Intelligent decision dynamic optimization method and system based on BLM and MM

Abstract

The invention relates to the technical field of enterprise strategic decision optimization, in particular to an intelligent decision dynamic optimization method and system based on BLM and MM, and the method comprises the steps: collecting a supply chain fluctuation index, a social media emotion polarity value, a competitive product price variation coefficient and channel inventory saturation through an MM system, and generating a structured market feature data set; inputting the data into a BLM model analysis engine, calculating a value claim weight offset, a task conflict matrix and market opportunity window probability distribution, outputting a strategic element state vector, analyzing the state vector, identifying a resource conflict item, generating a Pareto optimal strategy cluster, calculating a risk entropy value, forming an executable decision strategy set, and obtaining a decision-making result. And dynamically updating the BLM model parameters and the MM system acquisition weight based on the feedback data of the inventory turnover acceleration and the customer wastage rate change curvature. According to the method, efficient closed-loop optimization from market perception to strategy execution is realized, and the method has the advantages of high real-time performance, high risk control precision, excellent system adaptability and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise strategic decision optimization, and in particular to an intelligent decision dynamic optimization method and system based on BLM and MM. Background Art

[0002] In today's highly dynamic and uncertain business environment, companies face increasingly complex market dynamics. Factors influencing decision-making quality are no longer limited to internal resource allocation and traditional metrics, but increasingly come from multidimensional market dynamics signals such as supply chain fluctuations, shifts in social media sentiment, adjustments in competitive strategies, and channel inventory pressures. To achieve precise response and strategic adaptation, a growing number of companies are introducing Business Logic Modeling (BLM) methods and Market Monitoring (MM) systems, attempting to build intelligent decision-making mechanisms driven by real-time data. In theory, BLM can model a company's strategic goals, value propositions, and mission configurations, while MM systems can provide high-frequency signals from the external market for decision-making. The combined efforts of these two promise to achieve adaptive optimization of strategic behavior.

[0003] However, within the existing technology landscape, the integration of BLM and MM remains primarily at the static modeling and offline analysis stage, presenting significant technical limitations. On the one hand, multi-dimensional market dynamic data comes from heterogeneous sources and differing structures, lacking efficient real-time collection, structuring, and fusion mechanisms. This results in BLM model inputs lacking timeliness and operability. On the other hand, the strategy generation process is largely based on static weights and linear rules, making it difficult to dynamically identify resource conflicts or quantify risk entropy. This results in poor generalization of strategy outputs and weak risk control capabilities. Furthermore, the lack of a self-evolutionary mechanism driven by strategy effectiveness feedback prevents the system from making timely structural corrections and adjusting collection weights after strategy execution fails, severely impacting the iterative stability and closed-loop responsiveness of the intelligent decision-making system. Summary of the Invention

[0004] The present invention provides an intelligent decision-making dynamic optimization method and system based on BLM and MM, and provides a dynamic optimization method that integrates BLM and MM, which can perform value offset modeling, conflict matrix analysis, multi-objective strategy optimization and feedback-driven model update based on structured data, so as to realize a full-link intelligent optimization closed loop from data collection to strategy execution.

[0005] The intelligent decision-making dynamic optimization method based on BLM and MM includes the following steps:

[0006] S1, real-time market feature extraction: The MM system collects multi-dimensional market dynamic data, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation, to generate a market feature dataset;

[0007] S2, BLM element dynamic mapping: Input the market feature dataset into the BLM model parsing engine, calculate the value proposition weight offset, task conflict matrix, and market opportunity window probability distribution in real time, and output the strategic element state vector. The strategic element state vector includes the value proposition weight vector, task priority sequence, and strategic gap heat map;

[0008] S3, closed-loop optimization of decision strategies: Input the state vector of the strategic elements into the dynamic optimization engine, identify resource conflict items, generate Pareto optimal strategy clusters, and perform strategy risk entropy evaluation to generate an executable decision strategy set and activate the strategy executor;

[0009] S4, feedback-driven model iteration: Collect the strategy effect feedback data generated by the strategy executor. When the inventory turnover acceleration in the feedback data is lower than the threshold or the curvature of the customer churn rate change exceeds the limit, it triggers the update of the BLM model parameters and MM data collection weights.

[0010] Optionally, the S1 includes:

[0011] S11, real-time collection of multi-dimensional market dynamic data: The MM system collects multi-dimensional market dynamic data in real time, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation;

[0012] S12, constructing a structured data set: aligning the multi-dimensional market dynamic data collected in S11 by timestamp to generate a structured data set containing the supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation;

[0013] S13, generate a market characteristic data set: fill missing values ​​and remove outliers in the structured data set constructed in S12, and output a standardized market characteristic data set.

[0014] Optionally, the S2 includes:

[0015] S21, market feature analysis input: input the market feature dataset outputted in S1 into the BLM model analysis engine;

[0016] S22, real-time calculation of strategic elements: The following calculations are performed synchronously through the BLM model parsing engine:

[0017] Calculate the value proposition weight offset based on the supply chain volatility index and social media sentiment polarity value in the market characteristic dataset;

[0018] Based on the coefficient of variation of competitor product prices and channel inventory saturation in the market characteristic dataset, a task conflict matrix is ​​constructed;

[0019] Generate the probability distribution of market opportunity windows based on the dynamic distribution of four-dimensional indicators in the market characteristic data set;

[0020] S23, state vector encapsulation output: encapsulate the value proposition weight offset calculated in S22 into a value proposition weight vector, parse the task conflict matrix into a task priority sequence, map the market opportunity window probability distribution into a strategic gap heat map, and finally integrate and output the strategic element state vector including the value proposition weight vector, task priority sequence, and strategic gap heat map.

[0021] Optionally, the S3 includes:

[0022] S31, strategic element state vector analysis: The strategic element state vector output by S2 is input into the dynamic optimization engine and decomposed into three subcomponents: value proposition weight vector, task priority sequence, and strategic gap heat map;

[0023] S32, strategy generation and optimization: Executed sequentially through the dynamic optimization engine:

[0024] Identify resource conflict items: Based on the matching analysis between the value proposition weight vector and the task priority sequence, identify the resource conflict items that may cause resource allocation conflicts;

[0025] Generate Pareto optimal strategy clusters: Using resource conflict terms as constraints and combining them with the strategic gap heat map, perform multi-objective optimization to generate Pareto optimal strategy clusters.

[0026] Perform strategy risk entropy assessment: Calculate the risk entropy value of each strategy in the Pareto optimal strategy cluster on the strategy gap heat map;

[0027] S33, executable policy output and activation: screening the policies with risk entropy values ​​lower than the threshold to form an executable decision policy set, and activating the policy executor to load the executable decision policy set.

[0028] Optionally, the S4 includes:

[0029] S41, strategy effect feedback collection: real-time collection of strategy effect feedback data generated after the strategy executor executes the executable decision strategy set,

[0030] S42, dynamic threshold condition detection: real-time monitoring of the strategy effect feedback data collected in S41, and generating a trigger signal when the strategy effect feedback data exceeds a preset threshold;

[0031] S43, collaborative update of model parameters: after the trigger signal is generated, an update operation is performed.

[0032] Optionally, the strategy effect feedback data includes inventory turnover acceleration and customer churn rate change curvature.

[0033] Optionally, the trigger signal includes:

[0034] When the inventory turnover acceleration is lower than a preset acceleration threshold, a first trigger signal is generated;

[0035] When the curvature of the customer churn rate change exceeds a preset curvature threshold, a second trigger signal is generated.

[0036] Optionally, when the first trigger signal or the second trigger signal is generated, an update operation is performed, where the update operation includes updating BLM model parameters in the BLM model parsing engine and adjusting the MM data acquisition weight of the MM system data acquisition module.

[0037] The intelligent decision-making dynamic optimization system based on BLM and MM is used to implement the above-mentioned intelligent decision-making dynamic optimization method based on BLM and MM, and includes the following modules:

[0038] MM system data acquisition module: used to collect multi-dimensional market dynamic data from the external environment in real time, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient and channel inventory saturation, and send it to the data access module;

[0039] Data access module: used to align timestamps, unify formats, and structure data for the collected multi-dimensional market dynamic data to build a structured data set;

[0040] Natural language processing module: used to classify and quantify the sentiment polarity values ​​of social media and generate sentiment scores in the range of [-1, 1];

[0041] Market feature processing module: used to perform missing value filling, outlier removal and Z-score normalization on structured data sets, and output standardized market feature data sets;

[0042] BLM model parsing engine: includes value proposition modeling unit, task conflict parsing unit and market opportunity window analysis unit, which are used to:

[0043] Calculate the value proposition weight offset based on the supply chain volatility index and social media sentiment polarity value;

[0044] Construct a task conflict matrix based on the coefficient of variation of competitor prices and channel inventory saturation;

[0045] Generate market opportunity window probability distribution based on multi-dimensional indicator distribution;

[0046] Output the strategic element state vector, including the value proposition weight vector, task priority sequence, and strategic gap heat map;

[0047] Task conflict analysis module: used to analyze the task conflict matrix into a task priority sequence, supporting task sorting and resource allocation judgment;

[0048] Dynamic Optimization Engine: This engine receives the state vector of strategic elements, identifies resource conflicts, generates Pareto-optimal strategy clusters, and evaluates strategy risk entropy. It then filters and outputs executable decision strategy sets with risk entropy values ​​below a threshold.

[0049] Strategy executor: used to load the executable decision strategy set and execute corresponding operations, while outputting strategy effect feedback data;

[0050] Feedback analysis module: used to calculate inventory turnover acceleration and customer churn rate change curvature in real time, and generate trigger signals based on preset thresholds;

[0051] Model adaptive update module: used to update the BLM model parameters in the BLM model parsing engine when a trigger signal is detected, and adjust the collection weight of the MM system data collection module to achieve closed-loop coordination between strategy execution and model update.

[0052] Beneficial effects of the present invention:

[0053] The present invention, through the coordinated cooperation of the MM system data acquisition module and the data access module, constructs a multi-dimensional market dynamic data collection mechanism for supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient and channel inventory saturation. Combined with the sentiment polarity extraction and structured alignment of the natural language processing module, it realizes the fusion processing of unstructured public opinion signals and quantitative market indicators, effectively supports the real-time modeling needs of the BLM model in a complex market environment, and significantly improves the system's response sensitivity and forward-looking prediction capabilities to rapidly changing market situations.

[0054] Based on the analysis of strategic element state vectors, this paper utilizes a resource conflict identification model and a Pareto-optimal strategy cluster generation mechanism within a dynamic optimization engine to construct a multi-objective optimization system that considers task priorities, value deviations, and strategic gap heat maps. Furthermore, through a strategy risk entropy assessment method, it quantitatively controls the stability and risk concentration of strategy outputs in the heat space. This optimization-screening mechanism significantly improves the feasibility and reliability of strategy implementation, making it particularly suitable for decision-making and deployment in scenarios with limited resources or intense strategic conflicts.

[0055] This invention, through the collection of policy effect feedback and a dynamic threshold condition detection mechanism, can identify deviations in decision-making system execution in real time based on two core indicators: inventory turnover acceleration and customer churn rate curvature. This can then trigger the coordinated update of the BLM model parsing engine parameters and the adaptive adjustment of the MM system's data collection weights. This design establishes a closed-loop evolutionary path from "data-strategy-feedback-update," effectively improving the system's ability to recover from policy failures and its self-adjustment capabilities, ensuring the stability and robustness of the intelligent decision-making system in the long term. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention;

[0058] Figure 2 Schematic diagram of the system flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0060] like Figure 1 As shown in FIG, the intelligent decision-making dynamic optimization method based on BLM and MM includes the following steps:

[0061] S1, real-time market feature extraction: The MM system collects multi-dimensional market dynamic data, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation, to generate a market feature dataset;

[0062] S2, BLM dynamic mapping: The market feature dataset is input into the BLM model parsing engine. By real-time calculation of the value proposition weight offset, task conflict matrix, and market opportunity window probability distribution, the strategic element state vector is output. The strategic element state vector includes the value proposition weight vector, task priority sequence, and strategic gap heat map.

[0063] S3, closed-loop optimization of decision-making strategies: The state vector of strategic elements is input into the dynamic optimization engine. By identifying resource conflict items, generating Pareto optimal strategy clusters, and evaluating the risk entropy of the execution strategies, an executable decision strategy set is generated and the strategy executor is activated.

[0064] S4, feedback-driven model iteration: Collect the strategy effect feedback data generated by the strategy executor. When the inventory turnover acceleration in the feedback data is lower than the threshold or the curvature of the customer churn rate change exceeds the limit, it triggers the update of the BLM model parameters and MM data collection weights.

[0065] S1 includes:

[0066] S11, real-time collection of multi-dimensional market dynamic data: Utilize the data access module deployed in the MM system to collect multi-dimensional market dynamic data consisting of the following four key indicators in real time:

[0067] Supply Chain Volatility Index: We extract indicators such as order delay rate and logistics node lag time from the logistics monitoring platform and the purchase order fulfillment system, and use weighted standard deviation modeling to calculate the index. The calculation formula is as follows:

[0068]

[0069] Among them, w i represents the importance weight of the i-th type supply chain node, d i (t) is the number of days of delay of the i-th type node at time t, is the weighted average delay value.

[0070] Social media sentiment polarity value: The natural language processing module is used to perform sentiment analysis on social media platform comment data. Polarity scores are generated based on the BERT or LSTM sentiment classifier. The value range is [-1, 1], where positive values ​​indicate positive sentiment and negative values ​​indicate negative sentiment.

[0071] Competitive product price variation coefficient: Obtain competitive product price data for the target category from e-commerce platforms and B2B quotation systems, and calculate its coefficient of variation:

[0072]

[0073] Among them, σ p (t) is the standard deviation of the prices of competing products at time t, μ p (t) is its mean.

[0074] Channel inventory saturation: By connecting with the ERP system of each sales channel, the ratio of actual inventory to upper limit capacity is collected, which is defined as:

[0075]

[0076] Among them, I(t) is the current inventory of the channel, C max is the maximum inventory allowance for the channel;

[0077] The data collection cycle for each type ranges from 30 seconds to 5 minutes, and is automatically adjusted as needed to ensure real-time and continuity of the data.

[0078] S12, constructing a structured data set: aligning the multi-dimensional market dynamic data collected above by timestamps to construct a structured data set in a unified format. The following strategies are used for alignment and unification:

[0079] The time alignment method is sliding window aggregation, and the window size is set to 15 minutes by default.

[0080] For missing time periods, linear interpolation is used to fill them;

[0081] The construction result is in matrix form, and the structure is as follows:

[0082] M(t)=[SCVI(t),SPS(t),CPVC(t),CIS(t)];

[0083] Example: If the data collected at 10:00 on June 14, 2025 is: Supply Chain Volatility Index = 0.127, Social Media Sentiment Polarity = -0.35, Competitive Product Price Variation Coefficient = 0.18, Channel Inventory Saturation = 0.62;

[0084] Then the row vector of structured data at this moment is M(10:00)=[0.127,-0.35,0.18,0.62];

[0085] S13, generating a market feature dataset: performing the following data preprocessing operations on the structured data set constructed in S12 to generate a standardized market feature dataset:

[0086] 1. Missing value filling: In addition to linear interpolation, if more than 10 minutes of consecutive missing data are missing, the mean value based on similar historical periods is used to fill the missing data:

[0087]

[0088] Where ΔT is the cycle interval and k is the number of historical windows;

[0089] 2. Outlier removal: Use an IQR (interquartile range)-based method to detect and remove abnormal data points that deviate from the range by more than 1.5 times the IQR.

[0090] 3. Standardization: Use Z-score normalization to unify the data scale:

[0091] Among them, μ is the mean and σ is the standard deviation, which standardize the four indicators of supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation.

[0092] The final output market characteristic data set is a continuous, standardized multi-dimensional time series matrix, which serves as the input data of the BLM model parsing engine.

[0093] S2 includes:

[0094] S21, market feature analysis input: input the market feature dataset X={M(t)} output by S1 into the BLM model analysis engine, where the market feature vector M(t)∈R at each time t 4 , expressed as:

[0095] M(t)=[SCVI(t),SPS(t),CPVC(t),CIS(t)];

[0096] SCVI(t) represents the supply chain volatility index, SPS(t) represents the social media sentiment polarity value, CPVC(t) represents the coefficient of variation of competitor prices, and CIS(t) represents the channel inventory saturation.

[0097] S22 Real-time calculation of strategic elements: The BLM model parsing engine synchronously performs the following three subtasks based on different feature combinations:

[0098] 1. Calculation of value proposition weight offset: Based on the supply chain volatility index SCVI(t) and the social media sentiment polarity value SPS(t), a two-factor weighted model is constructed to calculate the value proposition weight offset Δw v (t):

[0099] Δw v (t)=α1·SCVI(t)+α2·SPS(t);

[0100] Among them, α1 and α2 are the sensitivity coefficients of the supply chain and social emotions to value perception, respectively. The empirical settings are such as α1 = 0.6 and α2 = 0.4. The output results are used as the weight offset estimation of the value proposition in the current market environment.

[0101] Example: If SCVI(t) = 0.12 and SPS(t) = -0.45 at a certain sampling time, then:

[0102] Δw v (t)=0.6·0.12+0.4·(-0.45)=0.072-0.18=-0.108;

[0103] Indicates that the value proposition direction should be less weighted.

[0104] 2. Task conflict matrix construction: Based on the competitive product price variation coefficient CPVC(t) and channel inventory saturation CIS(t), the task conflict matrix C(t)∈R is constructed. m×m , where m is the number of strategic tasks, and is constructed as follows:

[0105] C ij (t)=|γ i CPVC(t)-γ j ·CIS(t)|;

[0106] Among them, γ i and γ j is the sensitivity weight of the i-th and j-th tasks; the larger the value, the stronger the task resource competition. This matrix is ​​used to reveal the degree of resource allocation conflict between tasks under the current price-inventory conditions.

[0107] 3. Generation of market opportunity window probability distribution: Through clustering and kernel density estimation methods, the distribution state of the four-dimensional market characteristic indicator sequence in the past T time windows is modeled and the market opportunity window probability distribution function P is output. window (x), the construction process includes:

[0108] Using a sliding window dataset X t-T:t ∈R T×4 ;

[0109] Use Gaussian kernel density estimation to generate joint distribution P(x);

[0110] Calculate the probability mass within the threshold interval Ω of a strategic opportunity trigger:

[0111] P window =∫ Ω P(x)dx;

[0112] Ω is defined as the ideal opportunity window characteristic interval consisting of small market volatility, positive sentiment, low price difference, and moderate inventory.

[0113] S23, state vector encapsulation output: Encapsulate the three results in S22 into a strategic element state vector, as follows:

[0114] Δw v (t) is mapped to the value proposition weight vector W v (t), as a set of offsets on the multidimensional value dimension;

[0115] The task conflict matrix C(t) is converted into the task priority sequence Π(t) through the task conflict analysis module, and the conflict minimization sorting principle is adopted;

[0116] The market opportunity window probability distribution P window(x) Map different strategic paths and construct a strategic gap heat map H gap (t),The higher the heat value, the more urgent the opportunity window for this strategic direction.

[0117] The final integration is: S state (t)={W v (t),Π(t),H gap (t)};

[0118] As the basic data for dynamic optimization input to the S3 step.

[0119] S3 includes:

[0120] S31, strategic element state vector analysis: the strategic element state vector S output by S2 state (t) Input dynamic optimization engine, which is broken down into:

[0121] Value proposition weight vector W v (t)=[w1,w2,…,w m ], represents the weight offset value of each strategic direction;

[0122] Task priority sequence Π(t)=[T1,T2,…,T n ], sorted by execution priority from high to low;

[0123] Strategic Gap Heat Map H gap (t)∈R m×1 , reflecting the pressure intensity of the opportunity window faced by the current strategic task.

[0124] This step completes the structural decoupling of features and provides a data basis for subsequent resource conflict identification and strategy optimization.

[0125] S32, strategy generation and optimization: The dynamic optimization engine performs the following operations in sequence based on the analysis results:

[0126] 1. Identify resource conflict items: Calculate the matching matrix M between the value proposition weight vector and the task priority sequence ij , defined as: M ij =|w i -θ j |, where w i is the weight value of the i-th strategic direction, θ j is the position weight of the jth task in the priority sequence, which is usually modeled in an exponential decay manner and is expressed as: j =e -λj ; λ is the priority attenuation factor, which usually ranges from [0.1,1].

[0127] If M ij>δ (preset deviation threshold), then the combination is recorded as a resource conflict item R ij° .

[0128] 2. Generate Pareto-optimal strategy clusters: Based on the identified resource conflicts, use them as constraints and combine them with the strategic gap heat map to perform multi-objective optimization. The optimization objectives are as follows:

[0129] Minimize the number of resource allocation conflicts:

[0130] Maximize high-heat strategy coverage max∑ i w i ·H gap (i);

[0131] Form a Pareto front solution set, denoted as

[0132] Each S k Represents a strategy combination.

[0133] 3. Execute strategy risk entropy assessment: To quantify the uncertainty and risk of each strategy combination on the strategy gap heat map, define each strategy S k The risk entropy value Expressed as:

[0134]

[0135] in: Indicates that the i-th strategic direction is in strategy S k Normalize the importance weights in ;

[0136] This entropy value measures the uncertainty of the strategy structure. The smaller it is, the higher the risk concentration and the stronger the execution stability.

[0137] Example: There are three strategic directions, and the weight vector is W v =[0.5,0.3,0.2], the heat map is H gap =0.9,0.4,0.1], then:

[0138]

[0139] Calculate p2 and p3 in the same way, and finally get:

[0140]

[0141] S33, executable strategy output and activation: After the strategy risk entropy value calculation is completed, select the strategy that meets the following conditions from the Pareto optimal strategy cluster:

[0142] Risk entropy Where ε is a safety threshold (e.g. 0.75);

[0143] A strategy that satisfies both resource constraints and objective function evaluation.

[0144] Integrate screening results into an executable decision strategy set And loaded and activated by the policy executor:

[0145]

[0146] This strategy set will enter the next stage of execution and feedback analysis.

[0147] S4 includes:

[0148] S41, strategy effect feedback collection: by connecting with the strategy executor, real-time collection of executable decision strategy set S exec The feedback data on the strategy effect during operation includes the following two key indicators:

[0149] 1. Inventory turnover acceleration a I (y): Defined as the rate of change of inventory turnover per unit time, expressed as:

[0150]

[0151] Where C(t) is the sales cost per unit time, and I(t) is the average inventory value;

[0152] If the sampling is a discrete time series, the central difference approximation can be used:

[0153]

[0154] 2. Customer Churn Rate Curvature κ C (t): Defined as the absolute value of the second-order derivative of the customer churn rate function curve on the time axis, it reflects the sharpness of the churn trend and is expressed as:

[0155]

[0156] Where L(t) is the customer churn rate at time t (e.g., number of churned customers in the current period / total number of customers), and Δt is the sampling interval.

[0157] Example description: The sample data at a certain moment is as follows:

[0158] C(t)=12,000, I(t)=3,000, we get τ(t)=4.0;

[0159] τ(t+1)=4.2、τ(t-1)=3.8, then the inventory turnover acceleration is:

[0160]

[0161] If a I (t) If it continues to be lower than the preset threshold, a feedback response will be triggered.

[0162] S42, dynamic threshold condition detection: Perform real-time dynamic threshold judgment on the strategy effect feedback data collected in S41 to determine whether the current strategy operation effect is lower than expected:

[0163] The first trigger signal condition: If there is a time t that satisfies: a I (t)<θ a , then the first trigger signal is generated, where θ a It is the preset acceleration threshold, such as 0.1.

[0164] Second trigger signal condition: If there is a time t that satisfies: κ C (t)>θ κ , then generate the second trigger signal, where θ κ It is the preset curvature change threshold, such as 0.5.

[0165] To prevent false triggering, the system adopts a dual verification strategy of sliding average and anomaly detection.

[0166] S43, collaborative update of model parameters: If the system detects that the first trigger signal or the second trigger signal is established, the collaborative update mechanism of model parameters is immediately started, including:

[0167] 1. Update the BLM model parameters in the BLM model parsing engine: For the parameter sets used in the calculation of value proposition weight offset, task conflict matrix, and market opportunity window probability distribution, reconstruct the following parameters based on feedback data:

[0168] Retrain the weight sensitivity coefficients α1 and α2 in the value perception model;

[0169] Adjust the task sensitivity coefficient γ in the task conflict matrix i ;

[0170] Updates the kernel density bandwidth parameter in the opportunity window probability distribution model.

[0171] The update method can use weighted sliding window learning or Bayesian regression fine-tuning;

[0172] 2. Adjust the MM data collection weight of the MM system data collection module: redistribute the collection weight according to the market dimensions involved in the feedback abnormal indicators. For example:

[0173] If the curvature of customer churn rate changes is abnormal, the sampling frequency and accuracy of social media sentiment polarity values ​​should be increased;

[0174] If the inventory turnover rate continues to be low, then strengthen the real-time collection of channel inventory saturation and supply chain volatility index. The collection weight of each type of data in the MM system data collection module ω i The following update rules are used:

[0175] Among them, Δ i is the acquisition offset signal related to the feedback index, and η is the learning rate, which controls the update speed of the acquisition strategy.

[0176] like Figure 2 As shown, the intelligent decision-making dynamic optimization system based on BLM and MM is used to implement the above-mentioned intelligent decision-making dynamic optimization method based on BLM and MM, and includes the following modules:

[0177] MM system data acquisition module: used to collect multi-dimensional market dynamic data from the external environment in real time, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient and channel inventory saturation, and send it to the data access module;

[0178] Data access module: used to align timestamps, unify formats, and structure data for the collected multi-dimensional market dynamic data to build a structured data set;

[0179] Natural language processing module: used to classify and quantify the sentiment polarity values ​​of social media and generate sentiment scores in the range of [-1, 1];

[0180] Market feature processing module: used to perform missing value filling, outlier removal and Z-score normalization on structured data sets, and output standardized market feature data sets;

[0181] BLM model parsing engine: includes value proposition modeling unit, task conflict parsing unit and market opportunity window analysis unit, which are used to:

[0182] Calculate the value proposition weight offset based on the supply chain volatility index and social media sentiment polarity value;

[0183] Construct a task conflict matrix based on the coefficient of variation of competitor prices and channel inventory saturation;

[0184] Generate market opportunity window probability distribution based on multi-dimensional indicator distribution;

[0185] Output the strategic element state vector, including the value proposition weight vector, task priority sequence, and strategic gap heat map;

[0186] Task conflict analysis module: used to analyze the task conflict matrix into a task priority sequence, supporting task sorting and resource allocation judgment;

[0187] Dynamic Optimization Engine: This engine receives the state vector of strategic elements, identifies resource conflicts, generates Pareto-optimal strategy clusters, and evaluates strategy risk entropy. It then filters and outputs executable decision strategy sets with risk entropy values ​​below a threshold.

[0188] Strategy executor: used to load the executable decision strategy set and execute corresponding operations, while outputting strategy effect feedback data;

[0189] Feedback analysis module: used to calculate inventory turnover acceleration and customer churn rate change curvature in real time, and generate trigger signals based on preset thresholds;

[0190] Model adaptive update module: used to update the BLM model parameters in the BLM model parsing engine when a trigger signal is detected, and adjust the collection weight of the MM system data collection module to achieve closed-loop coordination between strategy execution and model update.

[0191] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0192] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. The intelligent decision-making dynamic optimization method based on BLM and MM is characterized by: The following steps are involved: S1, real-time market feature extraction: The MM system collects multi-dimensional market dynamic data, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation, to generate a market feature dataset; S2, BLM element dynamic mapping: Input the market feature dataset into the BLM model parsing engine, calculate the value proposition weight offset, task conflict matrix, and market opportunity window probability distribution in real time, and output the strategic element state vector. The strategic element state vector includes the value proposition weight vector, task priority sequence, and strategic gap heat map; S3, closed-loop optimization of decision strategies: Input the state vector of the strategic elements into the dynamic optimization engine, identify resource conflict items, generate Pareto optimal strategy clusters, and perform strategy risk entropy evaluation to generate an executable decision strategy set and activate the strategy executor; S4, feedback-driven model iteration: Collect the strategy effect feedback data generated by the strategy executor. When the inventory turnover acceleration in the feedback data is lower than the threshold or the curvature of the customer churn rate change exceeds the limit, it triggers the update of the BLM model parameters and MM data collection weights.

2. The intelligent decision-making dynamic optimization method based on BLM and MM according to claim 1 is characterized in that: Said S1 comprises: S11, real-time collection of multi-dimensional market dynamic data: The MM system collects multi-dimensional market dynamic data in real time, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation; S12, constructing a structured data set: aligning the multi-dimensional market dynamic data collected in S11 by timestamp to generate a structured data set containing the supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient, and channel inventory saturation; S13, generate a market characteristic data set: fill missing values ​​and remove outliers in the structured data set constructed in S12, and output a standardized market characteristic data set.

3. The intelligent decision-making dynamic optimization method based on BLM and MM according to claim 2 is characterized in that: The S2 includes: S21, market feature analysis input: input the market feature dataset outputted in S1 into the BLM model analysis engine; S22, real-time calculation of strategic elements: The following calculations are performed synchronously through the BLM model parsing engine: Calculate the value proposition weight offset based on the supply chain volatility index and social media sentiment polarity value in the market characteristic dataset; Based on the coefficient of variation of competitor product prices and channel inventory saturation in the market characteristic dataset, a task conflict matrix is ​​constructed; Generate the probability distribution of market opportunity windows based on the dynamic distribution of four-dimensional indicators in the market characteristic data set; S23, state vector encapsulation output: encapsulate the value proposition weight offset calculated in S22 into a value proposition weight vector, parse the task conflict matrix into a task priority sequence, map the market opportunity window probability distribution into a strategic gap heat map, and finally integrate and output the strategic element state vector including the value proposition weight vector, task priority sequence, and strategic gap heat map.

4. The intelligent decision-making dynamic optimization method based on BLM and MM according to claim 3 is characterized in that: The S3 includes: S31, strategic element state vector analysis: The strategic element state vector output by S2 is input into the dynamic optimization engine and decomposed into three subcomponents: value proposition weight vector, task priority sequence, and strategic gap heat map; S32, strategy generation and optimization: Executed sequentially through the dynamic optimization engine: Identify resource conflict items: Based on the matching analysis between the value proposition weight vector and the task priority sequence, identify the resource conflict items that may cause resource allocation conflicts; Generate Pareto optimal strategy clusters: Using resource conflict terms as constraints and combining them with the strategic gap heat map, perform multi-objective optimization to generate Pareto optimal strategy clusters. Perform strategy risk entropy assessment: Calculate the risk entropy value of each strategy in the Pareto optimal strategy cluster on the strategy gap heat map; S33, executable policy output and activation: screening the policies with risk entropy values ​​lower than the threshold to form an executable decision policy set, and activating the policy executor to load the executable decision policy set.

5. The intelligent decision-making dynamic optimization method based on BLM and MM according to claim 4 is characterized in that: The S4 includes: S41, strategy effect feedback collection: real-time collection of strategy effect feedback data generated after the strategy executor executes the executable decision strategy set, S42, dynamic threshold condition detection: real-time monitoring of the strategy effect feedback data collected in S41, and generating a trigger signal when the strategy effect feedback data exceeds a preset threshold; S43, collaborative update of model parameters: after the trigger signal is generated, an update operation is performed.

6. The intelligent decision-making dynamic optimization method based on BLM and MM according to claim 5 is characterized in that: The strategy effect feedback data includes inventory turnover acceleration and customer churn rate change curvature.

7. The intelligent decision-making dynamic optimization method based on BLM and MM according to claim 6 is characterized in that: The trigger signal includes: When the inventory turnover acceleration is lower than a preset acceleration threshold, a first trigger signal is generated; When the curvature of the customer churn rate change exceeds a preset curvature threshold, a second trigger signal is generated.

8. The intelligent decision-making dynamic optimization method based on BLM and MM according to claim 7 is characterized in that: When the first trigger signal or the second trigger signal is generated, an update operation is performed, and the update operation includes updating the BLM model parameters in the BLM model parsing engine and adjusting the MM data acquisition weight of the MM system data acquisition module.

9. An intelligent decision-making dynamic optimization system based on BLM and MM, used to implement the intelligent decision-making dynamic optimization method based on BLM and MM according to any one of claims 1 to 8, characterized in that: Includes the following modules: MM system data acquisition module: used to collect multi-dimensional market dynamic data from the external environment in real time, including supply chain volatility index, social media sentiment polarity value, competitor price variation coefficient and channel inventory saturation, and send it to the data access module; Data access module: used to align timestamps, unify formats, and structure data for the collected multi-dimensional market dynamic data to build a structured data set; Natural language processing module: used to classify and quantify the sentiment polarity values ​​of social media and generate sentiment scores in the range of [-1, 1]; Market feature processing module: used to perform missing value filling, outlier removal and Z-score normalization on structured data sets, and output standardized market feature data sets; BLM model parsing engine: includes value proposition modeling unit, task conflict parsing unit and market opportunity window analysis unit, which are used to: Calculate the value proposition weight offset based on the supply chain volatility index and social media sentiment polarity value; Construct a task conflict matrix based on the coefficient of variation of competitor prices and channel inventory saturation; Generate market opportunity window probability distribution based on multi-dimensional indicator distribution; Output the strategic element state vector, including the value proposition weight vector, task priority sequence, and strategic gap heat map; Task conflict analysis module: used to analyze the task conflict matrix into a task priority sequence, supporting task sorting and resource allocation judgment; Dynamic Optimization Engine: This engine receives the state vector of strategic elements, identifies resource conflicts, generates Pareto-optimal strategy clusters, and evaluates strategy risk entropy. It then filters and outputs executable decision strategy sets with risk entropy values ​​below a threshold. Strategy executor: used to load the executable decision strategy set and execute corresponding operations, while outputting strategy effect feedback data; Feedback analysis module: used to calculate inventory turnover acceleration and customer churn rate change curvature in real time, and generate trigger signals based on preset thresholds; Model adaptive update module: used to update the BLM model parameters in the BLM model parsing engine when a trigger signal is detected, and adjust the collection weight of the MM system data collection module to achieve closed-loop coordination between strategy execution and model update.

Images