Big data dynamic price index prediction system and method considering spatio-temporal correlation features

Abstract

The application provides a big data dynamic price index prediction system and method considering space-time correlation characteristics, and relates to the technical field of price index prediction, and comprises the following steps: collecting multi-channel price data through a specified channel interface, and preprocessing the multi-channel price data to obtain a standard space-time price data set; extracting space-time correlation characteristics of the standard space-time price data set; determining a space-time dynamic weight vector based on the price information entropy corresponding to different products in the standard space-time price data set, combining a market supply and demand coefficient and a market regulation influence coefficient as external factors; predicting a dynamic price index sequence based on the space-time correlation characteristics and the space-time dynamic weight vector through a pull-type price index prediction model; and sending the dynamic price index sequence to a specified associated terminal to enable a display component of the specified associated terminal to visualize the dynamic price index sequence. The application can realize dynamic and accurate price index prediction.

Description

Technical Field

[0001] This invention relates to the field of price index prediction technology, and in particular to a big data dynamic price index prediction system and method that considers spatiotemporal correlation characteristics. Background Technology

[0002] Price indices are core indicators for measuring the trend and magnitude of changes in market price levels, and are widely used in scenarios such as macroeconomic regulation, market supervision, and corporate pricing decisions. With the development of big data technology, multi-source data (such as transaction data from e-commerce platforms, POS (Point of Sale) data from offline supermarkets, and regional market research data) provide data support for the accurate calculation of price indices, but existing price index calculation methods still have significant shortcomings.

[0003] In existing technologies, price index calculations often employ traditional methods such as the Laspeyres formula, Paasche formula, or weighted averages. These methods typically suffer from two shortcomings: First, in the time dimension, they fail to adequately consider the temporal evolution of prices, making it difficult to capture long-term trends and short-term abrupt changes in price fluctuations, resulting in a lag in the index's response to market dynamics. Second, in the spatial dimension, they neglect the price correlation effects between different regions, such as the supply and demand transmission between adjacent regions and the price radiation effect of core regions on surrounding areas, making it impossible for the index to accurately reflect the coordinated fluctuation characteristics of regional markets. For example, existing land price index calculations often use simple weighted averages, which fail to characterize the spatial clustering effect and time non-stationary characteristics of land prices. Furthermore, conventional commodity price index calculations lack sufficient spatiotemporal collaborative analysis of online and offline omnichannel data, making it difficult for the index to adapt to the dynamic pricing needs of omnichannel retail scenarios.

[0004] Meanwhile, existing methods generally suffer from data fragmentation and low spatiotemporal alignment accuracy in multi-source data processing, failing to effectively integrate price data at different time granularities (daily, weekly, monthly) and spatial scales (province, city, district / county), thus limiting the coverage and accuracy of index calculations. Furthermore, existing dynamic price index models largely rely on single time-series forecasting algorithms, lacking deep integration of spatiotemporal correlation features, making it difficult to address the need for price fluctuation forecasting in complex market environments, such as the differentiated impacts of sudden events and policy adjustments on prices in different regions. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a big data dynamic price index prediction system and method that considers spatiotemporal correlation characteristics, which can achieve dynamic and accurate price index prediction.

[0006] In a first aspect, the present invention provides a big data dynamic price index prediction system that considers spatiotemporal correlation characteristics, comprising: The data acquisition and preprocessing module is used to: acquire multi-channel price data through specified channel interfaces, and preprocess the multi-channel price data to obtain a standard spatiotemporal price dataset; The spatiotemporal correlation feature extraction module is used to extract spatiotemporal correlation features from the standard spatiotemporal price dataset. The dynamic weight determination module is used to: determine the spatiotemporal dynamic weight vector based on the price information entropy of different products in the standard spatiotemporal price dataset, combined with the market supply and demand coefficient and the market regulation influence coefficient as external factors. The spatiotemporal dynamic weight vector includes the spatiotemporal dynamic weight coefficient corresponding to the price correlation region where the standard spatiotemporal price dataset is located within a specified time window. The dynamic price index prediction module is used to predict dynamic price index sequences based on spatiotemporal correlation features and spatiotemporal dynamic weight vectors using a Laplacian price index prediction model. The dynamic price index display module is used to send a dynamic price index sequence to a designated associated terminal so that the display component of the designated associated terminal can visualize the dynamic price index sequence.

[0007] Secondly, the present invention also provides a method for predicting dynamic price indices for big data that considers spatiotemporal correlation characteristics, comprising: Collect multi-channel price data through the specified channel interface, and preprocess the multi-channel price data to obtain a standard spatiotemporal price dataset; Extract spatiotemporal correlation features from standard spatiotemporal price datasets; Based on the price information entropy of different products in the standard spatiotemporal price dataset, and combined with the market supply and demand coefficient and the market regulation influence coefficient as external factors, the spatiotemporal dynamic weight vector is determined. The spatiotemporal dynamic weight vector includes the spatiotemporal dynamic weight coefficient corresponding to the price correlation region where the standard spatiotemporal price dataset is located within a specified time window. Using the Lagrange price index prediction model, dynamic price index sequences are predicted based on spatiotemporal correlation characteristics and spatiotemporal dynamic weight vectors. Send the dynamic price index sequence to a designated associated terminal so that the display component of the designated associated terminal can visualize the dynamic price index sequence.

[0008] Thirdly, the present invention also provides an electronic device including a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method provided in the second aspect.

[0009] Fourthly, the present invention also provides a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method provided in the second aspect.

[0010] The present invention provides a big data dynamic price index prediction system and method considering spatiotemporal correlation features, comprising: a data acquisition and preprocessing module, used to: acquire price data from multiple channels and preprocess the price data from multiple channels to obtain a standard spatiotemporal price dataset; a spatiotemporal correlation feature extraction module, used to: extract the spatiotemporal correlation features of the standard spatiotemporal price dataset; a dynamic weight determination module, used to: determine a spatiotemporal dynamic weight vector based on the price information entropy corresponding to different products in the standard spatiotemporal price dataset, combined with market supply and demand coefficients and market regulation influence coefficients as external factors, wherein the spatiotemporal dynamic weight vector includes the spatiotemporal dynamic weight coefficient corresponding to the price correlation region where the standard spatiotemporal price dataset is located within a specified time window; a dynamic price index prediction module, used to: predict a dynamic price index sequence based on the spatiotemporal correlation features and the spatiotemporal dynamic weight vector using a Laplacian price index prediction model; and a dynamic price index display module, used to: send the dynamic price index sequence to a designated associated terminal so that the display component of the designated associated terminal can visualize the dynamic price index sequence. The above method obtains spatiotemporal correlation features by deeply integrating the characteristics of standard spatiotemporal price datasets in the time and space dimensions, and determines the spatiotemporal dynamic weight vector by combining information entropy theory and external factors. Then, the dynamic price index sequence is predicted by the Laplacian price index prediction model based on the spatiotemporal correlation features and the spatiotemporal dynamic weight vector. This invention improves the ability of price index to characterize and predict market fluctuations by deeply integrating spatiotemporal correlation features, and can provide a reliable basis for macro-control and market decision-making.

[0011] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0012] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0013] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0014] Figure 1A schematic diagram of the structure of a big data dynamic price index prediction system considering spatiotemporal correlation features provided in an embodiment of the present invention; Figure 2 A schematic diagram of the specific structure of a big data dynamic price index prediction system considering spatiotemporal correlation features provided in an embodiment of the present invention; Figure 3 A flowchart illustrating a big data dynamic price index prediction method considering spatiotemporal correlation features provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Currently, existing technologies are insufficient to meet the demand for price fluctuation prediction in complex market environments. Based on this, the present invention provides a big data dynamic price index prediction system and method that considers spatiotemporal correlation characteristics, which can achieve dynamic and accurate price index prediction.

[0017] To facilitate understanding of this embodiment, a detailed description of a big data dynamic price index prediction system considering spatiotemporal correlation features disclosed in this embodiment of the invention will be provided first. (See [link to relevant documentation]). Figure 1 The diagram shows the structure of a big data dynamic price index prediction system that considers spatiotemporal correlation characteristics. The system mainly includes the following parts: The data acquisition and preprocessing module 102 is used to: acquire multi-channel price data through specified channel interfaces, and preprocess the multi-channel price data to obtain a standard spatiotemporal price dataset. The multi-channel price data may include online e-commerce platform transaction data, offline retail terminal POS data, regional wholesale market transaction data, and macroeconomic data. Preprocessing may include spatiotemporal stamp alignment, spatial coordinate system transformation, unification of data format and measurement standards, and outlier removal.

[0018] The spatiotemporal correlation feature extraction module 104 is used to extract spatiotemporal correlation features from a standard spatiotemporal price dataset. In one implementation, for the standard spatiotemporal price dataset, time dimension feature extraction, spatial dimension feature extraction, and spatiotemporal fusion feature construction can be performed separately to obtain spatiotemporal correlation features.

[0019] The dynamic weight determination module 106 is used to determine a spatiotemporal dynamic weight vector based on the price information entropy corresponding to different products in the standard spatiotemporal price dataset, combined with market supply and demand coefficients and market regulation influence coefficients as external factors. The spatiotemporal dynamic weight vector includes the spatiotemporal dynamic weight coefficients corresponding to the price correlation region where the standard spatiotemporal price dataset is located within a specified time window. In one implementation, a basic weight vector is constructed based on the price information entropy corresponding to different products in the standard spatiotemporal price dataset. This vector is then adjusted in both the time and spatial dimensions. Furthermore, the market supply and demand coefficients and market regulation influence coefficients are used as external factors to externally correct the adjusted weight vector, resulting in the final spatiotemporal dynamic weight vector.

[0020] The dynamic price index prediction module 108 is used to predict a dynamic price index sequence based on spatiotemporal correlation features and a spatiotemporal dynamic weight vector using a Laplace price index prediction model. The Laplace price index prediction model is an index prediction model based on the Laplace formula framework. In one implementation, the spatiotemporal dynamic weight vector and the current price dataset can be input into the Laplace price index prediction model to output an initial price index sequence. This initial price index sequence is then adjusted using spatiotemporal correlation features to obtain the final dynamic price index sequence.

[0021] The dynamic price index display module 110 is used to send the dynamic price index sequence to a designated associated terminal so that the display component of the designated associated terminal can visualize the dynamic price index sequence.

[0022] The big data dynamic price index prediction system considering spatiotemporal correlation features provided in this invention deeply integrates the features of a standard spatiotemporal price dataset in the time and space dimensions to obtain spatiotemporal correlation features. It then combines information entropy theory and external factors to determine the spatiotemporal dynamic weight vector. Subsequently, it uses a Laplacian price index prediction model to predict dynamic price index sequences based on the spatiotemporal correlation features and the spatiotemporal dynamic weight vector. By deeply integrating spatiotemporal correlation features, this invention improves the ability of price indices to characterize and predict market fluctuations, and can provide a reliable basis for macroeconomic control and market decision-making.

[0023] For ease of understanding, this invention provides a specific implementation of a big data dynamic price index prediction system that considers spatiotemporal correlation characteristics. See [link to relevant documentation]. Figure 2 The diagram shows a specific structure of a big data dynamic price index prediction system that considers spatiotemporal correlation features. It includes a data acquisition and preprocessing module 102, a spatiotemporal correlation feature extraction module 104, a dynamic weight determination module 106, a dynamic price index prediction module 108, and a dynamic price index display module 110.

[0024] (a) Data acquisition and preprocessing module 102, including a data acquisition unit and a preprocessing unit.

[0025] Data Acquisition Unit: Collects price data from multiple channels, including online e-commerce platform transaction data, offline retail terminal POS data, regional wholesale market transaction data, and macroeconomic data.

[0026] Preprocessing unit: Synchronize timestamps and convert spatial coordinate systems for collected price data from multiple channels to unify data formats and measurement standards; A sliding window mechanism combined with an outlier detection algorithm is used to remove noise and outliers from the data, resulting in a standard spatiotemporal price dataset. In one optional implementation, the outlier detection algorithm uses the Isolation Forest algorithm, and the window size of the sliding window is adaptively adjusted according to the data time granularity (the window size is set to 7 for daily data and 3 for monthly data).

[0027] (ii) Spatiotemporal correlation feature extraction module 104, including a time dimension feature extraction unit, a spatial dimension feature extraction unit, and a spatiotemporal fusion feature construction unit: The time dimension feature extraction unit is used to: extract price fluctuation features of the standard spatiotemporal price dataset within different time windows, and determine the time dimension features of the standard spatiotemporal price dataset based on the similarity between price fluctuation features within different time windows. The price fluctuation features include at least price trend features, price seasonal features, and price random fluctuation features.

[0028] In one example, the Sequence Decomposition (STL) algorithm can be used to extract price fluctuation characteristics of a standard spatiotemporal price dataset within different time windows. The specific implementation process is as follows: (1) The STL algorithm inner loop iterates to obtain the seasonal term (i.e., the seasonal price feature), including: low-pass filtering the standard spatiotemporal price data to obtain the initial trend term, taking the difference between the standard spatiotemporal price dataset and the initial trend term as the detrending sequence, grouping the detrending sequence according to the seasonal cycle, and performing Loess smoothing on each group of data to obtain the initial seasonal term, repeating this process until the preset conditions are met to obtain the seasonal term; (2) The STL algorithm outer loop iterates to obtain the trend term (i.e., the price trend feature), including: taking the difference between the standard spatiotemporal price dataset and the seasonal term as the deseasonal sequence; performing Loess smoothing on the deseasonal sequence to obtain the new initial trend term, feeding the new initial trend term back to the inner loop to recalculate the seasonal term, updating the trend term based on the new seasonal term, repeating this process until the preset conditions are met to obtain the trend term; (3) taking the difference between the standard spatiotemporal price dataset and the final seasonal term and the final trend term as the random fluctuation term (i.e., the random price fluctuation feature).

[0029] In one example, the temporal correlation of prices can be mined by using the similarity of price fluctuation characteristics within different time windows through the Dynamic Time Warping (DTW) algorithm, thus obtaining the temporal dimension characteristics of the standard spatiotemporal price dataset.

[0030] The spatial dimension feature extraction unit is used to: determine the regional price correlation coefficient based on the location information of the collection points corresponding to the standard spatiotemporal price dataset; perform clustering processing on the location information of the collection points according to the regional price correlation coefficient to obtain multiple price-related regions; and extract the spatial dimension features of the standard spatiotemporal price dataset according to the price correlation interval.

[0031] Specifically, it includes: (I) Construct a spatial weight matrix based on the geographical location information of each price collection point.

[0032] In one implementation, a combination of adjacency weighting and distance decay weighting is used. First, n sampling points are determined, and the data is initialized. An empty matrix is ​​obtained; for any two collection points, the geographical distance between them is calculated according to the spherical distance formula. If the geographical distance is less than the preset adjacency threshold, the two collection points are determined to be spatially adjacent; for adjacent collection points, the weight value is assigned according to the distance decay, and for non-adjacent collection points, the weight value is 0, thus obtaining the spatial weight matrix.

[0033] In another implementation, the spatial weight matrix is ​​constructed using the geographical distance weighting method, where the weight values ​​are inversely proportional to the geographical distance between regions.

[0034] (II) A spatial autoregressive model (SAR) is used to capture the spatial clustering effect and radiation effect of prices between different regions, and the price correlation coefficient between regions is calculated. The parameters of the spatial autoregressive model are solved by the maximum likelihood estimation method to ensure the model fitting accuracy.

[0035] (III) Combine the K-Nearest Neighbor (KNN) algorithm to divide price-related regions and extract the spatial distribution features (i.e., spatial dimension features) of prices within the regions. Specifically: construct the feature vector of each sampling point (including geographic coordinates, price dataset, and price correlation coefficient); determine the K (nearest neighbor number) value using the elbow rule; perform KNN clustering, including: calculating the Euclidean distance between the feature vectors corresponding to the sampling points, matching K nearest neighbor sampling points for each sampling point, and classifying the sampling points with the highest similarity into the same class to obtain multiple price-related regions; for each price-related region, calculate one or more of the following core spatial distribution features, including: price mean, price standard deviation, spatial center coordinates, price centroid offset, and spatial clustering degree, construct the spatial distribution features with the related regions as rows and the spatial features as columns, and normalize the matrix to obtain the final spatial distribution features.

[0036] The spatiotemporal fusion feature construction unit is used to perform attention fusion on time-dimensional features and spatial-dimensional features to obtain spatiotemporal correlation features. In one implementation, a spatiotemporal fusion model is constructed based on an attention mechanism, and time-dimensional features and spatial-dimensional features are weighted and fused to generate a spatiotemporal correlation feature matrix; wherein, the weight coefficients are dynamically optimized through a gradient descent algorithm to maximize the explanatory power of the features on price fluctuations.

[0037] (III) Dynamic Weight Determination Module 106: In one implementation, the information value of price data within each region and time window is quantified by combining information entropy theory to determine the basic weight vector; market supply and demand coefficients and policy influence coefficients are introduced as adjustment factors to dynamically correct the basic weight vector, resulting in a spatiotemporal dynamic weight vector. In specific implementation, this embodiment of the invention calculates the spatiotemporal dynamic weight w of the i-th type of commodity within the t-th time window (daily window, t=1,2,...,365) and the s-th region in four stages: "basic weight initialization - time dimension adjustment - spatial dimension adjustment - external factor correction". t , s , i The regions can be divided into provincial capital areas, eastern areas, western areas, southern areas, northern areas, southeastern areas, northeastern areas, southwestern areas, northwestern areas, etc., and the specific division can be based on actual needs. This embodiment of the invention does not impose any restrictions on this. Commodities can be divided into food, daily necessities, beauty and personal care products, digital and home appliances, baby and toy products, etc., and the specific settings can be based on actual needs. This embodiment of the invention does not impose any restrictions on this. For example, this embodiment of the invention provides a specific process for determining dynamic weights using regions (s=1 for the core area of ​​the provincial capital, s=2 for the eastern agricultural area, s=3 for the western mountainous area) and commodities (i=1 for wheat, i=2 for corn, i=3 for vegetables) as examples.

[0038] Optionally, the dynamic weight determination module 106 includes a weight determination unit and a weight adjustment unit.

[0039] In one example, the weight determination unit is used to: determine the basic weight vector based on the price information entropy of different products in the standard spatiotemporal price dataset.

[0040] Specifically, firstly, based on the price data of the t-th time window, the s-th region, and the i-th product category, the channel distribution probability of each product category is determined; then, the price information entropy H of each product category is determined using the information entropy calculation formula based on this channel distribution probability. i (For example, wheat H1=1.02, corn H2=1.05, vegetables H3=0.98); The price information entropy is normalized to map it to information value V. i (For example, V1=0.32, V2=0.30, V3=0.38); construct the basic weight vector directly based on information value. (For example =[0.32, 0.30, 0.38]).

[0041] In one example, the weight adjustment unit is used to: adjust the basic weight vector in terms of time dimension, space dimension, and external factors, using market supply and demand coefficients and market regulation influence coefficients as external factors, to obtain a spatiotemporal dynamic weight vector. Specifically, this includes: (I) Time dimension adjustment unit, used to: determine the current time decay coefficient and price fluctuation intensity, and use the current time decay coefficient and price fluctuation intensity to adjust the time dimension of the basic weight vector to obtain the time-adjusted weight.

[0042] The specific implementation process is as follows: (1) Determine the time decay coefficient: This time decay coefficient reflects the decay rate of the event correlation of price data. The larger the value, the faster the historical data decays and the higher the weight of the current window data. For example, the decay coefficient =0.1.

[0043] (2) Determine the time difference : The time interval between the current time window and the historical time window.

[0044] (3) Determine the time decay factor Corrects the time validity of the current window data, time decay factor. A higher value indicates a higher time value for the current window's data; this is the time decay factor. Based on attenuation coefficient and time difference Definitely obtained, for example Wherein, k is the time decay coefficient, used to control the rate at which price information decays over time: the larger k is, the faster the exponential decay, the more rapidly the weight of historical data decreases, and the model focuses more on recent prices; the smaller k is, the more gradual the decay, and the earlier data still retains a higher weight.

[0045] (4) Determine the intensity of price fluctuations : represents the price volatility intensity of the i-th type of agricultural product within the t-th time window, reflecting the degree of price volatility for that category within the current time window. A value greater than 1 indicates that the fluctuation is higher than the average level. <1 indicates volatility below average. Price volatility intensity. The determination process is as follows: Extract multi-channel price data (online e-commerce, offline wholesale markets, and the National Bureau of Statistics) for the i-th type of agricultural product within the t-th time window (daily window, i.e., 1 day), and calculate the average price for that day. Calculate the average daily price of this product category over the entire period (t=1~365). Price volatility intensity formula: (If the average price for the day is higher than the average price for the entire period,) >1 indicates stronger fluctuations; conversely, less than 1 indicates weaker fluctuations. Specific values: calculated for wheat... ,corn ,vegetable .

[0046] (5) Combining the meaning of the parameters, the correction method of "basic weight × time decay factor × price volatility intensity" is adopted to obtain the result for the time dimension. The adjusted time weights The formula is: .

[0047] Substitute parameters for calculation (taking the core area of ​​the provincial capital (s=1) and the first time window (t=1) as an example): Wheat (i=1): ; Corn (i=2): ; Vegetables (i=3): .

[0048] Among them, let In the first time window (t=1), using itself as the baseline, =0, therefore: This means that the current price data has no time decay and its validity remains complete; as the window iteration moves forward, The value will decrease exponentially and gradually decrease as the time interval increases.

[0049] Time-adjusted weights This represents the importance of a certain type of agricultural product in a specific region and time window, adjusted for the time dimension. The higher the value, the more significant the price fluctuations of that product category are in the current period and the greater its impact on the price index; the lower the value, the smaller the impact.

[0050] (II) Spatial dimension adjustment unit, used to: determine the regional economic scale coefficient, and use the regional price correlation coefficient and the regional economic scale coefficient to adjust the time adjustment weight spatially to obtain the spatial adjustment weight.

[0051] The specific implementation process is as follows: (1) Determine the parameters required for spatial dimension adjustment one by one, and clarify the value logic, calculation method or determination basis of each parameter, as follows (focusing on the core area of ​​the provincial capital s=1, t=1 window): (1.1) Determine the area division and numbering: This is the current area number. Number adjacent regions (i.e., regions that have a price relationship with the current region). For example, (Current area, core area of ​​the provincial capital) (Eastern agricultural region, adjacent area 1) (Western mountainous area, adjacent area 2) There are no other adjacent areas, so only these two related areas are considered.

[0052] (1.2) Determine the price correlation coefficient : For the current region With adjacent areas The price correlation coefficient reflects the degree of correlation between prices in two regions, and its value ranges from [0,1]. The closer a number is to 1, the higher the correlation and the stronger the mutual influence; the closer it is to 0, the lower the correlation. Specific calculation steps: Extract the price series (online + offline + statistical bureau multi-channel average) of the provincial capital core area (s=1), eastern agricultural area (s'=2), and western mountainous area (s'=3) at window t=1. The Pearson correlation coefficient formula is used to calculate the price correlation between regions. The formula is as follows: ;in, For data channel numbers, The price for the kth channel in region s. The average price in region s; For example, the core area of ​​the provincial capital and the eastern agricultural area With the western mountainous areas .

[0053] (1.3) Determine the basic weights of adjacent regions : represents the basic weights of the t-th time window, the adjacent region s', and the i-th type of agricultural product, which are calculated in the same logic as the basic weights of the current region (both are determined by the information value derived from information entropy).

[0054] (1.4) Determine the regional economic scale coefficient : This is the economic scale coefficient of the current region s, reflecting the economic influence of the current region, with a value range of [0,1]. The larger the value, the stronger the regional economic influence, and the greater the weighting adjustment range. For example, this coefficient is directly calculated as the proportion of the current region's GDP to the total GDP of all related regions (s=1, s'=2, s'=3), using the following formula: Substituting this into the fact that the GDP of the provincial capital's core area accounts for 42%, we get... That is, the GDP of the core area of ​​the provincial capital accounts for 42%.

[0055] (2) Spatial correlation weight is calculated by combining time-adjusted weight with the basic weight of adjacent areas and price correlation coefficient. The core logic is: the higher the correlation between the current area and adjacent areas, the higher the correlation between the current area and adjacent areas. The larger the value, the stronger the influence of the adjacent region's basic weight on the current region's weight. Specific calculation steps: Determine the spatial dimension Adjusted spatial correlation weights Formula: The formula uses a weighted summation method of "time-adjusted weight + basic weight of adjacent areas × price correlation coefficient", and the formula is as follows: .

[0056] Substitute parameters for calculation (taking the core area of ​​the provincial capital (s=1) and the first time window (t=1) as an example): Wheat (i=1): ; Corn (i=2): ; Vegetables (i=3): .

[0057] Spatial correlation weights of various agricultural products in the core area of ​​the provincial capital at window t=1 were obtained. .

[0058] Among them, spatial correlation weight This indicates the overall impact weight of a certain type of agricultural product in a certain region and time window, after considering the price linkages in surrounding areas. The higher the value, the stronger the influence of inter-regional price linkages on the product category, and the greater its contribution to the overall price index; the lower the value, the weaker the influence of regional linkages on the product category, and the smaller its contribution to the index.

[0059] (3) Introducing the regional economic scale coefficient The core logic for calibrating spatial correlation weights is: the larger the current regional economic scale ( The larger the value, the closer the weighted average reflects the regional economic influence. Normalization is not applied immediately after calibration (it will be uniformly normalized after subsequent external factor correction). Specific calculation steps: The calibration formula is determined as follows: The calibration method is "spatial correlation weight × (1 + regional economic scale coefficient)", and the formula is: The calibration coefficient is... The final output is the spatial adjustment weights of various agricultural products in the core area of ​​the provincial capital at window t=1. .

[0060] (III) External Factor Correction Unit, used to: externally correct the weights by adjusting the logarithmic space using the market supply and demand coefficient and the market regulation influence coefficient, to obtain a spatiotemporal dynamic weight vector. Among them, the market supply and demand coefficient is calculated by the ratio of commodity transaction volume to inventory in the region, and the policy influence coefficient is obtained by policy text mining and quantitative analysis.

[0061] Among them, external factors are divided into "supply and demand coefficients" "and market regulation influence coefficient" ". Let be the supply and demand coefficient of the i-th type of agricultural product in the t-th time window, the s-th region; Let be the policy impact coefficient for the t-th time window, the s-th region, and the i-th type of agricultural product. Both values ​​are ≥0. A coefficient >1 indicates a positive impact (increased weight), a coefficient =1 indicates no impact, and a coefficient <1 indicates a negative impact (decreased weight).

[0062] Based on this, the spatial adjustment weights are corrected according to the following formula: The calculation results of the modified formula are then normalized to obtain the final spatiotemporal dynamic weights. Among them, spatiotemporal dynamic weights This represents the comprehensive contribution and influence weight of a particular agricultural product to the overall price index within a specific time window and region, taking into account factors such as time decay, price fluctuations, regional spatial correlations, and economic scale. A higher value indicates a stronger price influence and greater market importance for the agricultural product within the current time and space, and a more significant effect of its price changes on the overall price index. A lower value indicates a weaker price influence and lower market importance for the agricultural product within the current time and space, and a relatively limited impact of its price fluctuations on the overall price index.

[0063] (iv) Dynamic price index prediction module 108, including: current data acquisition unit, index prediction unit and index correction unit.

[0064] (I) Current data acquisition unit, used to acquire the current price dataset.

[0065] (II) Index prediction unit, used to determine the initial price index sequence based on the current price dataset and spatiotemporal dynamic weight vector using the Laplace price index prediction model. In one implementation, a dynamic price index calculation model is constructed based on the Laplace formula framework and incorporates the spatiotemporal dynamic weight vector to obtain the initial price index sequence output by the model.

[0066] The expression for the Lagrange price index forecasting model is shown below: ; in, For the first Time window, first The initial price index sequence of the price correlation region, This represents the total number of products. For the first Time window, first Price-related area Spatiotemporal dynamic weighting coefficients for similar products For the first Time window, first Price-related area Current price data for similar products. For the first Time window, first Price-related area Base period price data for similar products.

[0067] (III) Index correction unit, used to determine correction coefficients based on spatiotemporal correlation characteristics, and use the correction coefficients to correct the initial price index sequence to obtain a dynamic price index sequence.

[0068] In one implementation, the initial price index sequence is corrected by combining trend and volatility characteristics from the spatiotemporal correlation feature matrix, and a sliding window is used for rolling calculation to obtain a real-time updated dynamic price index sequence. Specifically: First, a feedforward multilayer perceptron neural network model is constructed using the complete spatiotemporal correlation feature matrix (containing trend, seasonal and random fluctuation features in the time dimension, and structured representations such as price mean, standard deviation and spatial clustering in the spatial dimension) as input data. The network outputs a correction coefficient vector with the same length as the initial price index sequence, and each coefficient corresponds to the index calibration intensity under a time window-region combination.

[0069] Subsequently, a genetic algorithm was introduced to collaboratively optimize the neural network: the average absolute error between the predicted value of the dynamic price index and the actual price fluctuation over a historical period was used as the fitness function; the weight connections, number of hidden layer nodes, and activation function types of the neural network were encoded; and through selection, crossover, and mutation operations, the network structure and parameter combination that minimizes the prediction error were selected; finally, a set of correction coefficient generation mechanisms with strong robustness and good generalization ability determined by global optimization was obtained.

[0070] In actual operation, this mechanism can receive newly generated spatiotemporal correlation feature matrices in real time, automatically output correction coefficients that adapt to the current market state, and use them to adjust the initial price index sequence, thereby improving the accuracy and responsiveness of the dynamic price index.

[0071] (v) Dynamic price index display module 110: In one embodiment, the dynamic price index sequence is sent to a designated associated terminal, which is equipped with a display component, so that the dynamic price index sequence is visualized through the display component with a preset display effect.

[0072] (vi) System optimization module, used to optimize the parameters of the spatiotemporal correlation feature extraction module and / or dynamic weight determination module based on the dynamic price index sequence and actual price fluctuations through a pre-trained neural network model.

[0073] In one implementation, the dynamic price index sequence is backtested using historical price data to calculate the fitting error between the index and actual price fluctuations; an index prediction model is constructed based on a long short-term memory neural network (LSTM), and the weight coefficients of the spatiotemporal fusion model and the parameters of the dynamic weight model are adjusted by combining the prediction error feedback to optimize the index calculation accuracy.

[0074] This invention provides an application example of a big data dynamic price index prediction system that considers spatiotemporal correlation characteristics, used to calculate regional dynamic price indices for agricultural products, including: This invention uses the calculation of the regional dynamic price index of agricultural products (taking wheat, corn, and vegetables as examples) in a certain province as an example to illustrate the implementation process of this invention: (1) Multi-source price data collection and preprocessing: Data collection: Data on wholesale prices and transaction volumes of wheat, corn, and vegetables in various cities were collected through the offline wholesale market information system; online agricultural product transaction data (including transaction time, price, transaction volume, and seller's geographical location) from 10 major cities in the province were collected through the e-commerce platform API interface; and agricultural production data and macroeconomic data (CPI and PPI) released by the Provincial Bureau of Statistics were collected simultaneously.

[0075] Data preprocessing: The timestamps of all data were unified to Beijing time, and the geographical location information was converted to the WGS-84 coordinate system; the data format was standardized using the pandas library of Python, and the price unit was unified to "yuan / kg"; the sliding window size was set to 7 (daily data), and the isolated forest algorithm was used to remove abnormal price data (such as temporary sky-high vegetable prices caused by extreme weather) to obtain a standard spatiotemporal price dataset.

[0076] (2) Spatiotemporal correlation feature extraction: Time dimension feature extraction: The STL algorithm was used to decompose the standardized data to obtain the trend features (quarterly change trend), seasonal features (holiday fluctuations), and random fluctuation features (short-term weather impact) of wheat, corn, and vegetable prices. The DTW algorithm was used to calculate the similarity of agricultural product price fluctuation patterns among different cities. It was found that the price fluctuation patterns of provincial capital cities and surrounding cities had a similarity of more than 0.85, indicating a significant time-series correlation.

[0077] Spatial dimension feature extraction: Based on the latitude and longitude information of each city, a geographical distance weight matrix is ​​constructed, with the weight value being the reciprocal of the geographical distance between two cities; a spatial autoregressive model (SAR) is used to model the agricultural product price data of each quarter, and the spatial autoregressive coefficient ρ=0.62 is obtained through maximum likelihood estimation, indicating that there is a significant spatial price clustering effect between adjacent cities; the KNN algorithm (k=5) is used to divide the province into 3 price-related regions (the core area of ​​the provincial capital, the eastern agricultural area, and the western mountainous area).

[0078] Spatiotemporal fusion feature construction: A spatiotemporal fusion model is constructed based on the attention mechanism. The time dimension features and spatial dimension features are input, and the gradient descent algorithm is used to optimize and obtain the weight of the time feature (0.45) and the weight of the spatial feature (0.55), generating a spatiotemporal correlation feature matrix.

[0079] (3) Dynamic weight calculation: The information value of price data of various cities and agricultural products is calculated by information entropy. The basic weight of the provincial capital core area is determined to be 0.35, the eastern agricultural area is 0.4, and the western mountainous area is 0.25. The market supply and demand coefficient (transaction volume / inventory volume) of each region is calculated. The supply and demand coefficient of the eastern agricultural area is relatively high (1.2), so its weight is increased by 0.05. The policy impact coefficient is quantified by combining agricultural subsidy policy. The western mountainous area is more affected by the subsidy policy, so its weight is increased by 0.03. Finally, the spatiotemporal dynamic weight vector is obtained.

[0080] (4) Calculation of dynamic price index: Based on the Laplace formula framework, the initial price index is calculated by substituting the spatiotemporal dynamic weight vector and the current price and base period price data; the initial index is corrected by combining the trend characteristics and fluctuation characteristics in the spatiotemporal correlation feature matrix; the daily updated regional dynamic price index sequence of agricultural products is obtained by using the sliding window rolling calculation.

[0081] (5) Index verification and optimization: The index was backtested using historical price data from the past three years. The results showed that the fitting error between the price index calculated in this embodiment and the actual price fluctuation was 3.2%, which was lower than the 8.5% of the traditional weighted average method. An index prediction model was constructed based on LSTM to predict the price index for the next month, with a prediction error of 4.1%. The weight coefficients of the spatiotemporal fusion model were adjusted according to the prediction error feedback, and the spatial feature weight was adjusted to 0.58, which further improved the index accuracy.

[0082] This invention also provides another application example of a big data dynamic price index prediction system that considers spatiotemporal correlation characteristics, used to calculate the dynamic price index of retail goods across all channels, including: This invention takes the calculation of a dynamic price index for omnichannel products (taking daily necessities as an example) of a chain retail enterprise as an example. The implementation process is as follows: Collect daily necessities transaction data from online malls, offline stores, and community group buying platforms, including prices, transaction times, store / pickup point geographical locations, and user review data; after data preprocessing, extract intraday fluctuation characteristics and weekly trend characteristics in the time dimension, and store radiation range characteristics and regional consumption capacity characteristics in the spatial dimension; construct a spatiotemporal fusion model and a dynamic weight model to calculate the omnichannel dynamic price index; optimize the model parameters through price sensitivity feedback in user reviews, and the final price index can accurately reflect the price synergy fluctuations of online and offline channels, providing support for the enterprise's dynamic pricing.

[0083] In summary, the embodiments of the present invention have at least the following characteristics: This invention constructs a spatiotemporal fusion model by extracting the temporal evolution features and spatial correlation features of price data. This solves the problem of low index accuracy caused by neglecting the spatiotemporal correlation effect in existing methods, and can more comprehensively depict the dynamic fluctuation pattern of market prices.

[0084] The embodiments of this invention employ a dynamic weight model, combining information entropy theory with market and policy adjustment factors to achieve real-time dynamic optimization of weights. Compared with traditional fixed weight methods, this model is better able to adapt to price changes in complex market environments and improves the index's response speed to market dynamics.

[0085] This invention integrates multi-source price data, expands data coverage and improves data quality through standardized preprocessing and spatiotemporal alignment; combined with sliding window rolling calculation and feedback optimization mechanism, it further improves the calculation accuracy and predictive ability of price index, providing a more reliable basis for macro-control and market decision-making.

[0086] Based on the foregoing embodiments, this invention provides a method for predicting dynamic price indices in big data that considers spatiotemporal correlation characteristics. (See [link to previous document]). Figure 3 The diagram shows a flowchart of a big data dynamic price index prediction method that considers spatiotemporal correlation characteristics. The method mainly includes the following steps S302 to S310: Step S302: Collect multi-channel price data through the specified channel interface, and preprocess the multi-channel price data to obtain a standard spatiotemporal price dataset; Step S304: Extract the spatiotemporal correlation features of the standard spatiotemporal price dataset; Step S306: Based on the price information entropy of different products in the standard spatiotemporal price dataset, and combined with the market supply and demand coefficient and the market regulation influence coefficient as external factors, determine the spatiotemporal dynamic weight vector. The spatiotemporal dynamic weight vector includes the spatiotemporal dynamic weight coefficient corresponding to the price correlation region where the standard spatiotemporal price dataset is located within the specified time window. Step S308: Using the Lap price index prediction model, a dynamic price index sequence is predicted based on spatiotemporal correlation characteristics and spatiotemporal dynamic weight vectors; Step S310: Send the dynamic price index sequence to the designated associated terminal so that the display component of the designated associated terminal can visualize the dynamic price index sequence.

[0087] The big data dynamic price index prediction method considering spatiotemporal correlation features provided in this invention obtains spatiotemporal correlation features by deeply fusing the features of a standard spatiotemporal price dataset in the time and space dimensions. It then determines the spatiotemporal dynamic weight vector by combining information entropy theory and external factors. Finally, it predicts the dynamic price index sequence based on the spatiotemporal correlation features and the spatiotemporal dynamic weight vector through a Laplacian price index prediction model. By deeply fusing spatiotemporal correlation features, this invention improves the ability of the price index to characterize and predict market fluctuations, and can provide a reliable basis for macroeconomic control and market decision-making.

[0088] The method provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned system embodiment. For the sake of brevity, any part not mentioned in the method embodiment can be referred to the corresponding content in the aforementioned system embodiment.

[0089] This invention provides an electronic device, specifically, the electronic device includes a processor and a memory; the memory stores a computer program, which, when run by the processor, executes the method described in any of the above embodiments.

[0090] Figure 4 The present invention provides a schematic diagram of the structure of an electronic device 100, which includes a processor 40, a memory 41, a bus 42 and a communication interface 43. The processor 40, the communication interface 43 and the memory 41 are connected through the bus 42. The processor 40 is used to execute executable modules, such as computer programs, stored in the memory 41.

[0091] The memory 41 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 43 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0092] Bus 42 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0093] The memory 41 is used to store programs. After receiving an execution instruction, the processor 40 executes the program. The method executed by the device for defining the flow process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 40 or implemented by the processor 40.

[0094] Processor 40 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 40 or by instructions in software form. Processor 40 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 41. The processor 40 reads the information in memory 41 and, in conjunction with its hardware, completes the steps of the above method.

[0095] The computer program product of the readable storage medium provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For specific implementation, please refer to the foregoing method embodiments, which will not be repeated here.

[0096] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a portion 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 described in the various embodiments of the present invention. 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.

[0097] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, 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 the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A big data dynamic price index prediction system considering spatiotemporal correlation characteristics, characterized in that, include: The data acquisition and preprocessing module is used to: acquire multi-channel price data through a specified channel interface, and preprocess the multi-channel price data to obtain a standard spatiotemporal price dataset; The spatiotemporal correlation feature extraction module is used to: extract the spatiotemporal correlation features of the standard spatiotemporal price dataset; The dynamic weight determination module is used to: determine the spatiotemporal dynamic weight vector based on the price information entropy of different products in the standard spatiotemporal price dataset, combined with the market supply and demand coefficient and the market regulation influence coefficient as external factors. The spatiotemporal dynamic weight vector includes the spatiotemporal dynamic weight coefficient corresponding to the price correlation region where the standard spatiotemporal price dataset is located within a specified time window. The dynamic price index prediction module is used to: predict a dynamic price index sequence based on the spatiotemporal correlation features and the spatiotemporal dynamic weight vector using a Laplacian price index prediction model; A dynamic price index display module is used to: send the dynamic price index sequence to a designated associated terminal so that the display component of the designated associated terminal can visualize the dynamic price index sequence.

2. The big data dynamic price index prediction system considering spatiotemporal correlation characteristics according to claim 1, characterized in that, The spatiotemporal correlation feature extraction module is specifically used for: Extract price fluctuation features of the standard spatiotemporal price dataset within different time windows, and determine the time dimension features of the standard spatiotemporal price dataset based on the similarity between the price fluctuation features within different time windows. The price fluctuation features include at least price trend features, price seasonal features, and price random fluctuation features. Based on the location information of the collection points corresponding to the standard spatiotemporal price dataset, the regional price correlation coefficient is determined. Based on the regional price correlation coefficient, the location information of the collection points is clustered to obtain multiple price correlation regions, and the spatial dimension features of the standard spatiotemporal price dataset are extracted according to the price correlation intervals. Attention fusion is performed on the time dimension features and the spatial dimension features to obtain spatiotemporal correlation features.

3. The big data dynamic price index prediction system considering spatiotemporal correlation characteristics according to claim 1, characterized in that, The dynamic weight determination module includes: The weight determination unit is used to: determine the basic weight vector based on the price information entropy of different products in the standard spatiotemporal price dataset; The weight adjustment unit is used to adjust the basic weight vector in terms of time dimension, space dimension, and external factors, using the market supply and demand coefficient and the market regulation influence coefficient as external factors, to obtain a spatiotemporal dynamic weight vector.

4. The big data dynamic price index prediction system considering spatiotemporal correlation characteristics according to claim 3, characterized in that, The weight adjustment unit is specifically used for: The time dimension adjustment unit is used to: determine the current time decay coefficient and the price fluctuation intensity, and adjust the time dimension of the basic weight vector using the current time decay coefficient and the price fluctuation intensity to obtain the time-adjusted weight; The spatial dimension adjustment unit is used to: determine the regional economic scale coefficient, and adjust the time adjustment weight spatially using the regional price correlation coefficient and the regional economic scale coefficient to obtain the spatial adjustment weight; The external factor correction unit is used to: use the market supply and demand coefficient and the market regulation influence coefficient to externally correct the spatial adjustment weights mentioned in the logarithm to obtain the spatiotemporal dynamic weight vector.

5. The big data dynamic price index prediction system considering spatiotemporal correlation characteristics according to claim 1, characterized in that, The dynamic price index prediction module includes: The current data acquisition unit is used to acquire the current price dataset; The index prediction unit is used to determine the initial price index sequence based on the current price dataset and the spatiotemporal dynamic weight vector using the Laplacian price index prediction model. An index correction unit is used to determine a correction coefficient based on the spatiotemporal correlation characteristics, and to correct the initial price index sequence using the correction coefficient to obtain a dynamic price index sequence.

6. The big data dynamic price index prediction system considering spatiotemporal correlation characteristics according to claim 5, characterized in that, The expression for the Lagrange price index forecasting model is shown below: ； in, For the first Time window, first The initial price index sequence of the price correlation region, This represents the total number of products. For the first Time window, first Price-related area Spatiotemporal dynamic weighting coefficients for similar products For the first Time window, first Price-related area Current price data for similar products. For the first Time window, first Price-related area Base period price data for similar products.

7. The big data dynamic price index prediction system considering spatiotemporal correlation characteristics according to claim 1, characterized in that, It also includes a system optimization module, used for: The parameters of the spatiotemporal correlation feature extraction module and / or the dynamic weight determination module are optimized using a pre-trained neural network model based on the dynamic price index sequence and actual price fluctuations.

8. A method for predicting dynamic price indices in big data that considers spatiotemporal correlation characteristics, characterized in that, include: Collect multi-channel price data through designated channel interfaces, and preprocess the multi-channel price data to obtain a standard spatiotemporal price dataset; Extract the spatiotemporal correlation features from the standard spatiotemporal price dataset; Based on the price information entropy of different products in the standard spatiotemporal price dataset, and combined with the market supply and demand coefficient and the market regulation influence coefficient as external factors, the spatiotemporal dynamic weight vector is determined. The spatiotemporal dynamic weight vector includes the spatiotemporal dynamic weight coefficient corresponding to the price correlation region where the standard spatiotemporal price dataset is located within a specified time window. Based on the spatiotemporal correlation features and the spatiotemporal dynamic weight vector, the dynamic price index sequence is predicted using the Lagrange price index prediction model. The dynamic price index sequence is sent to a designated associated terminal so that the display component of the designated associated terminal can visualize the dynamic price index sequence.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to implement the method of claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method of claim 8.

Images