Sales strategy tendency real-time judgment and closed-loop optimization system based on edge nodes

Abstract

The invention discloses a sales strategy tendency real-time judgment and closed-loop optimization system based on edge nodes, and relates to the technical field of strategy judgment decision, and the system comprises a data obtaining module which is used for obtaining the business flow data and the spatial environment semantic data of the edge nodes corresponding to all sales terminals in the current sampling period; the field state mapping module is used for constructing a strategy influence factor field based on the spatial environment semantic data and projecting the service flow data into the strategy influence factor field to form strategy form distribution mapping; the evolutionary analysis module is used for performing differential evolutionary analysis based on the strategy form distribution mapping of the current period in the next sampling period, and generating a form drift vector set representing the change direction and the change amplitude of the strategy tendency; the method has the beneficial effects that error accumulation caused by dynamic environment change is effectively inhibited, and the long-term stability of sales strategy tendency judgment is further ensured.

Description

Technical Field

[0001] This invention relates to the field of strategy determination and decision-making technology, and in particular to a real-time determination and closed-loop optimization system for sales strategy preferences based on edge nodes. Background Technology

[0002] This invention relates to the fields of distributed computing and intelligent decision-making technology, specifically to a real-time sales strategy preference determination and closed-loop optimization system based on edge nodes. With the deep integration of new retail business and edge computing technology, processing massive amounts of business flow data in real time by deploying nodes on the edge of each sales terminal, and quickly determining sales preferences and issuing optimization strategies accordingly, has become a key means to improve business decision-making efficiency and respond to transient market changes.

[0003] However, traditional sales strategy decision-making schemes typically rely on pre-defined linear feedback mechanisms, which often exhibit significant lag and vulnerability when dealing with complex business flows with strong spatiotemporal correlations. Specifically, existing closed-loop optimization logics focus primarily on numerical compensation for output results, neglecting the structural distortions caused by external environmental disturbances to the underlying decision feature space. The lack of a technical means to adaptively constrain and geometrically correct the decision influence domain based on actual feedback makes the system highly susceptible to deviations from its pre-defined stable execution range when facing sudden nonlinear environmental disturbances, resulting in severe decision biases and making it difficult to achieve a balance between decision accuracy and control stability in dynamically changing business environments. Summary of the Invention

[0004] In view of the above-mentioned prior art, this application is made. Embodiments of this application provide a real-time sales strategy preference determination and closed-loop optimization system based on edge nodes, which effectively suppresses error accumulation caused by dynamic environmental changes and further ensures the long-term stability of sales strategy preference determination.

[0005] According to one aspect of this application, a real-time sales strategy preference determination and closed-loop optimization system based on edge nodes is provided, comprising:

[0006] Data acquisition module: used to acquire business flow data and spatial environment semantic data of the edge nodes corresponding to each sales terminal within the current sampling period;

[0007] Field mapping module: used to construct a strategy influence factor field based on spatial environment semantic data, and project business flow data onto the strategy influence factor field to form a strategy form distribution mapping;

[0008] Evolutionary Analysis Module: Used to perform differential evolutionary analysis based on the strategy morphology distribution mapping of the previous sampling period within the current sampling period, generating a set of morphological drift vectors that characterize the direction and magnitude of strategy tendency changes.

[0009] Regionalization identification module: used to identify abnormal evolution regions based on the evolution trend of morphological drift vectors, and generate policy trigger confidence based on the drift vector characteristics of abnormal evolution regions;

[0010] Judgment module: used to trigger confidence based on the generated strategy and determine whether intervention is needed;

[0011] If the judgment result is yes, a strategy intervention instruction is output. The business feedback data after the execution of the strategy intervention instruction is fed back into the strategy morphology distribution mapping. The evolution trend of the morphology drift vector is updated according to the fed-back business feedback data, and the morphology constraint boundary of the strategy influence factor field is adjusted to optimize the stability of the strategy tendency in the next sampling period.

[0012] If the judgment result is negative, no intervention is made, and the existing strategy is maintained.

[0013] According to another aspect of this application, an electronic device is provided, including a memory and a processor, the memory being used to store computer-executable instructions, and the processor being used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the system described above.

[0014] According to another aspect of this application, a computer storage medium is provided that stores computer-executable instructions thereon, which, when executed by a processor, implement the steps of the system described above.

[0015] Compared with existing technologies, the real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to the embodiments of this application can effectively limit the determination logic within the stable execution domain defined by the semantics of the physical environment by adjusting the morphological constraint boundary. This avoids the decision logic collapse caused by traditional solutions that only make numerical fine-tuning of the results, and significantly improves the decision robustness of the system in complex environments. This invention can not only identify changes in business volume, but also quantify the drift direction and acceleration of the strategy preference in the feature space, enabling earlier detection of abnormal evolution zones and providing a more forward-looking confidence basis for strategy intervention, thus reducing determination delay. By comparing the geometric deviation between actual feedback and evolution prediction, the morphological constraint boundary of the next sample is directly updated. This is no longer a lagging parameter correction, but a real-time correction of the determination logic basis, effectively suppressing the accumulation of errors caused by dynamic changes in the environment, and further ensuring the long-term stability of the sales strategy preference determination. Attached Figure Description

[0016] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 This is a schematic diagram of the overall process of the real-time determination and closed-loop optimization system for sales strategy preferences based on edge nodes according to the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the expansion of the feature dimensions of the real-time determination and closed-loop optimization system for sales strategy preferences based on edge nodes according to the present invention.

[0019] Figure 3 This is a schematic diagram illustrating the stability index judgment of the real-time sales strategy tendency determination and closed-loop optimization system based on edge nodes according to the present invention. Detailed Implementation

[0020] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0021] In existing technologies, edge computing-based real-time decision-making systems have been widely applied in retail terminals. However, in complex and ever-changing spatial environments, the accurate determination of sales strategies still faces significant challenges. For terminal scenarios with strong dynamic interference, external environmental fluctuations (such as local instantaneous hotspots and spatial semantic abrupt changes) can cause nonlinear distortions in the collected business flow features. Traditional adjustment methods based on fixed parameters or linear feedback are difficult to identify and correct the morphological shift of the underlying decision logic in the feature space. This makes it easy for the judgment results to deviate from the stable execution range, resulting in decision oscillations or cumulative biases, which seriously affect the stability and reliability of strategy execution.

[0022] To address the aforementioned issues, the applicant discovered that the core reason for the failure of sales strategy judgment lies in the lack of geometric constraints and evolutionary control capabilities within the physical feature space of the decision-making logic. Analysis revealed that business flow data and spatial environmental semantics, after coupling, exhibit specific field-state characteristics, and the evolution of these characteristics within adjacent sampling periods is continuous, allowing for the prediction of target morphology through evolutionary trends. Based on this, a technical approach is proposed: constructing a strategy influence factor field, performing differential evolution analysis, and introducing dynamic adjustment of morphological constraint boundaries. By projecting business flow onto the influence factor field to form a morphological distribution mapping, cross-period morphological drift vector sets are identified to locate abnormal evolutionary areas. Furthermore, the drift trend is corrected in real-time based on post-execution business feedback data, thereby dynamically adjusting the morphological constraint boundaries of the influence factor field. This forces the judgment logic to be anchored within a stable execution domain defined by the physical environment, solving the problem of distorted decision-making logic.

[0023] Example 1:

[0024] Reference Figures 1-3 As an embodiment of the present invention, a real-time sales strategy preference determination and closed-loop optimization system based on edge nodes is provided, including: a data acquisition module, a field state mapping module, an evolution analysis module, a zoning identification module, and a judgment module.

[0025] Specifically Figure 1 The figure illustrates a real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to an embodiment of this application, including:

[0026] Data acquisition module: used to acquire business flow data and spatial environment semantic data of the edge nodes corresponding to each sales terminal within the current sampling period.

[0027] Specifically, edge nodes extract real-time transaction data for each sales terminal within the current sampling period, including transaction timestamps, SKU identifiers, transaction amounts, and member interaction frequency. To ensure data accuracy, edge nodes perform cleaning and normalization on the raw transaction data, removing duplicate and abnormal transaction records, thereby obtaining business flow data that reflects the sales activity status of the terminals within the sampling period.

[0028] Edge nodes acquire contextual information about the physical space where the sales terminal is located through associated sensors and external service interfaces. Specifically, the system uses visual sensors to acquire real-time customer flow density around the terminal, uses meteorological data interfaces to acquire the current temperature, humidity and weather conditions, and obtains real-time promotional dynamics of surrounding competitors from the business district information platform. Subsequently, the system uses semantic extraction technology to label the above multi-source unstructured information, identify key spatial environmental factors that affect sales fluctuations, and thus obtain spatial environmental semantic data that reflects the potential constraints of the physical environment on sales behavior. The obtained spatial environmental semantic data will serve as the logical basis for subsequently constructing a strategy influencing factor field.

[0029] Field mapping module: Used to construct a strategy influence factor field based on spatial environment semantic data, and project business flow data onto the strategy influence factor field to form a strategy morphology distribution mapping. The specific implementation is as follows:

[0030] First, the acquired spatial environment semantic data is parsed, and the unstructured environmental features are transformed into the geometric constraint dimensions of the strategy influence factor field. Specifically, the system uses geographical location coordinates, population density level, weather influence coefficient, and competition intensity index as multi-dimensional spatial axes to construct a strategy influence factor field that reflects the interference of the external environment on sales behavior. In this strategy influence factor field, each coordinate point represents a specific combination of environmental features, and the interaction of different environmental factors is defined as the potential energy distribution within the field.

[0031] Secondly, the collected business flow data is dimensionally aligned. Based on the sales timestamp, terminal ID, and transaction type in the business flow data, discrete sales records are converted into data points in the strategy's influencing factor field through a preset mapping function. The projection process refers to using the characteristic attributes (such as average order value and conversion rate) in the business flow data as quality attributes in the strategy's influencing factor field, and accurately mapping them to the coordinate area corresponding to the strategy's influencing factor field based on the spatial semantic background (such as the weather and pedestrian flow at the time) when the business occurs.

[0032] Furthermore, the system performs spatial density aggregation on the massive business flow data points projected into the field, calculates the data clustering strength of each local area, and thus forms a strategy pattern distribution map. This strategy pattern distribution map, in the form of a cloud map or probability distribution map, intuitively presents the real-time pattern distribution of sales strategies under the current environmental constraints. This mapping not only includes the quantitative value of sales data, but also reflects the topological structure of sales behavior distribution in the environmental factor field, providing a data carrier with spatial logic for subsequent evolutionary analysis.

[0033] Evolution Analysis Module: Used to perform differential evolution analysis based on the strategy morphology distribution mapping of the previous sampling period within the current sampling period, generating a set of morphological drift vectors that characterize the direction and magnitude of strategy tendency changes.

[0034] It should be noted that the generation of the morphological drift vector set includes:

[0035] The system extracts the strategy pattern distribution mapping for the current sampling period and retrieves the strategy pattern distribution mapping stored in the previous sampling period. Specifically, the system extracts the strategy pattern distribution mapping for the current sampling period and attempts to retrieve the strategy pattern distribution mapping stored in the previous sampling period. It should be noted that if the current sampling period is the first period (i.e., the initial running phase) and the system has not yet stored historical mapping data, it will automatically call the preset benchmark pattern distribution mapping (such as the static field distribution preset based on historical averages or industry experience) as the replacement value for the strategy pattern distribution mapping stored in the previous sampling period to ensure the executability and continuity of the difference operation.

[0036] The strategy morphology distribution mappings of adjacent sampling periods are compared differentially, and the eigenvalue differences of each mapping point in the strategy morphology distribution mapping are calculated. The specific formula is as follows:

[0037] ;

[0038] in, Let i be the eigenvalue difference component of the mapping point i. This represents the feature value corresponding to the mapping point i within the current sampling period. This is the feature value corresponding to the mapping point i in the previous sampling period;

[0039] The eigenvalue differences of each mapping point are calculated. Then, the system is based on its spatial coordinates within the strategy influence factor field. Perform logical indexing and sorting, and summarize to generate a morphological difference matrix. The morphological difference matrix It fully records the temporal variation of feature density at each mapping point within the entire domain;

[0040] By utilizing the correlation between preset coordinate points in the strategy influence factor field, the instantaneous offset vector of each mapping point in the morphological difference matrix within the strategy influence factor field is extracted.

[0041] It should be further explained that the extraction of the instantaneous offset vector includes:

[0042] Based on the pre-defined physical location relationships between coordinate points in the strategy influence factor field, the spatial weight distribution corresponding to each mapping point is determined, as shown in the following formula:

[0043] ;

[0044] Where di(i,j) is the Euclidean distance between point i and the j-th adjacent coordinate point in the neighborhood. The index of the adjacent coordinates within the neighborhood is N, where N is the preset total number of adjacent points within the neighborhood.

[0045] By using spatial weight distribution to perform weighted cluster analysis on the eigenvalue differences of each mapping point in the morphological difference matrix, the geometric center displacement value representing the core region of morphological evolution is identified. The specific formula is as follows:

[0046] ;

[0047] in, This represents the displacement value of the geometric center of the core region of morphological evolution. Let be the position vector of mapping point i within the policy influence factor field, and M be the total number of mapping points participating in clustering calculation in the morphological difference matrix;

[0048] The initial displacement direction of each mapping point is corrected based on the displacement value of the geometric center to obtain the instantaneous offset vector that conforms to the physical distribution law of the strategy influence factor field. The specific formula is as follows:

[0049] ;

[0050] in, This is the corrected instantaneous offset vector. Let i be the initial displacement vector of each mapping point i. This is the preset correction adjustment coefficient. ;

[0051] Based on the instantaneous offset vector The geometric orientation and modulus are used to generate a set of morphological drift vectors representing the global evolutionary trajectory of the representation strategy. .

[0052] The zoning identification module is used to identify anomalous evolution zones based on the evolutionary trend of morphological drift vectors, and to generate policy trigger confidence scores based on the drift vector characteristics of the anomalous evolution zones. The specific implementation is as follows:

[0053] The system analyzes the evolution trend of each vector in the morphological drift vector set, identifies abnormal evolution zones, and establishes a grid-based spatial evaluation matrix to divide the strategy influencing factor field into several cells. It also counts the dispersion of the morphological drift vectors in each cell. When the consistency of vector direction in a certain region is lower than a preset threshold, or the average rate of change of vector magnitude exceeds a preset fluctuation threshold, the system marks the region as an abnormal evolution zone. This abnormal evolution zone represents that the sales strategy has experienced unexpected logical fluctuations under this environmental combination.

[0054] The system extracts drift vector features from the anomalous evolution region, which include the average magnitude of vectors within the anomalous evolution region. (Characteristic of deviation intensity) and the distribution entropy value of vector direction (Characterizing the degree of evolutionary disorder), followed by drift vector features, generating policy trigger confidence, the specific formula is as follows:

[0055] ;

[0056] in, For policy trigger confidence, The average magnitude of all instantaneous offset vectors within the anomalous evolution region. The preset baseline evolution radius serves as a reference for the displacement deviation strength. This represents the dimensionless distribution entropy value of vector directions within the anomalous evolution region, used to measure the certainty of the evolutionary trend. It is a natural constant; this formula describes the judgment logic of strategic intervention, when the displacement intensity of the abnormal evolution zone... Exceeding the baseline evolution radius When the ratio term increases, the strategy trigger confidence C rises, indicating that the strategy deviation has exceeded the safety threshold and intervention is highly necessary; if the distribution entropy value at this time... The large value indicates that the deviation is caused by disordered fluctuations, and the distribution entropy value is relatively large. This will act as a hedge against the exponential term, suppressing the blind rise of the strategy trigger confidence C, thereby achieving precise control over strategy triggering through the mutual constraints of physical features.

[0057] Judgment module: Used to trigger confidence based on the generated strategy and determine whether intervention is needed.

[0058] If the judgment result is yes, then output the strategy intervention instruction, collect the business feedback data after the strategy intervention instruction is executed, and backfeed it to the strategy morphology distribution mapping. Update the morphology drift vector evolution trend based on the backfeed business feedback data, and adjust the morphology constraint boundary of the strategy influence factor field to optimize the stability of the strategy tendency in the next sampling period.

[0059] The system collects business feedback data after the strategy instructions are executed. Specifically, edge nodes access the sales terminal's execution logs in real time through a pre-defined monitoring interface to obtain real-time transaction volume, conversion rate, and user retention fluctuations within a pre-defined observation period after the strategy instructions are issued. The system then performs differential processing on the above raw feedback information and the data before strategy execution to remove environmental background noise, thereby obtaining business feedback data that truly reflects the effect of the strategy intervention.

[0060] The business feedback data is fed back into the strategy pattern distribution mapping. Specifically, the system extracts the terminal identifier and its corresponding spatial attributes (such as the time when the feedback occurred, the geographical location, and the combination of environmental parameters at that time) contained in the business feedback data, retrieves the matching coordinate points in the constructed strategy influence factor field, and then uses the conversion rate increment in the business feedback data as the quality feature value. The system injects the feedback feature into the corresponding field coordinates through the coordinate alignment function, thereby superimposing a real-time feedback layer on the original strategy pattern distribution mapping.

[0061] It should be noted that the updating of evolutionary trends and the adjustment of morphological constraint boundaries include:

[0062] The business feedback data is mapped to the strategy morphology distribution mapping, and the morphological deviation entropy between the actual morphological position corresponding to the business feedback data and the target position predicted by the morphological drift vector set is calculated. The specific formula is as follows:

[0063] ;

[0064] in, For the form to deviate from entropy, This represents the true shape and position vector corresponding to the k-th feedback data point. Let k be the predicted target position vector corresponding to the k-th feedback data point. Here, Q is the index of the feedback points, and Q is the total number of feedback points. This is the ratio of the deviation of the actual feedback point from the predicted point in physical space to the system's tolerance radius. Used to measure the suddenness of a point's deviation from the predicted trajectory;

[0065] Based on the morphological deviation entropy, the correction weight of the morphological drift vector evolution trend is determined, the correction weight is used to compensate for the deviation of the evolution trend, and the evolution trend is updated by superimposing the correction factors.

[0066] Based on the magnitude of the morphological deviation entropy, the spatial scaling coefficient of the morphological constraint boundary of the strategy influencing factor field is determined, and the contraction or expansion state of the morphological constraint boundary is adjusted using the spatial scaling coefficient.

[0067] Specifically, the system determines the morphological deviation from entropy. The size of the value determines the spatial scaling coefficient of the policy influencing factor field. This spatial scaling coefficient is then used to directly act on the geometric equation of the current morphological constraint boundary to change the contraction or expansion state of the boundary. For example, when the morphological deviation entropy exceeds a preset stability threshold, the spatial scaling coefficient drives the boundary to contract towards the geometric center. By reducing the policy effective spatial domain in the next sampling period, the stability of the policy tendency is forced to be optimized.

[0068] It should be further explained that the determination of the corrected weights and spatial scaling coefficients includes:

[0069] The formula for calculating the rate of change of morphological deviation entropy over time with the sampling period is as follows:

[0070] ;

[0071] in, The time-domain rate of change of morphological deviation from entropy. It is a scalar representing the time interval between two adjacent sampling periods;

[0072] Based on the magnitude of the time-domain rate of change, a corresponding damping factor is matched within a preset damping function. The damping function characterizes the system's ability to suppress the severity of deviation fluctuations. When the value is large (indicating rapid spread of deviation), the system is matched with a higher damping factor to prevent overshoot in subsequent correction actions;

[0073] The shape deviation entropy is weighted by a damping factor to generate a smoothed and corrected deviation control variable, as shown in the following formula:

[0074] ;

[0075] in, These are deviation control variables, used for subsequent mapping to generate control parameters. Here, represents the damping factor, a constant coefficient with the dimension of time. The absolute value of the rate of change in the time domain. It is a preset reference time constant used to normalize the rate of change term; it is a constant with time dimensions.

[0076] Corrected weights and spatial scaling coefficients are generated by mapping the deviation control variables respectively;

[0077] The system utilizes a linear mapping function The correction weights used to correct the vector evolution trend are generated, and their calculation formula is as follows:

[0078] ;

[0079] in, The weighting is used to adjust the magnitude of the correction for the shape drift vector in the next cycle. The preset linear gain coefficient, As a deviation control variable, This is a preset base correction weight bias term;

[0080] The system utilizes a nonlinear mapping function The spatial scaling factor used to adjust the boundary morphology is calculated using the following formula:

[0081] ;

[0082] in, This is the spatial scaling factor, representing the scaling ratio of the shape constraint boundary relative to the current position. This is the maximum scaling limit factor. The nonlinear sensitivity coefficient It is an exponential function with the natural constant e as its base;

[0083] The above mapping logic establishes the response relationship between the deviation control variable and the actuator, and for the correction weight... The use of linear mapping ensures the stability of trend updates, enabling the system to linearly compensate for the evolution trajectory based on the magnitude of the morphological deviation from entropy. For the spatial scaling coefficient S, a nonlinear mapping in the form of a negative exponential form is used. The technical logic is that when the deviation control variable U is small, the spatial scaling coefficient S is close to... To maintain boundary stability, once U increases (i.e., feedback deviation or volatility surges), due to the characteristics of the exponential function, S will rapidly decay to a smaller value, driving the morphological constraint boundary to contract centripetally. This composite control logic of "linear compensation for small deviations and exponential contraction for large deviations" can greatly improve the system's decision stability in complex business environments. Finally, the system utilizes the generated correction weights... The evolution trend of the updated morphological drift vector is used, and the morphological constraint boundary of the influencing factor field is adjusted synchronously using the spatial scaling coefficient S, thereby completing the closed-loop optimization of the entire sampling period.

[0084] If the judgment result is negative, no policy intervention will be implemented, and the existing policy will continue to be executed, as follows:

[0085] First, the system performs data entry and status marking. Although no intervention instructions are output in the current period, the edge nodes will still store the policy pattern distribution mapping, pattern drift vector set and corresponding policy trigger confidence generated in the current sampling period in a structured manner, and mark them with the logical label of "normal evolution". These data will serve as historical reference benchmarks for subsequent periods to support trend prediction across long periods.

[0086] Secondly, the system performs quasi-steady-state maintenance of the policy influence factor field. Since there is no need to dynamically adjust the morphological constraint boundary, the system directly transmits the morphological constraint boundary parameters of the current period to the next sampling period. This maintenance operation avoids the edge nodes from performing meaningless boundary recalculation when the environmental fluctuation is extremely small, effectively reducing the computational overhead and energy consumption on the edge side.

[0087] Finally, the system initiates the pre-triggered monitoring cycle for the next period. While maintaining the execution of the existing strategy, the edge nodes keep real-time monitoring of business flow data and spatial environment semantic data to ensure that they can quickly switch from the maintenance state to the intervention state when a sudden shift occurs in the environment. Through the self-alignment mechanism in the non-intervention state, this invention ensures the continuity of system judgment and the economy of edge node operation when business evolution is in a stable range, and constructs a complete closed-loop discrimination system.

[0088] This application proposes a closed-loop adjustment mechanism based on morphological deviation entropy and deviation control variables. Its core concept originates from the adaptive disturbance compensation idea in industrial control theory. In complex edge sales scenarios, the execution environment of the strategy is dynamically changing, and the preset morphological drift vector set inevitably has a prediction bias. Traditional strategy adjustments are mostly discrete or delayed, that is, manual intervention is required after a significant decline in performance is detected. The operational logic of this solution aims to establish an automated closed loop of "instantaneous perception, smooth feedback, and gradient response". By calculating the "information entropy difference (morphological deviation entropy)" between the actual position and the predicted position, the disorder of strategy failure is quantified, thereby achieving nonlinear fine control.

[0089] In contrast, while conventional PID control can achieve feedback regulation, its linear gain cannot handle high-dimensional spatial expansion or contraction demands when facing "strategy morphology distribution mapping" with spatial distribution characteristics. Rule-based logic switching, which sets hard thresholds for strategy switching, lacks a buffer mechanism and is prone to frequent system oscillations at the threshold edge. This solution, by introducing a damping factor and deviation control variable, achieves a composite control strategy of "linear compensation for small deviations and exponential contraction for large deviations." This not only ensures the smoothness of strategy updates but also greatly improves the robustness of edge nodes in the unattended state by forcibly locking in risks at the contraction boundary when extreme anomalies occur.

[0090] For example, suppose that within a certain sampling period T, the system is in a two-dimensional policy influence factor field:

[0091] (1) Calculation of morphological deviation entropy and acquisition of time-domain change rate

[0092] In the current period T, the target position vector predicted by the morphological drift vector set After the strategy command is executed, the actual shape and location vector corresponding to the returned business feedback data. Let the system tolerance radius be... ;

[0093] Morphological deviation entropy calculation: The Euclidean distance between the actual morphological position and the predicted target position is: ;

[0094] The ratio of the deviation of the actual feedback point from the predicted point in physical space to the system's tolerance radius is: ;

[0095] Assuming there are multiple feedback points in the current period, the morphological deviation entropy of the current sampling period is calculated after logarithmic operation and summation. Given that the shape deviates from the entropy of the previous sampling period ;

[0096] Set sampling interval The time-domain rate of change of morphological deviation from entropy ;

[0097] (2) Generation of deviation control variables

[0098] Set damping factor Reference time constant ;

[0099] Deviation control variable calculation The morphological deviation entropy was smoothed by using a damping factor, transforming the original deviation value of 0.8 into a deviation control variable of 0.5.

[0100] (3) Execution of the mapping between weights and coefficients

[0101] Corrected weight generation: Let , Then adjust the weights. Spatial scaling factor generation: Let , Then the spatial scaling factor Because the morphological deviation from entropy is on an upward trend, the resulting spatial scaling coefficient... The system drives the current shape constraint boundary to shrink towards the geometric center to about 55% of the original boundary range, forcibly reducing the effective space domain of the strategy;

[0102] In the next sampling period, the system uses a correction weight of 0.3 to compensate for the deviation in the evolution trend of the morphological drift vector, so that the predicted position is corrected to the actual fluctuation path. At the same time, it uses enhanced morphological constraint boundaries to perform optimization. If the morphological deviation entropy decreases in the next period, the system will enter a non-intervention state to perform data entry and quasi-steady-state maintenance, ensuring the economic efficiency of system operation.

[0103] Figure 2 This is a schematic diagram illustrating the expansion of the feature dimensions of the real-time sales strategy tendency determination and closed-loop optimization system based on edge nodes according to the present invention.

[0104] This application further proposes that the system also includes:

[0105] Based on the distribution characteristics of the abnormal evolution zone, auxiliary feature sources are selected from the preset feature source set. The feature source set includes at least one or more of the following: social media sentiment index (reflecting consumers' subjective will), real-time regional economic indicators (reflecting macro purchasing power), and competitor dynamic information flow (reflecting the market competition environment) associated with the edge nodes.

[0106] Specifically, the edge nodes maintain a feature source label library. Each auxiliary feature source (such as social media sentiment index, regional economic indicators) is assigned a corresponding spatial influence category label and business relevance label. Based on the geospatial coordinates of the abnormal evolution zone and the product category attributes in the current business flow data of the zone, the system uses a similarity matching algorithm to select the feature source with the highest spatial overlap and the strongest business relevance as the auxiliary feature source from the feature source set.

[0107] The spatiotemporal correlation density between the morphological drift vector and the auxiliary feature source is calculated using the following formula:

[0108] ;

[0109] in, The spatiotemporal correlation density takes values ​​in the range [-1, 1]. Let be the magnitude of the shape drift vector at time t. This represents the average magnitude of the shape drift vector within the sliding time window W. To provide the eigenvalues ​​of the auxiliary feature source at time t, The average feature value of the auxiliary feature source within the sliding time window W is used;

[0110] The call weight is obtained using the following formula:

[0111] ;

[0112] Among them, the call weight directly inherits the spatiotemporal correlation density. The absolute value of , as a proportional factor for auxiliary feature injection, has the physical meaning that the higher the synergy between auxiliary features and business drift, the greater their say in the fusion mapping.

[0113] Based on the call weight, the feature data of the auxiliary feature source is injected into the strategy morphology distribution map to expand the feature dimension of the abnormal evolution zone and generate the fusion morphology distribution map. The specific back-feedback path is as follows: using the call weight as the gain coefficient, the value of the auxiliary feature source at the coordinate point of the abnormal evolution zone is added as a new dimension to the original strategy morphology distribution map, thereby realizing the normalized feature completion of the abnormal region.

[0114] And matching is performed using the fusion pattern distribution mapping with the preset strategy logic rule set;

[0115] It should be noted that the determination is made as to whether the change in the morphological drift vector exceeds a preset mutation threshold;

[0116] If the judgment result is yes, calculate the mutual information entropy between the morphological drift vector and the auxiliary feature source to obtain the semantic confidence, as shown in the following formula:

[0117] ;

[0118] in, For semantic confidence, The mutual information between the morphological drift vector and the auxiliary feature source represents the amount of effective information shared between them. The self-entropy of the morphological drift vector;

[0119] The call weight is then dynamically decayed based on semantic confidence to obtain the corrected call weight. The specific formula is as follows:

[0120] ;

[0121] Feature dimension augmentation is performed using the revised call weights, specifically:

[0122] The system utilizes the corrected call weights As a gain coefficient, the original numerical matrix of the auxiliary feature source is linearly scaled. Then, the system adds the scaled auxiliary feature values ​​as new feature channels to the tensor structure corresponding to the strategy morphology distribution map. For example, if the original mapping is a two-dimensional density distribution, the expanded version generates a multi-dimensional superimposed tensor containing "business density channel" and "external environment channel", which generates the fused morphology distribution map.

[0123] Next, the system uses the fusion morphological distribution mapping to match the preset strategy logic rule set. The strategy logic rule set contains multi-dimensional conditional discrimination operators (e.g., if the business density decreases and the social sentiment index decreases simultaneously, it is determined to be a normal fluctuation driven by the external environment and no intervention is triggered). The system inputs the fusion morphological distribution mapping into the strategy logic rule set, calculates the coupling relationship of each feature channel in the abnormal evolution zone, and outputs the final strategy correction suggestion, thereby realizing the attribution discrimination of the cause of abnormal evolution.

[0124] If the judgment result is negative, the auxiliary feature source is isolated from the feature dimension expansion calculation of the current sampling period. Specifically, the system sets the auxiliary feature source read pointer of the current sampling period to null, and in the generation logic of the fusion morphological distribution mapping, the corresponding auxiliary feature channel values ​​are uniformly filled with preset invalid values ​​(such as zero values ​​or mask values). Through this data masking at the logical level, the system stops the cross-correlation calculation and tensor splicing operation for auxiliary features, so that the processor resources of the edge nodes are fully focused on the judgment of the basic business flow, ensuring the steady-state operation of computing resources and energy consumption optimization under normal evolution.

[0125] This application proposes an attribution discrimination mechanism based on the expansion of auxiliary feature source dimensions. Its core concept is to break the information silo effect of single business data at edge nodes and provide an external reference system for the abnormal evolution of strategy forms through external environmental features (such as social sentiment, economic indicators, etc.). Traditional anomaly detection systems often only focus on the quantitative fluctuations of business data. However, in the actual sales environment, strategy drift is often driven by external unstructured factors. If only internal business indicators are relied upon, the system cannot determine whether a certain decline is due to the failure of the strategy itself or a normal fluctuation caused by macro-environmental factors (such as price wars among competitors or low consumer sentiment).

[0126] In contrast, alternative solutions include multivariate regression models, which attempt to use all external data as input features, but in edge computing scenarios, their computational overhead is enormous and they cannot identify spurious correlations; another example is manual labeling intervention, which relies on manual input of market dynamics, resulting in poor timeliness and an inability to achieve high-frequency spatial dimension coupling with strategy pattern distribution mapping. This solution employs a dual verification mechanism of spatiotemporal correlation density and semantic confidence, and dynamically decays feature weights under abrupt changes through mutual information entropy, ensuring that the system only performs dimensional expansion when external features have a substantial semantic correlation with business drift. This on-demand fusion operation not only improves the accuracy of attribution judgment but also ensures low-energy operation of edge nodes through the "feature isolation" mechanism.

[0127] For example, suppose the system detects an abnormal evolution zone within the current sampling period and has identified the social media sentiment index as an auxiliary feature source;

[0128] In the sliding time window Within this sequence, the magnitude sequence of the morphological drift vector is {5.0, 7.0, 9.0}, with an average magnitude of... The auxiliary feature source feature value sequence is {10.0, 14.0, 18.0}, and the average feature value is... ;

[0129] Substituting into the spatiotemporal correlation density formula, we get:

[0130] ,but ;

[0131] Assuming the change in the current morphological drift vector exceeds the preset mutation threshold (i.e., the judgment result is yes), the system initiates semantic verification, calculating the mutual information between the morphological drift vector and the auxiliary feature source I(V;Sa) = 0.6, the self-entropy of the morphological drift vector H(V) = 0.8; the semantic confidence η = 0.6 / 0.8 = 0.75; and the corrected call weight. The system uses the corrected call weights. =0.75 is the gain coefficient. The original numerical matrix of the auxiliary feature source is linearly scaled. The system adds the scaled feature values ​​as new feature channels to the tensor structure to generate a fused morphological distribution map. The system inputs the fused morphological distribution map into the policy logic rule set. If the rule is defined as: "If the business density decreases and the social sentiment index decreases simultaneously, it is determined to be driven by the external environment", the system matches the rule according to the value of the expanded channel to achieve accurate attribution.

[0132] Figure 3 This is a schematic diagram illustrating the stability index judgment of the real-time sales strategy tendency determination and closed-loop optimization system based on edge nodes according to the present invention.

[0133] This application further proposes that the system also includes:

[0134] Record the displacement step size of the shape constraint boundary generated within N consecutive historical sampling periods. This forms a historical sequence of boundary evolution. displacement step size The physical meaning is the spatial displacement distance of the geometric center of the morphological constraint boundary within the strategy influence factor field during adjacent sampling periods;

[0135] The statistical stability index for calculating the boundary evolution history sequence is given by the following formula:

[0136] ;

[0137] in, As a statistical stability indicator, The standard deviation of the displacement step size within the boundary evolution history sequence. This is the arithmetic mean of the displacement steps within the boundary evolution history sequence. Let be the boundary displacement step size within the i-th historical sampling period. The length of the historical sampling sequence;

[0138] And determine whether the statistical stability index is lower than the preset confidence threshold;

[0139] If the judgment result is yes, a global constraint operator representing the global convergence trend is generated based on the deviation pattern between historical business feedback data and historical strategy pattern distribution mapping. The global constraint operator is then weighted and fused with the pattern constraint boundary initially generated in the current sampling period. The fusion result is used as the enhanced pattern constraint boundary for the next sampling period. The specific formula is as follows:

[0140] ;

[0141] in, To enhance the morphological constraint boundary, It is a global constraint operator, which is represented by a preset steady-state reference boundary. This refers to the morphological constraint boundary initially generated based on the instantaneous morphological deviation entropy within the current sampling period. These are the global confidence weight coefficients;

[0142] If the judgment result is negative, the morphological constraint boundary initially generated in the current sampling period is directly used as the enhanced morphological constraint boundary in the next sampling period; the system considers the boundary evolution generated by the current instantaneous feedback to have high reliability, and directly uses the morphological constraint boundary initially generated in the current sampling period. As an enhanced form constraint boundary for the next sampling period, this operation ensures the system's ability to respond sensitively to minor market fluctuations in a steady state.

[0143] In the next sampling period, the strategy's tendency to stability is optimized using enhanced morphological constraint boundaries. The specific implementation process is as follows:

[0144] The system determines in real time whether the geometric center of the projected distribution is within the safe evolution zone enclosed by the enhanced morphological constraint boundary. If the evolution trajectory of the business flow data attempts to cross the enhanced morphological constraint boundary, the system will trigger the boundary locking mechanism, forcibly locking the strategy execution parameters within the sampling period within the steady-state range defined by the boundary, preventing the strategy from spreading to the abnormal and unstable area outside the boundary. Through this physical limit based on the enhanced boundary, the system forcibly eliminates the excessive oscillation of the sales strategy from the spatial constraint level, thereby achieving dynamic optimization of the stability of the strategy tendency.

[0145] When the system frequently experiences large boundary displacements over multiple consecutive periods, the initial boundary generated solely by instantaneous morphological deviation from entropy will lose its reference value. At this time, the system may fall into local optimum oscillations or experience evolutionary drift due to random noise. The system quantifies the dispersion of boundary evolution by calculating the ratio of the standard deviation to the arithmetic mean of the displacement step size in the historical sequence of boundary evolution (i.e., the statistical stability index Sta). When the stability index is abnormal, the system seeks a global constraint operator with global guiding effect and corrects the current instantaneous deviation through long-term steady-state experience.

[0146] In contrast, alternative technical solutions include low-pass filters, which can smooth fluctuations, but their fixed cutoff frequency can cause the system to respond slowly to real market changes, resulting in severe phase lag; and historical averages, which simply average the boundary positions and cannot distinguish between normal trend evolution and abnormal random oscillations.

[0147] This scheme achieves dynamic switching between local sensitive response and global steady-state constraints by judging statistical stability indicators, and preserves the morphological constraint boundary when the system is stable. High sensitivity; during system oscillations, through global constraint operators Introducing a global convergence trend effectively avoids overshooting at the policy execution boundary; the final enhanced morphological constraint boundary is not a simple numerical correction, but rather achieved through a global constraint operator. With the currently initially generated morphological constraint boundary Geometric weighted fusion provides physical-level constraint protection for the next sampling period; the introduced boundary locking mechanism can forcibly constrain the evolution trajectory within the safe evolution zone. This spatial hard constraint, compared with parameter correction, can more quickly eliminate excessive oscillations in the sales strategy and ensure the long-term stability of the strategy tendency.

[0148] Example 2:

[0149] In one embodiment of the present invention, which differs from the previous embodiment, the electronic device includes one or more processors and a memory.

[0150] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0151] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0152] In one example, the electronic device may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown). In addition, depending on the specific application, the electronic device may include any other suitable components.

[0153] Example 3:

[0154] Embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps described in the "Exemplary Methods" section above according to the various embodiments of this application.

[0155] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0156] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not restrict the application from being implemented using the specific details described above.

[0157] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0158] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0159] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0160] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A real-time sales strategy preference determination and closed-loop optimization system based on edge nodes, characterized in that, include: Data acquisition module: used to acquire business flow data and spatial environment semantic data of the edge nodes corresponding to each sales terminal within the current sampling period; Field mapping module: used to construct a strategy influence factor field based on spatial environment semantic data, and project business flow data onto the strategy influence factor field to form a strategy form distribution mapping; Evolutionary Analysis Module: Used to perform differential evolutionary analysis based on the strategy morphology distribution mapping of the previous sampling period within the current sampling period, generating a set of morphological drift vectors that characterize the direction and magnitude of changes in strategy tendencies; Regionalization identification module: used to identify abnormal evolution regions based on the evolution trend of morphological drift vectors, and generate policy trigger confidence based on the drift vector characteristics of abnormal evolution regions; Judgment module: used to trigger confidence based on the generated policy and determine whether policy intervention is needed; If the judgment result is yes, then output the strategy intervention instruction, collect the business feedback data after the execution of the strategy intervention instruction, and backfeed it to the strategy morphology distribution mapping. Update the morphology drift vector evolution trend according to the backfeed business feedback data, and adjust the morphology constraint boundary of the strategy influence factor field to optimize the stability of the strategy tendency in the next sampling period. If the judgment result is negative, no intervention is made, and the existing strategy is maintained.

2. The real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to claim 1, characterized in that, The system also includes: Based on the distribution characteristics of the abnormal evolution zone, auxiliary feature sources are selected from a preset feature source set. The feature source set includes at least one or more of the following: social media sentiment index, real-time stream of regional economic indicators, and dynamic information stream of competitors associated with the edge node. Calculate the spatiotemporal correlation density between the morphological drift vector and the auxiliary feature source to obtain the call weight; Based on the call weight, the feature data of the auxiliary feature source is injected into the strategy morphology distribution map to expand the feature dimension of the abnormal evolution region and generate a fusion morphology distribution map; and the fusion morphology distribution map is matched with a preset strategy logic rule set.

3. The real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to claim 2, characterized in that: Determine whether the change in the morphological drift vector exceeds a preset mutation threshold; If the judgment result is yes, calculate the mutual information entropy between the morphological drift vector and the auxiliary feature source to obtain the semantic confidence, and perform dynamic decay processing on the call weight according to the semantic confidence to obtain the corrected call weight. Then, use the corrected call weight to perform the feature dimension expansion. If the judgment result is negative, the auxiliary feature source is isolated from the feature dimension expansion calculation of the current sampling period.

4. The real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to claim 1, characterized in that, The system also includes: The displacement step size of the morphological constraint boundary generated in multiple consecutive historical sampling periods is recorded to form a boundary evolution history sequence; Calculate the statistical stability index of the boundary evolution history sequence and determine whether the statistical stability index is lower than a preset confidence threshold. If the judgment result is yes, a global constraint operator representing the global convergence trend is generated based on the deviation pattern between historical business feedback data and historical strategy pattern distribution mapping. The global constraint operator is then weighted and fused with the pattern constraint boundary initially generated in the current sampling period, and the fusion result is used as the enhanced pattern constraint boundary for the next sampling period. If the judgment result is negative, the morphological constraint boundary initially generated in the current sampling period will be directly used as the enhanced morphological constraint boundary in the next sampling period. In the next sampling period, the enhanced morphological constraint boundary is used to optimize the policy tendency stability.

5. The real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to claim 1, characterized in that: The generation of the morphological drift vector set includes: Extract the strategy morphology distribution mapping for the current sampling period and retrieve the strategy morphology distribution mapping stored in the previous sampling period. The strategy morphology distribution mapping of adjacent sampling periods is compared by difference, the feature value difference of each mapping point in the strategy morphology distribution mapping is calculated, and a morphology difference matrix representing the feature density change of each mapping point is generated. By utilizing the correlation between preset coordinate points in the strategy influence factor field, the instantaneous offset vector of each mapping point in the morphological difference matrix within the strategy influence factor field is extracted. The morphological drift vector set is generated by summing the geometric direction and magnitude of the instantaneous offset vector.

6. The real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to claim 5, characterized in that: The extraction of the instantaneous offset vector includes: Based on the physical positional relationship between each coordinate point in the strategy influence factor field, the spatial weight distribution corresponding to each mapping point is determined. By using the spatial weight distribution to perform weighted cluster analysis on the eigenvalue differences of each mapping point in the morphological difference matrix, the geometric center displacement value representing the core region of morphological evolution is identified. The initial displacement direction of each mapping point is corrected based on the displacement value of the geometric center to obtain the instantaneous offset vector that conforms to the physical distribution law of the strategy influence factor field.

7. The real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to claim 1, characterized in that: The updating of the evolutionary trend and the adjustment of the morphological constraint boundaries include: Map the business feedback data to the strategy morphology distribution mapping, and calculate the morphology deviation entropy between the actual morphology position corresponding to the business feedback data and the target position predicted by the morphology drift vector set. Based on the morphological deviation entropy, the correction weight of the morphological drift vector evolution trend is determined, and the deviation of the evolution trend is compensated using the correction weight to complete the update of the evolution trend. Based on the magnitude of the morphological deviation entropy, the spatial scaling coefficient of the morphological constraint boundary of the strategy influencing factor field is determined, and the contraction or expansion state of the morphological constraint boundary is adjusted using the spatial scaling coefficient.

8. The real-time sales strategy preference determination and closed-loop optimization system based on edge nodes according to claim 7, characterized in that: The determination of the correction weight and the spatial scaling coefficient includes: Calculate the time-domain rate of change of the morphological deviation entropy with the sampling period; Based on the magnitude of the time-domain rate of change, a corresponding damping factor is matched in a preset damping function; The damping factor is used to perform gain weighting on the morphological deviation entropy to generate a smoothed and corrected deviation control variable. The correction weights and spatial scaling coefficients are generated by mapping the deviation control variables respectively.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the system as described in any one of claims 1 to 8.

10. A computer storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by the processor, they implement the steps of the system as described in any one of claims 1 to 8.

Images