Park sales volume prediction method, system and device and storage medium

Abstract

The invention discloses a park sales volume prediction method, system and device and a storage medium, and relates to the technical field of park activity operation, and the method comprises the steps: obtaining the sales volume related data of a target park; performing data processing on the sales related data to construct a spatio-temporal feature vector sequence for model input; inputting the spatial-temporal feature vector sequence into a target sales volume prediction model, and obtaining a sales volume prediction result for the target park activity output by the target sales volume prediction model; and generating a decision suggestion report based on the sales prediction result.

Description

Technical Field

[0001] This application relates to the field of park activity operation technology, and in particular to a method, system, equipment and storage medium for predicting park sales. Background Technology

[0002] In the current field of park operation and management, effectively boosting overall sales and merchant vitality through commercial or cultural events is one of the core challenges faced by operators. Traditional event planning and decision-making processes heavily rely on the personal experience and qualitative judgment of operational staff, lacking scientific and quantitative decision-making basis. For example, when deciding on the type of event, its location, and the scale of resource investment, it is often difficult to accurately predict the potential customer traffic conversion and sales improvement, leading to inefficient resource allocation and situations where insufficient investment fails to achieve the expected results, or excessive investment results in wasted resources. At the same time, this vague decision-making model makes it difficult to clearly communicate the potential value of the event to merchants within the park, hindering their participation and long-term retention.

[0003] Furthermore, although modern industrial parks deploy various data collection devices and information systems capable of collecting data such as merchant sales, customer traffic, merchant attributes, and activity information, this data is typically scattered across different subsystems, varies in format, and contains missing or outlier values. Current technologies lack a systematic data processing and analysis framework, failing to effectively integrate this multi-source, heterogeneous data and transform it into dynamic characteristics that depict the complex relationships between activities, spaces, merchants, and sales. Therefore, it is difficult to build accurate predictive models using historical data, and reliable forward-looking analysis of future activity effects is impossible. This results in operational decisions consistently lagging behind actual business changes, making it difficult to form a data-driven closed loop of "planning-execution-evaluation-optimization." Summary of the Invention

[0004] The embodiments of this application provide a method, system, device, and storage medium for predicting sales revenue in a park, thereby enabling, to a certain extent, quantitative prediction of sales revenue from park activities without relying on human experience and making scientific decisions accordingly.

[0005] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0006] This application specifically includes the following aspects:

[0007] Firstly, this application proposes a method for forecasting sales revenue in industrial parks, including:

[0008] Obtain sales data related to the target park;

[0009] The sales data is processed to construct a spatiotemporal feature vector sequence for model input;

[0010] The spatiotemporal feature vector sequence is input into the target sales revenue prediction model to obtain the sales revenue prediction result for the target park activity output by the target sales revenue prediction model.

[0011] Based on the sales forecast results, a decision-making recommendation report is generated.

[0012] In one feasible implementation, obtaining sales-related data of the target park includes:

[0013] Real-time pedestrian flow statistics and store entry data are collected using sensors in various areas of the target park.

[0014] The sales data of each merchant is obtained through the merchant POS system of the target park.

[0015] The management system of the target park is used to obtain merchant attribute data, pedestrian flow information data, activity plan data, and geospatial data.

[0016] In one feasible implementation, the data processing of the sales-related data includes:

[0017] Identify whether there are any missing data in the sales-related data;

[0018] If missing pedestrian traffic data is identified during non-activity periods, the sales-related data will be filled in using the results of linear regression fitting based on historical data from the same period.

[0019] If missing sales data is identified during the event, a weighted regression model that integrates event intensity and foot traffic data is used to fill in the missing sales data.

[0020] In one feasible implementation, the data processing of the sales-related data to construct a spatiotemporal feature vector sequence for model input further includes:

[0021] Based on the imputed sales data, a format standardization operation is performed to obtain the standardized sales data. The format standardization operation includes geographic coordinate format conversion, time format unification, and data unit unification.

[0022] Based on preset feature classification rules, categorical data and numerical data are determined from the standardized sales data.

[0023] Based on the categorized data, perform a classification coding operation to determine the categorized feature vector;

[0024] Based on the numerical data, perform numerical standardization to determine the numerical feature vector;

[0025] Based on the categorical feature vector and the numerical feature vector, feature combination is performed to obtain the spatiotemporal feature vector sequence used for model input.

[0026] In one feasible implementation, the target sales forecasting model is a time-series forecasting model based on a long short-term memory network. Before inputting the spatiotemporal feature vector sequence into the target sales forecasting model, the method further includes:

[0027] Based on the spatiotemporal feature vector sequences of historical periods and their corresponding actual sales data, a model training set, a model validation set, and a model test set are constructed.

[0028] Based on the model training set and the preset root mean square error loss function, the time series prediction model is iteratively trained to determine the preliminary training model.

[0029] Based on the model validation set, the parameters of the initially trained model are adjusted to determine the adjusted model;

[0030] Based on the model test set, the generalization ability of the adjusted model is evaluated until the target requirements are met, thus obtaining the target sales forecast model.

[0031] In one feasible implementation, generating a decision recommendation report based on the sales forecast results includes:

[0032] Obtain the configuration information of the target activity plan corresponding to the sales forecast result, wherein the configuration information includes activity type, activity area and resource investment scale;

[0033] Based on the sales forecast results and the configuration information, the decision recommendation report is generated.

[0034] In one feasible implementation, after generating the decision recommendation report based on the sales forecast results, the method further includes:

[0035] Collect actual sales data and foot traffic data;

[0036] Based on the actual sales data and the sales forecast results, an error calculation operation is performed to determine the forecast error;

[0037] Based on the prediction error, a model optimization operation is performed on the target sales forecast model;

[0038] The actual sales data and pedestrian flow data are stored to update the construction of the spatiotemporal feature vector sequence.

[0039] Secondly, this application also proposes a sales forecasting system for industrial parks, the system comprising:

[0040] The data acquisition module is used to obtain sales data related to the target park.

[0041] The data processing module is used to process the sales-related data to construct a spatiotemporal feature vector sequence for model input.

[0042] The sales forecasting module is used to input the spatiotemporal feature vector sequence into the target sales forecasting model and obtain the sales forecasting results for the target park activities output by the target sales forecasting model.

[0043] The results generation module is used to generate a decision recommendation report based on the sales forecast results.

[0044] Thirdly, this application also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the park sales forecasting method as described in any of the first aspects above.

[0045] Fourthly, this application also proposes a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of any of the park sales forecasting methods of the first aspect.

[0046] This application proposes a method for predicting sales revenue in industrial parks. By systematically acquiring and integrating multi-source heterogeneous operational data within the park, and constructing a feature vector sequence that can deeply characterize spatiotemporal relationships, the method then uses a trained prediction model to accurately predict sales revenue. Finally, based on the quantitative prediction results, it generates decision-making suggestions, thereby transforming the original process of event planning and resource allocation, which relied on subjective experience, into a data-driven and prediction-oriented scientific decision-making process. This effectively solves the problems of insufficient decision-making basis, low resource allocation efficiency, and difficulty in assessing the expected value of activities under traditional methods, and significantly improves the accuracy of park operation and management and the return on investment of business activities.

[0047] This application proposes a method for forecasting sales revenue in industrial parks. Other advantages, objectives, and features of this application will be partly apparent from the following description and partly understood by those skilled in the art through research and practice of this application. Attached Figure Description

[0048] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0049] Figure 1 A flowchart illustrating a method for forecasting sales revenue in a park, provided as an embodiment of this application;

[0050] Figure 2 A schematic diagram of the functional modules of a park sales forecasting system provided in this application embodiment;

[0051] Figure 3 This is a schematic diagram of the structure of a park sales forecasting device provided in an embodiment of this application. Detailed Implementation

[0052] To better understand the technical solutions provided in the embodiments of this specification, the technical solutions of the embodiments of this specification will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this specification and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this specification, rather than limitations on the technical solutions of this specification. In the absence of conflict, the embodiments of this specification and the technical features in the embodiments can be combined with each other.

[0053] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The term "two or more" includes two or more cases.

[0054] Please see Figure 1 This is a flowchart illustrating a method for forecasting sales revenue in a park, provided in an embodiment of this application. Specifically, it may include:

[0055] S110. Obtain sales data related to the target park.

[0056] For example, within the target park, basic sales data and basic data of related sub-businesses are acquired through hardware data collectors and software subsystems. This facilitates more in-depth analysis of sales data and key business activities that influence it. This data is then used as sales-related data, and daily sales-related data is integrated into a unified database to form a historical database for subsequent analysis and processing.

[0057] In some examples, sales data related to the target park is retrieved, including:

[0058] Real-time pedestrian flow statistics and store entry data are collected through sensors in various areas of the target park.

[0059] Obtain sales data from each merchant through the merchant POS system in the target park;

[0060] By using the target park's management system, we can obtain merchant attribute data, pedestrian flow data, activity plan data, and geospatial data.

[0061] For example, given the current economic downturn and rising vacancy rates in commercial office buildings, the park urgently needs to promote the coordinated development of culture, industry, and the economy through various activities. Under this demand, the park's operation faces three core pain points: first, how to accurately target high-yield merchants for activities to maximize benefits; second, the lack of quantitative basis for setting activity duration and promotional intensity, easily leading to either "over-investment" wasting resources or "insufficient promotion" failing to meet targets; and third, the lack of data support for activity type selection, making it difficult to identify the optimal "cost-effective" solution.

[0062] To address these issues, this application utilizes hardware data collection devices linked to multiple systems to comprehensively collect sales data within the park. The hardware data collection devices include sensors and POS machines. Infrared sensors installed by merchants in various areas of the target park collect data such as store foot traffic (number of customers entering the store), total park foot traffic statistics (real-time foot traffic statistics), and customer dwell time. Sales data collection devices are installed on the POS machines within each merchant to collect sales data from each merchant.

[0063] The management system connects to multiple subsystems, including: a business data system, a pedestrian flow system, an event management subsystem, a space subsystem, and an integrated investment promotion and operation management subsystem. Based on these subsystems, it collects merchant information (i.e., merchant attribute data, including merchant name and business type classification), geographic coordinates (i.e., geospatial data, including park building geographic coordinates and event space area), event information (i.e., event plan data, including event type, number of participants, event location, and event scale), and pedestrian flow data (including the age and origin of people entering the park) to obtain daily sales data for merchants. This data serves as the foundation for park sales figures. Operations management can refer to the pedestrian flow, sales, and year-on-year trends for each merchant based on data from each event. Based on this data, it can predict sales figures for future park events, avoiding excessive investment.

[0064] Based on this, AI deep learning technology is used to conduct in-depth analysis of multi-source data, accurately predict the sales revenue that each event may generate, and ultimately provide scientific guidance for park operation from a data perspective, helping the park achieve the goal of "organizing the right events, empowering merchants, and increasing revenue".

[0065] Furthermore, after acquiring the basic data from the hardware and software subsystems, it is necessary to extract key historical data (such as merchant information, geographic coordinates, sales volume, activity time, type, number of participants, etc.) to construct a unified basic data matrix. This process aims to achieve data alignment from multiple data sources. Specifically, the data provided by each subsystem can be viewed as a feature vector. For example, data from a certain subsystem can be represented as a feature set X = {x1, x2, x3, x4, x5...., x...} n}, where n represents the number of attributes (features) provided by the subsystem, x1, x2, x3, x4, x5, ..., x n These are the relevant attribute values ​​obtained; similarly, the data of another subsystem can be represented as another feature set Y = {y1, y2, y3, y4, y5, ..., y6}. n}, y1, y2, y3, y4, y5...., y n These are the relevant attribute values ​​obtained for this system. By aligning and integrating these feature sets from different subsystems, a multi-dimensional basic data matrix can be constructed for subsequent analysis and modeling.

[0066] Arranging the feature vectors of basic data related to sales revenue across multiple parks is to effectively organize and represent the complex, multi-dimensional characteristic information of each merchant into a continuous data sequence, facilitating subsequent analysis, modeling, and application. Here, the target feature vectors for sales revenue across multiple classification dimensions typically contain various key features extracted from activity information and geographic coordinates. These features may cover aspects such as the merchant's spatial location, activity type, scale, business type, and daily sales revenue. By arranging these feature vectors, the characteristic information of merchants for each activity period can be organized in a certain order or time sequence, forming a time-space level data sequence. This helps in predicting the temporal order or correlation of park sales revenue, further reflecting the impact of park activities on sales revenue forecasting trends. For example, if the activity scale, category, and distance of merchants from the activity are known, the spatiotemporal feature vector sequence can display the merchant's sales revenue and its changes during different activity periods. This serialized representation helps in identifying the impact of upcoming activities on sales revenue, recognizing long-term and short-term behavioral patterns, and formulating personalized marketing and service strategies accordingly.

[0067] S120. Perform data processing on sales-related data to construct a spatiotemporal feature vector sequence for model input.

[0068] In some examples, data processing is performed on sales-related data, including:

[0069] Identify whether there are missing data in sales-related data;

[0070] If missing pedestrian traffic data is identified during non-activity periods, the sales-related data will be filled in using the results of linear regression fitting based on historical data from the same period.

[0071] If missing sales data is identified during the event, a weighted regression model that integrates event intensity and foot traffic data is used to fill in the missing sales data.

[0072] For example, to address the missing data in sales-related data, the sales-related data is cleaned, missing values ​​are filled, outliers are handled, and the data is standardized to provide a foundational data base for subsequent LSTM models (time series prediction models based on long short-term memory networks).

[0073] Specifically, during non-activity periods, pedestrian flow exhibits a stable linear change over time. Historical data is used to fit a linear relationship between time and pedestrian flow. The pedestrian flow filling formula is x. t = a.t+b, x t Let a be the time coefficient (slope) for filling the crowd at time t, and b be the intercept.

[0074] In cases where there are no holidays or events, and daily sales figures are occasionally missing, supplementary measures should be taken. N represents the number of days before and after the event. If it is the end of the month, the average of the previous N days is used.

[0075] Furthermore, for holidays without major events, missing sales figures are filled using linear regression based on pedestrian traffic data. That is, the formula for supplementing daily sales figures for holidays and events is... This is to fill in missing sales figures; X is the daily foot traffic data; β0 is the merchant's basic cost; β1 is the regression coefficient, representing the sales contribution per unit of foot traffic. n is the sample size; X represents the daily foot traffic data, and Y represents the sales revenue data. XY is the product of foot traffic and sales revenue. The average number of people, This represents the average sales revenue.

[0076] In cases where there is a lack of crowd data during major events with surges in visitor numbers, data processing needs to integrate historical event intensity with real-time visitor flow characteristics. First, intensity indicators such as event scale and promotional investment are extracted. Then, weighted regression is used to quantify the dynamic correlation of visitor flow. Finally, imputation values ​​are generated by combining the surge in visitor flow on the day with the event score, preserving both the explosive characteristics and ensuring data consistency.

[0077] Specifically, when the park hosts an event of known scale (large, medium, or small), if visitor flow data for a particular day during the event is missing, historical data from events of similar scale is used to fit a typical pattern of visitor flow changes over time, and intelligent predictions are made based on this. First, based on historical records of similar-scale events (all considered "large"), analysis reveals that the change in visitor flow over time (day 1, day 2, etc.) is not a simple linear relationship, but may exhibit a non-linear pattern of rapid initial growth followed by slow decline. Therefore, a quadratic polynomial is used... To fit this trend. Where d represents the day of the activity, and x... d (s) The pedestrian flow is used to fill the gaps for the large-scale event on day d, and the coefficients c2, c1, and c0 are obtained by performing multinomial regression fitting on historical data of similar events. Secondly, to further optimize the accuracy of the filling, an adjustment factor (i.e., the peak coefficient k of similar-scale events over the years) is incorporated. s This coefficient reflects the average peak level or growth intensity that a specific scale of activity can historically achieve. It is used to fine-tune the results of the polynomial forecast above so that the filler value conforms to both the short-term change curve and the long-term empirical level of this type of activity.

[0078] During major events, foot traffic can increase more than tenfold. In this case, the supplementary weighted regression formula is: Missing value of sales revenue Y = β0 + β1X + β2S, β1 = [n∑wXY) - Σ(wX)Σ(wY)] / [n∑(wX) 2 ))-(∑wX) 2 [; where Y is sales data; X is daily foot traffic data; β0 is the merchant's basic cost; β1 is the sales contribution per unit of foot traffic; β2 is the additional contribution per unit of activity intensity; S is the current activity intensity score (1-10 points).]

[0079] The above can be understood as follows: In response to the "surge in foot traffic - sales explosion" characteristic of major events, this application introduces an event intensity variable and weighted regression to both preserve the basic correlation between foot traffic and sales and quantify the additional boost the event itself provides to sales (e.g., during large-scale events, the sales generated per 100 people are 30% higher than on ordinary days). For missing values, a filler value reflecting both foot traffic conversion and event increment is calculated based on the actual surge in foot traffic and the scale of the event, avoiding the underestimation of the "event explosion effect" by conventional regression. The event classification supplementary rules include defining the event as a large-scale event with a minimum of 5000 participants; if the date is a national statutory holiday, the event date is automatically supplemented with the day before and after the holiday for event time determination.

[0080] In some examples, data processing of sales-related data to construct a sequence of spatiotemporal feature vectors for model input also includes:

[0081] Based on the imputed sales data, a format standardization operation is performed to obtain the standardized sales data. The format standardization operation includes geographic coordinate format conversion, time format unification, and data unit unification.

[0082] Based on preset feature classification rules, categorical data and numerical data are determined from the standardized sales data.

[0083] Based on the categorized data, perform classification coding operations to determine the categorized feature vectors;

[0084] Based on numerical data, numerical standardization is performed to determine numerical feature vectors;

[0085] Based on categorical and numerical feature vectors, feature combination is performed to obtain a spatiotemporal feature vector sequence for model input.

[0086] For example, perform format standardization operations on the supplemented sales data. This includes converting GIS coordinates to decimal; standardizing time series data based on time characteristics such as activity time, converting data across days and with inconsistent time formats to hours; converting sales data strings to floating-point numbers in ten thousand yuan; and quantifying the time synergy effect by setting feature values ​​0 and 1 for peak merchant periods and activity times, i.e., the activity time interval [T]. h_start ,T h_end Merchant peak time interval [T] s_start, T s_end ],

[0087] Furthermore, based on the categorized data, a classification coding operation is performed to determine the categorized feature vectors. Specifically, the categorized data is encoded, including data on activity type, merchant type, merchant region, age distribution, and origin distribution. By obtaining the transformed park sales data and feature vectors affecting sales data, it is easier to use for mathematical calculations and model training, helping park operators analyze the relationship between activities and industries. Through numerical vectors, park operators can classify and group sales based on the common characteristics of various park activities. For example, clustering algorithms can be used to group merchants of the same type together, analyze the impact of activities on this type of business, and thus provide decision-making suggestions for operators on the scale and investment of activities.

[0088] To avoid "type differences" in the matrix, such as event scale, event type, merchant business type, number of participants, etc., the event type is classified into unordered categories using one-hot encoding, from [0, 0, 1] to [1, 1, 0]. The event categories are ordered: large events = 1; small events = 0; holidays are unordered categories, marked 0 / 1, indicating holidays / non-holidays.

[0089] For categorical features with an inherent order, such as "time period" (e.g., morning, afternoon, evening) or "age segment" (e.g., youth, middle-aged, elderly), they need to be converted into numerical values ​​that the model can process. This embodiment uses a weighted encoding method, the core of which is to assign a representative value (e.g., the average index of that segment) to each ordinal category. Its proportion in the total P i Combined, a weighted average is calculated as the encoded value of this feature, using the following formula: in, This is the calculated weighted encoding value, which comprehensively reflects the overall level of ordered classification and can be used for subsequent modeling; n is the number of segments in the ordered classification. P represents the mean of a specific indicator for the i-th segment (e.g., an age group); i Let be the proportion of the i-th age group in the total population (satisfying ∑P)i =1).

[0090] Furthermore, to quantify the attractiveness of activities within the park to the surrounding space, this embodiment divides the area into three levels of radiation zones based on distance from the activity venue. A pedestrian density threshold is defined for each zone to assess the activity's popularity gathering effect or as a basis for constructing spatial characteristics: Core Zone (0-50 meters): Pedestrian density threshold ρ ≥ 15 (e.g., unit: people / 100 square meters). This zone is expected to have dense pedestrian traffic. Secondary Core Zone (51-100 meters): Pedestrian density threshold 8 ≤ ρ ≤ 15. Peripheral Radiation Zone (101-150 meters): Pedestrian density threshold 5 ≤ ρ ≤ 8. The pedestrian density threshold is mainly used to define business rules or construct spatial characteristics. For example, it can be used to determine whether a certain area reaches the expected popularity during an activity, or as a feature to characterize the popularity level of a merchant / area within the activity's radiation zone.

[0091] Furthermore, based on numerical data, numerical standardization is performed to eliminate differences in magnitude, including the number of participants in major events, sales revenue (holidays and non-holidays), and foot traffic. By eliminating the differences in the units of measurement between features, features of different magnitudes have equal weight in model training, thereby improving prediction accuracy. Based on the gradient changes in the analyzed business based on the quantitative situation, a standard score (Z-Score) normal distribution is used to provide numerical feature standardization, with the formula: r1 = (R - MIN) / (MAX - R), where r1 is the standard data; MAX is the maximum value; MIN is the minimum value; and R is the data being processed, such as the number of participants in an event, the area of ​​a merchant, etc.

[0092] The above can be understood as follows: In the scenario of predicting sales in industrial parks, major events (such as National Day and Spring Festival promotions) may lead to extreme peaks in daily sales and foot traffic (e.g., daily sales of 500,000 yuan, reaching 2 million yuan on the event day). These extreme values ​​represent real business data, maintaining the stability of the distribution of most regular data. This allows the model to learn the patterns of extreme scenarios without ignoring the characteristics of regular business. Simultaneously, sales predictions for multiple industrial parks on the platform show differences across different time periods and for each park annually. For example, comparing data from 2023 and 2025, industrial park sales were significantly less affected by the pandemic in 2023, but were severely impacted in 2024. To ensure fair comparison across dimensions regarding the impact of changes in average foot traffic, the platform balances the weights of multiple features, even when the event scale is around 5,000 people compared to a typical industrial park foot traffic of 30,000, preventing the scale from dominating the overall picture.

[0093] Furthermore, after completing the basic data processing, in order to further explore the temporal patterns and business characteristics contained in the data, feature derivation operations need to be performed. This process, based on the historical time-based matrix, constructs a combined feature matrix through business logic analysis, for example, forming a matrix such as XY = {x1, x2, x3, y4, y5, ..., y...}. n The combination representation of}, where XY is the combination matrix, x1, x2, x3, y4, y5, ..., y n This is a related historical combination matrix formed under the same identifier. Each element represents a summary of historical association features under the same identifier. Specific derivation methods include statistical feature calculations, such as calculating the 7-day average of historical sales, with the formula: 7-day average = ∑7-day sales ÷ 7; and calculating pedestrian flow volatility, with the formula: pedestrian flow volatility = ∑ ( (Daily foot traffic - average foot traffic over the past N days) ÷ average foot traffic over the past N days × 100%, where N is an adjustable window period, such as 3 days, 7 days, or 14 days.

[0094] For scenarios involving a surge in pedestrian traffic during holidays or specific events, the above volatility calculation can be adjusted to use the median pedestrian traffic over the past N days as a benchmark. The formula then becomes: Pedestrian Traffic Fluctuation = ∑ ( (Daily foot traffic - Median foot traffic over the past N days) ÷ Median foot traffic over the past N days × 100%. Furthermore, during the event, volatility analysis can be conducted by region, for example, using Z-score volatility calculation: Z-score volatility = ∑ ( (Daily pedestrian flow - average pedestrian flow over the past N days) ÷ median pedestrian flow over the past N days × 100%. These derived features, by integrating time window statistics, special scenario adjustments, and stability measures, transform basic data into higher-order features that can characterize short-term trends, activity burst effects, and regional anomalies, thereby providing richer and more business-interpretive input information for subsequent predictive models.

[0095] S130. Input the spatiotemporal feature vector sequence into the target sales revenue prediction model and obtain the sales revenue prediction results for the target park activities output by the target sales revenue prediction model.

[0096] For example, after processing sales-related data, feature extraction needs to be performed to construct the model input. Specifically, the goal of feature extraction is to identify key factors affecting sales from multi-source data, including: sales data of each merchant, activity plan data (activity information, such as type, scale, location, etc.), and real-time pedestrian flow statistics. From this data, basic features (describing the inherent attributes of the object, such as merchant coordinates, merchant type, and peak operating hours), time series features (reflecting the changes of indicators over time, such as historical sales series and pedestrian flow time series data), activity features (describing the attributes of the activity itself, such as activity type, classification, and scale), and spatial and relational features (characterizing geographical and business relationships, such as the distance between the merchant and the activity venue, and the clustering of merchants / sales within the region). These features, after being encoded and standardized, together constitute the components of the feature vector. To integrate the above features and mine their deep patterns, this embodiment uses a feature extraction network containing parallel convolutional layers. This network processes data using convolutional kernels of different scales (first-scale and second-scale one-dimensional convolutional kernels) to obtain first-scale feature vectors and second-scale feature vectors. These two feature vectors are then concatenated to generate a sequence of spatiotemporal feature vectors that comprehensively express the spatiotemporal correlation of park sales revenue. This high-level feature vector will be used as input to train and infer time-series prediction models such as LSTM, ultimately enabling the prediction of future park activity sales revenue.

[0097] Specifically, feature extraction aims to help operations personnel identify and capture potential patterns and regularities in park sales data. Park sales data contains various types of information and time-series dimensions. For example, merchant consumption data includes details of each transaction and the time of consumption; while data influencing park consumption covers large-scale events, the distance between merchants and event venues, business types, and pedestrian traffic. Different feature combinations can be extracted from this data: including basic features such as merchant coordinates, business type, and peak start and end times; time-series features such as event name, start time, and pedestrian traffic sequence; and basic event features such as event name, category, and scale. Time-series data specifically includes merchant sales time-series features (sales volume, merchants, region, time), daily regional sales time-series features, and daily event participant traffic. In addition, it is necessary to extract sequence pattern features, such as changes in pedestrian traffic before and after events, sales fluctuations on holidays or weekends; and spatial sequence features, such as the distance between merchants and event venues, and regional clustering features based on region, business type, merchants, sales volume, and time. These fundamental inherent properties constitute the basic classification of features, while time-space temporal features provide a dynamic perspective on data changes.

[0098] To integrate this information, a time-space multi-dimensional data table needs to be constructed. Its elements include the activity coordinate grid, merchants within the grid, business types, peak operating times, coordinates, activity start and end times, merchant peak operating times, the matching degree between the activity and merchant peak times, activity duration, and foot traffic. By extracting business logic, specific sequence features can be established, such as: linking the activity with historical foot traffic data from multiple time points, regional and merchant foot traffic data, and combining this with information such as activity start time, end time, day of the activity, merchant coordinates within the grid, and sales revenue.

[0099] To achieve the aforementioned feature extraction, this embodiment employs a feature extraction network containing multi-scale convolutional layers. This network structure includes: a first convolutional layer, a second convolutional layer running parallel to the first convolutional layer, and a cascaded layer connected to these two convolutional layers. The first convolutional layer uses a one-dimensional convolutional kernel with a first scale, and the second convolutional layer uses a one-dimensional convolutional kernel with a second scale. Specifically, the first convolutional layer of the feature extraction network performs one-dimensional convolutional encoding on sales-related data items to obtain a first-scale feature vector; simultaneously, the second convolutional layer of the same network performs one-dimensional convolutional encoding on multiple merchant activity data items to obtain a second-scale feature vector; finally, the first-scale feature vector and the second-scale feature vector are concatenated to obtain a spatiotemporal correlation feature vector of park sales that comprehensively reflects complex relationships. This feature vector will be used as input to a deep learning model such as LSTM for processing, to learn from the data and generate a more advanced business feature representation, ultimately achieving the prediction of park activity sales at a specific future point in time.

[0100] Sales forecasts for park activities can be constructed using specific matrices based on different analytical granularities (i.e., forecast targets). Specifically: Daily total sales forecast for the park: its feature matrix mainly includes total park foot traffic, activity type, number of participants, and advertising investment. Daily sales forecast for each area: based on the overall park forecast features, additional features such as specific area foot traffic, distance between the area and the activity venue, and the business composition of merchants within the area are added. Sales forecast for each business type in the park: its features include total park foot traffic, activity type, number of participants, and advertising investment. To quantify the forecast results, this embodiment provides a forecast formula as follows: in, Forecasted sales revenue (e.g., total sales revenue of the park, regional sales revenue, or sales revenue of a specific business type); ΔX represents the park's foot traffic; n represents the total number of merchants in the park; C i Let M be the business format conversion coefficient for the i-th merchant, used to quantify the consumer conversion capability of that business format; i R represents the business type matching score for the i-th merchant, used to measure the degree of fit between the merchant's business type and the current activity type; iLet be the historical foot traffic capture rate of the i-th merchant, reflecting its historical efficiency in converting visitors into actual customers. Based on the total foot traffic in the park, this model comprehensively estimates the potential total sales revenue by weighted summing of the business conversion capabilities, activity matching rates, and historical capture rates of all merchants within the park.

[0101] In some examples, the target sales forecasting model is a time-series forecasting model based on a long short-term memory network. Before inputting the spatiotemporal feature vector sequence into the target sales forecasting model, the following steps are also included:

[0102] Based on the spatiotemporal feature vector sequences of historical periods and their corresponding actual sales data, a model training set, a model validation set, and a model test set are constructed.

[0103] Based on the model training set and the preset root mean square error loss function, the time series prediction model is iteratively trained to determine the initial training model.

[0104] Based on the model validation set, the parameters of the initially trained model are adjusted to determine the adjusted model;

[0105] Based on the model test set, the generalization ability of the adjusted model is evaluated until the target sales forecast model is obtained.

[0106] In some examples, after generating a decision-making recommendation report based on sales forecasts, the following is also included:

[0107] Collect actual sales data and foot traffic data;

[0108] Based on the actual sales data and the sales forecast results, perform an error calculation operation to determine the forecast error;

[0109] Based on the prediction error, perform model optimization operations on the target sales forecast model;

[0110] The actual sales data and pedestrian flow data are stored to update the construction of the spatiotemporal feature vector sequence.

[0111] For example, after predicting sales revenue for the park and various regions based on the LSTM model, the following error analysis and model interpretation process can be performed to further analyze prediction errors and optimize the model. The core purpose of this process is to help locate the reasons for the deviations in the LSTM predictions and to optimize the feature vectors by constructing a more easily interpretable auxiliary model, thereby deducing the relationship from macro-level park sales revenue to micro-level merchant sales revenue. First, calculate and analyze the residuals. For all training or validation samples, calculate the difference between the model's predicted values ​​and the true values ​​to obtain the residual sequence e = [e1, e2, e3, ..., e...]. nSubsequently, we analyzed whether these residuals were systematically correlated with certain features (such as the increase in pedestrian flow and the duration of activities) in order to determine whether the model had overlooked some important patterns.

[0112] Secondly, an interpretative auxiliary model is constructed. To quantify the impact of each feature on sales revenue and understand the composition of the error, a multiple linear relationship model can be built. This model will predict the park's sales revenue. Represented as a series of features x = [x1, x2, ... x n ] T A linear combination of (for example, x1 can represent the number of people) is in the form of: Where w = [w1, w2, ... w n ] T For the weight vector, each w i Represents the i-th feature pair The magnitude of the influence, where b is the global intercept.

[0113] Next, the model parameters are solved. To find the optimal weight coefficients w and b, the least squares method is used. The formula is expressed as: W = (X... T X) -1 X T Here, X is an n×(d+1) feature matrix (n is the number of samples, d is the number of features, and each row is the feature of one sample + 1), and y is an n×1 true target value vector.

[0114] Finally, the optimization effect is evaluated. Root mean square error (RMSE) is used as the key evaluation metric to measure the deviation between the predicted and actual values. The formula for calculating RMSE is: The smaller the RMSE value, the smaller the overall deviation between the model's predicted value and the true value, and the better the model performance. Through the above analysis, feedback can guide the feature selection and structure adjustment of the LSTM model, forming an optimization closed loop.

[0115] S140. Based on the sales forecast results, generate a decision recommendation report.

[0116] In some examples, decision-making recommendation reports are generated based on sales forecasts, including:

[0117] Obtain the configuration information of the target activity plan corresponding to the sales forecast results, including the activity type, activity area and resource investment scale;

[0118] Based on sales forecasts and configuration information, a decision-making recommendation report is generated.

[0119] In summary, this application first obtains basic data collected by the subsystem and sales data related to the park collected by the database. Then, it uses AI deep learning technology to extract features and perform correlation analysis on the two. Finally, it uses time-space time series features to predict the sales of a future activity, avoiding "over-investment" or "insufficient promotion"; and how to select "high-cost-performance" activity types.

[0120] Furthermore, this application also proposes a park sales forecasting system, an embodiment for executing any of the above-mentioned park sales forecasting methods, specifically as follows: Figure 2 The diagram shown is a functional module illustration of a park sales forecasting system proposed in this application. The system includes:

[0121] Data acquisition module 21 is used to acquire sales data related to the target park;

[0122] The data processing module 22 is used to process sales-related data to construct a spatiotemporal feature vector sequence for model input;

[0123] Sales forecasting module 23 is used to input the spatiotemporal feature vector sequence into the target sales forecasting model and obtain the sales forecasting results for the target park activities output by the target sales forecasting model.

[0124] Result generation module 24 is used to generate a decision recommendation report based on the sales forecast results.

[0125] It should be noted that the above embodiments are merely best examples and are not intended to limit the implementation of this application.

[0126] Furthermore, such as Figure 3 As shown, this application embodiment also provides an electronic device 300, including a processor 310, a memory 320, and a computer program 321 stored in the memory 320 and executable on the processor. When the processor 310 executes the computer program 321, it implements the steps of any of the above-mentioned park sales forecasting methods.

[0127] Since the electronic device described in this embodiment is the device used to implement a method for predicting sales in a park according to the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiments of this application is within the scope of protection of this application.

[0128] In practical implementation, when the computer program 321 is executed by the processor, it can achieve the following: Figure 1Any of the corresponding implementation methods in the embodiments.

[0129] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0130] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-readable program code.

[0131] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0132] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0133] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0134] This application also provides a computer program product, which includes computer software instructions that, when executed on a processing device, cause the processing device to execute a process of an automatic vehicle lighting control method.

[0135] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0136] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0137] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0138] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0139] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0140] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0141] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions 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 this application.

[0142] Although preferred embodiments have been described in this specification, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this specification.

[0143] Obviously, those skilled in the art can make various modifications and variations to this specification without departing from its spirit and scope. Therefore, if such modifications and variations fall within the scope of the claims and their equivalents, this specification is also intended to include such modifications and variations.

Claims

1. A park sales amount prediction method characterized by, include: Obtain sales data related to the target park; The sales data is processed to construct a spatiotemporal feature vector sequence for model input; The spatiotemporal feature vector sequence is input into the target sales revenue prediction model to obtain the sales revenue prediction result for the target park activity output by the target sales revenue prediction model. Based on the sales forecast results, a decision-making recommendation report is generated.

2. The method according to claim 1, characterized in that, The acquisition of sales data related to the target park includes: Real-time pedestrian flow statistics and store entry data are collected using sensors in various areas of the target park. The sales data of each merchant is obtained through the merchant POS system of the target park. The management system of the target park is used to obtain merchant attribute data, pedestrian flow information data, activity plan data, and geospatial data.

3. The method according to claim 1, characterized in that, The data processing of the sales-related data includes: Identify whether there are any missing data in the sales-related data; If missing pedestrian traffic data is identified during non-activity periods, the sales-related data will be filled in using the results of linear regression fitting based on historical data from the same period. If missing sales data is identified during the event, a weighted regression model that integrates event intensity and foot traffic data is used to fill in the missing sales data.

4. The method according to claim 3, characterized in that, The step of processing the sales-related data to construct a spatiotemporal feature vector sequence for model input further includes: Based on the imputed sales data, a format standardization operation is performed to obtain the standardized sales data. The format standardization operation includes geographic coordinate format conversion, time format unification, and data unit unification. Based on preset feature classification rules, categorical data and numerical data are determined from the standardized sales data. Based on the categorized data, perform a classification coding operation to determine the categorized feature vector; Based on the numerical data, perform numerical standardization to determine the numerical feature vector; Based on the categorical feature vector and the numerical feature vector, feature combination is performed to obtain the spatiotemporal feature vector sequence used for model input.

5. The method according to claim 1, characterized in that, The target sales revenue prediction model is a time-series prediction model based on a long short-term memory network. Before inputting the spatiotemporal feature vector sequence into the target sales revenue prediction model, the method further includes: Based on the spatiotemporal feature vector sequences of historical periods and their corresponding actual sales data, a model training set, a model validation set, and a model test set are constructed. Based on the model training set and the preset root mean square error loss function, the time series prediction model is iteratively trained to determine the preliminary training model. Based on the model validation set, the parameters of the initially trained model are adjusted to determine the adjusted model; Based on the model test set, the generalization ability of the adjusted model is evaluated until the target requirements are met, thus obtaining the target sales forecast model.

6. The method according to claim 1, characterized in that, The step of generating a decision recommendation report based on the sales forecast results includes: Obtain the configuration information of the target activity plan corresponding to the sales forecast result, wherein the configuration information includes activity type, activity area and resource investment scale; Based on the sales forecast results and the configuration information, the decision recommendation report is generated.

7. The method according to claim 1, characterized in that, After generating the decision recommendation report based on the sales forecast results, the method further includes: Collect actual sales data and foot traffic data; Based on the actual sales data and the sales forecast results, an error calculation operation is performed to determine the forecast error; Based on the prediction error, a model optimization operation is performed on the target sales forecast model; The actual sales data and pedestrian flow data are stored to update the construction of the spatiotemporal feature vector sequence.

8. A sales forecasting system for industrial parks, characterized in that, The system includes: The data acquisition module is used to obtain sales data related to the target park. The data processing module is used to process the sales-related data to construct a spatiotemporal feature vector sequence for model input. The sales forecasting module is used to input the spatiotemporal feature vector sequence into the target sales forecasting model and obtain the sales forecasting results for the target park activities output by the target sales forecasting model. The results generation module is used to generate a decision recommendation report based on the sales forecast results.

9. An electronic device, comprising: The memory and processor are characterized in that the processor is used to implement the steps of the park sales forecasting method as described in any one of claims 1 to 7 when executing a computer program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the park sales forecasting method as described in any one of claims 1 to 7.

Images