invention belongs to the technical field of e-commerce data processing, and particularly provides a sales prediction and strategy optimization method based on big data analysis and related equipment thereof, and the method mainly comprises the steps: collecting the historical sales volume of a commodity, the price of a competitive product, marketing and reform data, dynamically constructing a competitive product feature matrix, training a neural network to generate an influence weight vector, and carrying out the calculation of the influence weight vector; a comprehensive influence factor is obtained through dot product operation, a sales volume prediction value is output by combining an LSTM time sequence prediction model, and finally, price and marketing resource allocation are intelligently adjusted through a nonlinear mapping function based on the target sales volume gap direction and strength. According to the method, real-time quantitative evaluation of dynamic influences of competitive products can be realized, the response sensitivity of a prediction result to sudden competition events can be remarkably improved, a price adjustment and marketing resource allocation strategy can be intelligently generated according to the gap direction and strength, the demand can be actively stimulated to fill the sales gap, and excessive supply can be reversely inhibited to maintain profit balance.

Description

Technical Field

[0001] The present invention belongs to the technical field of e-commerce data processing, and particularly relates to a sales prediction and strategy optimization method based on big data analysis and related devices. Background Art

[0002] Predicting sales data can help enterprises optimize inventory, stock up according to demand while preventing overstocking, contribute to reasonable production planning, adapt to the market rhythm and reduce costs, set directions for marketing strategies, improve the effectiveness of promotion, be the foundation of financial budgets, accurately predict revenues and expenditures, ensure the stable flow of enterprise funds, and achieve long-term development.

[0003] Currently, sales prediction models based on big data mainly rely on the historical sales data of products themselves for analysis, and it is difficult to analyze the non-linear relationship between the prices of competing products and the sales volume of target products in real time, resulting in lagged or deviated prediction results. It is difficult to capture the sudden impacts of product model changes or marketing activities of competing products, which limits the adaptability of the model in complex market environments. Summary of the Invention

[0004] This application effectively solves the problems in the prior art that it is difficult to analyze the non-linear relationship between the prices of competing products and the sales volume of target products in real time, resulting in lagged or deviated prediction results, and it is difficult to capture the sudden impacts of product model changes or marketing activities of competing products, which limits the adaptability of the model in complex market environments, by providing a sales prediction and strategy optimization method based on big data analysis and related devices. It can capture these changes in the first time, dynamically adjust the nutritional support strategy, ensure that the nutritional intervention always fits the current state of the patient, and effectively improve the quality of medical services to ensure that the patient receives the most appropriate care.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, this application provides a sales prediction and strategy optimization method based on big data analysis, including: obtaining the historical sales data of products; obtaining competing product data, including price data, marketing activity data, and product model change data; constructing a competing product feature vector according to the price data, marketing activity data, and product model change data; using the competing product feature vector as the input and the historical sales data of products as the output to complete the training of a neural network model, and obtaining a feature influence weight vector according to the trained neural network model; performing a dot product operation on the competing product features and the feature influence weight vector to obtain a comprehensive influence factor; using the trained long short-term memory network to process the comprehensive influence factor and the historical sales data, and outputting the predicted sales volume of products; determining the target sales volume, and optimizing the strategy according to the target sales volume and the predicted sales volume.

[0007] Further, the price data includes a benchmark price, a current price, and a price reduction range; the marketing activity data includes the remaining time of the marketing activity; the model change data includes model change information and technical parameter annotation information; determine competing products, and use web crawler management software to obtain the price information, promotional activity information, and publicity pictures of the competing products at a specified time; determine the benchmark price and the current price according to the price information, and calculate the price reduction range; determine the remaining time of the marketing activity according to the promotional activity information; extract the appearance of the competing products from the publicity pictures, and calculate the similarity of the appearance of the competing products in the publicity pictures obtained twice in a row; set a similarity threshold, and if the similarity is less than the similarity threshold, record the model change information; identify the text in the publicity pictures, match it with the ISO standard term library, and record the technical parameter annotation information that matches the standard terms.

[0008] Further, according to the price data, marketing activity data, and model change data, construct a competing product feature vector, including: constructing a price matrix with the benchmark price, current price, and price reduction range; constructing a marketing matrix with the remaining time of the marketing activity; constructing a model change matrix with the model change information and technical parameter annotation information; constructing a competing product feature vector including the price matrix, marketing matrix, and model change matrix.

[0009] Further, use the competing product feature vector as the input and the historical sales data of the product as the output to complete the training of the neural network model, including: aligning the competing product feature vector data and the historical sales data of the product in time; preprocessing the competing product feature vector, including normalizing the benchmark price, current price, price reduction range, and the remaining time of the marketing activity, and encoding the model change information and technical parameter annotation information; constructing a neural network model, using the preprocessed competing product feature vector data as the input data, and using the historical sales data of the product as the output label to train the neural network model until the training is completed.

[0010] Further, the feature influence weight vector includes: price weight, marketing weight, and model change weight; the price weight includes the weights of the benchmark price, current price, and price reduction range; the marketing weight includes the weight of the remaining time of the marketing activity; the model change weight includes the weights of the model change information and technical parameter annotation information.

[0011] Further, train a long short-term memory network, including: aligning the comprehensive influence factor and the historical sales data in time to obtain combined data; constructing a three-dimensional tensor data including the number of samples, time steps, and number of features; constructing an LSTM model, using the three-dimensional tensor data as the input data, and the predicted value of the product sales volume as the output to complete the training of the LSTM model.

[0012] Further, perform strategy optimization based on the target sales volume and the predicted sales volume, including: calculating a gap value based on the target sales volume and the predicted sales volume, and determining a gap direction symbol; wherein, if the gap value is not negative, the gap direction symbol is positive, otherwise the gap direction symbol is negative; calculating a historical average sales volume fluctuation value based on historical sales data, and taking the ratio of the absolute value of the gap to the historical average sales volume fluctuation value as the gap intensity; taking the product obtained by multiplying the comprehensive influence factor, the price weight, and the gap intensity as the price base quantity; multiplying the gap direction symbol, the hyperbolic tangent function value of the price base quantity, and a preset scaling coefficient, and constraining the obtained product within a specified range to obtain a price adjustment ratio; taking the product obtained by multiplying the comprehensive influence factor, the marketing weight, and the gap intensity as the marketing base quantity; inputting the product of the gap direction symbol and the marketing base quantity into an S-shaped function for non-linear mapping, and constraining the mapping result within a specified interval to obtain a marketing adjustment coefficient.

[0013] In a second aspect, the present application provides a sales prediction and strategy optimization system based on big data analysis, which adopts the sales prediction and strategy optimization system based on big data analysis described in the first aspect, and includes: a data collection module, which is used to obtain historical sales data of commodities; obtain competitor data, including price data, marketing activity data, and model change data; a competitor feature construction module, which is used to construct a competitor feature vector according to the price data, marketing activity data, and model change data; a neural network training and weight generation module, which is used to take the competitor feature vector as an input, and take the historical sales data of commodities as an output to complete the training of a neural network model, and obtain a feature influence weight vector according to the trained neural network model; a comprehensive influence evaluation module, which is used to perform a dot product operation on the competitor features and the feature influence weight vector to obtain a comprehensive influence factor; a sales volume prediction module, which is used to process the comprehensive influence factor and the historical sales data by using a trained long short-term memory network, and output the predicted sales volume of the commodity; a strategy optimization decision module, which is used to determine the target sales volume and perform strategy optimization based on the target sales volume and the predicted sales volume.

[0014] In a third aspect, the present application provides a device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of the sales prediction and strategy optimization method based on big data analysis described in the first aspect when executing the computer program.

[0015] In a fourth aspect, the present application provides a readable storage medium, in which computer program instructions are stored, and when the computer program instructions are read and run by a processor, the steps of the sales prediction and strategy optimization method based on big data analysis described in the first aspect are executed.

[0016] Advantages of the present invention:

[0017] This application constructs a competitor feature vector through heterogeneous data fusion, extracts feature influence weights using neural network gradient analysis, combines LSTM time series modeling to achieve coupled prediction of internal and external factors, introduces a two-way strategy optimization algorithm, and dynamically generates price adjustment and marketing resource allocation strategies based on gap analysis. It effectively solves the problems in the prior art that it is difficult to analyze the non-linear relationship between competitor prices and the sales volume of target products in real time, resulting in lagging or deviated prediction results, and it is difficult to capture the sudden impacts of competitor model changes or marketing activities, limiting the adaptability of the model in complex market environments. It can realize real-time quantitative evaluation of the dynamic impacts of competitors, significantly improve the response sensitivity of prediction results to sudden competition events, and can also intelligently generate price adjustment and marketing resource allocation strategies based on the gap direction and intensity. It can not only actively stimulate demand to fill the sales volume gap, but also reverse suppress excessive supply to maintain profit balance.

[0018] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 Shows a schematic flow chart of the sales prediction and strategy optimization method based on big data analysis of the present invention;

[0021] Figure 2 Shows a schematic diagram of the sales prediction and strategy optimization system based on big data of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] To solve the problems raised in the background art, the present disclosure constructs a competitor feature vector through heterogeneous data fusion, extracts feature influence weights using neural network gradient analysis, combines LSTM time series modeling to achieve coupled prediction of internal and external factors, introduces a two-way strategy optimization algorithm, and dynamically generates price adjustment and marketing resource allocation strategies based on gap analysis. It can realize real-time quantitative evaluation of the dynamic impacts of competitors, significantly improve the response sensitivity of prediction results to sudden competition events, and can also intelligently generate price adjustment and marketing resource allocation strategies based on the gap direction and intensity. It can not only actively stimulate demand to fill the sales volume gap, but also reverse suppress excessive supply to maintain profit balance.

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] As Figure 1 shown, the present disclosure provides a sales prediction and strategy optimization method based on big data analysis, including:

[0025] S100. Obtain the historical sales data of the product; obtain competitor data, including price data, marketing activity data, and model change data.

[0026] S200. Construct a competitor feature vector based on the price data, marketing activity data, and model change data.

[0027] S300. Use the competitor feature vector as the input and the historical sales data of the product as the output to complete the training of the neural network model, and obtain the feature influence weight vector according to the trained neural network model.

[0028] S400. Perform a dot product operation on the competitor features and the feature influence weight vector to obtain a comprehensive influence factor.

[0029] S500. Use the trained long short-term memory network to process the comprehensive influence factor and the historical sales data, and output the predicted sales volume of the product.

[0030] S600. Determine the target sales volume and optimize the strategy according to the target sales volume and the predicted sales volume.

[0031] S100. Obtain the historical sales data of the product; obtain competitor data, including price data, marketing activity data, and model change data. In S100:

[0032] In some embodiments, the price data includes the benchmark price, the current price, and the price reduction range; the marketing activity data includes the remaining time of the marketing activity; the model change data includes the model change information and the technical parameter annotation information.

[0033] Obtaining the price data, marketing activity data, and model change data respectively includes the following steps:

[0034] Determine the competitors, and use web crawler management software to obtain the price information, promotion activity information, and publicity pictures of the competitors at a specified time.

[0035] Price data: Determine the benchmark price and the current price according to the price information, and calculate the price reduction range.

[0036] Marketing campaign data: Determine the remaining time of a marketing campaign based on promotional information.

[0037] Model change data: extract the appearance of competing products from promotional pictures, and calculate the similarity of the appearance of competing products in two consecutive promotional pictures; set a similarity threshold, and if the similarity is less than the similarity threshold, record the model change information; identify the text in the promotional pictures and match it with the ISO standard terminology library, and record the technical parameter annotation information that matches the standard terminology.

[0038] For example, based on the industry classification code to which the target product belongs, such as NAICS or GB / T standard, competitive products with the same functions, price range and target user group can be screened through a semantic matching algorithm to build a dynamic competitive product list;

[0039] A distributed web crawler framework can be deployed to perform targeted crawling of competing official stores and mainstream e-commerce platforms within a preset time window. Price information includes the real-time price marked in the page DOM node and the historical highest price in the hidden field. Promotional information is obtained by parsing the countdown component and activity rule pop-up data dynamically generated by JavaScript. Promotional images can be obtained by downloading the main visual image of the product details page and the technical parameter display board image.

[0040] Based on the sliding time window algorithm, the price series of the competing products in the past 30 natural days are denoised, and the median of the price concentration area is taken as the benchmark price; the price reduction range can be calculated by the following formula: Among them, D represents the price reduction, P b represents the base price, P c Represents the current price.

[0041] The promotional text can be parsed by combining regular expression matching with natural language processing. For example, the explicit timestamp "until 2023-12-31 23:59:59" is directly converted to a UTC timestamp, and the relative time description "3 days left" can be used to calculate the deadline through the server time at the time of web page crawling, thereby obtaining the remaining time of the marketing activity.

[0042] After preprocessing the promotional pictures crawled twice in succession, the structural similarity index can be used to calculate the image similarity. When the similarity is less than the similarity threshold, the modification timestamp and the comparison matrix of the new and old images are recorded.

[0043] The attention mechanism model based on ResNet-50 can be used to detect the text area in the promotional picture, extract the text content through the OCR engine, and establish an ISO term mapping pipeline: construct a keyword library for technical parameters, and use the BERT semantic similarity model to calculate the cosine similarity between the extracted text and the standard terms. When the similarity is greater than 0.92, write the matching standard terms into the technical annotation field, so as to obtain the technical parameter annotation information.

[0044] S200. Construct a competitor feature vector based on price data, marketing activity data, and model change data. In S200:

[0045] In some embodiments, constructing a competitor feature vector based on price data, marketing activity data, and model change data includes: constructing a price matrix with the benchmark price, current price, and price reduction range; constructing a marketing matrix with the remaining time of the marketing activity; constructing a model change matrix with the model change information and technical parameter annotation information; and constructing a competitor feature vector including the price matrix, marketing matrix, and model change matrix.

[0046] Take the benchmark price, current price, and price reduction range as three independent dimensions and arrange them in the order of collection time to form a three-dimensional price matrix; the remaining time of the marketing activity can be converted into a countdown value to generate a marketing matrix.

[0047] S300. Use the competitor feature vector as the input and the historical sales data of the product as the output to complete the training of the neural network model, and obtain the feature influence weight vector according to the trained neural network model. In S300:

[0048] In some embodiments, using the competitor feature vector as the input and the historical sales data of the product as the output to complete the training of the neural network model includes:

[0049] S310. Align the competitor feature vector data and the historical sales data of the product in time.

[0050] Extract the collection timestamp corresponding to each price matrix in the competitor feature vector, and segment and aggregate the historical sales data of the product at the same time granularity.

[0051] S320. Preprocess the competitor feature vector, including normalizing the benchmark price, current price, price reduction range, and the remaining time of the marketing activity, and encoding the model change information and technical parameter annotation information.

[0052] The benchmark price and the current price can be normalized to the [0, 1] interval using Min-Max normalization, such as: where P bnorm The normalized benchmark price, P b represents the current benchmark price, P bminRepresents the historical lowest benchmark price, P bmax Represents the historical highest benchmark price; the same applies to the current price.

[0053] The price reduction range can be logarithmically compressed and then normalized, e.g.: where D norm Represents the normalized price reduction range, D represents the actual price reduction range, D max Represents the maximum allowable discount rate.

[0054] The remaining marketing time can be converted according to the negative exponential, e.g.: where T norm Represents the normalized remaining time, T represents the remaining marketing campaign time, T c Represents the total time of the marketing campaign.

[0055] The model change and technical parameter coding contains two parts, namely the model change information coding and the technical parameter annotation coding. For the model change information coding, the similarity difference can be quantified to obtain three-level discrete values. When the similarity difference is less than or equal to 0.1, the coding value is 0; when the similarity difference is greater than 0.1 and less than or equal to 0.3, the coding value is 1; when the similarity difference is greater than 0.3, the coding value is 2. For the technical parameter annotation coding, a multi-dimensional binary vector can be generated, e.g., 200-dimensional. For the i-th bit of this vector, if the i-th type of parameter of the ISO standard is detected, the value of this bit is 1; if the i-th type of parameter of the ISO standard is not detected, the value of this bit is 0.

[0056] S330. Construct a neural network model, use the preprocessed competitor feature vector data as input data, and use the historical sales data of the product as the output label to train the neural network model until the training is completed.

[0057] Exemplarily, the constructed neural network can include an input layer, a hidden layer, and an output layer, where the hidden layer includes a first fully connected layer, a second fully connected layer, and a Dropout layer.

[0058] The input layer can accept the preprocessed competitor feature vector.

[0059] The first fully connected layer is set with 128 neurons and uses the ReLU activation function. The ReLU function can effectively perform non-linear transformation on data and enhance the model's ability to express complex data. The second fully connected layer is set with 64 neurons and uses the Sigmoid activation. The Sigmoid function maps the data to a value between 0 and 1, which helps the model output values that meet specific range requirements and further extracts and transforms the features of the data. To prevent the model from overfitting, a Dropout layer with a specified dropout rate is set. For example, when the dropout rate is 0.2, during the training process, this layer will randomly set the outputs of 20% of the neurons to 0, thereby avoiding excessive dependence between neurons and improving the generalization ability of the model.

[0060] The output layer contains only 1 neuron and uses a linear activation method. The model outputs the absolute value of the predicted sales volume through this output layer.

[0061] When training the model, the Huber loss function is selected. The reference formula is:

[0062]

[0063] where δ takes a value of 1.0. When the absolute value of the difference between the true sales volume y and the predicted sales volume is less than or equal to δ, the loss function is When the absolute value of the difference between the true sales volume y and the predicted sales volume is greater than δ, the loss function is

[0064] The Nadam optimizer can be used. The initial learning rate is set to 0.001. After every 10 rounds of training, the learning rate decays by 50%. Each batch processes 32 samples. The maximum number of training rounds is set to 100 rounds, and the early stopping method is used to prevent the model from overfitting. The patience value of the early stopping method is set to 5 rounds, that is, if the mean absolute error of the validation set does not decrease in 5 consecutive rounds of training, the training will be stopped.

[0065] The training execution process includes the following steps: divide the dataset into a training set, a validation set, and a test set according to 7:2:1; perform forward propagation to calculate the predicted sales volume and backward propagation to update the weights; calculate the mean absolute error of the validation set after each round of training. When the validation set has not decreased for 5 consecutive rounds, save the best model and terminate the training.

[0066] In some embodiments, the feature influence weight vector includes: price weight, marketing weight, and model change weight. Among them, the price weight includes the weights of the benchmark price, the current price, and the price reduction range; the marketing weight includes the weight of the remaining time of the marketing activity; the model change weight includes the weights of the model change information and the technical parameter annotation information.

[0067] The price weight includes three sub - weights: the benchmark price, the current price, and the price reduction range, which respectively reflect the influence degrees of the long - term pricing strategies of competing products, the real - time price changes, and the promotion intensity on the sales volume of the target product. The marketing weight consists of a single weight of the remaining time of the marketing activity, quantifying the driving force of the promotion timeliness on consumers' purchase decisions. The model change weight contains two sub - weights: the model change information and the technical parameter annotation information. The weight of the model change information measures the influence of the design iteration of competing products through the change value of appearance similarity, and the weight of the technical parameter is based on the ISO standard term matching results to evaluate the competition intensity of the technical upgrade of competing products.

[0068] After completing the training of the neural network model, the feature influence weight vector is directly extracted from the model parameters in the following way: First, locate the weight matrix between the input layer and the first hidden layer. The column vectors of this matrix correspond to different modules of the input features in sequence. For the price module, according to the original input order of the three features of the benchmark price, the current price, and the price reduction range, extract the first weight value corresponding to each as a sub - weight; for the marketing module, directly extract the first weight value of the remaining time feature; for the model change module, extract the first weight values of the appearance similarity change feature and the technical parameter matching feature respectively. Combine these weight values in the order of price, marketing, and model change, and then form a complete feature influence weight vector.

[0069] S400. Calculate the dot product of the competing product features and the feature influence weight vector to obtain the comprehensive influence factor. The reference formula is: I1 = (P bnorm ×w P +C bnorm ×w C +D norm ×w D )+T norm ×w T +(F×w F +S×w S ; where, I1 represents the comprehensive influence factor, P bnorm , C bnorm , D norm respectively represent the benchmark price, the current price, and the price reduction range of the competing product, w P , w C , w D represent the benchmark price weight, the current price weight, and the price reduction range weight, T norm represents the remaining time of the competing product's marketing activity, w T represents the weight of the remaining time of the competing product's marketing activity, F represents the model change information, S represents the technical parameter annotation information, w F and w S respectively represent the weights of the model change information and the technical parameter annotation information.

[0070] S500. Process the comprehensive impact factor and the historical sales data using the trained long short-term memory network, and output the predicted sales volume of the commodity. In S500:

[0071] In some embodiments, training the long short-term memory network includes:

[0072] S510. Align the comprehensive impact factor and the historical sales data in time to obtain merged data.

[0073] If the prediction target is the time point t + k, where k is an arbitrary integer, the alignment of the input features and the label needs to satisfy:

[0074] 1. The comprehensive impact factor and the historical sales volume data must be strictly taken from the time period before t + k - Δt; where Δt represents the feature window length, such as 30 days, and t + k - Δt > 0, ensuring that the feature window does not contain future data.

[0075] 2. The label (historical sales volume) is always the true value at the time of t + k.

[0076] 3. If the input data is the feature data within the time period of [t s , t e , then t s is: t s = t + k - Δt + 1, and t e is: t + k - L, where L represents the prediction lag period, such as the impact of competitors needs to lag for 2 days to take effect, and the output data is the sales at t + k.

[0077] S520. Construct a three-dimensional tensor data including the number of samples, the time step, and the number of features.

[0078] Construct an output tensor with dimensions (N1, T1, F1), where N1 represents the number of samples, T1 represents the time step, and F1 represents the number of features. The number of samples is the total number of data days minus the time step plus 1; the time step covers the price matrix time window, such as 30 days; the number of features is 2-dimensional, including the comprehensive impact factor value and the historical sales volume of the target commodity.

[0079] S530. Construct an LSTM model, input the three-dimensional tensor data, and use the predicted value of the commodity sales volume as the output to complete the training of the LSTM model.

[0080] The LSTM model is a standard time series prediction network, including an input layer, an LSTM layer, a Dropout layer, a fully connected layer, and an output layer.

[0081] The input layer is used to receive three-dimensional tensors; the LSTM layer can contain 64 memory units, which capture temporal dependencies in the data through a gating mechanism, that is, it can process the correlation information generated as the data changes over time; the Dropout layer randomly discards a specified proportion of neurons, such as 20%; the fully connected layer can be set with 32 neurons, and its main role is to integrate the features extracted by the previous layers to form high-order features; the output layer contains only a single neuron, which is used to output the predicted value of future sales volume.

[0082] During training, historical sales data and comprehensive impact factors over a past period can be used as input data, and the historical sales data over a past period can be used as the true label. The loss function can choose mean squared error, and the optimizer can use the Adam algorithm. The data is input in batches, such as each batch containing 32 samples, and iterative training is carried out for a specified number of rounds, such as 200 rounds. After each round, the performance is evaluated using the validation set. When the validation loss does not decrease for a specified number of consecutive rounds, such as 15 rounds, the early stopping mechanism is triggered to terminate the training.

[0083] S600. Perform strategy optimization based on the target sales volume and the predicted sales volume. In S600:

[0084] In some embodiments, S600 includes:

[0085] S610. Calculate the gap value based on the target sales volume and the predicted sales volume, and determine the gap direction symbol; where, if the gap value is not negative, the gap direction symbol is positive, otherwise the gap direction symbol is negative.

[0086] Exemplarily, G = S t -S p ; where, G represents the gap value, S t represents the preset target sales volume, and S p represents the future sales volume predicted by the model. If Sign represents the gap direction symbol, when G ≥ 0, Sign is 1, otherwise Sign is -1.

[0087] If the prediction fails to reach the target, positive stimulation of sales volume is required. If the prediction exceeds the target, negative suppression of excessive demand is required to prevent situations such as insufficient inventory and out-of-control market prices.

[0088] S620. Calculate the historical average sales volume fluctuation value based on the historical sales data, and take the ratio of the absolute value of the gap to the historical average sales volume fluctuation value as the gap intensity.

[0089] Exemplarily, take the average absolute deviation of the sales volume data in the most recent specified number of days as the historical average sales volume fluctuation value. Taking 90 days as an example, the reference formula: where, σ h represents the historical average sales volume fluctuation value, N represents 90 days, and Si representing the sales volume data of the i-th day, and μ representing the average daily sales volume data; the gap intensity is: where I2 represents the gap intensity and |G| represents the absolute value of the gap value; normalizing the gap amplitude can eliminate the dimensional differences in the sales volumes of different products.

[0090] S630. Multiply the comprehensive influence factor, price weight, and gap intensity to obtain the product as the price base quantity; multiply the gap direction symbol, the hyperbolic tangent function value of the price base quantity, and the preset scaling coefficient, and constrain the obtained product within a specified range to obtain the price adjustment ratio.

[0091] Exemplarily, the price base quantity is: B p = I1 × ||w p || × I2, where B p represents the price base quantity, ||w p || represents the modulus of the price weight vector, and the calculation formula is: where w P 、w C 、w D represent the benchmark price weight, current price weight, and price reduction amplitude weight. The price adjustment ratio is: ΔP = Sign × k p × tanh(B p ), where ΔP represents the price adjustment ratio, k p represents the preset scaling coefficient, which can be 0.15, tanh(·) represents the hyperbolic tangent function, compresses the input to the interval [-1, 1], and ΔP can be constrained to [-15%, 15%].

[0092] The non-linear mapping prevents extreme price adjustments (such as a sudden drop of 30%), maintaining market stability. A positive gap (Sign = 1) triggers a price cut, and a negative gap (Sign = -1) triggers a price increase.

[0093] S640. Multiply the comprehensive influence factor, marketing weight, and gap intensity to obtain the product as the marketing base quantity; input the product of the gap direction symbol and the marketing base quantity into the S-shaped function for non-linear mapping, and constrain the mapping result within a specified interval to obtain the marketing adjustment coefficient.

[0094] Exemplarily, the marketing base quantity is: B m = I1 × ||w T || × I2, where B m represents the marketing base quantity, and w T represents the weight of the remaining time of the competitor's marketing activities. The marketing adjustment coefficient is: where ΔM represents the marketing adjustment coefficient, and k m represents the curvature coefficient, which can be 0.5, Represents the scaled Sigmoid function, with an output range of [-1, 1].

[0095] When B m > 3, ΔM approaches ±1, achieving saturated control of marketing resource investment. For a positive gap (Sign = 1), increase marketing investment, such as increasing the marketing activity time. For a negative gap (Sign = -1), cut the budget, such as reducing the marketing activity time.

[0096] Such as Figure 2 As shown, in some embodiments, the present disclosure provides a sales prediction and strategy optimization system based on big data analysis, which includes: a data collection module, a competitor feature construction module, a neural network training and weight generation module, a comprehensive impact evaluation module, a sales volume prediction module, and a strategy optimization decision-making module.

[0097] The data collection module is used to obtain the historical sales data of the product; obtain competitor data, including price data, marketing activity data, and model change data; the competitor feature construction module is used to construct a competitor feature vector according to the price data, marketing activity data, and model change data; the neural network training and weight generation module is used to take the competitor feature vector as input and the historical sales data of the product as output to complete the training of the neural network model, and obtain a feature impact weight vector according to the trained neural network model; the comprehensive impact evaluation module is used to perform a dot product operation on the competitor feature and the feature impact weight vector to obtain a comprehensive impact factor; the sales volume prediction module is used to process the comprehensive impact factor and the historical sales data by using the trained long short-term memory network and output the predicted sales volume of the product; the strategy optimization decision-making module is used to determine the target sales volume and optimize the strategy according to the target sales volume and the predicted sales volume.

[0098] The sales prediction and strategy optimization system based on big data has the advantages of the sales prediction and strategy optimization method based on big data analysis and can automatically implement all steps of the sales prediction and strategy optimization method based on big data analysis.

[0099] In some embodiments, the present disclosure provides a device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of the sales prediction and strategy optimization method based on big data analysis when executing the computer program.

[0100] In some embodiments, the present disclosure provides a readable storage medium, in which computer program instructions are stored. When the computer program instructions are read and run by a processor, the steps of the sales prediction and strategy optimization method based on big data analysis are executed.

[0101] Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0102] It should be noted that, in this text, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0103] Although the present invention 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 described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A sales forecasting and strategy optimization method based on big data analysis, characterized in that, Including: Obtain the historical sales data of the product; obtain competitor data, including price data, marketing activity data, and model change data; Construct a competitor feature vector based on the price data, marketing activity data, and model change data; Use the competitor feature vector as the input and the historical sales data of the product as the output to complete the training of the neural network model, and obtain a feature influence weight vector according to the trained neural network model; Perform a dot product operation on the competitor features and the feature influence weight vector to obtain a comprehensive influence factor; Use the trained long short-term memory network to process the comprehensive influence factor and the historical sales data, and output the predicted sales volume of the product; Determine the target sales volume, and optimize the strategy according to the target sales volume and the predicted sales volume.

2. The sales prediction and strategy optimization method based on big data analysis according to claim 1, wherein The price data includes the benchmark price, the current price, and the price reduction range; the marketing activity data includes the remaining time of the marketing activity; the model change data includes the model change information and the technical parameter annotation information; Identify the competitors, and use web crawler management software to obtain the price information, promotional activity information, and promotional pictures of the competitors at a specified time; determine the benchmark price and the current price according to the price information, and calculate the price reduction range; Determine the remaining time of the marketing activity according to the promotional activity information; Extract the appearance of the competitor from the promotional pictures, and calculate the similarity of the appearance of the competitor in the promotional pictures obtained twice in a row; Set a similarity threshold. If the similarity is less than the similarity threshold, record the model change information; identify the text in the promotional pictures, and match it with the ISO standard terminology library, and record the technical parameter annotation information that matches the standard terms.

3. The sales prediction and strategy optimization method based on big data analysis according to claim 2, characterized in that, Construct a competitor feature vector according to the price data, marketing activity data, and model change data, including: Construct a price matrix with the benchmark price, the current price, and the price reduction range; Construct a marketing matrix with the remaining time of the marketing activity; Construct a model change matrix with the model change information and the technical parameter annotation information; Construct a competitor feature vector including the price matrix, the marketing matrix, and the model change matrix.

4. The sales prediction and strategy optimization method based on big data analysis according to claim 3, characterized in that Use the competitor feature vector as the input and the historical sales data of the product as the output to complete the training of the neural network model, including: Align the competitor feature vector data and the historical sales data of the product in time; Preprocess the competitor feature vector, including normalizing the benchmark price, the current price, the price reduction range, and the remaining time of the marketing activity, and encoding the model change information and the technical parameter annotation information; Construct a neural network model, use the preprocessed competitor feature vector data as the input data, and use the historical sales data of the product as the output label to train the neural network model until the training is completed.

5. The sales prediction and strategy optimization method based on big data analysis according to claim 3, characterized in that The feature influence weight vector includes: price weight, marketing weight, and model change weight; The price weight includes the weights of the benchmark price, the current price, and the price reduction range; The marketing weight includes the weight of the remaining time of the marketing activity; The model change weight includes the weights of the model change information and the technical parameter annotation information.

6. The sales prediction and strategy optimization method based on big data analysis according to claim 1, characterized in that Train the long short-term memory network, including: Align the comprehensive influence factor and the historical sales data in time to obtain combined data; Construct a three-dimensional tensor data including the number of samples, the time step, and the number of features; Build an LSTM model, input three-dimensional tensor data, and use the predicted value of the commodity sales volume as the output to complete the training of the LSTM model.

7. The method for sales prediction and strategy optimization based on big data analysis according to claim 5, characterized in that, Perform policy optimization based on the target sales volume and the predicted sales volume, including: Calculate the gap value based on the target sales volume and the predicted sales volume, and determine the gap direction symbol; where, if the gap value is not negative, the gap direction symbol is positive, otherwise the gap direction symbol is negative; Calculate the historical average sales volume fluctuation value based on the historical sales data, and use the ratio of the absolute value of the gap to the historical average sales volume fluctuation value as the gap intensity; Use the product obtained by multiplying the comprehensive influence factor, the price weight, and the gap intensity as the price base quantity; multiply the gap direction symbol, the hyperbolic tangent function value of the price base quantity, and the preset scaling coefficient, and constrain the obtained product within a specified range to obtain the price adjustment ratio; Use the product obtained by multiplying the comprehensive influence factor, the marketing weight, and the gap intensity as the marketing base quantity; input the product of the gap direction symbol and the marketing base quantity into the S-shaped function for non-linear mapping, and constrain the mapping result within a specified interval to obtain the marketing adjustment coefficient.

8. A sales forecasting and strategy optimization system based on big data analysis, characterized in that, It includes: A data collection module, which is used to obtain the historical sales data of the commodity; obtain competitor data, including price data, marketing activity data, and model change data; A competitor feature construction module, which is used to construct a competitor feature vector based on the price data, marketing activity data, and model change data; A neural network training and weight generation module, which is used to use the competitor feature vector as the input and the historical sales data of the commodity as the output to complete the training of the neural network model, and obtain a feature influence weight vector according to the trained neural network model; A comprehensive influence evaluation module, which is used to perform a dot product operation on the competitor features and the feature influence weight vector to obtain a comprehensive influence factor; A sales volume prediction module, which is used to use the trained long short-term memory network to process the comprehensive influence factor and the historical sales data, and output the predicted sales volume of the commodity; A policy optimization decision module, which is used to determine the target sales volume and perform policy optimization based on the target sales volume and the predicted sales volume.

9. A device, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of the sales prediction and policy optimization method based on big data analysis according to any one of claims 1-7 when executing the computer program.

10. A readable storage medium, characterized in that, Computer program instructions are stored in a readable storage medium. When the computer program instructions are read and run by a processor, the steps of the sales prediction and policy optimization method based on big data analysis according to any one of claims 1-7 are executed.

Images