Data sharing and analysis method and system based on multi-dimensional association

Abstract

The invention relates to the technical field of bulk commodity operation, in particular to a data sharing and analysis method and system based on multi-dimensional association, and the method comprises the steps: collecting multi-source heterogeneous data through a mixed mode of interface synchronization and manual reporting, and quantifying an unstructured text into a standardized index through a bidirectional LSTM semantic analysis algorithm; a cosine similarity and grey relational degree fusion algorithm and an LSTM dynamic weight optimization model are adopted, association rules are mined in combination with an improved Apriori algorithm, and market pre-judgment is achieved through three sub-models of supply and demand, price and risk; and finally, establishing a closed-loop mechanism, and outputting an inventory adjustment scheme and a futures hedging strategy. The system correspondingly comprises four functional modules, full-process data processing and decision support are achieved, and scientificity and operability of operation decision are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bulk commodity operation, more particularly, it relates to a data sharing and analysis method and system based on multi-dimensional correlation. BACKGROUND

[0002] The present application relates to the technical field of bulk commodity operation data processing and decision support, specifically focusing on the market analysis, inventory management and futures operation linkage scenarios of chemical and energy bulk commodities, aiming to solve the precision and synergy problems in operation decision-making through data sharing and multi-dimensional analysis.

[0003] In the prior art, the data processing and analysis method related to bulk commodity operation has obvious deficiencies: firstly, data collection is mostly limited to a single structured data source, lacking effective quantitative means for manually reported unstructured text data, resulting in incomplete data coverage and insufficient value mining; secondly, correlation analysis mostly focuses on single type correlation mining, making it difficult to capture linear and implicit non-linear relationships between data, and unable to dynamically adapt to market changes, resulting in insufficient analysis results; thirdly, market prediction and inventory and futures decision-making are mutually isolated, lacking a systematic closed-loop mechanism, making it difficult to directly convert prediction results into actionable plans, resulting in lagging or biased operation decisions.

[0004] To solve the above problems, the present application provides a data sharing and analysis method and system based on multi-dimensional correlation, which integrates full-dimensional data, accurately quantifies semantics, conducts comprehensive correlation analysis and outputs closed-loop decisions, to make up for the deficiencies of the prior art. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application aims to provide a data sharing and analysis method and system based on multi-dimensional correlation.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] The data sharing and analysis method based on multi-dimensional correlation specifically includes the following steps:

[0008] Step 1: Multi-source heterogeneous data collection: through a mixed collection mode of interface synchronization and manual reporting import, integrate full-dimensional data related to chemical and energy bulk commodity operation, including manually reported unstructured data, enterprise internal structured data and futures business data;

[0009] Step 2: Semantic quantification: preprocess the collected data, and quantize the manually reported unstructured text data into demand index, competitor influence coefficient and regional market activity level through a bidirectional LSTM semantic analysis algorithm;

[0010] Step three, multi-dimensional correlation analysis and market prediction: a hybrid model of cosine similarity, grey correlation degree fusion algorithm and LSTM dynamic weight optimization is used, and an improved Apriori algorithm is used to mine association rules; sub-models are constructed from the supply and demand, price and risk dimensions to realize market supply and demand state analysis, price trend prediction and core risk assessment;

[0011] Step four, inventory and futures linkage decision output: based on the market prediction results of step three, a closed-loop mechanism of market prediction, inventory adjustment and futures hedging is established, and finally, the inventory adjustment scheme and futures hedging strategy are output.

[0012] The application also provides a data sharing and analysis system based on multi-dimensional correlation, comprising:

[0013] Multi-source heterogeneous data acquisition module: used for mixed acquisition mode of interface synchronization and manual reporting import, integrating manual reporting unstructured data, enterprise internal structured data and futures business data;

[0014] Semantic quantification module: used for preprocessing the collected data, quantifying the unstructured text data into standardized indexes and grades through a bidirectional LSTM semantic parsing algorithm;

[0015] Multi-dimensional correlation analysis and prediction module: used for mining linear and nonlinear correlations of data using a hybrid model, mining association rules using an improved Apriori algorithm, and realizing market prediction through three sub-models;

[0016] Inventory and futures linkage decision output module: used for outputting inventory adjustment scheme and futures hedging strategy based on market prediction results, and establishing a closed-loop decision mechanism.

[0017] Further, the semantic quantification process of step two is as follows:

[0018] Determine the data object to be quantified: the unstructured text data is the text data manually reported in the bulk commodity operation scenario, and at least contains regional name, demand description, competitor name and competitor price change related description information;

[0019] Construct an industry dictionary: the industry dictionary contains special terms in the field of bulk commodities, competitor names and demand description keywords as basic entries, and training samples are automatically supplemented after the addition of entries to fine-tune subsequent quantification models;

[0020] Perform quantification using a bidirectional long short-term memory semantic parsing algorithm: the bidirectional LSTM semantic parsing algorithm includes an input layer, a bidirectional LSTM layer, a full connection layer and an output layer, wherein the input layer dimension matches the industry dictionary word vector dimension, and the sequence length adapts to the maximum number of characters of the unstructured text data to be quantified;

[0021] The result output in combination with the quantification mapping rule: based on the preset quantification mapping rule, the parsing result of the bidirectional LSTM semantic parsing algorithm on the unstructured text data is mapped to the demand index, the competitive product influence coefficient and the regional market activity level that can participate in the bulk commodity operation analysis, and the quantification mapping rule clearly defines the correspondence between different text description contents and the quantization intervals of the three indicators.

[0022] Further, the multi-dimensional correlation analysis in step three specifically includes:

[0023] Linear correlation mining: mining the linear correlation relationship between core indicators through the cosine similarity algorithm;

[0024] Nonlinear correlation mining: capturing the implicit nonlinear correlation between data through the grey correlation degree algorithm;

[0025] Dynamic weight optimization: using the LSTM neural network, dynamically adjusting the weight proportion of the cosine similarity algorithm and the grey correlation degree algorithm based on recent operation feedback data, and the weight update period is adapted to the market change frequency.

[0026] Further, the market prediction of step three is realized by constructing sub-models of supply and demand, price, and risk dimensions, specifically as follows:

[0027] Supply and demand analysis sub-model: constructing a supply and demand gap coefficient through the downstream operating load rate, demand index and inventory coefficient, quantifying the market supply and demand relationship, and presetting the supply and demand gap coefficient interval to determine the three supply and demand states of supply exceeding demand, supply and demand balance, and supply exceeding demand;

[0028] Price prediction sub-model: inputting core factors affecting the price, the core factors at least including regional price trend slope, futures and forward contract price pressure, raw material cost fluctuation factor, supply and demand gap coefficient and futures position change rate, obtaining the final price prediction result through the prediction mechanism, and determining the confidence interval of the predicted price;

[0029] Risk assessment sub-model: for the core risk types in the bulk commodity operation, the core risk types at least including price decline risk, inventory accumulation risk and demand reduction risk, respectively calculating the risk score based on the corresponding influence factors of each risk, and presetting the total risk score interval to divide the three risk levels of low risk, medium risk and high risk.

[0030] Further, the improved Apriori algorithm of step three optimizes the frequent item set mining process through transaction compression and hash pruning, presets minimum support degree and minimum confidence threshold; the standardized data after semantic quantification is converted into transaction items to construct a transaction dataset, the frequent item set is generated based on the transaction dataset, and the association rules meeting the minimum confidence threshold are screened out from the frequent item set, and the rules are output in the form of natural language combined with logical expressions in priority of confidence.

[0031] Further, the inventory adjustment scheme of step four includes safety inventory benchmark value determination and inventory operation suggestion generation:

[0032] Safety inventory benchmark value determination: combined with the weighted features of historical sales, seasonal fluctuation rules and future demand growth trend, the safety inventory benchmark value is determined, and the calculation logic is configured to adapt to the seasonal characteristics of different categories of bulk commodities;

[0033] Inventory operation suggestion generation: according to the market supply and demand state determination result, the safety inventory threshold under the corresponding supply and demand state is set, the operation direction of replenishment or clearing is clarified, and the replenishment or clearing batch is calculated based on the proportional factor corresponding to the supply and demand gap coefficient.

[0034] Further, the futures hedging strategy of step four includes hedging ratio calculation, operation opportunity determination and strategy parameter output:

[0035] Hedging ratio calculation: based on the coefficient corresponding to the risk level, the price fluctuation relative intensity coefficient and the preset adjustment coefficient, the hedging ratio adapting to the current market is calculated;

[0036] Operation opportunity determination: combined with the market supply and demand state, the price prediction trend amplitude and the hedging ratio, the trigger conditions of long hedging and short hedging are set respectively;

[0037] Strategy parameter output: output the core parameters required for futures operation, and the core parameters at least include futures contract identification, position direction, building price interval, stop loss point and stop profit point.

[0038] Compared with the prior art, the present application has the following beneficial effects:

[0039] 1. Solve the problems of multi-source heterogeneous data integration and unstructured data utilization, and improve the data support capability. The present application adopts a hybrid acquisition mode combining interface synchronization and manual reporting, and integrates unstructured text, enterprise internal structured and futures business data comprehensively; through constructing an industry dictionary and a bidirectional LSTM semantic analysis algorithm, the unstructured text is quantified into standardized indexes and grades, realizing effective utilization of unstructured data; combined with a hybrid algorithm of cosine similarity and gray correlation degree and LSTM dynamic weight optimization, linear and nonlinear correlations are fully mined, ensuring the comprehensiveness and adaptability of correlation analysis;

[0040] 2. Improve market prediction accuracy and provide reliable basis for decision-making. Based on the correlation analysis results, the supply and demand, price, risk three sub-models are constructed to realize market state analysis, price trend prediction and risk level evaluation from multiple dimensions. Meanwhile, the improved Apriori algorithm is introduced, the high confidence correlation rules are mined through transaction compression and hash pruning optimization, and the prediction results are output in natural language combined with logical expressions, so that the prediction results have scientificity and readability, and accurate and clear reference basis is provided for subsequent inventory and futures decision-making, avoiding decision-making blindness;

[0041] 3. Establish a closed-loop decision-making mechanism to realize inventory and futures linkage and improve the operability of business decision-making. Based on the market prediction results, a closed-loop mechanism of market prediction, inventory adjustment and futures hedging is established, personalized safety stock calculation logic is designed for different categories of bulk commodities, and replenishment or inventory clearing suggestions are output combined with supply and demand state. Meanwhile, according to the risk level and price prediction, the hedging ratio, operation opportunity and core parameters are calculated, and the strategy can be directly imported into the trading software, solving the problem of disconnection between inventory and futures decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The flowchart of the data sharing and analysis method based on multi-dimensional correlation of the present application;

[0043] Figure 2 The flowchart of multi-dimensional correlation analysis and market prediction of the present application;

[0044] Figure 3 The module block diagram of the data sharing and analysis system based on multi-dimensional correlation of the present application. DETAILED DESCRIPTION

[0045] Example 1, refer to Figure 1 The data sharing and analysis method based on multi-dimensional correlation of the present embodiment specifically includes the following steps:

[0046] Step 1, multi-source heterogeneous data acquisition.

[0047] Through the mixed acquisition mode of interface synchronization and manual reporting import, the full-dimensional data integration related to chemical, energy and other bulk commodity operation is realized, covering the data link from first-line market feedback to enterprise internal operation to futures market dynamics, providing complete data basis for subsequent quantitative analysis;

[0048] S11, data type and field definition:

[0049] S111. Manually submitted unstructured data: includes region name, demand description, competitor name, competitor price change range, downstream feedback, submitter, and submission time; where demand description is in text format, competitor price change range is a numerical value (in %), and submission time is in date format. The data comes from account managers or regional managers filling in the system's built-in templates. For example: East China region, agricultural film demand increases by 20% daily, competitor A's price decreases by 5%;

[0050] S112. Enterprise Internal Structured Data: This includes CRM customer order volume (in tons), customer credit rating (divided into five AE levels), ERP inventory quantity (in tons), ERP capacity (monthly average tons), capacity utilization rate (%), regional real-time price (in yuan / ton), and upstream and downstream operating load rate (in%). The data comes from CRM systems (such as Salesforce), ERP systems (such as SAP S / 4HANA), BW data warehouse, and enterprise transaction systems. Among them, CRM system refers to Customer Relationship Management System; ERP system refers to Enterprise Resource Planning System; and BW refers to Data Warehouse.

[0051] S113. Futures Business Data: This includes futures contract codes, real-time market prices (in yuan / ton), forward contract prices (covering a 1-3 month period, in yuan / ton), open interest (in lots), open interest change rate (in %), and premium / discount amount (in yuan / ton). The data is sourced from the API of commodity exchanges (such as the Shanghai Futures Exchange) and futures trading software.

[0052] S12, Data Mixed Acquisition Implementation:

[0053] S121, Interface Synchronous Acquisition:

[0054] API interface unit configuration: The protocol version adopts HTTP / HTTPS1.1 and also supports WebSocket protocol for real-time push of futures market data; the platforms to be connected include but are not limited to the market data interface of Zhengzhou Commodity Exchange, Shanghai Futures Exchange, and price interface of coal chemical platform; the request header must carry the API key issued by the third-party platform, and the request frequency is set to 1 time / second for futures data and 1 time / 5 minutes for price and operating rate data;

[0055] Database synchronization unit configuration: The selected tool is the lightweight ETL tool Apache NiFi 1.23.0, which supports visual process orchestration; compatible databases include Oracle 19c, SQL Server 2022, and MySQL 8.0, and connections are established via JDBC drivers; the synchronization strategy adopts incremental synchronization and full verification. Incremental synchronization captures newly added or modified data based on database logs, and a full data verification is performed at 2:00 AM every day to ensure data consistency;

[0056] S122, Manual Reporting and Import Data Collection:

[0057] Standardized template design: Template formats support Excel 2007-2021 (.xlsx) and CSV; required fields include region name (covering 31 provincial-level administrative regions nationwide), requirement description, competitor name (manually entered), and submission time (automatically filled with the current date); validation rules include requirement description cannot be empty, competitor price fluctuation range must be within the "-50% to 50%" range, and the submitter must link an internal company account;

[0058] Data collection frequency configuration: Supports two modes: scheduled collection and manual trigger collection. Scheduled collection can be customized within the range of 1-24 hours. Manual trigger collection requires the user to upload a template and click the "Collect Now" button. Data parsing and database entry will be completed within 10 minutes after collection.

[0059] Step 2: Semantic quantization.

[0060] The collected data was cleaned and deduplicated to remove invalid information and duplicate data; numerical data was standardized using Z-score, and categorical data was encoded using one-hot encoding to eliminate differences in units and formats; unstructured text data submitted manually still needed to be further quantified before it could participate in subsequent association analysis, so a bidirectional LSTM semantic parsing algorithm was used to achieve semantic quantization.

[0061] S21. Network structure of the bidirectional LSTM semantic parsing algorithm:

[0062] The input dimension of the input layer is equal to the word vector dimension of the industry dictionary (300 dimensions), and the sequence length is equal to the maximum number of characters in the manually submitted text (500 in this embodiment);

[0063] The bidirectional LSTM layer consists of two layers, each with 256 hidden units, a dropout rate of 0.2, a recurrent_dropout rate of 0.1, and the activation function is ReLU. The fully connected layer has 128 hidden units (with ReLU activation function) and 3 output units, corresponding to the demand index (0-10 points), competitor influence coefficient (0-5 points), and regional market activity level (1-5 levels), respectively.

[0064] In the output layer activation function, the demand index and competitor influence coefficient are Sigmoid, and the regional market activity level is Softmax.

[0065] S22, Industry Dictionary and Quantitative Mapping:

[0066] The industry dictionary consists of: basic entries including chemical commodity terms, competitor names, and demand description keywords; the extension mechanism allows users to add entries through the system's dictionary management function, filling in the entry name, category, and associated quantitative range. After adding, the model performs fine-tuning, and the system automatically supplements the training set with the added entries. The amount of training data = number of added entries × 100 labeled samples.

[0067] Quantitative mapping rules: A demand index score of 0-3 corresponds to textual descriptions such as "sluggish demand, with a 30% decrease in procurement volume" and "downstream enterprises have suspended operations and have no intention to purchase"; a score of 4-7 corresponds to descriptions such as "stable demand, with procurement volume remaining the same as last month" and "downstream enterprises are purchasing as needed, with no significant fluctuations"; a score of 8-10 corresponds to descriptions such as "strong demand, with procurement volume surging by 50%" and "downstream enterprises are actively stocking up, with orders scheduled for next month".

[0068] A competitor impact coefficient of 0-2 points corresponds to descriptions such as "competitors have no price changes and their market share is stable" and "competitors have insufficient production capacity and have no impact on our company"; a coefficient of 3-5 points corresponds to descriptions such as "competitors reduce prices by 5% and some customers are lost" and "competitors launch new products and seize 10% of the market share".

[0069] Regional market activity levels 1-2 correspond to descriptions such as "sluggish transactions, with an average daily transaction volume of less than 100 tons" and "traders are hesitant and have few inquiries"; levels 3-4 correspond to descriptions such as "stable transactions, with an average daily transaction volume of 200-300 tons" and "traders replenish their stock as needed and have moderate inquiries"; level 5 corresponds to descriptions such as "booming transactions, with an average daily transaction volume of over 500 tons" and "traders are actively stockpiling and have frequent inquiries".

[0070] Step 3: Multidimensional correlation analysis and market forecasting.

[0071] After semantic quantization, standardized and computable full-dimensional data has been obtained. Through association analysis, deep relationships between data are mined to achieve market prediction. A hybrid model combining cosine similarity, grey relational analysis, and LSTM dynamic weight optimization is employed, along with an improved Apriori algorithm to mine association rules. A four-layer analysis architecture is constructed that considers linear association, non-linear association, dynamic adaptation, and rule verification, ensuring both the comprehensiveness of association analysis and the accuracy of market prediction. Figure 2 As shown, the specific steps are as follows:

[0072] S31. Linear association mining based on cosine similarity algorithm:

[0073] For linear relationships in numerical data, a cosine similarity algorithm is used for data mining. The calculation focuses on 15 core numerical indicators, including inventory quantity, operating rate, regional price, futures real-time price, open interest, demand index, competitor influence coefficient, raw material cost, order volume, capacity utilization rate, transaction amount, futures premium / discount, inventory turnover rate, customer credit rating, and downstream purchasing frequency. The calculation process consists of three steps:

[0074] The first step is to construct a vector matrix M from the standardized data of the 15 indicators (based on the Z-score standardization results). n×15 (n is the number of data samples);

[0075] The second step is to calculate the cosine similarity S between any two index vectors;

[0076] The third step is to output the linear correlation matrix, which provides a basis for subsequent analysis of linear correlations.

[0077] S32. Nonlinear association mining based on grey relational algorithm:

[0078] To compensate for the shortcomings of the cosine similarity algorithm in nonlinear association mining, the grey relational algorithm is used to capture latent associations: the resolution coefficient ρ = 0.5 (the industry's conventional optimal value) is used in the algorithm parameters, the reference sequence is set as: price fluctuation range (dependent variable), and the comparison sequences are set as: demand index, competitor influence coefficient, futures premium / discount, open interest change rate, and inventory coefficient (independent variables); the calculation steps are divided into three steps:

[0079] The first step is to test the reference sequence X0 and the comparison sequence X. i (i = 1, 2, ... 5) are initialized by eliminating dimensions and using an initialization method:

[0080] The second step is to calculate the difference between the reference sequence X0 and the comparison sequence X at the k-th data point. i The absolute difference sequence Δ after initialization i (k)=|x'0(k)-x' i (k)|, find the minimum absolute difference Δ among all comparison sequences i and all data points k. min and the maximum absolute difference Δ max ;

[0081] The third step is to calculate the grey relational degree r, using the following formula:

[0082]

[0083] n1 represents the length of the data sequence, i.e., the number of data points involved in the calculation; k is the index in the data sequence;

[0084] S33. Market-adaptive dynamic weight optimization:

[0085] To enable the hybrid algorithm to adapt to dynamic market changes, an LSTM neural network is used to dynamically adjust the weights of the two algorithms. The input features are operational feedback data from the past 6 months, consisting of 8 features, including the deviation rate between the order price and the actual price, the negative effect coefficient of inventory overdraft procurement, the order fulfillment rate, the customer complaint rate, the profit from futures hedging, the inventory turnover rate, the competitor's price following rate, and the demand forecast deviation rate.

[0086] The deviation rate between the order price and the actual price is defined as the degree of deviation between the actual settlement price and the price agreed in the order, reflecting the accuracy of price execution; the negative effect coefficient of inventory overdraft procurement is defined as the degree of negative impact of emergency procurement on operating costs and supply stability when inventory is insufficient. It is calculated by weighting the emergency procurement premium rate, delivery delay days, and supplier default risk, and finally normalized to a coefficient of 0-1. The higher the value, the more severe the negative impact. The supplier default risk is obtained by mapping based on historical data; the order fulfillment rate is defined as the proportion of orders completed on time and with good quality to the total number of orders; the customer complaint rate is defined as the proportion of the number of complaint orders to the total number of orders within a certain period (6 months in this embodiment); the period The profit from hedging is defined as the amount of profit or loss generated by hedging spot price fluctuations through futures instruments. It is calculated as: (futures closing price - opening price) × open position - transaction fee. A positive number indicates profit, and a negative number indicates loss. The inventory turnover rate is defined as the number of times inventory turns over within a certain period (6 months in this example), reflecting inventory liquidity. The competitor price following rate is defined as the degree of matching between the frequency or magnitude of price adjustments of one's own product and those of major competitors, and can be measured by the absolute value of (own price change rate - competitor price change rate). The demand forecast deviation rate is defined as the degree of deviation between the demand forecast and actual demand, where the demand forecast is mapped from the demand index of S22.

[0087] The LSTM model parameters are set as follows: The network structure is input layer (8-dimensional) → LSTM layer (64 hidden units, dropout = 0.2) → fully connected layer (32 units, ReLU activation) → output layer (2 units, Softmax activation, corresponding to cosine similarity weight ω1 and gray relational weight ω2, and ω1 + ω2 = 1);

[0088] The training parameters are: optimizer = Adam (learning rate = 0.001), loss function = cross-entropy loss, batch size = 32, number of iterations = 100 rounds; the weight update mechanism is to perform incremental training every 7 days based on the latest business feedback data (adding 7 days of samples), and update ω1 and ω2 to ensure that the algorithm adapts to market changes;

[0089] The overall correlation degree R = ω1×S + ω2×r, where ω1 and ω2 are the dynamic weights of the output, S is the linear correlation degree calculated by S31, and r is the non-linear correlation degree calculated by S32. Based on domain knowledge, a strong correlation relationship is determined when the overall correlation degree is ≥0.7. Finally, an overall correlation degree matrix of all-dimensional index pairs is generated, and strong correlation relationships with an overall correlation degree ≥0.7 are marked.

[0090] S34, Core Modeling System for Market Forecasting:

[0091] Based on the above correlation analysis results, three sub-models are constructed from the dimensions of supply and demand, price, and risk to achieve market prediction:

[0092] S341. Supply and Demand Analysis Model:

[0093] The core formula is:

[0094] Among them, the maximum value of the supply-demand gap coefficient is set at 2.0 according to industry practice, and the downstream operating load rate (%) is obtained from the API of a third-party platform and the average value of the past 7 days is taken to ensure the timeliness of the data.

[0095] Inventory coefficient = current inventory divided by safety stock baseline, where current inventory is read in real time from the ERP system; safety stock baseline = weighted average of historical sales over the past 12 months × (1 + demand growth coefficient), taking into account both historical data and future demand forecasts; weighted average of monthly historical sales = monthly sales × monthly weight, based on industry experience, peak season weight = 1.2, off-season weight = 0.8; demand growth coefficient is obtained by calculating the trend slope of historical monthly sales data over the past 12 months. When calculating the trend slope, the month is used as the independent variable and monthly sales as the dependent variable, and the slope obtained by fitting a linear regression model is the demand growth coefficient;

[0096] The rules for determining supply and demand status are as follows:

[0097] When the supply-demand gap coefficient is ≤0.3, it indicates that demand exceeds supply, requiring an increase in inventory and suggesting replenishment.

[0098] When the supply-demand gap coefficient is 0.3 < 0.8, supply and demand are in balance, and safety stock should be maintained.

[0099] When the supply-demand gap coefficient is greater than 0.8, it indicates oversupply, requiring a reduction in inventory. It is recommended to release inventory or promote sales, and to clearly define the supply-demand status and corresponding operational direction.

[0100] S342, Price Forecasting Model:

[0101] Five core factors were input, including the regional price trend slope, futures forward contract price pressure, raw material cost volatility factor, supply and demand gap coefficient, and open interest change rate, comprehensively covering key factors affecting prices. The regional price trend slope was calculated by fitting a linear regression to the regional commodity price time series data over the past 30 days; the resulting slope is the regional price trend slope. Futures forward contract price pressure = forward contract price - real-time price of the main contract. The forward contract price refers to futures contracts expiring in 1-3 months, and the real-time price of the main contract refers to the contract with the highest current trading volume. The raw material cost volatility factor was quantified by the change rate of the average raw material cost over the past 15 days and the previous 15 days. The open interest change rate was quantified by the change rate of the average open interest over the past 7 days and the previous 7 days.

[0102] The price forecasting process consists of three steps:

[0103] The first step involves using multiple linear regression, with five factors as independent variables X and historical price fluctuations as the dependent variable Y, to establish the regression equation Y = β0 + β1×X1 + β2×X2 + ... + β5×X5, outputting the preliminary predicted value Y. linear ;

[0104] The second step involves correcting the LSTM residuals and calculating the historical forecast residuals: Actual Price - Y. linear The future residual e is predicted using a typical LSTM model (input is the residual sequence, hidden units 32, iterations 50 times) to correct the prediction bias of linear regression.

[0105] The third step is to determine the final predicted value Y. final =Y linear +e, based on Y final Calculate the standard deviation σ of the forecast residuals over the past 3 months to determine the fluctuation range: [Y final -1.28σ,Y final +1.28σ] (90% confidence interval); meanwhile, the preset accuracy indicators are short-term (1 month) forecast deviation rate ≤8%, medium-term (2-3 months) forecast deviation rate ≤20%, and average deviation rate ≤10%, to ensure the accuracy of price forecasts;

[0106] S343, Risk Assessment Model;

[0107] To address the core risks in commodity trading, a risk assessment model is constructed:

[0108] Risk factors for price downside risk include the standard deviation of price volatility and the rate of change in futures open interest, and their scores are calculated as follows: λ1 and λ2 are weights, determined based on 1000 sets of sample experiments. In this embodiment, the risk identification accuracy is highest when λ1 = 0.6 and λ2 = 0.4. Among them, the standard deviation of price fluctuation is taken as the standard deviation of the commodity price over the past 30 days, reflecting the dispersion of price fluctuation; the historical mean of price fluctuation is taken as the average of the standard deviation of price fluctuation over the past 12 months, as the standardization benchmark; the rate of decline in open interest is taken as the percentage decrease in current futures open interest compared to the previous trading day; the maximum rate of decline in open interest is taken as the maximum percentage decrease in a single day of futures open interest over the past 12 months, as the standardization benchmark.

[0109] The risk factors for inventory backlog risk include the supply-demand gap coefficient and inventory turnover rate, and the scores are calculated as follows: λ3 and λ4 are weights, determined based on 1000 sets of sample experiments. In this embodiment, the risk identification accuracy is highest when λ3 = 0.7 and λ4 = 0.3. The inventory turnover rate is the inventory turnover rate of the past 30 days. Inventory turnover rate = cost of sales divided by average inventory. The historical mean is the average of the inventory turnover rate of the past 12 months, which is used as the standardization benchmark.

[0110] The risk factors for a sharp drop in demand include the magnitude of the decline in the demand index and the rate of decrease in customer orders, with scores calculated as follows: λ5 and λ6 are weights, determined based on 1000 sets of sample experiments. In this embodiment, the risk identification accuracy is highest when λ5 = 0.5 and λ6 = 0.5. The decline in the demand index is taken as the decrease in the current demand index compared to the previous period (e.g., the previous week); the maximum decline is taken as the maximum decline in the demand index in a single period over the past 12 months, as a standardization benchmark; the customer order reduction rate is taken as the percentage decrease in the current order volume compared to the previous period (e.g., the previous week); the maximum reduction rate is taken as the maximum percentage decrease in the order volume in a single period over the past 12 months, as a standardization benchmark.

[0111] Based on the above three types of risks, the total risk score is calculated, and the risk level classification rules are as follows:

[0112] A total risk score of ≤0.3 indicates low risk, meaning a risk probability of ≤30%.

[0113] A total risk score of 0.3 < 0.7 indicates medium risk, representing a risk probability of 30% to 70%.

[0114] A total risk score > 0.7 indicates high risk, meaning a risk probability ≥ 70%.

[0115] S35. Association rule mining based on the improved Apriori algorithm:

[0116] S351. Improved Strategy: To verify the association analysis results and provide a clear basis for decision-making, an improved Apriori algorithm is used to mine association rules. The specific operation is as follows:

[0117] Transaction compression: After the frequent 1-itemset mining is completed, the original transaction dataset is compressed—all infrequent 1-itemsets contained in the transaction are removed, and only items that may participate in the generation of frequent k-itemsets (k≥2) are retained;

[0118] Hash pruning: When generating candidate k-itemsets, a hash table is used to record the support count of each candidate set. If the support count of a candidate set is lower than the threshold of minimum support × total number of transactions, the candidate set is pruned directly to prevent it from participating in subsequent frequent itemset judgments and association rule generation.

[0119] S352, Algorithm Parameter Definition:

[0120] Minimum support (15%):

[0121] Definition: The itemset corresponding to a certain association rule appears in all transaction samples at a rate ≥ 15%;

[0122] Value selection basis: Based on the sample distribution statistics of commodity trading data over the past 12 months, it was found that a 15% threshold can filter out occasional weak correlation patterns while retaining more than 80% of strong correlation patterns with business significance.

[0123] Minimum confidence level (70%):

[0124] Definition: The probability of a certain association rule's "premise → conclusion" being true is ≥70%;

[0125] Basis for value selection: Through backtesting experiments of 1000 sets of historical business scenarios, when the confidence level is ≥70%, the accuracy of the rule in actual business decision-making is stable at over 85%; if the confidence level is below 70%, the accuracy will plummet to below 50%. Therefore, 70% is determined to be the critical threshold for the effectiveness and reliability of the rule.

[0126] S353, Rule Generation and Output:

[0127] Data preprocessing: The standardized data and semantic quantization results from Sections S31-S34 are transformed into transaction items to construct a transaction dataset. Specifically, the standardized categorical data is directly used to generate transaction items based on whether the categories match. The semantically quantified and standardized continuous data (such as downstream operating rate, demand index, futures premium, etc.) are combined with industry experience and thresholds verified by 1000 sets of historical samples (such as downstream operating rate ≥85%, demand index ≥8 points, futures premium ≥3%) to be transformed into Boolean transaction items based on whether the values ​​meet the thresholds. All transaction items of each data point are combined to form a complete transaction, and finally, a transaction dataset is constructed to improve the Apriori algorithm mining.

[0128] Improve Apriori execution:

[0129] Step 1: Optimize using transaction compression and hash pruning to generate frequent 1-itemsets, frequent 2-itemsets, ... up to frequent k-itemsets, where k is the number of transaction items contained in the itemset;

[0130] Step 2: Generate association rules based on frequent itemsets, and filter out rules with a confidence level of ≥70%;

[0131] Formatted output according to rules:

[0132] Using Python's natural language processing library (spaCy) and logical expression parsing library (PyParsing), the data is presented in the form of natural language and logical expressions, and sorted from high to low confidence, making it easy for users to use as a reference for inventory and futures decisions.

[0133] Step 4: Outputting decisions based on the linkage between inventory and futures.

[0134] Based on the results of multidimensional correlation analysis and market forecasting, this step establishes a closed-loop mechanism of market forecasting → inventory adjustment → futures hedging, outputting an operational plan that can be directly implemented, ensuring that those skilled in the art can execute it without additional optimization, while also supporting personalized adjustments, solving the pain point of the disconnect between inventory and futures in existing technologies, and providing full-process decision support for proactive inventory management of bulk commodities.

[0135] S41. Inventory Adjustment Plan:

[0136] Based on the results of supply and demand analysis models and price forecasting models, develop targeted inventory adjustment plans:

[0137] S411, Safety Stock Calculation Template:

[0138] The general formula is: Safety stock baseline value = weighted average of historical sales over the past 12 months × seasonal fluctuation coefficient × demand growth coefficient, taking into account historical sales, seasonal characteristics and future demand growth.

[0139] Taking aromatics, rubber, and synthetic fibers as examples, in the product-specific templates, the weighted average weight of aromatics is 1.3 for peak season, 1.0 for off-season, and 0.7 for off-season, with seasonal fluctuation coefficients of 1.2 for January-March, 1.0 for April-June, 0.8 for July-September, and 1.1 for October-December;

[0140] The weighted average weight for rubber products is 1.4 for peak season, 1.0 for off-season, and 0.6 for low-season. The seasonal fluctuation coefficients are 1.3 for January-March, 1.0 for April-June, 0.7 for July-September, and 1.2 for October-December.

[0141] The weighted average weight for synthetic fibers is 1.2 for peak season, 1.0 for off-season, and 0.8 for low-season. The seasonal fluctuation coefficients are 1.1 for January-March, 1.0 for April-June, 0.9 for July-September, and 1.2 for October-December, to suit the seasonal characteristics of different products.

[0142] The weighted average weight and seasonal fluctuation coefficient are determined based on industry demand patterns and historical data statistics. Industry demand patterns determine the peak and off-peak seasons of products, and historical data statistics calculate the increase or decrease in sales during peak and off-peak seasons compared to the flat seasons using monthly sales data.

[0143] The adjustment mechanism allows users to modify weights and coefficients in the template configuration, and the safety stock benchmark value is updated in real time after modification to meet personalized business needs.

[0144] S412. Inventory Operation Recommendations: Based on the supply and demand situation assessment, formulate clear inventory operation recommendations:

[0145] In a supply-demand imbalance, the safety stock threshold is the baseline value × 1.3. The recommended action is immediate replenishment, prioritizing futures delivery, followed by spot purchases. The replenishment batch size is calculated as: Replenishment Batch = (Baseline Value × 1.3 - Current Inventory) × ψ1, where ψ1 is the stocking ratio factor corresponding to the supply-demand gap coefficient output by the supply-demand analysis model. A linear mapping formula is used to derive ψ1: ψ1∈[0.8,1.2], where the supply-demand gap coefficient is the normalized value calculated in S341, and satisfies the condition that the supply-demand gap coefficient ≤0.3 under the condition of supply shortage; when the supply-demand gap coefficient = 0.3 (the critical value of supply shortage), ψ1 = 0.8 (minimum stocking ratio); when the supply-demand gap coefficient = 0 (extreme supply shortage), ψ1 = 1.2 (maximum stocking ratio);

[0146] Under a balanced supply and demand situation, the safety stock threshold is the baseline value × (0.9-1.1). The recommended operation is to maintain inventory and replenish as needed. The batch standard is that the replenishment batch = baseline value × 0.1 (each replenishment shall not exceed 10% of the baseline value) to avoid inventory backlog or shortage.

[0147] In a supply-demand imbalance, the safety stock threshold is the baseline value × 0.7. The recommended action is to expedite inventory release, potentially with promotions, to avoid stockpiling. The batch size standard is: release batch = (current inventory - baseline value × 0.7) × ψ2, where ψ2 is the inventory clearance ratio corresponding to the supply-demand gap coefficient output by the supply-demand analysis model. A linear mapping formula is used to derive ψ2: ψ2 is the outbound ratio factor corresponding to the supply-demand gap coefficient output by the supply-demand analysis model. ψ2∈[0.6,1.0], where the maximum value of the supply-demand gap coefficient is set to an upper limit of 2.0 according to industry convention. When the supply-demand gap coefficient = 0.8 (the critical value of oversupply), ψ2 = 0.6 (the minimum inventory clearance ratio); when the supply-demand gap coefficient = 2.0 (extreme oversupply), ψ2 = 1.0 (the maximum inventory clearance ratio).

[0148] S42. Futures Hedging Strategy:

[0149] By combining the results of the risk assessment model and the price prediction model, a futures hedging strategy is output:

[0150] S421. Calculation of hedging ratio:

[0151] The core formula is: Hedging ratio = Risk level coefficient × Price volatility coefficient × 0.8; 0.8 is the adjustment coefficient obtained from the fitting of multiple sets of experiments; where the risk level coefficient corresponds to 0.3 for low risk, 0.5 for medium risk, and 0.8 for high risk, and is directly related to the risk level; the price volatility coefficient is specifically the predicted price volatility of the past 15 days output by the price prediction model divided by the historical average price volatility of the past 6 months statistically analyzed by the price prediction model, with a value range of 0.5 to 1.5, reflecting the relative strength of the current price volatility relative to the historical level;

[0152] Example: For high risk (coefficient 0.8) and price volatility coefficient 1.2, the hedging ratio = 0.8 × 1.2 × 0.8 = 0.768 → 77%, which is within the adjustable range of 12% to 96%, ensuring the rationality and flexibility of the ratio;

[0153] S422. Determining the timing of operations: Identifying the appropriate timing for futures hedging operations;

[0154] Long hedging is suitable when supply is insufficient, the price is expected to rise by ≥5%, and the hedging ratio is ≥40%, and it is applicable to scenarios with high price increase risk.

[0155] Short hedging is suitable when supply exceeds demand, the price is expected to fall by ≥5%, and the hedging ratio is ≥50%, and it is applicable to scenarios with high price decline risk.

[0156] The timing alert method involves the system monitoring and predicting the results in real time. When the timing conditions are met, a futures operation alert will be sent within 10 minutes to ensure that users can seize the opportunity to operate in a timely manner.

[0157] S423, Hedging Strategy Parameter Output: Outputs hedging strategy parameters that can be directly implemented;

[0158] The parameters include: contract code, position direction (long / short), entry price range (based on price prediction range), stop-loss point (entry price × 97%), and take-profit point (entry price × 105%). The parameters are clear and specific. The interface supports APIs of mainstream futures trading software such as WenHua Finance Winshun Cloud and Boyi Master. The parameters can be directly imported into the software (in JSON format), eliminating the need for manual input by the user and improving operational convenience.

[0159] Example 2, refer to Figure 3 The data sharing and analysis system based on multi-dimensional correlation in this embodiment includes a multi-source heterogeneous data acquisition module, a semantic quantification module, a multi-dimensional correlation analysis and prediction module, and an inventory and futures linkage decision output module;

[0160] Multi-source heterogeneous data acquisition module: Through a hybrid acquisition mode of interface synchronization and manual reporting and import, it integrates all-dimensional data related to bulk commodity operations. The interface synchronization acquisition part is configured with adapted API interfaces and database synchronization tools to connect with various external platforms and internal enterprise systems to obtain futures business data and internal structured data. The manual reporting and import acquisition part is designed with standardized templates to receive unstructured data reported by account managers or regional managers. Ultimately, it realizes the integration of various data sources and provides a foundation for subsequent quantitative analysis.

[0161] Semantic Quantization Module: First, the collected multi-source data is preprocessed. For unstructured text data submitted manually, a bidirectional LSTM semantic parsing algorithm is used. Combined with an industry dictionary containing industry terms and keywords and preset quantization mapping rules, the text description is transformed into standardized indices and levels that can participate in subsequent association analysis, completing the quantization transformation of all-dimensional data and preparing for association analysis.

[0162] Multi-dimensional correlation analysis and prediction module: Based on standardized data after semantic quantification, it first uses different algorithms to mine linear and non-linear correlations between data, and then uses an LSTM neural network to dynamically adjust the weights of the two algorithms according to recent business feedback data to adapt to market changes. At the same time, it uses an improved Apriori algorithm to mine correlation rules with business significance. On this basis, it constructs sub-models from three dimensions: supply and demand, price, and risk, to analyze the market supply and demand status, predict price trends, and assess core risks in operations. Finally, it outputs correlation analysis results and multi-dimensional market prediction conclusions to provide a basis for subsequent decision-making.

[0163] The inventory and futures linkage decision-making output module establishes a closed-loop mechanism from market forecasting to inventory adjustment and futures hedging, based on market forecasting results derived from multi-dimensional correlation analysis and prediction modules. Regarding inventory adjustment, it combines supply and demand conditions with safety stock calculation logic adapted to different product characteristics to formulate inventory adjustment plans, including inventory threshold settings, replenishment or clearance operation directions, and batch recommendations. For futures hedging, it calculates appropriate hedging ratios based on risk levels and price forecasts, clarifies the timing of long or short hedging operations, and outputs specific strategy parameters including contract information, position opening range, and stop-loss and take-profit settings. Finally, it outputs actionable operational plans, providing full-process decision support for proactive inventory management of bulk commodities.

[0164] Through the detailed description of the above embodiments, the data sharing and analysis method and system based on multi-dimensional correlation of the present invention breaks through the limitations of traditional data processing by comprehensively integrating multi-source heterogeneous data and accurately quantifying semantics; it achieves in-depth mining of data correlation by using hybrid algorithms and dynamic weight optimization; it improves the comprehensiveness and reliability of market prediction by relying on three sub-models and the improved Apriori algorithm; and finally, through a closed-loop decision-making mechanism, it transforms the prediction results into directly implementable inventory adjustment and futures hedging solutions. This method and system spans the entire process from data collection, quantification, analysis, prediction to decision output, effectively solving problems such as insufficient data utilization, one-sided correlation analysis, and disconnected decision-making in bulk commodity trading. It provides scientific and efficient decision support for the trading of chemical and energy bulk commodities, helping enterprises improve operational stability and market competitiveness.

[0165] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0166] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0167] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0168] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0169] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0170] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0171] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0172] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data sharing and analysis method based on multi-dimensional correlation, characterized in that, The method flow is as follows: Step 1: Multi-source heterogeneous data acquisition: Through a hybrid acquisition mode of interface synchronization and manual submission, integrate all-dimensional data related to the operation of chemical and energy commodities. The all-dimensional data includes manually submitted unstructured data, internal structured data of enterprises, and futures business data. Step 2, Semantic Quantization: The collected data is preprocessed and the unstructured text data submitted by humans is quantified into demand index, competitor influence coefficient and regional market activity level through a bidirectional LSTM semantic parsing algorithm; Step 3, Multidimensional Association Analysis and Market Forecasting: A hybrid model combining cosine similarity, grey relational degree fusion algorithm and LSTM dynamic weight optimization is adopted, and the improved Apriori algorithm is used to mine association rules; sub-models are constructed from the dimensions of supply and demand, price, and risk to realize market supply and demand status analysis, price trend prediction and core risk assessment. Step 4, Inventory and Futures Linked Decision Output: Based on the market forecast results in Step 3, establish a closed-loop mechanism from market forecasting and inventory adjustment to futures hedging, and finally output the inventory adjustment plan and futures hedging strategy.

2. The data sharing and analysis method based on multi-dimensional correlation according to claim 1, characterized in that, The semantic quantization process in step two is as follows: Identify the data objects to be quantified: The unstructured text data is text data manually submitted in the context of bulk commodity trading, and includes at least the region name, demand description, competitor name, and description information related to competitor price changes; Constructing an industry dictionary: The industry dictionary includes specialized terms in the commodity field, competitor names, and keywords describing demand as basic entries. As new entries are added, training samples are automatically supplemented to fine-tune the subsequent quantitative model. The quantization is performed using a bidirectional long short-term memory semantic parsing algorithm: the bidirectional LSTM semantic parsing algorithm includes an input layer, a bidirectional LSTM layer, a fully connected layer and an output layer, wherein the dimension of the input layer matches the dimension of the word vectors in the industry dictionary, and the sequence length is adapted to the maximum number of characters in the unstructured text data to be quantized; The output results of the quantization mapping rule are combined as follows: Based on the preset quantization mapping rule, the parsing results of the bidirectional LSTM semantic parsing algorithm on unstructured text data are mapped to the demand index, competitor influence coefficient and regional market activity level that can be used for commodity business analysis. The quantization mapping rule clarifies the correspondence between different text descriptions and the quantization intervals of the above three indicators.

3. The data sharing and analysis method based on multi-dimensional correlation according to claim 1, characterized in that, The multidimensional correlation analysis in step three specifically includes: Linear correlation mining: using cosine similarity algorithm to mine linear correlations between core indicators; Nonlinear association mining: capturing implicit nonlinear associations between data using the grey relational analysis algorithm; Dynamic weight optimization: Using an LSTM neural network, the weight ratio of the cosine similarity algorithm and the grey relational algorithm is dynamically adjusted based on recent business feedback data, and the weight update cycle is adapted to the frequency of market changes.

4. The data sharing and analysis method based on multi-dimensional correlation according to claim 1, characterized in that, The market forecast described in step three is achieved by constructing sub-models based on the dimensions of supply and demand, price, and risk, as detailed below: Supply and demand analysis sub-model: The supply and demand gap coefficient is constructed by using downstream operating rate, demand index and inventory coefficient to quantify the market supply and demand relationship. The supply and demand gap coefficient range is preset to determine the three supply and demand states: supply shortage, supply and demand balance and supply surplus. Price prediction sub-model: Input the core factors affecting prices, including at least the regional price trend slope, futures forward contract price pressure, raw material cost volatility factor, supply and demand gap coefficient, and futures open interest change rate. Obtain the final price prediction result through the prediction mechanism and determine the confidence interval of the predicted price. Risk assessment sub-model: For the core risk types in commodity trading, the core risk types include at least price decline risk, inventory backlog risk and sudden demand reduction risk. The risk score is calculated based on the influencing factors corresponding to each risk, and a preset total risk score range is used to divide the risk into three risk levels: low risk, medium risk and high risk.

5. The data sharing and analysis method based on multi-dimensional correlation according to claim 1, characterized in that, The improved Apriori algorithm described in step three optimizes the frequent itemset mining process through transaction compression and hash pruning, and presets minimum support and minimum confidence thresholds. It transforms the semantically quantized standardized data into transaction items to construct a transaction dataset, generates frequent itemsets based on the transaction dataset, and selects association rules that meet the minimum confidence threshold from the frequent itemsets. The rules are then output in the form of natural language combined with logical expressions according to confidence priority.

6. The data sharing and analysis method based on multi-dimensional correlation according to claim 4, characterized in that, Step four describes an inventory adjustment plan that includes determining the safety stock baseline and generating inventory operation recommendations. Safety stock benchmark value determination: Based on the weighted characteristics of historical sales, seasonal fluctuation patterns and future demand growth trends, a safety stock benchmark value is determined, and calculation logic adapted to the seasonal characteristics is configured for different categories of bulk commodities. Inventory operation suggestion generation: Based on the market supply and demand status assessment, set the safety inventory threshold for the corresponding supply and demand status, clarify the operation direction of replenishment or clearance, and calculate the replenishment or clearance batch based on the proportional factor corresponding to the supply and demand gap coefficient.

7. The data sharing and analysis method based on multi-dimensional correlation according to claim 4, characterized in that, Step four describes a futures hedging strategy that includes calculating the hedging ratio, determining the timing of operations, and outputting strategy parameters. Hedging ratio calculation: Based on the coefficient corresponding to the risk level, the relative strength coefficient of price fluctuation, and the preset adjustment coefficient, the hedging ratio adapted to the current market is calculated. Determining the timing of operations: Based on the market supply and demand situation, the predicted price trend and the hedging ratio, set the trigger conditions for long and short hedging respectively. Strategy parameter output: Outputs the core parameters required for futures trading. These core parameters include at least the futures contract identifier, position direction, entry price range, stop-loss point, and take-profit point.

8. A data sharing and analysis system based on multi-dimensional correlation, used to implement the data sharing and analysis method based on multi-dimensional correlation as described in any one of claims 1-7, characterized in that, The system includes: Multi-source heterogeneous data acquisition module: used for a hybrid acquisition mode of interface synchronization and manual reporting, integrating manually reported unstructured data, internal enterprise structured data and futures business data; Semantic quantization module: used to preprocess the collected data, quantizing unstructured text data into standardized indices and levels through a bidirectional LSTM semantic parsing algorithm; Multi-dimensional correlation analysis and prediction module: It is used to mine linear and non-linear correlations in data using a hybrid model, and combines an improved Apriori algorithm to mine correlation rules. It achieves market prediction through three sub-models. Inventory and Futures Linked Decision Output Module: Based on market forecasts, this module outputs inventory adjustment plans and futures hedging strategies to establish a closed-loop decision-making mechanism.

Images