Intelligent branch prediction-based correlation analysis method and device, equipment and medium

Abstract

The application provides a correlation analysis method and device based on intelligent branch prediction, equipment and medium, including: constructing a historical weighted transaction set; each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time and space dimensions; obtaining future time prediction data, determining a prediction confidence decay coefficient based on the prediction step and model historical accuracy of the future time prediction data; encapsulating a virtual risk item and the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight; fusing the virtual transaction with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information; and performing association rule mining based on the weighted transaction database to output an association rule containing early warning information. In this way, the prediction data is seamlessly integrated into the mining process by generating a virtual transaction with confidence decay, and the future risk early warning improves the early warning accuracy.

Description

Technical Field

[0001] This invention relates to the field of association analysis technology, and in particular to an association analysis method, apparatus, device and medium based on intelligent branch prediction. Background Technology

[0002] Association rule mining is widely used in environmental risk analysis. Traditional methods, which only mine co-occurrence patterns from historical data, cannot provide forward-looking warnings of future risks. Related technologies, such as the FP-Growth algorithm, can only describe what has already happened and lack the ability to integrate predictive data. Their drawback is that they cannot utilize the model's output of future weather and pollution diffusion predictions, leading to delayed warnings and failing to meet the timeliness requirements of environmental emergency response. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method, apparatus, device and medium for association analysis based on intelligent branch prediction, which seamlessly integrates prediction data into the mining process by generating virtual transactions with confidence decay, thereby realizing future risk warning and improving the accuracy of warning.

[0004] In a first aspect, embodiments of the present invention provide a correlation analysis method based on intelligent branch prediction. The method includes: acquiring historical monitoring data and preprocessing the historical monitoring data to construct a historical weighted transaction set; each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time, and spatial dimensions; acquiring future time-based prediction data output by a prediction model, and determining a prediction confidence decay coefficient based on the prediction step size of the future time-based prediction data and the model's historical accuracy; mapping the future time-based prediction data to virtual risk items and encapsulating them with the prediction confidence decay coefficient to generate virtual transactions with second transaction weights; the second transaction weight is less than the first transaction weight; merging the virtual transactions with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information; and performing correlation rule mining based on the weighted transaction database to output correlation rules containing early warning information.

[0005] In a preferred embodiment of the present invention, the above-mentioned preprocessing of historical monitoring data to construct a historical weighted transaction set includes: spatiotemporally aligning the historical monitoring data to construct standardized transactions based on driving factors and response factors; dynamically thresholding the monitoring values ​​of continuous indicators in the standardized transactions and mapping them to discrete risk level identifiers and intensity factors; determining the comprehensive weight of the transactions based on time decay factors, spatial distance factors, and intensity factors, as the first transaction weight; and encapsulating the risk level identifiers and the first transaction weights to generate a historical weighted transaction set.

[0006] In a preferred embodiment of the present invention, the above-mentioned dynamic threshold determination of the monitoring values ​​of continuous indicators in standardized transactions includes: determining a statistical threshold based on statistical distribution characteristics and determining a regulatory threshold based on environmental regulatory standards; for positive risk indicators, taking the minimum value between the statistical threshold and the regulatory threshold; for negative risk indicators, taking the maximum value between the statistical threshold and the regulatory threshold; setting a three-level risk level boundary point based on the positive risk indicators and the negative risk indicators; and mapping the monitoring values ​​of continuous indicators to multiple discrete risk levels.

[0007] In a preferred embodiment of the present invention, the determination of the prediction confidence decay coefficient based on the prediction step size and the historical accuracy of the model based on the prediction data at future time points includes: obtaining the basic confidence of the prediction model; the basic confidence is the theoretical maximum confidence of the prediction model at zero prediction step size; calculating the step size decay confidence using an exponential decay model based on the prediction step size; multiplying the step size decay confidence by the historical accuracy of the model to obtain the prediction confidence decay coefficient; and using the prediction confidence decay coefficient as a decay factor in the second transaction weight of the virtual transaction.

[0008] In a preferred embodiment of the present invention, the determination of the prediction confidence decay coefficient further includes: obtaining historical measured data corresponding to the prediction data and calculating the systematic deviation of the prediction data; based on the systematic deviation, using a sign function to perform reverse dynamic correction on the value of the prediction data at future times to obtain the corrected prediction value; calculating the corrected residual and updating the prediction confidence decay coefficient a second time based on the residual to obtain the final decay coefficient.

[0009] In a preferred embodiment of the present invention, the above-mentioned association rule mining based on a weighted transaction database to output association rules containing early warning information includes: scanning the weighted transaction database, calculating the weighted support of each item, and constructing an item header table based on the weighted support; scanning the weighted transaction database a second time, rearranging the items in each transaction based on the priority order of the item header table; inserting the rearranged transactions into a frequent pattern tree, and accumulating the count value of the node during the insertion process by adding the first transaction weight or the second transaction weight of the corresponding transaction to construct a weighted frequent pattern tree; recursively mining frequent itemsets from the weighted frequent pattern tree and generating association rules.

[0010] In a preferred embodiment of the present invention, the method further includes an incremental update step: when new monitoring data flows in, the data growth rate is calculated; if the data growth rate is less than a preset growth rate threshold, the new monitoring data is converted into a new transaction and directly inserted into the existing frequent pattern tree; if the data growth rate is greater than or equal to the growth rate threshold or the number of tree nodes exceeds a preset threshold, the weighted frequent pattern tree is reconstructed in a background thread based on all historical data and new monitoring data, and switched by atomic pointers.

[0011] Secondly, embodiments of the present invention also provide a correlation analysis device based on intelligent branch prediction. The device includes: a historical weighted transaction set construction module, used to acquire historical monitoring data and preprocess the historical monitoring data to construct a historical weighted transaction set; each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time, and spatial dimensions; a prediction confidence decay coefficient determination module, used to acquire future time prediction data output by the prediction model, and determine the prediction confidence decay coefficient based on the prediction step size of the future time prediction data and the historical accuracy of the model; a virtual transaction generation module, used to map the future time prediction data into virtual risk items and encapsulate them with the prediction confidence decay coefficient to generate virtual transactions with a second transaction weight; the second transaction weight is less than the first transaction weight; a weighted transaction database construction module, used to merge the virtual transactions with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information; and a correlation rule output module, used to perform correlation rule mining based on the weighted transaction database and output correlation rules containing early warning information.

[0012] Thirdly, embodiments of the present invention also provide an electronic device, including a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the association analysis method based on intelligent branch prediction described in the first aspect above.

[0013] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are invoked and executed by a processor, the computer-executable instructions cause the processor to implement the association analysis method based on intelligent branch prediction described in the first aspect above.

[0014] The embodiments of the present invention bring the following beneficial effects: This invention provides a method, apparatus, device, and medium for association analysis based on intelligent branch prediction. It acquires historical monitoring data and preprocesses it to construct a historical weighted transaction set. Each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time, and spatial dimensions. Future time-based prediction data output by the prediction model is acquired. Based on the prediction step size of the future time-based prediction data and the model's historical accuracy, a prediction confidence decay coefficient is determined. The future time-based prediction data is mapped to a virtual risk item and encapsulated with the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight, which is less than the first transaction weight. The virtual transaction is then merged with the historical weighted transaction set to construct a weighted transaction database containing historical and future prediction information. Association rule mining is performed based on the weighted transaction database to output association rules containing early warning information. In this method, by generating virtual transactions with confidence decay, the prediction data is seamlessly integrated into the mining process, achieving improved early warning accuracy for future risks.

[0015] Other features and advantages of this disclosure will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above.

[0016] To make the above-mentioned objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an association analysis method based on intelligent branch prediction provided in an embodiment of the present invention; Figure 2 A flowchart of another association analysis method based on intelligent branch prediction provided in an embodiment of the present invention; Figure 3 A flowchart illustrating another association analysis method based on intelligent branch prediction provided in an embodiment of the present invention; Figure 4 A schematic diagram of the structure of an association analysis device based on intelligent branch prediction provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Association rule mining is widely used in environmental risk analysis. Traditional methods, which only mine co-occurrence patterns from historical data, cannot provide forward-looking warnings of future risks. Related technologies, such as the FP-Growth algorithm, can only describe what has already happened and lack the ability to integrate predictive data. Their drawback is that they cannot utilize the model's output of future weather and pollution diffusion predictions, leading to delayed warnings and failing to meet the timeliness requirements of environmental emergency response.

[0021] Based on this, the present invention provides a method, apparatus, device, and medium for association analysis based on intelligent branch prediction. This method acquires historical monitoring data, preprocesses the data to construct a historical weighted transaction set, where each transaction has a first transaction weight determined based on risk intensity, time, and spatial dimensions. It then acquires future-time prediction data output by a prediction model, determines a prediction confidence decay coefficient based on the prediction step size and historical model accuracy of the future-time prediction data, maps the future-time prediction data to virtual risk items, and encapsulates these virtual items with the prediction confidence decay coefficient to generate virtual transactions with a second transaction weight (less than the first). These virtual transactions are then merged with the historical weighted transaction set to construct a weighted transaction database containing both historical and future prediction information. Association rule mining is performed based on this database to output association rules containing early warning information. In this approach, by generating virtual transactions with confidence decay, the prediction data is seamlessly integrated into the mining process, improving the accuracy of future risk warnings.

[0022] To facilitate understanding of this embodiment, a detailed description of an association analysis method based on intelligent branch prediction disclosed in this embodiment of the invention will be provided first.

[0023] Example 1 This invention provides a correlation analysis method based on intelligent branch prediction. Figure 1 This is a flowchart illustrating an association analysis method based on intelligent branch prediction, provided as an embodiment of the present invention. Figure 1 As shown, this association analysis method based on intelligent branch prediction may include the following steps: Step S101: Obtain historical monitoring data, preprocess the historical monitoring data, and construct a historical weighted transaction set.

[0024] Each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time, and spatial dimensions.

[0025] The historical monitoring data covers multiple sources, including meteorological, hydrological, sewage discharge, and ecological monitoring.

[0026] Preprocessing can include data cleaning, spatiotemporal alignment, and numerical standardization.

[0027] The historical weighted transaction set consists of a collection of multiple weighted transactions, each representing an environmental risk event or scenario and having a weight.

[0028] The first transaction weight is a comprehensive weight calculated based on three dimensions: risk intensity, time, and space, reflecting the importance of the transaction in the mining process.

[0029] Step S102: Obtain the future time prediction data output by the prediction model, and determine the prediction confidence decay coefficient based on the prediction step size and the historical accuracy of the model.

[0030] Among them, the prediction model is a numerical simulation model based on physical equations, used to predict the future state of the environment.

[0031] Among them, the future time prediction data are the predicted values ​​of environmental parameters for future time periods (such as the next 24 hours) output by the model.

[0032] The prediction step size is the time interval from the current time to the prediction target time.

[0033] Among them, the historical accuracy of the model is the average accuracy calculated by comparing the model's past predictions with the actual values.

[0034] Among them, the prediction confidence decay coefficient is used to reduce the weight of virtual transactions.

[0035] Step S103: Map the predicted data for future time moments to virtual risk items and encapsulate them with the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight.

[0036] In this process, the weight of the second transaction is less than that of the first transaction. This ensures that the influence of predicted data on data mining is lower than that of actual historical data.

[0037] Among them, the virtual risk item is an item identifier with a special prefix generated after discretizing continuous prediction data.

[0038] Among them, the second transaction weight is the weight of the virtual transaction.

[0039] Step S104: Merge the virtual transactions with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information.

[0040] The weighted transaction database is a database that stores all transactions (including historical and virtual transactions) and their weights.

[0041] Step S105: Perform association rule mining based on the weighted transaction database and output association rules containing early warning information.

[0042] In this embodiment, a weighted FP-Growth algorithm is used to construct the tree structure. The core of this approach lies in employing the aforementioned weighted counting mechanism, introducing a comprehensive transaction weight to correct the tree node count. Input: Transaction sequence and their corresponding comprehensive weights Process: When inserting nodes along the path, the node count is no longer simply incremented, but rather the weight of the transaction is accumulated, as shown in the following formula (14). To prevent any changes in the order of the instructions, the order of the formula labels and table numbers in the entire text was adjusted after the final draft was completed: (14) This mechanism preserves both the categorical characteristics of risk indicators (identified by node IDs) and the timeliness and intensity characteristics of risks at the statistical level (retained by floating-point values ​​of node counts).

[0043] Specifically, association rule mining based on a weighted transaction database, outputting association rules containing early warning information, may include: scanning the weighted transaction database, calculating the weighted support of each item, and constructing an item header table based on the weighted support; scanning the weighted transaction database a second time, rearranging the items in each transaction based on the priority order of the item header table; inserting the rearranged transactions into a frequent pattern tree, accumulating the node count value with the first or second transaction weight of the corresponding transaction during the insertion process to construct a weighted frequent pattern tree; recursively mining frequent itemsets from the weighted frequent pattern tree and generating association rules.

[0044] Among them, for any risk indicator Its weighted support is defined as the sum of the weights of all transactions that include the indicator, as shown in the following formula (15): (15) in, This represents the total number of transactions in the transaction database. Indicates the first Transaction record; For logical indicator functions, when the condition Establishment (i.e., indicator) Existing in transactions The value is 1 when the condition is met, and 0 otherwise. For the first The overall weight of each transaction (i.e., the weight calculated above) The relative weighted support is then normalized, as shown in the following formula (16): (16) The resulting logical consistency mechanism ensures that the screening of frequent items takes into account both the frequency of occurrence and the risk contribution. For example, a single extremely high-risk event ( In terms of statistical power, it is equivalent to 5 low-risk events. This effectively avoids the omission of high-risk sparse patterns.

[0045] Based on this, a dynamic threshold strategy is adopted when constructing the item header table. The theoretical basis is that the inherent sparsity of environmental risk events requires lowering the support threshold to retain key patterns. Accordingly, this framework sets thresholds based on the characteristics of the medium: 0.05 for aquatic environment (high mobility leads to pattern dispersion), 0.03 for atmospheric environment (high volatility), 0.04 for ecological environment (moderate complexity), and 0.08 for soil environment (low mobility).

[0046] DRIVER_RAIN_ACC_LEVEL4 (weighted support 0.30) is preferentially selected as the root node, while RC_SOIL_PH_LEVEL1 (0.06) is located as a leaf node. The node linked list structure (e.g., Node1→Node4→Node7) explicitly maintains the topological association of high-frequency risk patterns. Finally, through a global support descending order sorting mechanism, high-frequency basic items (such as season and geographical basis) are placed at the front of the transaction sequence.

[0047] Table 2 below shows an example of the weighted support statistics for environmental risk indicators and the construction of the FP tree header table: Table 2:

[0048] Among them, the system calculates the global weighted support based on the first scan ( The item header table is constructed in descending order, establishing the global priority of all risk indicators. Before inserting any transaction (such as T={RC_FISH_DO_LEVEL4,DRIVER_RAIN_ACC_LEVEL4,TIME_RECENT}) into the FP-tree, the items within the transaction must be rearranged according to the priority order of the item header table.

[0049] Assume the priority is: If TIME_RECENT > DRIVER_RAIN_ACC_LEVEL4 > RC_FISH_DO_LEVEL4, then the transaction is reorganized into an ordered sequence. T'=[TIME_RECENT,DRIVER_RAIN_ACC_LEVEL4,RC_FISH_DO_LEVEL4].

[0050] This step is crucial: it ensures that high-frequency items (such as spatiotemporal labels and primary driving factors) are always located at the root or shallow level of the tree, thus forcing different transactions to share the same path prefix.

[0051] Among them, the FP-tree construction process dynamically achieves three optimizations based on the rearranged ordered sequence: Path sharing compression: When inserting an ordered sequence T', the algorithm prioritizes searching for the shared prefix starting from the root node. If the path Root→TIME_RECENT→DRIVER_RAIN_ACC_LEVEL4 already exists in the tree, the system does not create a new node, but directly applies the transaction's comprehensive weight. The increments are added to the counter of each node on the existing path (i.e. A new child node is created only when the path branches at a point (such as RC_FISH_DO_LEVEL4). This mechanism compresses massive amounts of redundant data into a compact tree topology.

[0052] Node-linked list acceleration: The item header table maintains linked list pointers to items with the same risk level, reducing the query complexity of a specific risk item's conditional pattern base query from... Down to ( (where the length is the linked list length), significantly improving the efficiency of high-frequency traversal.

[0053] Furthermore, the association rules mined in the embodiments of this application strictly follow the standard paradigm of driver factor set → receptor response, which logically clearly defines the causal chain between environmental drivers and responses. Specifically, the antecedent of a rule is characterized as a combination of environmental driving factors, such as the synergistic effect of upstream continuous cumulative rainfall exceeding 200mm (DRIVER_RAIN_ACC_LEVEL4) and temperature above 30℃ (DRIVER_TEMP_ACC_LEVEL3). In the physical structure of the FP tree, such "parallel combination relationships" are not presented as parallel directed arrows, but are mapped as a vertical sequence path of parent node → child node (e.g., Root → spatiotemporal context → rainfall node → high temperature node).

[0054] Furthermore, node positions are optimized using a descending support ranking strategy: frequently occurring common driving factors are typically located in internal nodes near the root, thus playing a core role in path guidance. Based on this, the consequent of a rule corresponds to the receptor response state, such as extremely low dissolved oxygen at the cross-section (RC_FISH_DO_LEVEL4). In the FP-tree, such responses are all located at the end of the aforementioned driving path (i.e., leaf nodes or deep nodes), serving as the ultimate environmental result led by that path. It is important to note that NodeLinks in this structure only act as a horizontal indexing mechanism, connecting nodes with the same name in different branches to accelerate algorithm computation; they themselves do not carry any causal logical meaning. In summary, this embodiment, through the above structured design, ensures that all mining rules conform to the rigorous academic norms of driving factor set → receptor response in both form and content.

[0055] Furthermore, to comprehensively quantify rule quality, this application's embodiments employ the following four dimensions for evaluation. It should be specifically noted that, to maintain consistency in the algorithmic logic, all the following metric calculations are based on the weighted count (Sum of Weights) defined in PHASE III, rather than traditional frequency counting: Weighted Support: Measures the prevalence (i.e., frequency) of a rule. It reflects the proportion of a specific environmental risk pattern in the historical dataset, as shown in the following formula (18). (18) The numerator is the sum of the weights of all transactions that simultaneously include the predecessor X and the successor Y, and the denominator is the sum of the weights of all transactions in the database.

[0056] Weighted confidence reflects the reliability of a rule (i.e., conditional probability). It is characterized by the driving factors. Under the condition that the (antecedent) occurs, the receptor response The probability of the (subsequent) event occurring.

[0057] (19) Weighted Lift: Reflects the relevance of a rule. Used to determine whether a rule is better than random guessing. When... Time indicates and A positive correlation is observed, with larger values ​​indicating a stronger causal relationship; if This indicates that the two are independent (i.e., pseudo-association) and should be eliminated, as shown in the following formula (20): (20) Weighted Conviction: As a quality auxiliary verification metric, it measures the independence between the antecedent and consequent of a rule, and is used to exclude rules that are ruled out due to the consequent itself occurring with an extremely high frequency (i.e., The spurious strong association caused by the large number of large numbers is shown in the following formula (21): (twenty one) Furthermore, to overcome the inherent limitations of single-indicator assessment in environmental risk analysis, this application constructs a multi-dimensional rule scoring system based on the Analytic Hierarchy Process (AHP). This system scientifically weights and integrates four core indicators—support, confidence, lift, and certainty—to generate a comprehensive score, thereby achieving a systematic and quantitative assessment of the quality of associated rules.

[0058] Within this framework, due to the range of values ​​for lift and confidence (Conv) ) and support / confidence ( Inconsistency can lead to scaling bias if directly weighted. Therefore, in this embodiment, before substituting the two indicators into the scoring formula, the arctangent function is used to normalize and map them, compressing them to... Interval, define normalization operator The following formulas (22) and (23) are: (twenty two) (twenty three) Weight vector The settings are strictly based on the physical meaning and priority of each indicator in environmental decision-making: among which, the support weight ( =0.539) had the highest proportion, aiming to ensure that the rules were statistically significant and effectively exclude spurious associations caused by random perturbations; confidence weight ( =0.297) Secondly, the focus is on ensuring the accuracy of the rule's prediction of the receptor response; the degree weight ( =0.119) is used to identify and remove invalid strong associations (i.e., independent events with a Lift value close to 1); confidence weight ( =0.045) is used as an auxiliary verification indicator to further strengthen the evaluation of rule independence. Based on this weight allocation, any rule The comprehensive score calculation formula can be expressed as the following formula (24): (twenty four) To ensure the reliability and adaptability of the rules, the system first sets a global minimum weighted confidence benchmark. As a basic screening criterion, and considering the significant differences in risk sensitivity and data characteristics of different environmental media (such as the high immediate responsiveness of the water environment to pollution events, while the soil environment has a cumulative lag effect), this system introduces a differentiated confidence threshold mechanism, as shown in Table 3 below. Table 3 is the differentiated confidence threshold table.

[0059] For example, in water environment risk management, given the high sensitivity of drinking water safety, the confidence threshold is strictly set at 0.70; while in the soil environment, due to the cumulative effect of pollution, a certain buffer space is allowed, and the threshold is correspondingly relaxed to 0.55. To intuitively illustrate the operation process of this mechanism, let's take water environment risk mining as an example: Assume that in the weighted transaction set, the total weight contribution of "upstream pollution exceeding standards" (the antecedent, corresponding to DRIVER_DISCHARGE_LEVEL3) is... The sum of the transaction weights that co-occur with "DO (Dissolved Oxygen) Anomaly" (the consequent, corresponding to RC_FISH_DO_LEVEL4) is: The weighted confidence level is then calculated as follows: Because this value reaches the water environment threshold ( The system determines that this rule is a valid strong rule and retains it.

[0060] In summary, this scoring system, through hierarchical weight design and contextualized threshold adaptation, not only overcomes the one-sidedness of single-indicator evaluation, but also realizes a logical closed loop from statistical verification to environmental decision-making, providing a scientific basis for risk warning of complex environmental systems.

[0061] Table 3:

[0062] Furthermore, the method may also include an incremental update step: when new monitoring data flows in, the data growth rate is calculated; if the data growth rate is less than a preset growth rate threshold, the new monitoring data is converted into a new transaction and directly inserted into the existing frequent pattern tree; if the data growth rate is greater than or equal to the growth rate threshold or the number of tree nodes exceeds a preset threshold, the weighted frequent pattern tree is reconstructed in a background thread based on all historical data and new monitoring data, and switched via atomic pointers.

[0063] In response to the continuously growing nature of environmental monitoring data, this application proposes an incremental update framework based on a dynamic growth rate threshold to overcome the efficiency bottleneck of traditional full recalculation. Its core lies in first defining a data growth rate indicator. To quantify the scale of the new data, see formula (17): (17) in, For the amount of new transactions within the update cycle, This represents the total number of historical transactions. Based on this, the system abandons the static update mode and instead adopts a dual-track adaptive strategy: when... (In daily real-time monitoring scenarios), activate the append-based lightweight update mechanism. This mechanism directly embeds new transactions into the existing FP-tree: if the new path (e.g., A→B→C) matches the existing structural prefix, the transaction weight of the corresponding node is accumulated. Create a new branch only at the fork point. Conversely, when (Quarterly archive / history import) or the number of tree nodes exceeds the preset balance threshold When the system triggers a full, refactoring update mechanism, it merges all historical and new data in a separate background thread, recalculates weighted support, and optimizes the item header table sorting. Once the new tree is built, seamless replacement is achieved through atomic pointer switching, instantly releasing resources from the old structure. The key value of this design lies in avoiding tree imbalance caused by long-term appending while maintaining globally optimal mining efficiency. A 20% threshold is used as a baseline reference value, and in practical applications, it is adaptively adjusted according to the environmental medium type.

[0064] The association analysis method based on intelligent branch prediction provided in this invention can acquire historical monitoring data, preprocess the historical monitoring data to construct a historical weighted transaction set, where each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time, and spatial dimensions, acquire future time-based prediction data output by the prediction model, determine the prediction confidence decay coefficient based on the prediction step size of the future time-based prediction data and the model's historical accuracy, map the future time-based prediction data to a virtual risk item, and encapsulate it with the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight, which is less than the first transaction weight. The virtual transaction is then merged with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information. Association rule mining is performed based on the weighted transaction database to output association rules containing early warning information. In this method, by generating virtual transactions with confidence decay, the prediction data is seamlessly integrated into the mining process, achieving future risk early warning and improving the accuracy of early warning.

[0065] Example 2 This invention also provides another correlation analysis method based on intelligent branch prediction; this method is implemented on the basis of the method in the above embodiments; the method focuses on describing the specific implementation of preprocessing historical monitoring data and constructing a historical weighted transaction set.

[0066] Figure 2 A flowchart of another association analysis method based on intelligent branch prediction provided in an embodiment of the present invention is shown below. Figure 2As shown, the preprocessing of historical monitoring data to construct a historical weighted transaction set may include the following steps: Step S201: Perform spatiotemporal alignment on historical monitoring data to construct standardized transactions based on driving factors and response factors.

[0067] Before performing spatiotemporal alignment, a rationality review can be conducted based on professional knowledge to identify and remove abnormal records that do not conform to physical and chemical laws. For missing values, different interpolation strategies are adopted according to data characteristics: cubic spline interpolation is used to process smooth time series data, and kriging interpolation is used to process missing spatial dimensions.

[0068] To eliminate discrepancies in sampling frequencies from different monitoring sources, a minimum common time unit can be used as a benchmark to align the data with a unified time granularity. For high-frequency monitoring data, such as online monitoring, mean pooling or max pooling strategies are used to aggregate it into a standard time granularity, such as hourly. For low-frequency data, such as manually sampled data, interpolation algorithms are used to fill in the gaps. Simultaneously, by establishing a unique monitoring point identifier (SiteID) and a standardized timestamp, it is ensured that each transaction has an equal statistical benchmark in both spatiotemporal dimensions.

[0069] Furthermore, in order to eliminate the influence of dimensions and meet the requirement of non-negativity of data for subsequent entropy weight method calculation, the spatiotemporally aligned multi-source data is subjected to Min-Max normalization, as shown in the following formula (1): (1) in, For the normalized data, The original data, and These are the minimum and maximum values ​​of the indicator within the statistical period, respectively. It should be noted that, in order to maintain the intuitiveness of the physical semantics of the environmental standard, the subsequent dynamic threshold determination (Formulas 2 and 3) and "risk level mapping" (Formula 4) are still based on the raw physical quantity data, and the normalized data here is not used.

[0070] In order to fully explore the driving role of meteorological and hydrological conditions on receptor response and capture causal relationships without explicitly modeling complex transport paths, a sliding window technique and a statistically based spatiotemporal backtracking mechanism can be introduced to construct composite features: Extracting statistical features using the sliding window technique can include: Cumulative characteristics: such as cumulative rainfall in the past 24 hours and duration of continuous rainfall, are used to capture the cumulative effect of environmental stress.

[0071] Statistical features, such as daily maximum temperature and daily temperature range, are used to capture the impact of extreme weather conditions.

[0072] To address the complexity of environmental risk transmission, a two-level mapping mechanism combining spatial topological constraints and temporal cross-correlation optimization can be constructed, achieving precise end-to-end association between the source set and the receptor set. The first step is spatial mapping based on physical topology: The system pre-configures a physical topology matrix based on GIS geographic information system and hydrodynamic / gas field data. This matrix defines the unidirectional influence relationships between monitoring points. Let the set of points be... Matrix elements If the point lie in If the upstream catchment area or the upwind direction of the prevailing wind direction is within the effective influence threshold, then it is marked. ,Will Included The set of candidate driving sources. Conversely, if the physical path is unreachable, such as if located in different watersheds or downstream, then Through this step, the system uses domain knowledge to filter out more than 90% of invalid combinations before data mining, such as using downstream data to explain upstream anomalies, ensuring the physical compliance of the causal direction.

[0073] The second step is time lag optimization based on cross-correlation analysis: Within a defined set of candidate driving sources, for the upstream-to-downstream time lag problem, pure data-driven cross-correlation analysis can replace the traditional estimation method based on physical flow velocity, automatically finding the optimal lag amount. Lag Calculation: For any candidate driving factor (e.g., the discharge sequence of an upstream chemical plant), and receptor response. Cross-correlation analysis was performed on historical time series data (such as downstream cross-sectional water quality series). The cross-correlation function was calculated. The system searches within a preset time window (e.g., 0 to 72 hours) and retrieves... When the maximum value is reached, the corresponding Optimal transport lag time as the driving factor .

[0074] Feature backtracking and virtual transaction construction: based on the moment when the receptor anomaly occurs Using the reference anchor point, the calculated Targeted backtracking extraction The upstream driving data at any given moment. Through this end-to-end mapping mechanism that determines direction based on spatial topology and step size based on temporal correlation, the system reorganizes the originally discrete and asynchronous multi-point monitoring data into logically synchronous virtual transactions.

[0075] For example: {Upstream heavy rain (T-4h) → Chemical plant wastewater discharge (T-2h)} → {Downstream fish mortality (T)}. This method avoids complex physical modeling errors, greatly simplifies model dimensions, and improves the efficiency of association rule mining.

[0076] Step S202: Dynamically determine the monitoring values ​​of continuous indicators in standardized transactions and map them to discrete risk level identifiers and intensity factors.

[0077] Specifically, dynamic threshold determination for monitoring values ​​of continuous indicators in standardized transactions can include: determining statistical thresholds based on statistical distribution characteristics and determining regulatory thresholds based on environmental regulations and standards; for positive risk indicators, taking the minimum value between the statistical threshold and the regulatory threshold; for negative risk indicators, taking the maximum value between the statistical threshold and the regulatory threshold; setting three-level risk level boundary points based on positive and negative risk indicators; and mapping the monitoring values ​​of continuous indicators to multiple discrete risk levels.

[0078] It should be noted that, in order to maintain the intuitiveness of the physical semantics of environmental standards, the dynamic threshold determination is still based on the original physical quantity data, and the normalized data mentioned above is not used.

[0079] Among them, by calculating statistical thresholds and regulatory thresholds The strictest value of both will be used as the final risk assessment benchmark. .

[0080] Statistical threshold Calculated based on a moving time window, such as raw monitoring data from the past 3 months. Where μ is the moving average, σ is the standard deviation, and k is the statistical multiple, typically k=3.0. Regulatory threshold. Based on the national / local standard limit (S) corresponding to the environmental functional zoning of the monitoring point, and introducing the regional safety factor η.

[0081] This application's embodiments can automatically identify the risk direction of indicators and calculate the final risk assessment benchmark. : Positive risk indicators, the higher the value, the worse, such as COD and PM2.5: adopt the strict principle and take the minimum value between the statistical threshold and the regulatory threshold, as shown in the following formula (2): (2) If the original monitoring value x> If so, it is considered a risk.

[0082] The lower the value of the adverse risk indicator, the worse it is. For example, dissolved oxygen (DO): adopt the strict principle and take the maximum value of the statistical threshold and the regulatory threshold, as shown in the following formula (3): (3) If the original monitoring value x < If so, it is considered a risk.

[0083] This mechanism ensures dual security: in areas with high environmental background values, regulatory thresholds serve as mandatory constraints; in areas with excellent environmental quality, statistical thresholds can sensitively detect early signals of water quality deterioration (or abnormal decline in dissolved oxygen). The embodiments of this application, through the aforementioned adaptive logic, transform complex and ever-changing environmental monitoring data into a structured set of risk transactions.

[0084] Among them, the setting of the three-level risk level dividing point: basic threshold (Risk Initiation Line), Early Warning Threshold (Warning line) and high-risk threshold (High-risk line), thereby constructing a continuous hierarchical spectrum covering no risk to extremely high risk. Based on this, the mapping function... Data transformation is achieved through dual-channel processing: input variables The aim is to handle environmental characteristics with varying dimensions through a unified interface. In practical implementation, it can be based on the indicator attributes... The construction can be divided into the following two cases: Direct mapping to instantaneous values ​​of instantaneous physicochemical indicators: For water quality parameters such as COD and ammonia nitrogen, Directly take the physical monitoring value at the current sampling time, such as At this point, the aforementioned basic threshold... Warning threshold and high-risk threshold The corresponding water quality concentration standard limit for this functional area is loaded.

[0085] Statistical mapping of time-series characteristics for cumulative statistical indicators: For environmental factors with cumulative effects, a statistical counting mode can be enabled. Taking the construction of composite characteristics of high-temperature heat accumulation risk as an example, the specific execution process is as follows: Feature definition and window calculation: First, the indicator definition (RuleID:TEMP_ACC_HIGH) is loaded according to the preset rule base, and the sliding window is dynamically configured according to the characteristic action cycle of environmental pressure. Regarding this indicator, considering the temperature rise lag (thermal inertia) caused by the specific heat capacity of water and the tolerance limit of aquatic organisms, the following parameters are set: The criterion is set as a daily maximum temperature > 30℃. The preprocessing module scans the data of the past 7 days and calculates the number of consecutive days that meet the criterion up to the current time as the input variable. Assuming monitoring data shows that the highest temperature has exceeded 30℃ for the past three consecutive days, then x=3, unit: days.

[0086] Threshold system adaptive switching: If the indicator is cumulative, it automatically loads the corresponding time-sensitive threshold set from the database, rather than the concentration threshold. (Settings) Day (the beginning of accumulated risk, i.e., the thermal stress begins to manifest), Day (Severe cumulative warning) Day (extremely high risk, i.e., reaching the tolerance threshold).

[0087] Mapping execution: Substitute x=3 into the following formula (4) for judgment. Since 2<3≤4 (i.e., satisfies the condition) The system determines this record to be of high risk level (Code=3) and assigns an intensity factor I(x) = 0.8, ultimately generating a semantic discrete identifier DRIVER_TEMP_ACC_LEVEL3. This function makes the determination based on the adapted threshold system described above, simultaneously outputting the discrete level code Code and intensity factor. .

[0088] (4) The application logic of the mapping results is clearly layered: First, the system uses a prefix-based semantic naming convention to generate unique identifiers (ItemIDs) to automatically distinguish between driver factors and receptor responses. The specific format follows "[ROLE]_[INDICATOR]_[LEVEL]:".

[0089] [ROLE] (role prefix): DRIVER_ represents historical / real-time driving factor, RC_ represents receptor response, and PRED_ represents virtual driving factor generated by the prediction model (used for branch prediction).

[0090] [INDICATOR] (Indicator Code): A standardized and unique code representing the monitoring parameter (e.g., RAIN represents rainfall, TEMP represents air temperature, and COD represents chemical oxygen demand).

[0091] [LEVEL] (Level suffix): A standardized suffix generated based on the Code value output by the above formula (4), in the format of LEVEL plus the corresponding number (such as LEVEL0 to LEVEL4).

[0092] The parsing logic is as follows: Driver Factor: When the system resolves to the prefix DRIVER_ (e.g., DRIVER_RAIN_ACC_LEVEL4), it automatically categorizes it into the historical / real-time driver predecessor set. .

[0093] Virtual driver factor: When the system resolves to the prefix PRED_ (e.g., PRED_RAIN_ACC_LEVEL4), it automatically classifies it into the predictive driver antecedent set. And apply the confidence decay coefficient.

[0094] Receptor response: When the prefix RC_ is parsed (e.g., RC_COD_LEVEL3), it is automatically classified into the consequent set. .

[0095] Furthermore, intensity factor numerical set Employing a nonlinear heuristic mapping, its design strictly adheres to the response prioritization principle in environmental management practice: low risk (0.2) corresponds to controllable impact, requiring only routine monitoring; medium risk (0.5) corresponds to 1.0–1.5 times the exceedance (i.e., to (Range) requires enhanced monitoring and intervention; high risk (0.8) corresponds to 1.5–2.0 times the limit (i.e., to (range), triggering special governance; extremely high risk (1.0) corresponds to >2.0 times the standard (i.e., greater than) This requires the activation of emergency plans to avoid irreversible ecological damage. Therefore, this mechanism not only highlights the decision-making weight of high-level events through non-linear amplification of risk intensity, but also effectively suppresses low-risk noise interference at the algorithmic level, significantly improving the detection accuracy and explanatory power of key risk patterns in association rule mining.

[0096] Step S203: Determine the comprehensive weight of the transaction based on the time decay factor, spatial distance factor, and intensity factor, and use it as the first transaction weight.

[0097] Addressing the core challenge of unifying the quantification of multi-source heterogeneous environmental data (including historical monitoring data and predictive simulation data) in the construction of frequent-pattern FP-trees, a general transaction weight calculation model is proposed. This model introduces a dynamic confidence decay mechanism to achieve adaptive weighting of the confidence of different data sources. Specifically, for any transaction... Update count weight Defined as follows (5): (5) in, For comprehensive risk scoring, To predict the confidence decay coefficient. Its core mechanism lies in: assigning a certain value to historical / real-time monitoring data (hard data). The full confidence weight, where the transaction weight is determined solely by the risk score. The decision is to ensure that highly reliable data dominates pattern mining; for intelligent branch prediction data (soft data), the following settings are made. The dynamic decay range is determined by a coefficient based on the quantitative assessment of the uncertainty of the prediction model, ensuring that the weight of the predicted data is strictly lower than that of the measured data (1.0). As a result, low-confidence prediction results will automatically reduce their transaction weight through the multiplier effect, thereby significantly weakening their contribution to the support count during the FP-tree construction process.

[0098] The theoretical value of this design lies in its ability to retain the supplementary value of predictive data for risk forecasting through a confidence-weight coupling mechanism, while effectively avoiding the interference of uncertain data on association rule mining. (Predictive confidence coefficients of historical / real-time monitoring data) The value is always 1.0.

[0099] Among them, dynamic coding can be carried out by comprehensively considering spatiotemporal characteristics and risk intensity, as shown in the following formula (6): (6) Among them, weight parameters Based on the dynamic determination of entropy weight, it can adaptively adjust according to historical risk data, ensuring that the coding results are more objective and accurate.

[0100] Time decay factor: This reflects the timeliness of the data. This is the attenuation coefficient. For example, sudden risk... (Half-life 7 days), ongoing risk (Half-life 23 days).

[0101] Spatial distance factor: This reflects the distance attenuation effect. The impact thresholds are set (500m for aquatic environment and 2000m for atmospheric environment). The attenuation index is 1.5-2.0 for aquatic environments and 0.5-1.0 for atmospheric environments.

[0102] Intensity factor: A non-linear mapping based on risk level. (Using...) The numerical set reflects the response priorities for different risk levels.

[0103] Table 1 below is a risk indicator discretization and intensity mapping table (based on a unified five-level classification system).

[0104] Table 1:

[0105] Note: In the architectural design of this application, the functional boundaries of topology construction and weight calculation are strictly distinguished. The discrete identifier (ItemID) is specifically used to construct the topology node structure of the FP-tree; while the strength factor... It only participates in the calculation of the comprehensive risk score in the above formula (6), thereby affecting the transaction weight in the above formula (5). This separation design not only ensures the simplicity of the FP tree structure, but also ensures that the risk intensity information can be accurately quantified in the mining process.

[0106] To address the limitations of single weighting methods in simultaneously considering both statistical data patterns and environmental physical characteristics, this application constructs a three-level dynamic correction algorithm for attribute weights to generate a scenario-based weight vector that ultimately participates in risk scoring calculations. Its core design follows the principle of incremental optimization: First, based on the entropy weighting method, the information entropy of historical samples in the three dimensions of time, space, and intensity is calculated to generate a basic objective weight vector. As shown in formula (6) above, this vector exhibits an equilibrium distribution in the absence of prior knowledge. This provides a statistical benchmark for subsequent corrections. However, the basic weights fail to reflect the physical transport characteristics of the environmental medium. To address this limitation, a medium characteristic correction mechanism is introduced in the second stage: a preset environmental medium correction coefficient matrix is ​​used. This implicitly incorporates physical laws such as hydrological diffusion and atmospheric transport into the weight allocation. For example, due to the highly mobile nature of the water environment, the spatial weight coefficient is increased to 1.3. This automatically suppresses interference in non-hydraulic connected areas; the atmospheric environment, due to its strong time sensitivity, has a time weighting coefficient set to 1.4. The system prioritizes capturing meteorologically driven pollution transport paths. Therefore, it can dynamically select driving factors based on physical accessibility. For example, in river monitoring, when a water conservancy project blocks the hydraulic connection between the upstream pollution source and the receptor, its spatial weight will be significantly reduced, ensuring the algorithm focuses on real and effective causal chains. Furthermore, a third-level risk situation dynamic optimization mechanism (…) In real-time data mining, the system responds to changes in risk type: when identifying highly toxic emergencies, the intensity dimension weighting coefficient is automatically increased to 1.5. To avoid high-risk signals being submerged due to distance attenuation, the three-level weights are finally fused through element-wise multiplication, as shown in formula (7) below, and then normalized to generate a scenario-specific weight vector, as shown in formula (8) below. (7) (8) Where ⊙ represents the Hadamard product, which is the element-wise multiplication of vectors; The three components of the vector correspond to the formula (6) above. .

[0107] Step S204: Encapsulate the risk level identifier with the first transaction weight to generate a historical weighted transaction set.

[0108] Before constructing the FP-tree, the spatiotemporally weighted environmental risk data needs to be converted into a standard transaction format. This process requires not only discretizing continuous numerical values ​​but also fusing the causal chain of driving factors and response results through a spatiotemporal alignment mechanism, laying the data foundation for association rule mining. Its core lies in a four-stage collaborative algorithm of discretization-alignment-assembly-sorting. First, based on the threshold system determined by formulas (2) and (3) above, continuous monitoring values ​​are mapped to discrete item identifiers (ItemID), and the time factor T(t) and spatial factor S(s) are transformed into semantic context labels (such as TIME_RECENT, GEO_NEAR), fully preserving the characteristics of risk metadata. On this basis, the system abandons the traditional time-based packaging strategy and instead uses the abnormal time of the receptor. As a spatiotemporal anchor point, the time lag is estimated based on flow velocity / wind speed. Backtracking and crawling The upstream driving state within the window. This process internalizes the physical transport process into data preprocessing logic, accurately preserving cross-scale causal temporality while avoiding the complexity of hydrodynamic models.

[0109] Furthermore, the transaction structure adopts a set-based representation. In order to adapt to the requirements of the weighted FP-Tree algorithm for flattened input and to clearly distinguish between logical itemsets and numerical weights, the embodiment of this application defines the standard transaction as the following binary tuple structure, as shown in equation (9): (9) in, A flattened itemset is a set of driving factors. Receptor response A list of unique identifiers is generated by taking the union of the three spatiotemporal context labels and flattening them.

[0110] For example: {DRIVER_RAIN_ACC_LEVEL4, RC_FISH_DO_LEVEL4, TIME_RECENT}. These labels are mixed together and will be rearranged uniformly based on global support in subsequent steps to construct the path of the FP-tree. Transaction weight: This is a floating-point value calculated using formula (5) or formula (10) above. It is attached to the transaction as an independent attribute and is only used for the weighted accumulation of node counts (i.e., Node.count += ). ), does not participate in the sorting and topology construction of itemsets.

[0111] The system automatically removes redundant intermediate features and assembles the above identifiers into an itemset portion of a standard transaction: ["DRIVER_RAIN_ACC_LEVEL4", "DRIVER_TEMP_ACC_LEVEL3", "DRIVER_DISCHARGE_LEVEL3", "RC_FISH_DO_LEVEL4", "TIME_RECENT"].

[0112] Subsequently, the system performs a sorting operation based on global support in descending order. Since context labels such as TIME_RECENT or GEO_BASIN_A typically cover a large number of records and have extremely high support, they are automatically sorted to the head of the transaction (e.g., TIME_RECENT→DRIVER_...). This sorting mechanism allows the FP-tree to naturally form a topology guided by the "spatiotemporal context" as the root, maximizing prefix sharing rate (i.e., compression efficiency) and supporting fast slice mining along the spatiotemporal dimension.

[0113] Taking a typical risk event as an example: when the dissolved oxygen at the downstream fish breeding station is abnormal (RC_FISH_DO_LEVEL4), the system traces back the upstream 24-hour window and assembles the accumulated rainfall exceeding 200mm (DRIVER_RAIN_ACC_LEVEL4), the high temperature exceeding 30℃ (DRIVER_TEMP_ACC_LEVEL3), and the sewage discharge event (DRIVER_DISCHARGE_LEVEL3) with the time tag (TIME_RECENT) into a standard transaction: ["DRIVER_RAIN_ACC_LEVEL4", "DRIVER_TEMP_ACC_LEVEL3", "DRIVER_DISCHARGE_LEVEL3", "RC_FISH_DO_LEVEL4", "TIME_RECENT"].

[0114] The resulting data-driven implicit attribution mechanism replaces explicit physical modeling with spatiotemporal alignment logic, transforming complex transmission constraints into data preprocessing rules.

[0115] Example 3 This invention also provides another association analysis method based on intelligent branch prediction; this method is implemented on the basis of the method in the above embodiments; this method focuses on describing the specific implementation of determining the prediction confidence decay coefficient based on the prediction step size and the historical accuracy of the model based on the prediction data at future time.

[0116] Figure 3 A flowchart of another association analysis method based on intelligent branch prediction provided in an embodiment of the present invention.

[0117] To ensure the reliability boundary of the predicted data, the system is designed with a dual weighting control mechanism: First, a confidence decay coefficient is introduced for virtual transactions. Its value is dynamically calibrated by the historical accuracy of the physical model; secondly, the transaction weight is defined by the following formula (10). : (10) in, To predict the confidence decay coefficient, unlike formula (5), the virtual transaction... It needs to be dynamically calculated based on the spatiotemporal decay model (see Formula 11 below), while historical / real-time monitoring data The value is always 1.0. This ensures that the contribution of the predicted data to the support in the FP-tree construction is strictly lower than that of the measured data.

[0118] Furthermore, in order to accurately quantify the reliability of virtual transactions, this application embodiment establishes a coupled processing mechanism of initial confidence calculation, numerical deviation correction, and final weight update.

[0119] like Figure 3 As shown, determining the prediction confidence decay coefficient based on the prediction step size and historical model accuracy of future time prediction data can include the following steps: Step S301: Based on the predicted step size, calculate the step size decay confidence using an exponential decay model.

[0120] The system does not consider specific numerical deviations, but only calculates the theoretical confidence decay based on the prediction step size, as shown in the following formula (11): (11) in, This is the baseline confidence level for the model (usually taken as 0.9). To predict the time span, To predict the divergence coefficient. It is important to distinguish that... This describes the process by which the accuracy of a numerical model drops sharply as the prediction step size increases (the "butterfly effect"), and its value is usually much larger than the time decay coefficient of historical data. ; This represents the relative error from the most recent closed-loop verification. The resulting dynamic decay mechanism accurately characterizes the exponential growth of prediction uncertainty over time.

[0121] Step S302: Multiply the step size decay confidence by the model's historical accuracy to obtain the prediction confidence decay coefficient.

[0122] Step S303: The predicted confidence decay coefficient is used as the decay factor in the second transaction weight of the virtual transaction.

[0123] Step S304: Obtain historical measured data corresponding to the predicted data and calculate the systematic deviation of the predicted data.

[0124] Formula (11) above only addresses the timeliness issue and cannot eliminate the inherent systematic biases of the physical model in local environments (e.g., always overestimating pollutant diffusion under specific terrain). Therefore, error classification is required: Calculate historical moments Predicted value Compared with measured values relative error ,when If the model is found to have a significant systematic bias, a correction procedure is triggered.

[0125] Step S305: Based on systematic bias, the numerical values ​​of the predicted data at future time moments are dynamically corrected using a sign function to obtain the corrected predicted values.

[0126] This also requires dynamic numerical correction: adjusting the predicted values ​​themselves. Based on the most recent... Moving average of error over a time window Using symbolic functions Determine the direction of the deviation and the original predicted value. Reverse compensation is performed using the following formula (12): (12) Here, For the sign function: when historical predicted values ​​are too high ( When the value is positive, it is +1, causing the correction term to decrease the predicted value; conversely, it is -1, causing the predicted value to increase. The corrected value is... The original value will be replaced by the discretization encoding of the above formula (4).

[0127] Step S306: Calculate the corrected residuals and update the prediction confidence decay coefficient a second time based on the residuals to obtain the final decay coefficient.

[0128] This requires a confidence level closed-loop update: while numerical correction reduces bias, it introduces new correction uncertainties. Therefore, the system will update the corrected residuals... Feedback is sent to the weighting model to make a final correction to the initial confidence level calculated by formula (11), as shown in formula (13) below: (13) It should be noted that this final attenuation coefficient This is the final coefficient that is substituted into formula (10) to calculate the transaction weight.

[0129] The association analysis method based on intelligent branch prediction provided in this invention constructs a closed-loop strategy of error identification, numerical correction, and weight update: First, the relative error between the predicted and measured values ​​is calculated in real time, and correction is triggered when the error exceeds a preset threshold; second, dynamic numerical correction is performed, using the sign function direction of historical errors to compensate the original predicted values ​​in reverse, eliminating systematic overestimation or underestimation; finally, a closed-loop confidence update is performed, feeding the corrected residuals back to the weight model to further reduce the confidence coefficient of the corrected data. This mechanism ensures that even if the physical model has biases, the virtual transactions input into the FP-Tree still possess statistical reliability, effectively preventing erroneous predictions from misleading the mining results.

[0130] Example 4 Corresponding to the above method embodiments, this invention provides an association analysis device based on intelligent branch prediction. Figure 4 This is a schematic diagram of the structure of an association analysis device based on intelligent branch prediction provided in an embodiment of the present invention, as shown below. Figure 4 As shown, the association analysis device based on intelligent branch prediction may include: The historical weighted transaction set construction module 401 is used to acquire historical monitoring data, preprocess the historical monitoring data, and construct a historical weighted transaction set; each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time and space dimensions.

[0131] The prediction confidence decay coefficient determination module 402 is used to obtain the future time prediction data output by the prediction model, and determine the prediction confidence decay coefficient based on the prediction step size and the historical accuracy of the model.

[0132] The virtual transaction generation module 403 is used to map the predicted data of future time moments into virtual risk items and encapsulate them with the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight; the second transaction weight is less than the first transaction weight.

[0133] The weighted transaction database construction module 404 is used to merge virtual transactions with historical weighted transaction sets to construct a weighted transaction database containing historical information and future prediction information.

[0134] The association rule output module 405 is used to mine association rules based on the weighted transaction database and output association rules containing early warning information.

[0135] The association analysis device based on intelligent branch prediction provided in this invention can acquire historical monitoring data, preprocess the historical monitoring data to construct a historical weighted transaction set, where each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time, and spatial dimensions, acquire future time-based prediction data output by the prediction model, determine the prediction confidence decay coefficient based on the prediction step size of the future time-based prediction data and the model's historical accuracy, map the future time-based prediction data to a virtual risk item, and encapsulate it with the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight, which is less than the first transaction weight. The virtual transaction is then merged with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information. Association rule mining is performed based on the weighted transaction database to output association rules containing early warning information. In this method, by generating virtual transactions with confidence decay, the prediction data is seamlessly integrated into the mining process, achieving future risk early warning and improving the accuracy of early warning.

[0136] In some embodiments, the historical weighted transaction set construction module is also used to perform spatiotemporal alignment of historical monitoring data to construct standardized transactions based on driving factors and response factors; to dynamically determine the monitoring values ​​of continuous indicators in the standardized transactions and map them to discrete risk level identifiers and intensity factors; to determine the comprehensive weight of the transactions based on time decay factors, spatial distance factors, and intensity factors, which serves as the first transaction weight; and to encapsulate the risk level identifiers and the first transaction weights to generate a historical weighted transaction set.

[0137] In some embodiments, the historical weighted transaction set construction module is further configured to determine statistical thresholds based on statistical distribution characteristics and regulatory thresholds based on environmental regulatory standards; for positive risk indicators, the minimum value between the statistical threshold and the regulatory threshold is taken; for negative risk indicators, the maximum value between the statistical threshold and the regulatory threshold is taken; three-level risk level boundary points are set based on positive risk indicators and negative risk indicators; and the monitoring values ​​of continuous indicators are mapped to multiple discrete risk levels.

[0138] In some embodiments, the prediction confidence decay coefficient determination module is further configured to obtain the basic confidence of the prediction model; the basic confidence is the theoretical maximum confidence of the prediction model at zero prediction step size; based on the prediction step size, the step size decay confidence is calculated using an exponential decay model; the step size decay confidence is multiplied by the historical accuracy of the model to obtain the prediction confidence decay coefficient; and the prediction confidence decay coefficient is used as a decay factor in the second transaction weight of the virtual transaction.

[0139] In some embodiments, the prediction confidence decay coefficient determination module is further configured to acquire historical measured data corresponding to the prediction data, calculate the systematic bias of the prediction data; based on the systematic bias, use a sign function to perform reverse dynamic correction on the value of the prediction data at future time to obtain the corrected prediction value; calculate the corrected residual, and update the prediction confidence decay coefficient a second time based on the residual to obtain the final decay coefficient.

[0140] In some embodiments, the association rule output module is further configured to scan the weighted transaction database, calculate the weighted support of each item, and construct an item header table based on the weighted support; scan the weighted transaction database a second time, and rearrange the items in each transaction based on the priority order of the item header table; insert the rearranged transactions into the frequent pattern tree, and during the insertion process, accumulate the count value of the node with the first transaction weight or the second transaction weight of the corresponding transaction to construct the weighted frequent pattern tree; recursively mine frequent itemsets from the weighted frequent pattern tree and generate association rules.

[0141] In some embodiments, the association rule output module is also used to calculate the data growth rate when new monitoring data flows in; if the data growth rate is less than the preset growth rate threshold, the new monitoring data is converted into a new transaction and directly inserted into the existing frequent pattern tree; if the data growth rate is greater than or equal to the growth rate threshold or the number of tree nodes exceeds the preset threshold, the weighted frequent pattern tree is reconstructed in the background thread based on all historical data and new monitoring data, and switched by atomic pointers.

[0142] The device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0143] Example 5 This invention also provides an electronic device for running the above-described association analysis method based on intelligent branch prediction; see also Figure 5 The diagram shows the structure of an electronic device, which includes a memory 500 and a processor 501. The memory 500 stores one or more computer instructions, which are executed by the processor 501 to implement the aforementioned association analysis method based on intelligent branch prediction.

[0144] Furthermore, Figure 5 The electronic device shown also includes a bus 502 and a communication interface 503. The processor 501, the communication interface 503, and the memory 500 are connected via the bus 502.

[0145] The memory 500 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 503 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 502 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0146] Processor 501 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 501 or by instructions in software form. Processor 501 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a readily available storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 500, and processor 501 reads information from memory 500 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0147] This invention also provides a computer-readable storage medium storing computer-executable instructions. When these computer-executable instructions are called and executed by a processor, they cause the processor to implement the aforementioned association analysis method based on intelligent branch prediction. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0148] The computer program product for performing association analysis based on intelligent branch prediction provided in this embodiment of the invention includes a computer-readable storage medium storing non-volatile program code executable by a processor. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0149] 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.

[0150] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0151] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0152] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0153] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, 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 invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0154] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A correlation analysis method based on intelligent branch prediction, characterized in that, The method includes: Historical monitoring data is acquired and preprocessed to construct a historical weighted transaction set; each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time and space dimensions. Obtain the future time prediction data output by the prediction model, and determine the prediction confidence decay coefficient based on the prediction step size and the historical accuracy of the model based on the future time prediction data; The predicted data for future moments is mapped to a virtual risk item and encapsulated with the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight; the second transaction weight is less than the first transaction weight. The virtual transactions are merged with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information; Based on the weighted transaction database, association rule mining is performed to output association rules containing early warning information.

2. The method according to claim 1, characterized in that, The preprocessing of the historical monitoring data to construct a historical weighted transaction set includes: The historical monitoring data is spatiotemporally aligned to construct a standardized transaction based on driving factors and response factors. The monitoring values ​​of continuous indicators in the standardized transactions are dynamically thresholded and mapped to discrete risk level identifiers and intensity factors. Based on the time decay factor, spatial distance factor, and intensity factor, a comprehensive transaction weight is determined and used as the first transaction weight. The risk level identifier is encapsulated with the first transaction weight to generate the historical weighted transaction set.

3. The method according to claim 2, characterized in that, The dynamic threshold determination of the monitoring values ​​of continuous indicators in the standardized transactions includes: Statistical thresholds are determined based on statistical distribution characteristics, and regulatory thresholds are determined based on environmental regulations and standards. For positive risk indicators, the minimum value between the statistical threshold and the regulatory threshold is taken; for negative risk indicators, the maximum value between the statistical threshold and the regulatory threshold is taken. Three-level risk level demarcation points are set based on the positive risk indicators and the negative risk indicators; The monitoring values ​​of the continuous indicators are mapped to multiple discrete risk levels.

4. The method according to claim 1, characterized in that, The determination of the prediction confidence decay coefficient based on the prediction step size and historical model accuracy of the future time prediction data includes: Obtain the base confidence level of the prediction model; the base confidence level is the theoretical maximum confidence level of the prediction model at zero prediction step size; Based on the predicted step size, the confidence level of step size decay is calculated using an exponential decay model. Multiply the step size decay confidence by the model's historical accuracy to obtain the prediction confidence decay coefficient; The predicted confidence decay coefficient is used as the decay factor in the second transaction weight of the virtual transaction.

5. The method according to claim 4, characterized in that, The determination of the prediction confidence decay coefficient also includes: Obtain historical measured data corresponding to the predicted data, and calculate the systematic deviation of the predicted data; Based on the systematic bias, the numerical values ​​of the future time prediction data are dynamically corrected in reverse using a sign function to obtain the corrected prediction values. Calculate the corrected residuals, and update the prediction confidence decay coefficient a second time based on the residuals to obtain the final decay coefficient.

6. The method according to claim 2, characterized in that, The step of mining association rules based on the weighted transaction database and outputting association rules containing early warning information includes: Scan the weighted transaction database, calculate the weighted support of each item, and construct an item header table based on the weighted support. The second scan of the weighted transaction database rearranges the items in each transaction based on the priority order of the item header table; The rearranged transactions are inserted into the frequent pattern tree. During the insertion process, the count value of the node is accumulated with the first transaction weight or the second transaction weight of the corresponding transaction to construct a weighted frequent pattern tree. Frequent itemsets are recursively mined from the weighted frequent pattern tree, and association rules are generated.

7. The method according to claim 6, characterized in that, The method also includes an incremental update step: When new monitoring data is received, calculate the data growth rate. If the data growth rate is less than the preset growth rate threshold, the newly added monitoring data will be converted into a new transaction and directly inserted into the existing frequent pattern tree. If the data growth rate is greater than or equal to the growth rate threshold or the number of tree nodes exceeds the preset threshold, then based on all historical data and newly monitored data, the weighted frequent pattern tree is reconstructed in the background thread and switched through atomic pointers.

8. A correlation analysis device based on intelligent branch prediction, characterized in that, The device includes: A historical weighted transaction set construction module is used to acquire historical monitoring data, preprocess the historical monitoring data, and construct a historical weighted transaction set; each transaction in the historical weighted transaction set has a first transaction weight determined based on risk intensity, time and space dimensions; The prediction confidence decay coefficient determination module is used to acquire the future time prediction data output by the prediction model, and determine the prediction confidence decay coefficient based on the prediction step size and the historical accuracy of the model based on the future time prediction data. The virtual transaction generation module is used to map the future time prediction data into virtual risk items and encapsulate them with the prediction confidence decay coefficient to generate a virtual transaction with a second transaction weight; the second transaction weight is less than the first transaction weight. The weighted transaction database construction module is used to merge the virtual transaction with the historical weighted transaction set to construct a weighted transaction database containing historical information and future prediction information. The association rule output module is used to mine association rules based on the weighted transaction database and output association rules containing early warning information.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to implement the association analysis method based on intelligent branch prediction as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the association analysis method based on intelligent branch prediction as described in any one of claims 1 to 7.

Images