An all-channel product price intelligent monitoring and early warning method

Abstract

The application provides a full-channel product price intelligent monitoring and early warning method, relates to the technical field of price intelligent monitoring and early warning, and comprises the following steps: multi-source multi-channel standardized price data set construction, dynamic threshold regulation and multi-dimensional price anomaly analysis, and multi-level early warning response and data archiving and storage. The application can build a data acquisition architecture covering multiple channels, capture price-related data and preprocess and standardize the price-related data to generate a standardized price data set; dynamic threshold regulation is constructed based on the data set, anomaly analysis is carried out, normal fluctuations and malicious low prices are distinguished, and an analysis report is generated; and multi-level early warning is constructed based on the report, regulation measures are taken, and related data is classified and stored.

Description

Technical Field

[0001] This invention relates to the field of intelligent price monitoring and early warning technology, and in particular to a method for intelligent monitoring and early warning of product prices across all channels. Background Technology

[0002] In today's digital business environment, product sales have evolved into an omnichannel landscape integrating online e-commerce platforms, offline physical stores, community marketing, and cross-border channels. Price, as a core element of market competition, directly impacts a company's market competitiveness, profit margins, and brand image. Therefore, intelligent monitoring and early warning of product prices across all channels has become a critical requirement for enterprise operations management. However, existing omnichannel price monitoring and early warning technologies still face numerous unresolved issues, failing to meet the refined and intelligent management needs of enterprises. Specific shortcomings include: incomplete monitoring coverage and difficulty in guaranteeing data quality. Existing technologies primarily monitor single-channel types, making comprehensive coverage of all channels challenging, particularly regarding dynamic pricing on cross-border e-commerce platforms, mixed pricing by third-party sellers, and real-time price adjustments in offline stores. Furthermore, the continuous upgrading of anti-scraping technologies on various platforms makes traditional data collection methods susceptible to interception, resulting in poor data acquisition stability. The significant differences in data structures across different channels, including inconsistent product identification and diverse price display formats, lead to missing, incorrect, and duplicate data, failing to provide reliable support for subsequent early warning analysis. In addition, insufficient quality adjustments can cause biases in price analysis, further affecting data validity. Lack of precise correlation analysis capabilities. Existing technologies mostly only achieve simple collection and display of price data, making it difficult to deeply mine multi-channel price data, accurately identify price linkages between different channels, price fluctuation patterns, and the causes of abnormal prices. Furthermore, they have poor adaptability to complex scenarios such as product specification variations, regional pricing differences, and promotional activities, making it difficult to distinguish between normal price fluctuations and abnormal situations such as malicious low prices and price chaos. Moreover, many analytical models make unreasonable assumptions about fixed coefficients of characteristic variables, leading to significant biases in price analysis results and failing to provide valuable decision-making references for enterprises. Delayed response and insufficient targeting. Existing early warning technologies mostly rely on fixed thresholds for simple warnings, failing to build dynamic early warning models that incorporate multi-dimensional factors such as product lifecycle, market supply and demand changes, and competitor dynamics, easily resulting in false and missed warnings. Early warning responses are untimely, with a lag in capturing instantaneous price fluctuations and hidden price-cutting behaviors, preventing enterprises from taking timely control measures. At the same time, early warning information lacks hierarchical classification, failing to provide targeted handling suggestions based on the urgency and scope of impact of the warning, making it difficult to meet the differentiated early warning management needs of enterprises and adapt to large-scale product monitoring requirements. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method for intelligent monitoring and early warning of product prices across all channels. This method can build a data collection architecture covering multiple channels, capture price-related data and preprocess and standardize it to generate a standardized price dataset. Based on this dataset, dynamic threshold control is constructed, anomaly analysis is carried out and normal fluctuations are distinguished from malicious low prices, and an analysis report is generated. Based on the report, multi-level early warning is constructed, control measures are taken, and relevant data is classified, archived, and stored.

[0004] This invention provides a method for intelligent monitoring and early warning of product prices across all channels. The method includes: S1, establishing a multi-source, multi-channel data acquisition architecture to achieve real-time capture of product price-related data from various channels; preprocessing the captured raw data; standardizing product identifiers across different channels through a unified product coding mapping; and performing standardization and normalization processing on the preprocessed data to generate a standardized price dataset. The price-related data includes basic product information, real-time selling price, promotional price, pricing rules, inventory data, and competitor price data. S2, based on the standardized price dataset, constructing a dynamic price threshold control system to achieve real-time adjustment of dynamic feature thresholds. The initial feature thresholds are dynamically adjusted to ensure that they match market dynamics in real time. After real-time control and dynamic adjustment, multi-dimensional anomaly analysis is performed on the standardized price dataset. The real-time product price is compared with the dynamic feature thresholds to distinguish between normal price fluctuations and malicious low-price behavior, generating a complete anomaly price analysis report. S3: Based on the generated anomaly price analysis report, relevant anomaly information related to malicious low prices is extracted, a multi-level early warning mechanism is constructed, and early warning levels are divided according to the magnitude of price anomalies. Corresponding control measures are taken, and the original price data, price analysis results, and early warning records are classified, archived, and stored to support long-term retention and rapid query of historical price data.

[0005] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: By establishing a multi-source, multi-channel data acquisition architecture, real-time capture and standardized processing of price-related data across all channels are achieved, generating a unified and standardized price dataset. This effectively eliminates dimensional differences and labeling confusion among data from different channels, providing reliable data support for subsequent price monitoring. Dynamic adjustment of the price sample window size and correction of initial feature thresholds enable real-time matching of dynamic price thresholds with market fluctuations, improving threshold adaptability and accuracy. Based on multi-dimensional anomaly analysis, normal price fluctuations and malicious low-price behavior are accurately distinguished, avoiding misjudgments and omissions, ensuring the accuracy of price monitoring. By constructing a three-level early warning mechanism and implementing differentiated control measures, price anomalies can be responded to quickly, malicious low-price behavior can be controlled in a timely manner, and price order across all channels can be maintained. Simultaneously, the entire process data is classified, archived, and stored, enabling historical data traceability and queryability, providing data support for subsequent price optimization, threshold iteration, and regulatory verification. Overall, this improves the efficiency, accuracy, and practicality of intelligent monitoring and early warning of product prices across all channels, adapting to the multi-channel and dynamic market price control needs. Attached Figure Description

[0006] Figure 1 This is a flowchart of an intelligent monitoring and early warning method for product prices across all channels provided by an embodiment of the present invention; Figure 2 This is a flowchart illustrating the construction of a method for intelligent monitoring and early warning of product prices across all channels, as provided in this embodiment of the invention, and the dynamic threshold control of prices. Detailed Implementation

[0007] This invention provides a method for intelligent monitoring and early warning of product prices across all channels, such as... Figure 1 The flowchart shown is a method for intelligent monitoring and early warning of product prices across all channels. The processing flow of this method may include the following steps: S1 establishes a multi-source, multi-channel data acquisition architecture to achieve real-time capture of product price-related data from various channels. The captured raw data is preprocessed, and product identification across different channels is standardized through a unified product coding mapping. The preprocessed data is then standardized and normalized to generate a standardized price dataset. Price-related data includes basic product information, real-time selling price, promotional price, pricing rules, inventory data, and competitor price data.

[0008] It should be understood that the multi-source, multi-channel data collection architecture covers online e-commerce platforms, offline physical stores, community marketing channels, cross-border e-commerce platforms, and third-party seller channels.

[0009] It should be noted that the specific process for generating the standardized price dataset is as follows: Through a unified product coding mapping mechanism, product identifiers from different channels are mapped one by one to standard codes, achieving unique correspondence and standardization of product identifiers from different channels and in different formats. Product identifiers include online e-commerce platform product IDs, offline store barcodes, cross-border product codes, and third-party seller product codes. After the product identification standardization is completed, the data is normalized and the product price data of each channel is mapped to the preset range of [0,1] to eliminate the difference in dimensions between different channels and different data levels; The data that has undergone standardization and normalization is validated. If the validation passes, the data is classified and archived according to the product's unified code, channel type, and data collection time to generate a standardized price dataset. If the validation fails, the data that fails the validation is removed, and the missing valid fields are filled in using an interpolation algorithm.

[0010] In this embodiment, a unified product coding mapping mechanism is first used to standardize multi-source heterogeneous price data. Product identifiers in different formats, such as online e-commerce platform product IDs, offline store barcodes, cross-border product codes, and third-party seller product codes, are mapped one by one to a unified standard SKU code through a preset globally unique coding mapping table. This achieves unique identification and alignment of product entities across channels, avoiding data matching errors caused by inconsistent identifiers. Subsequently, an improved minimum-maximum normalization process is applied to the standardized data. A normalization algorithm based on dynamic extrema is constructed, mapping price-related data such as prices, inventory, and promotional discounts from various channels to the [0,1] interval. This adaptively removes extreme outliers and dynamically updates the extremum interval, effectively eliminating dimensional differences and numerical offsets between data from different channels and at different scales. The processed data is then subjected to integrity, consistency, and rationality checks. These checks include verifying missing data fields, the rationality of pricing logic, and the validity of channel timestamps. If the checks pass, the data is structured and archived according to the product's unified code, channel type, and collection timestamp to form a time-series, standardized price dataset. If the checks fail, abnormal or dirty data is directly removed, and the missing fields are filled in using a Lagrange interpolation algorithm combined with a time-series weighted average to ensure the continuity and availability of the dataset, providing a high-quality data foundation for subsequent dynamic threshold control and abnormal price analysis.

[0011] S2, based on a standardized price dataset, constructs a dynamic price threshold control system to achieve real-time adjustment of dynamic feature thresholds. It dynamically corrects initial feature thresholds to ensure real-time matching with market dynamics. After real-time control and dynamic correction, it performs multi-dimensional anomaly analysis on the standardized price dataset, comparing real-time product prices with dynamic feature thresholds to distinguish between normal price fluctuations and malicious low-price behavior, generating a complete anomaly price analysis report. The report includes anomaly price identification results, information on channels and products involved in malicious low-price behavior, the magnitude of price anomalies, anomaly cause analysis, and feature threshold control records.

[0012] It should be understood that Figure 2 This invention provides a flowchart of a method for intelligent monitoring and early warning of product prices across all channels, including dynamic price threshold control. The specific process is as follows: Starting with calculating the volatility coefficient of the current price sample window, the obtained volatility coefficient is compared with a preset volatility coefficient threshold range. When the volatility coefficient is less than or equal to the lower limit, the price fluctuation within the window is considered smooth and the dispersion is low. The window adjustment amount is obtained through a preset correspondence between the volatility coefficient and the window size. The current size is added to the adjustment amount, and the result is rounded up to obtain the target window size. Then, it is checked whether the target size is greater than or equal to a preset maximum value. If so, the maximum value is used as the target window size. The final window size is determined by the following criteria: if the volatility coefficient is greater than or equal to the upper limit, the price fluctuation within the window is considered to be drastic and highly dispersed. Based on the corresponding relationship, the window size is adjusted downwards. The current size is subtracted from the downward adjustment and then rounded down to obtain the target window size. The target window size is checked to see if it is less than or equal to the preset minimum value. If it is, the minimum value is set as the final window size; otherwise, the target size is retained. When the volatility coefficient is within the threshold range, the current price fluctuation is considered to be within a reasonable range. The original price sample window size is maintained unchanged, thus completing the full-scene dynamic adaptive adjustment of the price sample window size.

[0013] It needs to be explained that the specific process for constructing dynamic price threshold control is as follows: The basic parameters of the preset price sample window include the initial size of the price sample window, the minimum size of the price sample window, the maximum size of the price sample window, and the adjustment step size of the sample window. The price sample window is used to select valid historical price samples from the standardized price dataset. Valid samples are price data after abnormal quotes have been removed by the sample filtering rules. Real-time acquisition of current price data from standardized price datasets; extraction of valid historical price samples within the window based on the current size of the price sample window; calculation of statistical characteristic parameters of the valid historical price samples, including sample mean and sample standard deviation; and recording the result of processing the ratio of sample standard deviation to sample mean as the volatility coefficient of the current price sample window. The volatility coefficient of the current price sample window is used to characterize the dispersion of valid historical price samples within the current price sample window. A preset volatility coefficient threshold range is defined. The volatility coefficient of the current price sample window is compared with the volatility coefficient threshold range. The size of the price sample window is dynamically adjusted based on the comparison result. The volatility coefficient threshold range represents the open interval formed by the lower limit and upper limit of the volatility coefficient.

[0014] The specific process for dynamically adjusting the size of the price sample window based on the comparison results is as follows: A pre-constructed correspondence between volatility coefficient and sample window size is used to characterize the unique correspondence between the volatility coefficient of the price sample window and the adjustment amount of the price sample window size. If the volatility coefficient is less than or equal to the lower limit of the volatility coefficient, it is determined that the price fluctuation within the current price sample window is smooth and the dispersion is low. Then, the volatility coefficient of the current price sample window is input into the volatility coefficient-sample window size correspondence, and the price sample window size is adjusted upward. The result of coupling the current price sample window size and the price sample window size adjustment is rounded up as the target price sample window size. It is determined whether the adjusted target price sample window size is greater than or equal to the preset maximum price sample window size. If so, the maximum price sample window size is used as the final target price sample window size. If not, the target price sample window size remains unchanged.

[0015] The size of the price sample window is dynamically adjusted based on the comparison results, and the following are also included: If the volatility coefficient is greater than or equal to the upper limit of the volatility coefficient, it is determined that the price fluctuation within the current price sample window is drastic and the dispersion is high. Then, the volatility coefficient of the current price sample window is input into the volatility coefficient-sample window size correspondence, and the price sample window size is reduced by an amount. The difference between the current price sample window size and the price sample window size reduction amount is processed and rounded down to the nearest integer as the target price sample window size. It is determined whether the adjusted target price sample window size is less than or equal to the preset minimum price sample window size. If so, the minimum price sample window size is used as the final target price sample window size. If not, the target price sample window size remains unchanged. If the volatility coefficient is within the volatility coefficient threshold range, it is determined that the price fluctuation within the current price sample window is within a reasonable range, and the price sample window size remains unchanged.

[0016] In this embodiment, the basic parameters of the preset price sample window are determined through multiple experiments and optimizations. Specifically, these parameters include: the initial size of the price sample window is set to N0 (where N0 is a positive integer, and in this embodiment, N0=30 is preferred, i.e., 30 sets of continuous historical price data are initially selected as the initial sample set), and the minimum size of the price sample window is Nmin (in this embodiment, N0=30). m ᵢn =10, ensuring the sample size meets the minimum requirements for statistical feature calculation and avoiding statistical distortion due to insufficient sample size; maximum price sample window size Nmax (in this embodiment, Nmax=60, to prevent excessive sample size from introducing redundant data, affecting computational efficiency and sample specificity); sample window adjustment step size ΔN (in this embodiment, ΔN=5, the step size setting balances adjustment sensitivity and stability, avoiding excessive fluctuations in window size due to frequent adjustments); the price sample window is used to select valid historical price samples from the standardized price dataset. The standardized price dataset is pre-normalized from the original price data (normalization formula is x'=(x-xmin) / (xmax-xmin), where x is the original price data, xmin is the minimum value in the original price data, and xmax is the maximum value in the original price data). The dataset after the maximum value in the grid data is used. The effective samples are the price data after abnormal quotes have been removed by the sample filtering rules. The sample filtering rules adopt a composite filtering algorithm that combines the improved 3σ criterion with the Grubbs test. Specifically, the 3σ criterion is used to screen out suspected abnormal quotes that deviate from three times the standard deviation of the sample mean. Then, the suspected abnormal quotes are subjected to the Grubbs test (test statistic G=|xᵢ-μ| / σ, where xᵢ is the suspected abnormal quote, μ is the sample mean, and σ is the sample standard deviation. When G>G0 (G0 is the critical value of the Grubbs test, determined according to the sample size n and the significance level α=0.05), the quote is determined to be an abnormal quote and removed. The remaining quotes are considered as effective samples. Compared with a single filtering rule, this composite filtering algorithm can remove abnormal data more accurately and improve the purity of the effective samples. Real-time acquisition of current price data from a standardized price dataset, synchronous updating of historical price data sequences, and extraction of all historical price data within the updated historical price data sequence based on the current size N_current of the price sample window. Abnormal quotes are then removed using the aforementioned composite sample filtering rules, resulting in a valid historical price sample set S={s1,s2,...,s...} k} (where k is the number of valid samples, k≤N_current); then calculate the statistical characteristic parameters of the historical price valid sample set S. The statistical characteristic parameters include the sample mean μ and the sample standard deviation σ. The sample standard deviation σ and the sample mean μ are processed by ratio to obtain the volatility coefficient CV of the current price sample window. The formula for calculating the volatility coefficient CV is CV=σ / μ (when μ=0, CV=0 by default to avoid calculation abnormalities with a denominator of 0). The volatility coefficient CV is used to characterize the dispersion of the historical price valid samples within the current price sample window. The larger the CV value, the more drastic the price data fluctuation and the higher the dispersion within the window. The smaller the CV value, the smoother the price data fluctuation and the lower the dispersion within the window. A preset volatility coefficient threshold range is defined as the open interval (CVmin, CVmax) formed by the lower limit (CVmin) and the upper limit (CVmax) of the volatility coefficient. In this embodiment, through statistical analysis of a large amount of historical price data, CVmin = 0.05 and CVmax = 0.20 are optimized and determined. This threshold range can adapt to the price volatility characteristics in most scenarios, while reserving adjustment space. The volatility coefficient CV of the current price sample window is then compared with the volatility coefficient threshold range (0.05, CVmax, CVmax). A comparison is performed on 0.20, and the size of the price sample window is dynamically adjusted based on the comparison results. The specific adjustment process, combined with an improved adaptive adjustment algorithm, is detailed as follows: A correspondence between volatility coefficient and sample window size is pre-constructed. This correspondence is used to characterize the unique relationship between the volatility coefficient CV of the price sample window and the price sample window size adjustment amount ΔN_current. This correspondence is constructed based on an adaptive algorithm, breaking through the limitations of the traditional fixed adjustment amount. Specifically, the construction method is as follows: with the volatility coefficient CV as the independent variable and the size adjustment amount ΔN_current as the dependent variable, a nonlinear mapping relationship is established: ΔN_current = f(CV) = ΔN × (CVmin / CV) (when CV ≤ CVmin), where ΔN is the preset sample window adjustment step size. This mapping relationship makes the smaller the volatility coefficient CV (the smoother the price fluctuation), the larger the size adjustment amount ΔN_current, realizing adaptive matching of the adjustment amount and improving the rationality of the size adjustment; when CV > CVmin, the adjustment amount ΔN_current is dynamically adjusted according to the difference between CV and CVmax. Here, the adjustment process when CV ≤ CVmin is emphasized.If the calculated volatility coefficient CV is less than or equal to the lower limit of volatility coefficient CVmin (i.e., CV≤0.05), it is determined that the price fluctuation within the current price sample window is smooth and the dispersion is low. At this time, the representativeness of the effective samples within the window is insufficient, and the window size needs to be expanded to include more effective historical price samples and improve the comprehensiveness of the samples. The volatility coefficient CV of the current price sample window is input into the pre-constructed volatility coefficient-sample window size correspondence mentioned above. The upward adjustment amount ΔN_current of the price sample window size is calculated and output through the nonlinear mapping formula ΔN_current=ΔN×(CVmin / CV) (in this embodiment, ΔN=5, CVmin=0.05. If CV=0.03, then ΔN_current=5×(0.05 / 0.03)≈8.33, rounded to 8). Then, the current price sample window size N_current and the upward adjustment amount ΔN_current of the price sample window size are coupled and calculated. The coupling formula is N_target=ceil(N_current + ΔN_current), where ceil() is the floor function to ensure that the window size is a positive integer, and the target price sample window size N_target is obtained; then it is determined whether the adjusted target price sample window size N_target is greater than or equal to the preset maximum price sample window size Nmax (i.e., N_target≥60). If so, it means that the current adjusted window size has exceeded the preset upper limit. In order to avoid sample redundancy, the maximum price sample window size Nmax(60) is used as the final target price sample window size; if not, the target price sample window size N_target is kept unchanged and used as the final target price sample window size to complete the dynamic adjustment of the window size; after the adjustment is completed, the effective historical price samples are extracted again based on the new window size, and the above statistical feature calculation, volatility coefficient comparison and size adjustment process are repeated to realize the real-time dynamic adaptation of the price sample window size.

[0017] It should be further explained that the specific process for dynamically correcting the initial feature threshold is as follows: Preset initial feature threshold parameters, including initial feature threshold baseline value, initial feature threshold minimum threshold, initial feature threshold maximum threshold, and feature threshold adjustment coefficient. The initial feature threshold is used to make preliminary anomaly judgment on price features in the standardized price dataset. Price features include price volatility, price month-on-month change rate, and price year-on-year change rate. The price deviation threshold is obtained in real time. The current price data and corresponding historical price valid samples are collected from the standardized price dataset. The current price deviation is calculated as the absolute value of the difference between the current price and the mean of the historical price valid samples in the corresponding price sample window. Calculate the price deviation threshold utilization rate, which is the ratio of the current price deviation to the price deviation threshold, used to characterize the proportion of the current price deviation relative to the price deviation threshold; The preset price deviation threshold utilization rate range is compared with the calculated price deviation threshold utilization rate. Based on the comparison result, the initial feature threshold is dynamically adjusted according to the preset feature threshold adjustment coefficient. The preset price deviation threshold utilization rate range represents the open interval formed by the lower bound and the upper bound of the price deviation threshold utilization rate.

[0018] The specific process for dynamically adjusting the initial feature threshold based on the preset feature threshold adjustment coefficient is as follows: A threshold utilization rate-feature threshold adjustment coefficient matrix is ​​pre-constructed. This matrix is ​​used to characterize the unique correspondence between the price deviation threshold utilization rate and the initial feature threshold adjustment coefficient. If the price deviation threshold utilization rate is less than or equal to the lower bound of the price deviation threshold utilization rate, it is determined that the current price deviation is far below the price deviation threshold, and the price fluctuation is in an extremely flat state. The current price deviation threshold utilization rate is input into the threshold utilization rate-feature threshold adjustment coefficient matrix, and the initial feature threshold gain coefficient is output. The current initial feature threshold and the initial feature threshold gain coefficient are combined to obtain the target initial feature threshold. It is determined whether the target initial feature threshold is greater than or equal to the preset initial feature threshold maximum value. If so, the preset initial feature threshold maximum value is used as the final target initial feature threshold. If not, the target initial feature threshold is kept unchanged.

[0019] Dynamically adjusting the initial feature threshold also includes: If the price deviation threshold utilization rate is greater than or equal to the upper bound of the price deviation threshold utilization rate, it is determined that the current price deviation is close to or has reached the price deviation threshold, and the price fluctuation is in a critical abnormal state. The current price deviation threshold utilization rate is input into the threshold utilization rate-feature threshold adjustment coefficient matrix, and the initial feature threshold reduction coefficient is output. The current initial feature threshold and the initial feature threshold reduction coefficient are combined to obtain the target initial feature threshold. It is determined whether the target initial feature threshold is less than or equal to the preset initial feature threshold minimum value. If so, the preset initial feature threshold minimum value is used as the final target initial feature threshold. If not, the target initial feature threshold is kept unchanged. If the utilization rate of the price deviation threshold is within the preset price deviation threshold utilization rate range, it is determined that the current price deviation is within a reasonable range, the price fluctuation is stable, and the current initial feature threshold remains unchanged.

[0020] In this embodiment, preset initial feature threshold parameters are established. These parameters are determined through training and optimization using a large amount of historical price data to ensure that the parameters can adapt to most price scenarios. Specifically, these parameters include an initial feature threshold baseline value, an initial feature threshold minimum value, an initial feature threshold maximum value, and a feature threshold adjustment coefficient. The initial feature threshold baseline value is set separately for three price features: price volatility, month-on-month price change rate, and year-on-year price change rate. The thresholds for these three features are adjusted independently and do not interfere with each other. The initial feature threshold minimum value is also set separately for each of the three price features to prevent excessively low thresholds from causing overly sensitive anomaly detection and resulting in numerous false positives. The initial feature threshold maximum value is also set separately for each of the three price features to avoid excessively high thresholds that lead to delayed anomaly detection and failure to identify potential anomalies in a timely manner. The feature threshold adjustment coefficient is set to an adjustable range to provide a basis for subsequent dynamic adjustments. An adaptive coefficient correction algorithm is also introduced to allow the adjustment coefficient to be dynamically optimized in line with price fluctuation trends. Initial feature thresholds are used to make preliminary anomaly judgments on price features in the standardized price dataset. The standardized price dataset is a dataset after pre-normalizing the original price data. Normalization is to regularize the original price data to a uniform range by matching the minimum and maximum values ​​in the original price data. The price features are specifically defined as follows: price volatility is the percentage converted from the absolute value of the difference between the current price and the price of the previous period to the price of the previous period; price month-on-month change rate is the percentage converted from the difference between the current price and the price of the same period of the previous period to the price of the same period of the previous period; and price year-on-year change rate is the percentage converted from the difference between the current price and the price of the same period of the previous year to the price of the same period of the previous year. These three types of features work together to screen for price anomalies and improve the comprehensiveness of anomaly identification. In this embodiment, the price deviation threshold is not a fixed value, but is adaptively calculated based on the statistical characteristics of historical price valid samples. Specifically, it is calculated by combining the mean and unbiased standard deviation of historical price valid samples within the corresponding price sample window with a reasonable algorithm combination. This algorithm enables the price deviation threshold to dynamically adapt to historical price fluctuations, significantly improving the rationality of deviation judgment compared to a fixed threshold. Simultaneously, current price data from the standardized price dataset and the set of historical price valid samples within the corresponding price sample window are collected. The valid samples are obtained after removing abnormal quotes using a composite filtering algorithm combining the improved 3σ criterion and Grubbs' test, ensuring sample purity. The current price deviation is calculated as the absolute value of the difference between the current price and the mean of historical price valid samples within the corresponding price sample window. The larger the current price deviation value, the higher the deviation between the current price and the historical price average level, and vice versa.The price deviation threshold utilization rate is calculated as the ratio of the current price deviation to the price deviation threshold. When the price deviation threshold is zero, the default price deviation threshold utilization rate is zero to avoid calculation anomalies where the denominator is zero. The price deviation threshold utilization rate is used to characterize the proportion of the current price deviation relative to the price deviation threshold. Its value ranges from zero to positive infinity. The closer the value is to 1, the closer the current price deviation is to the upper limit of the threshold. The value is greater than 1, indicating that the current price deviation exceeds the threshold range. The smaller the value is, the lower the current price deviation is below the threshold range. A preset price deviation threshold utilization rate range is defined. This range is an open interval formed by the lower and upper bounds of the price deviation threshold utilization rate. In this embodiment, the upper and lower bounds of the range are optimized and determined by statistically analyzing a large amount of historical price deviation data. This range can accurately define the reasonable range of price deviation and provide a clear basis for threshold adjustment. The calculated price deviation threshold utilization rate is compared with the preset price deviation threshold utilization rate range. Based on the comparison results, combined with the preset feature threshold adjustment coefficient and the improved adaptive adjustment algorithm, the initial feature threshold is dynamically adjusted. The specific adjustment process is as follows, where the core improvement is the pre-construction of a threshold utilization rate-feature threshold adjustment coefficient matrix, which breaks through the limitations of the traditional single adjustment coefficient and achieves accurate matching between the adjustment coefficient and the utilization rate. A threshold utilization rate-feature threshold adjustment coefficient matrix is ​​pre-constructed. This matrix represents the unique correspondence between the price deviation threshold utilization rate and the initial feature threshold adjustment coefficient. The matrix construction employs a piecewise nonlinear mapping algorithm to improve the accuracy of coefficient matching. Specifically, the value range of the price deviation threshold utilization rate is divided into three intervals, corresponding to three cases: the price deviation threshold utilization rate is less than or equal to the lower bound, between the upper and lower bounds, and greater than or equal to the upper bound. An independent nonlinear mapping relationship is established within each interval. Specifically, when the price deviation threshold utilization rate is less than or equal to the lower bound, a gain coefficient function is used; the smaller the price deviation threshold utilization rate, the larger the gain coefficient, achieving adaptive matching of the threshold upward adjustment. When the price deviation threshold utilization rate is greater than or equal to the upper bound, a reduction coefficient function is used; the larger the price deviation threshold utilization rate, the smaller the reduction coefficient, achieving adaptive matching of the threshold downward adjustment. When the price deviation threshold utilization rate is between the upper and lower bounds, the adjustment coefficient is set to a fixed value, i.e., the threshold is not adjusted. This matrix allows the adjustment coefficient to be dynamically optimized according to changes in the price deviation threshold utilization rate.The first adjustment scenario: If the price deviation threshold utilization rate is less than or equal to the lower bound of the price deviation threshold utilization rate, it is determined that the current price deviation is far below the price deviation threshold, and the price fluctuation is in an extremely flat state. In this case, an excessively high initial feature threshold will lead to overly lenient anomaly detection, making it impossible to identify minor anomalies in a timely manner. It is necessary to increase the initial feature threshold to improve the sensitivity of anomaly detection. The current price deviation threshold utilization rate is input into the threshold utilization rate - feature threshold adjustment coefficient matrix mentioned above. The initial feature threshold gain coefficient is calculated and output through the corresponding gain coefficient function. The current initial feature threshold and the initial feature threshold gain coefficient are combined and processed using a multiplicative combination method to ensure the smoothness of the threshold adjustment and avoid abrupt changes, resulting in the target initial feature threshold. Then, it is determined whether the target initial feature threshold is greater than or equal to the preset maximum value of the initial feature threshold. If so, it means that the current increased threshold has exceeded the preset upper limit. To avoid overly sensitive anomaly detection, the preset maximum value of the initial feature threshold is used as the final target initial feature threshold. If not, the target initial feature threshold is kept unchanged and used as the final target initial feature threshold, completing the threshold adjustment in this scenario. The second adjustment scenario: If the price deviation threshold utilization rate is greater than or equal to the upper bound of the price deviation threshold utilization rate, it is determined that the current price deviation is close to or has reached the price deviation threshold, and the price fluctuation is in a critical abnormal state. At this time, if the initial feature threshold is too low, the anomaly detection will be too sensitive, resulting in a large number of false judgments. It is necessary to lower the initial feature threshold to reduce the false judgment rate. The current price deviation threshold utilization rate is input into the threshold utilization rate - feature threshold adjustment coefficient matrix. The initial feature threshold reduction coefficient is calculated and output through the corresponding reduction coefficient function. The current initial feature threshold and the initial feature threshold reduction coefficient are combined using the same multiplication combination method to obtain the target initial feature threshold. It is determined whether the target initial feature threshold is less than or equal to the preset minimum value of the initial feature threshold. If so, it means that the current lowered threshold is already lower than the preset lower limit. To avoid the lag in anomaly detection, the preset minimum value of the initial feature threshold is used as the final target initial feature threshold. If not, the target initial feature threshold is kept unchanged and used as the final target initial feature threshold, thus completing the threshold adjustment in this scenario. The third adjustment scenario: If the utilization rate of the price deviation threshold is within the preset price deviation threshold utilization rate range, it is determined that the current price deviation is within a reasonable range and the price fluctuation is stable. At this time, the current initial feature threshold can accurately adapt to the price fluctuation state and no adjustment is required. The current initial feature threshold remains unchanged to ensure the stability of the anomaly detection.After each initial feature threshold adjustment, based on the adjusted final target initial feature threshold, a preliminary anomaly judgment is made on the three types of price features of the current price data. If the value of a certain price feature exceeds the corresponding adjusted initial feature threshold, it is determined that the price feature has a preliminary anomaly and enters the subsequent accurate verification process. At the same time, the above-mentioned price deviation calculation, threshold utilization calculation, comparison and threshold adjustment process is repeated in real time to achieve real-time dynamic adaptation of the initial feature threshold.

[0021] It is important to understand that the specific process for performing multi-dimensional anomaly analysis on a standardized price dataset is as follows: The adjusted target initial feature threshold is used as the new initial feature threshold and applied to the preliminary judgment of price feature anomalies in the standardized price dataset. If the price volatility is greater than the new initial feature threshold, it is judged as a preliminary anomaly and enters the second judgment stage; if the price volatility is less than or equal to the new initial feature threshold, it is judged as a normal price feature. The specific process for entering the second judgment stage is as follows: By comparing the real-time selling price of products in each channel with the normal price fluctuation range threshold and the malicious low-price judgment benchmark threshold corresponding to that channel, a preliminary judgment result is obtained. If the real-time selling price of a product is within the threshold of the normal price fluctuation range corresponding to the channel, then the fluctuation of the real-time selling price of the product in that channel is determined to be a normal price fluctuation. If the real-time selling price of a product is less than the threshold for malicious low-price judgment, then the behavior corresponding to the real-time selling price of the product on that channel is determined to be malicious low-price behavior.

[0022] In this embodiment, the target initial feature threshold obtained through dynamic adjustment mentioned above is updated as the new initial feature threshold. This new initial feature threshold has been dynamically adapted through the utilization rate of the price deviation threshold, enabling it to accurately match the current price fluctuation state and avoid the misjudgment and omission problems caused by traditional fixed thresholds. This new initial feature threshold is applied to the preliminary judgment of price feature anomalies in the standardized price dataset. The judgment object focuses on the core price feature of price volatility. The judgment logic is clear and targeted: when analyzing each set of current price data in the standardized price dataset, the corresponding price volatility is first calculated, and then the price volatility is compared with the new initial feature threshold. If the calculated price volatility is greater than the new initial feature threshold, the price feature corresponding to the price data is judged to be initially abnormal, and further investigation is needed to check whether there are abnormal behaviors such as malicious low prices. Therefore, it directly enters the secondary judgment stage. If the calculated price volatility is less than or equal to the new initial feature threshold, the price feature corresponding to the price data is judged to be a normal price feature. It does not need to enter the subsequent secondary judgment and is directly included in the category of normal price data. This achieves the preliminary screening of abnormal data and improves the efficiency of subsequent anomaly analysis. The specific process for entering the secondary judgment stage is based on combining the differentiated characteristics of multiple channels and introducing a multi-channel adaptive threshold algorithm and a malicious low-price identification optimization algorithm. This breaks through the limitations of the traditional single judgment standard and achieves accurate differentiation of price anomalies in different channels. The specific operation logic is as follows: First, a multi-channel adaptive threshold calculation model is pre-built. For different sales channels (such as online e-commerce channels, offline physical store channels, distributor channels, etc.), the corresponding normal price fluctuation range threshold and malicious low-price judgment benchmark threshold are calculated respectively. The core improvement of this model is that the threshold calculation is not a fixed value, but is dynamically generated by combining multiple dimensions such as the effective historical price samples of each channel, channel operating costs, and market competition intensity through a weighted fusion algorithm to ensure that the threshold is accurately adapted to the price characteristics of each channel. Specifically, the normal price fluctuation range threshold for each channel is obtained by statistically analyzing the mean, standard deviation, and reasonable fluctuation range of historical price samples for that channel, combined with a weighted adjustment based on the channel's operating cost ratio. This accurately reflects the price fluctuation range under normal operating conditions for that channel. The malicious low-price judgment benchmark threshold for each channel is calculated using a weighted summation algorithm, taking into account multiple factors such as the lowest historical normal selling price of that channel, the product cost price, and the industry's malicious low-price warning line. The product cost price has the highest weight, ensuring the rationality of malicious low-price judgment and avoiding misjudging normal promotional low prices as malicious low prices. The second step involves real-time collection of product selling prices from each channel. The real-time selling price of each product from each channel is then compared with both the normal price fluctuation range threshold and the malicious low-price judgment benchmark threshold obtained from the aforementioned multi-channel adaptive threshold calculation model. The comparison results are then used to obtain a preliminary judgment result. This dual comparison effectively eliminates the limitations of single-threshold judgment and improves the accuracy of the judgment.The third step involves making a precise judgment based on the results of the dual comparison: If the real-time selling price of a product in a certain channel is within the normal price fluctuation range threshold corresponding to that channel, and is not lower than the malicious low-price judgment benchmark threshold of that channel, then the fluctuation corresponding to the real-time selling price of the product in that channel is determined to be a normal price fluctuation, and the initial anomaly judgment is a misjudgment. The price data is corrected to normal price data, and no further processing is required; If the real-time selling price of a product in a certain channel is lower than the malicious low-price judgment benchmark threshold corresponding to that channel, regardless of whether it is within the normal price fluctuation range threshold, the behavior corresponding to the real-time selling price of the product in that channel is determined to be malicious low-price behavior. The initial anomaly judgment is established, and the anomaly warning mechanism is triggered at the same time. The real-time selling price information of the channel and the product, the time of the anomaly, and the degree of deviation are recorded to provide data support for subsequent anomaly handling. To further improve the accuracy of malicious low-price judgment, this embodiment introduces an additional time-dimensional auxiliary judgment algorithm in the malicious low-price judgment process. That is, if the real-time price of a product in a certain channel is lower than the malicious low-price judgment benchmark threshold, and the duration of the low-price state exceeds the preset warning duration (the warning duration is dynamically set according to the operational characteristics of each channel), it will be finally judged as malicious low-price behavior. If it is only a short-term instantaneous low price (such as limited-time flash sales, temporary promotions and other normal marketing behaviors), it will be judged as normal price fluctuation, avoiding misjudging normal marketing behaviors as malicious low-price, and further improving the accuracy of anomaly analysis. The entire multi-dimensional anomaly analysis process is based on dynamically adjusted feature thresholds to achieve initial anomaly screening of price features. Then, through multi-channel adaptive threshold algorithms, malicious low-price identification optimization algorithms, and time-dimensional auxiliary judgment, a second precise judgment is completed, forming a two-layer architecture of initial screening of single features and precise verification of multiple dimensions. Compared with traditional single-dimensional, fixed-threshold anomaly analysis methods, it can accurately adapt to the price characteristics of different channels, effectively distinguish between normal price fluctuations and malicious low-price behavior, significantly reduce the false positive rate and false negative rate, and improve the comprehensiveness and accuracy of anomaly analysis of standardized price datasets. At the same time, through the fusion of multiple algorithms, the creativity and practicality of the method are enhanced, providing reliable technical support for price anomaly control.

[0023] S3, based on the generated abnormal price analysis report, extracts abnormal information related to malicious low prices, constructs a multi-level early warning mechanism, classifies early warning levels according to the magnitude of price anomalies, and takes corresponding control measures. It also classifies and archives the original price data, price analysis results, and early warning records to support the long-term retention and rapid query of historical price data.

[0024] It should be explained that the specific process for classifying warning levels based on the magnitude of price anomalies is as follows: If the price deviates from the benchmark by less than or equal to the lower limit of the benchmark deviation, it is judged as a Level 1 warning, and the abnormality type is normal fluctuation abnormality of non-malicious low price; If the price deviates from the benchmark range within the benchmark deviation range, a Level 2 warning is issued, and the anomaly type is suspected malicious low price. The benchmark deviation range refers to the open interval formed by the lower limit of the benchmark deviation and the upper limit of the benchmark deviation. If the price deviates from the benchmark by a greater than or equal to the upper limit of the benchmark deviation, it is judged as a Level 3 warning, and the anomaly type is confirmed malicious low price; The specific procedure for taking corresponding regulatory measures is as follows: Level 1 alerts only record and monitor in real time; Level 2 alerts send reminder notifications to the corresponding channel manager; Level 3 alerts suspend the display of abnormal prices on that channel.

[0025] In this embodiment, after completing the multi-dimensional anomaly analysis of the standardized price dataset and determining that the product price deviates from the benchmark, a graded early warning judgment needs to be made based on the specific numerical value of the price deviation from the benchmark. Corresponding control measures are then taken for different early warning levels to ensure that abnormal prices are reasonably controlled while avoiding excessive control that could affect normal operations. The specific implementation process is as follows: Preset parameters related to price deviation from the benchmark, including a lower limit and an upper limit. The lower limit is used to define the boundary between slight and moderate deviation, while the upper limit is used to define the boundary between moderate and severe deviation. The two form a deviation range, which is an open interval formed by the lower and upper limits of the deviation. All of the above parameters are determined through training and optimization using historical price data from various channels, product costs, industry standards, and other multi-dimensional factors to ensure the rationality and accuracy of the early warning judgment. The calculation determines the deviation of the current product price from the benchmark price. This deviation quantifies the degree of deviation between the current real-time selling price and the benchmark price (i.e., the benchmark value of the normal price fluctuation range for the corresponding channel). After calculation, the deviation is compared with preset lower, upper, and lower limits of the benchmark deviation, and the benchmark deviation range. Based on the comparison results, a graded warning is issued, and the corresponding anomaly type is identified: If the calculated deviation is less than or equal to the preset lower limit of the benchmark deviation, it indicates that the current price deviation is minor and does not meet the standard for suspected malicious low pricing. This is classified as a Level 1 warning, and the corresponding anomaly type is a normal fluctuation anomaly of non-malicious low pricing. Such anomalies are mostly due to normal market supply and demand changes and short-term promotions. If the price deviation is due to non-malicious factors such as price fluctuations, excessive intervention is unnecessary. If the price deviation is within the preset deviation range, it indicates that the current price deviation is moderate and there are signs of suspected malicious underpricing, but it has not yet reached the standard for confirming malicious underpricing. This is classified as a Level 2 warning, and the corresponding anomaly type is suspected malicious underpricing. Such anomalies should be promptly alerted to relevant personnel to prevent the deviation from further expanding. If the price deviation is greater than or equal to the preset upper limit of the deviation range, it indicates that the current price deviation is serious and has clearly constituted malicious underpricing behavior. This is classified as a Level 3 warning, and the corresponding anomaly type is confirmed malicious underpricing. Such anomalies will disrupt market order, damage brand image and industry interests, and control measures must be taken immediately.After determining the warning level and anomaly type, corresponding control measures are taken according to the preset control rules for different warning levels. The specific process is as follows: For a Level 1 warning, since the anomaly type is a normal fluctuation anomaly without malicious low prices, the deviation is minor and there is no malicious intent. Only recording and real-time monitoring measures are implemented, that is, recording relevant information about the anomaly in detail, including the time of occurrence, the corresponding channel, the product name, the price deviation range, etc., and tracking the price change trend of the product in real time. If the price deviation range further expands and reaches a higher warning level, the corresponding level of control measures are immediately triggered. For a Level 2 warning, since the anomaly type is suspected malicious low prices, timely intervention is required to avoid the risk from escalating. The control measures taken are to send a reminder notification to the person in charge of the corresponding channel. The notification clearly outlines the abnormal product information, price deviations, warning levels, and the criteria for determining suspected malicious low-price pricing. Channel managers are required to investigate the cause of the anomaly within a specified timeframe and report the results and corrective measures. They are also required to continuously monitor product prices in the channel and track the effectiveness of corrective actions. For Level 3 warnings, since the anomaly is confirmed as malicious low-price pricing, which has already caused adverse effects, strict control measures are necessary. The sale and display of products with abnormal prices in the channel must be immediately suspended, and the continued circulation of such products in the channel is prohibited. The channel manager is notified to immediately rectify the situation. The product can only be sold and displayed again after rectification is completed, prices return to normal, and verification is passed. This ensures that malicious low-price behavior is promptly curbed, maintaining normal market order and the product price system. The entire warning judgment and control process achieves tiered control of abnormal price deviations, avoiding excessive intervention in normal price fluctuations while promptly curbing malicious low-price behavior, ensuring the scientific, targeted, and efficient nature of price control.

Claims

1. A method for intelligent monitoring and early warning of product prices across all channels, characterized in that, The method includes: S1. Establish a multi-source, multi-channel data acquisition architecture to achieve real-time capture of product price-related data from various channels. Preprocess the captured raw data, standardize product identifiers from different channels through a unified product coding mapping, and perform standardization and normalization on the preprocessed data to generate a standardized price dataset. The price-related data includes basic product information, real-time selling price, promotional price, pricing rules, inventory data, and competitor price data. S2, based on a standardized price dataset, constructs a dynamic price threshold control system to achieve real-time control of dynamic feature thresholds, dynamically corrects the initial feature thresholds to ensure that the feature thresholds match the market dynamics in real time, and performs multi-dimensional anomaly analysis on the standardized price dataset after real-time control and dynamic correction, compares the real-time price of the product with the dynamic feature thresholds, distinguishes between normal price fluctuations and malicious low-price behavior, and generates a complete anomaly price analysis report. S3, based on the generated abnormal price analysis report, extracts abnormal information related to malicious low prices, constructs a multi-level early warning mechanism, classifies early warning levels according to the magnitude of price anomalies, and takes corresponding control measures. It also classifies and archives the original price data, price analysis results, and early warning records to support the long-term retention and rapid query of historical price data.

2. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 1, characterized in that, The specific process for generating the standardized price dataset is as follows: Through a unified product coding mapping mechanism, product identifiers from different channels are mapped one by one to standard codes, achieving unique correspondence and standardization of product identifiers from different channels and in different formats. The product identifiers include online e-commerce platform product IDs, offline store barcodes, cross-border product codes, and third-party seller product codes. After the product identification standardization is completed, the data is normalized and the product price data of each channel is mapped to the preset range of [0,1] to eliminate the difference in dimensions between different channels and different data levels; The data that has undergone standardization and normalization is validated. If the validation passes, the data is classified and archived according to the product's unified code, channel type, and data collection time to generate a standardized price dataset. If the validation fails, the data that fails the validation is removed, and the missing valid fields are filled in using an interpolation algorithm.

3. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 1, characterized in that, The specific process for constructing dynamic price threshold control is as follows: The basic parameters of the preset price sample window include the initial size of the price sample window, the minimum size of the price sample window, the maximum size of the price sample window, and the adjustment step size of the sample window. The price sample window is used to select valid historical price samples from the standardized price dataset. The valid samples are price data after abnormal quotes have been removed by sample filtering rules. Real-time acquisition of current price data from the standardized price dataset; extraction of historical price valid samples within the window based on the current size of the price sample window; calculation of statistical characteristic parameters of the historical price valid samples, including sample mean and sample standard deviation; recording the result of processing the ratio of sample standard deviation to sample mean as the volatility coefficient of the current price sample window; the volatility coefficient of the current price sample window is used to characterize the dispersion of historical price valid samples within the current price sample window. A preset volatility coefficient threshold range is defined. The volatility coefficient of the current price sample window is compared with the volatility coefficient threshold range. The size of the price sample window is dynamically adjusted based on the comparison result. The volatility coefficient threshold range represents the open interval formed by the lower limit and upper limit of the volatility coefficient.

4. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 3, characterized in that, The specific process for dynamically adjusting the size of the price sample window based on the comparison results is as follows: A pre-constructed correspondence between volatility coefficient and sample window size is used to characterize the unique correspondence between the volatility coefficient of the price sample window and the adjustment amount of the price sample window size. If the volatility coefficient is less than or equal to the lower limit of the volatility coefficient, it is determined that the price fluctuation within the current price sample window is smooth and the dispersion is low. Then, the volatility coefficient of the current price sample window is input into the volatility coefficient-sample window size correspondence, and the price sample window size is adjusted upward. The result of coupling the current price sample window size and the price sample window size adjustment is rounded up as the target price sample window size. It is determined whether the adjusted target price sample window size is greater than or equal to the preset maximum price sample window size. If so, the maximum price sample window size is used as the final target price sample window size. If not, the target price sample window size remains unchanged.

5. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 4, characterized in that, The method of dynamically adjusting the size of the price sample window based on the comparison results also includes: If the volatility coefficient is greater than or equal to the upper limit of the volatility coefficient, it is determined that the price fluctuation within the current price sample window is drastic and the dispersion is high. Then, the volatility coefficient of the current price sample window is input into the volatility coefficient-sample window size correspondence, and the price sample window size is reduced by an amount. The difference between the current price sample window size and the price sample window size reduction amount is processed and rounded down to the nearest integer as the target price sample window size. It is determined whether the adjusted target price sample window size is less than or equal to the preset minimum price sample window size. If so, the minimum price sample window size is used as the final target price sample window size. If not, the target price sample window size remains unchanged. If the volatility coefficient is within the volatility coefficient threshold range, it is determined that the price fluctuation within the current price sample window is within a reasonable range, and the price sample window size remains unchanged.

6. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 1, characterized in that, The specific process for dynamically correcting the initial feature threshold is as follows: Preset initial feature threshold basic parameters, including initial feature threshold baseline value, initial feature threshold minimum threshold, initial feature threshold maximum threshold, and feature threshold adjustment coefficient. The initial feature threshold is used to make preliminary anomaly judgment on price features in the standardized price dataset. The price features include price volatility, price month-on-month change rate, and price year-on-year change rate. Real-time acquisition of price deviation threshold, collection of current price data and corresponding historical price valid samples in standardized price dataset, calculation of current price deviation, wherein the current price deviation is the absolute value of the difference between the current price and the mean of the historical price valid samples in the corresponding price sample window; Calculate the price deviation threshold utilization rate, which is the ratio of the current price deviation to the price deviation threshold, used to characterize the proportion of the current price deviation relative to the price deviation threshold; A preset price deviation threshold utilization rate range is defined. The calculated price deviation threshold utilization rate is compared with the preset price deviation threshold utilization rate range. Based on the comparison result, the initial feature threshold is dynamically adjusted according to the preset feature threshold adjustment coefficient. The preset price deviation threshold utilization rate range represents the open interval formed by the lower bound of the price deviation threshold utilization rate and the upper bound of the price deviation threshold utilization rate.

7. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 6, characterized in that, The specific process for dynamically adjusting the initial feature threshold based on a preset feature threshold adjustment coefficient is as follows: A threshold utilization rate-feature threshold adjustment coefficient matrix is ​​pre-constructed, which is used to characterize the unique correspondence between the price deviation threshold utilization rate and the initial feature threshold adjustment coefficient; If the price deviation threshold utilization rate is less than or equal to the lower bound of the price deviation threshold utilization rate, it is determined that the current price deviation is far below the price deviation threshold, and the price fluctuation is in an extremely flat state. The current price deviation threshold utilization rate is input into the threshold utilization rate-feature threshold adjustment coefficient matrix, and the initial feature threshold gain coefficient is output. The current initial feature threshold and the initial feature threshold gain coefficient are combined to obtain the target initial feature threshold. It is determined whether the target initial feature threshold is greater than or equal to the preset initial feature threshold maximum value. If so, the preset initial feature threshold maximum value is used as the final target initial feature threshold. If not, the target initial feature threshold is kept unchanged.

8. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 7, characterized in that, The dynamic adjustment of the initial feature threshold also includes: If the price deviation threshold utilization rate is greater than or equal to the upper bound of the price deviation threshold utilization rate, it is determined that the current price deviation is close to or has reached the price deviation threshold, and the price fluctuation is in a critical abnormal state. The current price deviation threshold utilization rate is input into the threshold utilization rate-feature threshold adjustment coefficient matrix, and the initial feature threshold reduction coefficient is output. The current initial feature threshold and the initial feature threshold reduction coefficient are combined to obtain the target initial feature threshold. It is determined whether the target initial feature threshold is less than or equal to the preset initial feature threshold minimum value. If so, the preset initial feature threshold minimum value is used as the final target initial feature threshold. If not, the target initial feature threshold is kept unchanged. If the utilization rate of the price deviation threshold is within the preset price deviation threshold utilization rate range, it is determined that the current price deviation is within a reasonable range, the price fluctuation is stable, and the current initial feature threshold remains unchanged.

9. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 1, characterized in that, The specific process for performing multi-dimensional anomaly analysis on the standardized price dataset is as follows: The adjusted target initial feature threshold is used as the new initial feature threshold and applied to the preliminary judgment of price feature anomalies in the standardized price dataset. If the price volatility is greater than the new initial feature threshold, it is judged as a preliminary anomaly and enters the second judgment stage; if the price volatility is less than or equal to the new initial feature threshold, it is judged as a normal price feature. The specific process for entering the secondary judgment stage is as follows: By comparing the real-time selling price of products in each channel with the normal price fluctuation range threshold and the malicious low-price judgment benchmark threshold corresponding to that channel, a preliminary judgment result is obtained. If the real-time selling price of a product is within the threshold of the normal price fluctuation range corresponding to the channel, then the fluctuation of the real-time selling price of the product in that channel is determined to be a normal price fluctuation. If the real-time selling price of a product is less than the threshold for malicious low-price judgment, then the behavior corresponding to the real-time selling price of the product on that channel is determined to be malicious low-price behavior.

10. The method for intelligent monitoring and early warning of product prices across all channels as described in claim 1, characterized in that, The specific process for classifying early warning levels based on the magnitude of price anomalies is as follows: If the price deviates from the benchmark by less than or equal to the lower limit of the benchmark deviation, it is judged as a Level 1 warning, and the abnormality type is normal fluctuation abnormality of non-malicious low price; If the price deviates from the benchmark by a certain amount within the deviation range, a Level 2 warning is issued, and the anomaly type is suspected malicious low price. The deviation range refers to the open interval formed by the lower limit of the deviation from the benchmark and the upper limit of the deviation from the benchmark. If the price deviates from the benchmark by a greater than or equal to the upper limit of the benchmark deviation, it is judged as a Level 3 warning, and the anomaly type is confirmed malicious low price; The specific procedure for taking corresponding control measures is as follows: The Level 1 warning only records and monitors in real time; the Level 2 warning sends a reminder notification to the corresponding channel manager; the Level 3 warning suspends the sales display of abnormal prices on that channel.

Images