Method and system for dynamic flow regulation and anti-blocking of intravenous catheter based on fuzzy algorithm

Abstract

The present invention relates to the technical field of intelligent medical equipment management, and in particular to a method and system for dynamic flow regulation and anti-blocking of intravenous catheters based on fuzzy algorithms. The method comprises the following steps: acquiring historical parameter data of the intravenous catheter system, and constructing fuzzy input variables for the historical parameter data to obtain historical variable fuzzy data; constructing fuzzy rules based on the historical variable fuzzy data to obtain fuzzy rule data; performing fuzzy reasoning on the fuzzy rule data to obtain fuzzy reasoning data; and interpretably defuzzifying the fuzzy reasoning data to obtain fuzzy control data. The present invention utilizes fuzzy input modeling and a rule construction mechanism based on support and information gain to achieve high-dimensional state perception and dynamic control strategy generation for the operating status of intravenous catheters. The use of interpretable reasoning path extraction and control information encapsulation enhances the transparency and structured expression of the control output, making it suitable for intelligent management scenarios of infusion equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medical equipment management, and in particular to a method and system for dynamic flow regulation and anti-blocking of an intravenous indwelling needle based on a fuzzy algorithm. Background Art

[0002] With the continuous improvement of the intelligence level of medical equipment, intravenous infusion systems have gradually achieved digital integration in information perception, data recording, and process management. However, in actual use, intravenous catheter systems often experience operational instability due to factors such as abnormal flow, excessive use time, or environmental disturbances, such as clogging tendencies and decreased infusion efficiency, affecting the continuity and reliability of the equipment. Currently, common infusion parameter adjustment mechanisms are mostly based on fixed threshold judgments or static rule settings. They lack the ability to adapt to changes in operating status and cannot effectively model the changing patterns in historical data. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a method and system for dynamic flow regulation and anti-blocking of intravenous catheters based on fuzzy algorithm to solve at least one of the above technical problems.

[0004] The present application provides a method for dynamic flow regulation and anti-blocking of an intravenous catheter based on a fuzzy algorithm, comprising the following steps:

[0005] Step S1: acquiring historical parameter data of the intravenous catheter system, and constructing fuzzy input variables for the historical parameter data to obtain historical variable fuzzy data;

[0006] Step S2: construct fuzzy rules based on historical variable fuzzy data to obtain fuzzy rule data;

[0007] Step S3: performing fuzzy reasoning on the fuzzy rule data to obtain fuzzy reasoning data;

[0008] Step S4: Perform interpretable defuzzification on the fuzzy inference data to obtain fuzzy control data.

[0009] The present invention introduces a rule learning mechanism driven by fuzzy input variables and historical data, which can effectively reduce the subjectivity and uncertainty of manually set rules and achieve flexible modeling of complex environmental factors and nonlinear infusion states. The present invention uses a fuzzy inference engine to calculate the correlation of multi-dimensional fuzzy variables, and combines it with an interpretable defuzzification module to output quantifiable flow regulation parameters, which significantly enhances the system's ability to predict and respond to potential blockage risks. At the same time, compared with traditional methods that rely on fixed threshold control, the present invention has the ability to adaptively learn historical evolution patterns, can adapt to different usage scenarios and equipment status changes, and has stronger real-time performance, robustness and deployment flexibility, which improves the intelligence level and safety of intravenous catheter equipment management.

[0010] Preferably, step S1 is specifically:

[0011] Obtain historical parameter data of intravenous catheter system;

[0012] Perform event sequence extraction on historical parameter data to obtain event sequence data;

[0013] Perform parameter repair on event sequence data to obtain input parameter data;

[0014] Perform input dimension screening based on input parameter data to obtain input screening data;

[0015] Fuzzy linguistic variables are constructed for the input screening data to obtain historical variable fuzzy data.

[0016] In the present invention, event sequence extraction is used to convert raw data into a sequence expression with temporal logic and state transition relationships, which facilitates the subsequent model to identify potential patterns; parameter repair operations can fill in missing values, smooth abnormal fluctuations, and improve data continuity and confidence; input dimension screening can screen out redundant variables based on statistical characteristics, correlation metrics or information gain, reducing model calculation complexity and improving reasoning efficiency; through fuzzy language variable construction, continuous or discrete data is mapped into fuzzy sets with semantic labels, effectively enhancing the system's perception and flexible representation capabilities of uncertain states.

[0017] Preferably, step S2 is specifically:

[0018] Matching the historical variable fuzzy data with the preset fuzzy rule template library to obtain preliminary rule data;

[0019] Perform hypersurface support processing and asymmetric information entropy processing on the preliminary rule data to obtain support rule data and informative rule data respectively;

[0020] Correcting rule conflicts on preliminary rule data according to support rule data and informative rule data to obtain rule correction data;

[0021] The rule correction data is encapsulated in a knowledge graph structure to obtain fuzzy rule data.

[0022] The present invention uses historical variable fuzzy data to match with a preset template library to quickly obtain preliminary rule data with general heuristic features, thereby improving the efficiency and generalization ability of rule construction; by introducing hypersurface support processing, the coverage ability of rules in multidimensional input space can be quantified, and asymmetric information entropy processing is used to evaluate the information increment and discriminative power carried by the rules. These two types of evaluation mechanisms jointly improve the objectivity and accuracy of rule screening; the rule conflict correction mechanism ensures the consistency and enforceability of the rule set by semantically coordinating and adjusting the priority of conflicting rules; by encapsulating the rule correction results into knowledge graphs, it not only enhances the hierarchical organization and scalability of the rule set, but also provides structured retrieval support for subsequent reasoning modules.

[0023] Preferably, the hypersurface support processing is specifically as follows:

[0024] Constructing a rule premise hypersurface based on preliminary rule data to obtain rule premise hypersurface data;

[0025] Fuzzy support data is obtained by calculating fuzzy support based on historical variable fuzzy data and rule premise hypersurface data;

[0026] The preliminary rule data is labeled according to the fuzzy support data to obtain the support rule data.

[0027] In the present invention, by constructing a hypersurface for the premise conditions in the preliminary rule data, the rule premises can be mapped into a continuous multi-dimensional fuzzy space boundary, avoiding the local error and dimensionality explosion problems caused by traditional single-point sample judgment; combining historical variable fuzzy data to calculate fuzzy support, the degree to which each rule is supported by actual data can be evaluated in the entire fuzzy space, fully reflecting the wide applicability and stability of the rule under the distribution of historical data; the quantitative annotation of fuzzy support not only enhances the credibility of rule set screening, but also provides a data basis for rule weight allocation, conflict resolution and reasoning priority sorting.

[0028] Preferably, the asymmetric information entropy processing is specifically as follows:

[0029] Calculate the joint probability of premise and conclusion and the information entropy of conclusion for the preliminary rule data to obtain joint probability data and information entropy data of conclusion respectively;

[0030] The premise information entropy is calculated based on the joint probability data and the conclusion information entropy data to obtain the premise information entropy data;

[0031] Calculate information gain based on conclusion information entropy data and premise information entropy to obtain information gain data;

[0032] The preliminary rule data is labeled according to the information gain data to obtain the information rule data.

[0033] In the present invention, through the joint probability analysis of premise and conclusion and the calculation of conclusion information entropy, the ontological uncertainty of rule output can be clarified, providing a basis for information structure analysis; through the joint probability inference of the information entropy of the premise conditions, the "explanatory power" of the premise variables on the conclusion variables can be characterized, thereby identifying premise combinations with high information carrying capacity in the logical chain; the information gain measurement constructed based on the difference between the conclusion information entropy and the premise condition information entropy can effectively mine those rules with high causal significance and judgment value, thereby avoiding redundant or inefficient rules from being erroneously retained; the information gain results are annotated to the preliminary rules to form information rule data, so that the system has a data-driven evaluation basis when screening rules.

[0034] Preferably, step S3 is specifically:

[0035] Calculate the rule activation degree according to the fuzzy rule data to obtain the rule activation degree data;

[0036] Perform language-type fuzzy reasoning based on rule activation data to obtain fuzzy language variable data;

[0037] Perform functional fuzzy reasoning on fuzzy rule data to obtain functional fuzzy numerical data;

[0038] The fuzzy language variable data and the functional fuzzy numerical data are double-encapsulated by interpretation and control to obtain fuzzy reasoning data.

[0039] The rule activation calculation in the present invention can quantify the response intensity of fuzzy rules under current input conditions, effectively screen out highly relevant rules to participate in reasoning, thereby improving reasoning efficiency and stability; obtaining fuzzy language variable results (such as "fine-tuning", "rapid increase", etc.) through language-type fuzzy reasoning can enhance the readability and human-computer interaction friendliness of the system, and is suitable for nursing auxiliary decision-making and graphical display scenarios; at the same time, a functional fuzzy reasoning channel (such as Sugeno-type reasoning) is introduced to generate continuous numerical output to support the fine control requirements of flow regulation parameters; the two types of reasoning results are double-encapsulated through interpretation and control to form unified fuzzy reasoning data, which not only retains the semantic clues of the reasoning process, but also provides a quantitative control basis for the execution system.

[0040] Preferably, step S4 is specifically:

[0041] Perform interpretable centroid processing on fuzzy reasoning data to obtain key reasoning path data;

[0042] Perform control output extraction on key reasoning data to obtain control output data;

[0043] A control interpretation diagram is constructed based on the key reasoning path data and control output data to obtain fuzzy control data.

[0044] In the present invention, fuzzy reasoning data is aggregated through interpretable centroid processing, which not only retains the weight distribution relationship between the rule activation intensity and the language output, but also extracts the core path that affects the reasoning result, realizing the explicit identification of the "decision factor"; the control output extraction module extracts numerical adjustment suggestions or control parameters from the above-mentioned key paths, providing accurate instruction basis for the downstream equipment execution module or manual intervention; by combining the key path data and the control output, a structured control interpretation diagram is constructed, which not only outputs the "adjustment amount" at the result level, but also visualizes "why the adjustment" and "according to which rules" at the logic chain level, thereby taking into account the decision executability and traceability explanation capabilities. Compared with the "black box" fuzzification-defuzzification process of the traditional fuzzy control system, this step has the advantages of clear information links, visual and traceable outputs, and easy manual intervention and system auditing, which improves the credibility, controllability and deployment security of the entire system.

[0045] Preferably, the interpretable centroid processing is specifically:

[0046] Perform tailoring and fusion based on fuzzy inference data to obtain fuzzy graphic data;

[0047] Perform path tracing integration on the fuzzy graphic data to obtain the output center of gravity data;

[0048] Calculate the contribution interval based on the output center of gravity data to obtain the contribution interval data;

[0049] The reasoning paths of fuzzy reasoning data are aggregated according to the contribution interval data to obtain the key reasoning path data.

[0050] The fuzzy graphics clipping and fusion operation in the present invention can remove fuzzy subsets with low activation or redundancy while maintaining the semantic integrity of reasoning, making subsequent calculations more targeted and accurate; the fuzzy graphics area is continuously summed up by path tracing integral method to generate output center of gravity data, which can accurately locate the concentrated section of the reasoning output in the fuzzy space and provide a numerical basis for quantitative adjustment; then the contribution interval calculation mechanism is introduced to segment and calibrate the key areas in the reasoning space to reveal the actual contribution degree of each input condition or rule in the final reasoning; based on the above contribution interval information, the fuzzy reasoning data is aggregated by reasoning paths to output a "key reasoning path" with readability and logical coherence, which not only improves the transparency of the reasoning output, but also provides a basis for manual review and system tracing of control suggestions.

[0051] Preferably, the tailoring and fusion are specifically:

[0052] Calculate the activation degree based on the fuzzy reasoning data to obtain the rule activation degree data;

[0053] Output extraction is performed based on fuzzy reasoning data to obtain rule output data, wherein the output extraction is based on the activated rule set in the fuzzy reasoning data, and the fuzzy language variables and membership function expressions of each rule output are extracted;

[0054] Minimize the rule output data according to the rule activation data to obtain rule-cut data;

[0055] The regular clipping data is aggregated and fused to obtain fuzzy graphic data.

[0056] In the present invention, by performing precise calculation of the activation degree of the activated rules in the fuzzy reasoning data, the rule subset with actual contribution under the current input conditions can be accurately identified, thereby avoiding noise interference of invalid or marginal rules on the reasoning results; in the output extraction process, the corresponding fuzzy linguistic variables and membership function expressions are extracted based on the rule activation set, providing a basic semantic configuration for graphics processing; the rule output is trimmed through the minimum stage processing strategy, which is essentially a fuzzy output trimming threshold mechanism, which only retains the fuzzy subset with activation degree higher than the threshold, thereby significantly compressing the fuzzy graphics processing space and improving the graphics aggregation efficiency; through the aggregation and fusion operation of the rule trimming data, multiple fuzzy outputs are integrated into unified fuzzy graphics data, providing a unified input basis for centroid positioning, path integral and contribution analysis.

[0057] Preferably, the present application further provides a fuzzy algorithm-based intravenous catheter dynamic flow regulation and anti-blocking system, which is used to execute the fuzzy algorithm-based intravenous catheter dynamic flow regulation and anti-blocking method as described above. The fuzzy algorithm-based intravenous catheter dynamic flow regulation and anti-blocking system includes:

[0058] A fuzzy variable generation module is used to obtain historical parameter data of the intravenous catheter system and construct fuzzy input variables for the historical parameter data to obtain historical variable fuzzy data;

[0059] A fuzzy rule construction module is used to construct fuzzy rules based on historical variable fuzzy data to obtain fuzzy rule data;

[0060] A fuzzy reasoning engine module is used to perform fuzzy reasoning on fuzzy rule data to obtain fuzzy reasoning data;

[0061] The interpretable defuzzification module is used to perform interpretable defuzzification on fuzzy inference data to obtain fuzzy control data.

[0062] The beneficial effects of the present invention are: a set of fuzzy decision-making mechanisms with structured semantics and data interpretability is constructed around the three core links of operating state perception, rule learning and control output. By reconstructing the event sequence and mapping the fuzzy variables of the historical infusion data, the unified encoding of static parameters and dynamic trends is achieved; based on the hypersurface support and asymmetric information entropy indicators, a set of fuzzy rules with high contribution and low redundancy is extracted from the candidate rules, which effectively improves the stability and coverage of reasoning. In the reasoning link, the present invention introduces a linguistic and functional dual-channel fuzzy reasoning structure, which takes into account the clarity of decision semantics and the accuracy of control numerical values, and is adaptable to a variety of output scenarios; and by locating the center of mass position through fuzzy graph clipping fusion and path integral, the explanation link tracing and critical path calibration of the reasoning results are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0064] Figure 1 A flowchart showing the steps of a method for dynamic flow regulation and anti-blocking of an intravenous catheter based on a fuzzy algorithm according to one embodiment is shown;

[0065] Figure 2 A flowchart showing the steps of a fuzzy variable generation method according to an embodiment is shown;

[0066] Figure 3 A flowchart showing the steps of a fuzzy rule construction method according to an embodiment is shown;

[0067] Figure 4 A flowchart of the steps of a fuzzy reasoning method according to an embodiment is shown. DETAILED DESCRIPTION

[0068] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0069] Furthermore, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0070] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0071] See also Figures 1 to 4 The present application provides a method for dynamic flow regulation and anti-blocking of an intravenous catheter based on a fuzzy algorithm, comprising the following steps:

[0072] Step S1: acquiring historical parameter data of the intravenous catheter system, and constructing fuzzy input variables for the historical parameter data to obtain historical variable fuzzy data;

[0073] In one embodiment, the non-blocking intravenous catheter system of the present invention is composed of a three-way stopcock assembly, an anticoagulant storage unit, a catheter and control unit, a puncture introduction needle, and an injection operation unit. In this system, an intelligent edge control box is provided in the catheter section between the anticoagulant storage unit or the output end of the main infusion pump and the input end of the three-way stopcock assembly. The control box is connected in series with the infusion catheter via a flexible connector to form a relay node that can monitor fluid changes. The series connection refers to inserting the intelligent edge control box into the middle section of the infusion catheter via a flexible lock interface or a Luer connector to form a structurally sealed continuous fluid channel, so that the infused liquid flows through the micro-circulation channel in the edge control box before flowing to the three-way stopcock assembly. The intelligent edge control box is equipped with a micro-circulation channel and multiple sensing components integrated into the channel wall, including but not limited to a pressure sensor for real-time measurement of micro-pressure changes generated by liquid flow in the channel to reflect the system backpressure trend; a thermosensitive thin film sensor for detecting fluid temperature fluctuations, thereby indirectly determining changes in liquid viscosity or environmental interference; and a micro-flow rate sensing module that estimates the fluid flow rate per unit time through electrothermal time response or thermal film bridge technology. This structure enables indirect modeling of fluid state changes in the infusion pipeline, without interfering with the structure of the intravenous catheter itself or affecting the sterile path. Instead, it constructs proxy feature data that reflects the operating status of the infusion system solely through the dynamic trends of the liquid flowing through the edge control box. The system uses the collected parameters such as flow rate, pressure, and temperature as inputs to the fuzzy inference engine, executes the fuzzy control process in steps S1 to S4, and outputs the control variable through the centroid defuzzification method.

[0074] In one embodiment, the operating data of the intravenous catheter system is collected, and the historical parameter data includes but is not limited to the following dimensions, such as flow rate Q (mL / h), the flow rate of the liquid in the needle per unit time; pressure P (mmHg), the static or dynamic pressure value of the liquid in the tube; resistance R (Ω), the fluid resistance index that simulates the degree of pipeline blockage; pipeline temperature T (℃), the fluid or pipe wall temperature; event flag E (0 or 1), whether a blockage event occurs. Based on the above data, a set of fuzzy input variables is set to construct fuzzy linguistic variables. For example: Q→{low, medium, normal, high}, defining a triangular membership function; R→{unobstructed, critical, blocked}; T→{slightly low, normal, slightly high}; P→{slightly low, stable, slightly high}; E→ is only used for label supervision (non-fuzzy variable). Taking the flow rate Q as an example, the triangular membership function corresponding to its fuzzy linguistic variable "low" is defined as: 。The remaining variables are used to construct membership functions in turn to form a set of fuzzy input variables. Other levels such as "medium" are defined in a similar way, using [25, 35, 45], [40, 55, 70], etc. as the control parameters of trigonometric functions respectively to form a complete set of fuzzy language mapping. Similarly, the system constructs a set of fuzzy language variables for pressure P {low, stable, high}, where "stable" uses a symmetric trigonometric function (a symmetric triangular membership function is defined by three parameters: the left endpoint a, the central peak point b, and the right endpoint c, where a < b and b < c. This function increases linearly (x - a) / (b - a) between a and b, and the membership degree increases from 0 to 1; it decreases linearly (c - x) / (c - b) between b and c, and the membership degree decreases from 1 to 0; outside the interval [a, c], the membership degree is 0), the domain is 80 to 120 mmHg, and the peak value is 100 mmHg. A set of language variables for resistance R {unobstructed, critical, blocked} is constructed, where "blocked" uses a trapezoidal membership function (a trapezoidal membership function is controlled by four key parameters: a is the left starting point of the function, b is the end point of the rising segment, c is the starting point of the falling segment, and d is the right end point of the function. The function increases linearly (x - a) / (b - a) within the interval [a, b], maintains a membership degree of 1 between [b, c], decreases linearly (d - x) / (d - c) between [c, d], and the membership degree outside the intervals [d, +∞) and (-∞, a] is 0), and is defined as the membership degree is 1 when the R value is greater than or equal to 2 Ω, and it rises linearly between 1.5 and 2.0 Ω. A set of language variables for temperature T {low, normal, high} is constructed, where "normal" is a trigonometric function (controlled by three parameters, namely the left endpoint a, the central peak b, and the right endpoint c, where a < b < c. Within the interval from a to b, the membership degree increases linearly with the input value (x - a) / (b - a), and decreases linearly (c - x) / (c - b) within the interval from b to c, and the membership degree outside the interval is zero.), the central value is 37 °C, and the left and right boundaries are 35 °C and 39 °C respectively.

[0075] Step S2: Construct fuzzy rules based on historical variable fuzzy data to obtain fuzzy rule data;

[0076] In one embodiment, the system uses a combination of expert knowledge and data-driven methods to establish a fuzzy control rule base for fuzzy control reasoning regarding intravenous catheter adjustment strategies. These fuzzy rules are expressed using a standard IF-THEN format, where the IF component is the fuzzy input condition, composed of fuzzy linguistic variables; the THEN component is the fuzzy output variable, the result of which represents the level of adjustment intensity. Input variables include, but are not limited to, flow rate Q, resistance R, and temperature T; the output variable is the adjustment level A, whose fuzzy linguistic variable set includes no adjustment, fine adjustment, medium adjustment, and high adjustment. Each fuzzy linguistic variable is modeled using a membership function. For example, the fuzzy linguistic value "fine-tuning" of the output variable "Adjustment Degree A" can be defined using a triangular membership function with a fuzzy domain of [0.1, 0.3, 0.5], meaning it reaches a membership of 1 at 0.3 and 0 at 0.1 and 0.5. Another output value, "high-key," can be modeled using a trapezoidal membership function with a domain of [0.6, 0.8, 1.0, 1.0], meaning it maintains a membership of 1 between 0.8 and 1.0, increases linearly between 0.6 and 0.8, and becomes 0 below 0.6. Based on these fuzzy variables, the system constructs a preliminary set of fuzzy rules. The rule examples are as follows: Rule R1: If the flow rate Q is "low", the resistance R is "blocked", and the temperature T is "normal", then the adjustment degree A is "high adjustment"; Rule R2: If the flow rate Q is "normal", the resistance R is "smooth", and the temperature T is "high", then the adjustment degree A is "no adjustment"; Rule R3: If the flow rate Q is "medium", the resistance R is "critical", and the temperature T is "normal", then the adjustment degree A is "fine adjustment". The above rules constitute the fuzzy rule table R, which is recorded as rule table ,in is the total number of rules. Each rule consists of a set of premise fuzzy conditions and a fuzzy output, and the output label corresponds to the corresponding fuzzy membership function.

[0077] Step S3: performing fuzzy reasoning on the fuzzy rule data to obtain fuzzy reasoning data;

[0078] In one embodiment, the system fuzzifies input variables and matches them with the preconditions of each fuzzy rule, calculating the activation of each rule under the current input. For any rule, its preconditions typically include multiple fuzzy input variables (such as flow rate, resistance, and temperature). The system calculates the membership of each input variable under the fuzzy linguistic value corresponding to the rule's precondition and takes the minimum of all memberships as the rule's activation. For example, if a rule's preconditions are "flow rate is medium, resistance is blocked, and temperature is normal," and the membership of the current input sample under these three conditions is 0.8, 0.7, and 0.9, respectively, then the rule's activation is the minimum of these three values, namely, 0.7. The system clips the output of all activated rules and then performs a fusion operation. For each activated rule, its output corresponds to a fuzzy linguistic variable (such as "fine tune" or "high tune") and has a set of defined fuzzy membership functions. The system clips these output membership functions according to the corresponding rule's activation, limiting the function's maximum membership to no more than the activation. The system synthesizes all the clipped output fuzzy sets. The synthesis process uses the maximum synthesis rule, that is, for each value point in the output variable's domain, the maximum membership of all the clipped functions at that point is selected as the synthesized membership. This process realizes the set synthesis of the fuzzy outputs of multiple rules at the functional level, forming a unique fuzzy control output function. After completing the synthesis of the fuzzy outputs, the system plots the synthesis result as an output membership function curve, namely the fuzzy control chart. This chart shows the comprehensive membership corresponding to each value of the control output variable in its domain. This chart not only serves as an intuitive carrier for explaining the reasoning results, but also serves as the input basis for defuzzification processing (such as the centroid method) to generate specific control values. The system forms fuzzy reasoning data containing the output fuzzy set structure.

[0079] Step S4: Perform interpretable defuzzification on the fuzzy inference data to obtain fuzzy control data.

[0080] In one embodiment, the system performs a weighted integration of the membership function within the domain of the fuzzy output variable to determine a numerical value representing the meaning of the overall fuzzy output. The output variable A of the adjustment degree takes values ​​within the domain [0,1], and the synthetic output membership function obtained by the fuzzy inference process is μoutput (x). The system then performs the following two integral operations within the interval: calculating the integral value of the product of x and μoutput (x), which represents the "torque" of the fuzzy function; calculating the integral value of μoutput (x) itself, which represents the "fuzzy quality"; the system uses the ratio of the above two integral values ​​as the control output result, that is, the centroid position of the fuzzy set. The integral operation is implemented using numerical approximation methods, such as discrete sampling integration or trapezoidal rule integration. The system divides the entire domain into multiple equidistant small intervals, and calculates the membership value at each sampling point, thereby constructing an approximate integral expression of the fuzzy output function. Through the above calculations, the system can obtain an accurate numerical control output, recorded as the actual control value , ,in is the value point of the fuzzy output variable, and its domain is the numerical range of the output variable. is the fuzzy output membership value at point x, that is, the fuzzy response intensity corresponding to each value point of the fused fuzzy output graph. This function is synthesized by cutting multiple rule outputs. The upper and lower limits of the integral are the value range covered by the integral operation, that is, the complete numerical range of the output variable. This value is used to guide the dynamic adjustment of the flow control device (such as the pump control module or flow rate valve) of the intravenous catheter system, thereby achieving accurate response and prevention of pipeline blockage trends. Output As a bridge between fuzzy reasoning and physical control, it constitutes the key control quantity of the interpretable fuzzy control closed-loop system.

[0081] Preferably, step S1 is specifically:

[0082] Step S11: Acquire historical parameter data of the intravenous catheter system;

[0083] In one embodiment, the system collects raw monitoring data from monitoring modules (such as pressure sensors, flowmeters, and temperature probes) during the operation of an intravenous catheter, generating a time series parameter record. Collected parameters include, but are not limited to, flow rate Q(t), the fluid flow rate per unit time (mL / h); lumen pressure P(t), the static and dynamic pressures within the catheter (mmHg); resistance R(t), the resistance coefficient estimated from Q and P (simulating thrombosis or crystallization formation, in Ω); tubing temperature T(t), the tubing wall or fluid temperature (in °C); alarm event E(t), indicating whether an alarm event has been triggered, such as an occlusion alarm or abnormal flow rate alarm, with a 0 / 1 flag; manual adjustment flag M(t), marking the time of manual adjustment operations; and timestamp t, a system time record accurate to the second. All parameters must be fully time-stamped and aligned with a uniform time resolution (e.g., 1 second) to form a preliminary multidimensional time series dataset.

[0084] Step S12: extracting event sequence from historical parameter data to obtain event sequence data;

[0085] In one embodiment, the extraction process of the event sequence includes the following aspects: key event definition, event window extraction, and event marking and encapsulation. Specifically, the key event definition is a set of key event conditions defined by the system that can be used as the basis for event triggering, mainly including the occurrence of an alarm event, that is, at a certain time point, the alarm event mark value is 1 (ie, E(t)=1), indicating that the device has an alarm situation such as pipeline blockage, abnormal flow rate, etc. at that moment; manual operation behavior, that is, manually adjusting the mark value to 1 (ie, M(t)=1), indicating that the medical staff has performed manual intervention at that time point; flow rate mutation, that is, the flow rate change between adjacent time points exceeds the set threshold. Suppose the two time points before and after are respectively and , if the corresponding flow rate value 、 Satisfy the change If the change in resistance between adjacent time points exceeds the set threshold, it is considered a sudden change in flow rate; if the change in resistance between adjacent time points exceeds the set threshold, it indicates that there is intraluminal blockage. If the value is greater than the set threshold, it can be identified as a resistance mutation event. When any of the above key events is detected, the system takes the time point corresponding to the key event as the center, and extends the preset time length forward and backward (for example, 60 seconds each) to form a time window containing the key event context. This window is used to comprehensively record the state evolution process before and after the event occurs. The window size can be adjusted according to actual needs, such as setting it to forward. Seconds, backward Seconds. Event tagging and packaging assigns a unique event number tag to each extracted event window and records the following information: event number, which uniquely identifies the event, such as "Event 001"; start time and end time, which represent the start and end time points of the event window, respectively; and parameter trajectory, which contains the time series trajectory of all or selected key parameters within the window, such as flow rate Q(t), resistance R(t), temperature T(t), pressure P(t), etc.

[0086] Step S13: performing parameter repair on the event sequence data to obtain input parameter data;

[0087] In one embodiment, when the duration of consecutive missing values ​​is less than a set threshold (e.g., less than 5 seconds), a neighborhood interpolation method is used to estimate the missing value. Specifically, linear or spline interpolation is used to infer the estimated value at the missing time using data from two known time points before and after the missing value. For example, if the flow velocity value at a given time t is missing, the system can add half the difference between the flow velocities at the previous and next times to the value at the previous time to obtain the interpolated value. When the missing time is longer (e.g., greater than or equal to 5 seconds) and the interpolation is unstable, the system uses a local average estimate or uniformly fills the missing segments with an "unavailable" mark. If the missing percentage exceeds the event window tolerance threshold, the window is deemed invalid and discarded during the processing phase. A sliding calculation is performed on key parameters such as flow velocity, pressure, and resistance within each event window, and the local mean and standard deviation are calculated. If a data point deviates from its local mean by more than three standard deviations (i.e., outside the range of "mean plus or minus three standard deviations"), the data point is identified as an outlier. For data points identified as abnormal, the system can use local average value replacement or median filtering to smooth and repair them based on the neighborhood data of the previous and next time points.

[0088] Step S14: performing input dimension screening according to the input parameter data to obtain input screening data;

[0089] In one embodiment, the system first uses whether an alarm event occurs (i.e., the alarm event is marked as 1) as the target variable to evaluate the correlation between each input parameter and the target event. Specifically, it includes calculating the mutual information index to characterize the information dependence strength between each parameter and the alarm event. The larger the mutual information value, the more information the parameter contains when predicting the target event. If the mutual information value of a parameter is greater than 0.2, it is determined that the parameter has a significant statistical correlation with the target event. In some embodiments, a shallow classification model, such as a random forest model, can be further introduced to perform modeling training on all input parameters, with the goal of predicting the occurrence of alarm events. By ranking the feature importance after model training, several parameters with the highest contribution to alarm prediction (such as the top k) are selected as key input dimensions. Based on the above correlation evaluation and feature importance ranking, the system screens out a group of input parameters that are highly correlated with congestion risk to form the final input dimension set. For example, common retained parameters may include flow rate (Q), resistance (R), temperature (T), etc., which are closely related to the blockage state; while dimensions that are irrelevant to the modeling objectives or have high noise, such as manual operation marks or some pressure parameter items, are eliminated at this stage.

[0090] Step S15: construct fuzzy language variables for the input screening data to obtain historical variable fuzzy data.

[0091] In one embodiment, for each input dimension parameter, several fuzzy language values ​​with clinical interpretation significance are divided according to its physical meaning and empirical range. For example, for the flow rate parameter (Q, in milliliters per hour), it is divided into three fuzzy language values ​​based on the empirical range of common infusion rates, namely "low", "medium" and "high". This division is used to express the impact of different flow rate levels on the patency of the pipeline, such as , to obtain the fuzzy language value corresponding to the low situation. For the resistance parameter (R, in ohms, analogous to the mathematical modeling idea of ​​Ohm's law in the circuit system, used to abstractly represent the size of the fluid resistance in the infusion system), it is divided into three levels: "unobstructed", "critical" and "blocked", among which "blocked" means that the system is close to or has reached the state of pipeline blockage, such as "Unobstructed" indicates low fluid resistance and smooth flow. For the temperature parameter (T, in degrees Celsius), the normal range is 35 to 38 degrees Celsius. It is divided into three fuzzy linguistic values: "low," "normal," and "high" to capture the potential impact of ambient or fluid temperature on system status. The system assigns a membership function to each fuzzy linguistic value to achieve fuzzy mapping of continuous values. The membership function can be either triangular or trapezoidal, calculating the degree of membership under a given fuzzy linguistic value based on the parameter's position. For example, when the flow rate is near a medium level, its membership under the "medium" fuzzy linguistic value is high. If the resistance value is significantly above the critical point, its membership under the "blocked" fuzzy linguistic value is close to 1. When the temperature is in the middle of the normal range, its membership under the "normal" linguistic value is the highest. The system fuzzifies the multidimensional input parameters at each time point, outputting a fuzzy data set of historical variables. Each record includes the timestamp of the moment and the fuzzy linguistic value and its membership degree for the corresponding dimension. For example, the fuzzy expression of a certain time point can be expressed as: flow rate is "medium", membership is 0.7; resistance is "blocked", membership is 0.9; temperature is "normal", membership is 0.6. All fuzzification results together constitute a time series fuzzy dataset.

[0092] Preferably, step S2 is specifically:

[0093] Step S21: matching the historical variable fuzzy data with the preset fuzzy rule template library to obtain preliminary rule data;

[0094] In one embodiment, the system pre-builds a library of universal fuzzy rule templates to guide the flow regulation strategy of the intravenous catheter system during operation. These fuzzy rule templates are expressed in an "if-then" format and are used to describe the regulation response under different fuzzy input conditions. For example, one rule can be expressed as "If the flow rate is medium, the resistance is obstructed, and the temperature is normal, then the regulation intensity is high." In this example, the flow rate (Q), resistance (R), and temperature (T) are the input parameters; the regulation intensity (A) is the output parameter. Each input parameter is represented by a fuzzy linguistic variable, such as "flow rate is medium" or "resistance is obstructed." The output is a fuzzy output label, such as "high regulation," "medium regulation," "fine regulation," or "no regulation." The system uses collected historical fuzzy variable data as input and matches it against the fuzzy rule templates one by one. For example, if the corresponding fuzzy value of a historical data is "flow velocity is medium (membership degree 0.7), resistance is blocked (membership degree 0.8), temperature is normal (membership degree 0.6)", then for each fuzzy rule, the system will calculate its matching activation degree. Specifically, the matching activation degree is defined as the minimum value of the membership values ​​of each input fuzzy variable, denoted as For example, in the above example, the activation The minimum of the three membership values ​​is min{0.7, 0.8, 0.6}, which results in a value of 0.6. The system sets an activation threshold (e.g., 0.5). When the activation of a rule is greater than or equal to this threshold, the historical data is considered to have successfully matched the rule. The system traverses all historical samples and rule templates, counting the number of times each rule is matched and the activation value corresponding to each match. The system generates a preliminary rule dataset, including each rule's number, the total number of matches, the average activation value, and the corresponding adjustment output label. For example, rule number R_05 was matched 120 times, with an average activation of 0.78 and a corresponding output label of "high profile"; rule number R_09 was matched 80 times, with an average activation of 0.65 and an output label of "medium profile." The above process completes the preliminary matching of the fuzzy rule template with the historical fuzzy data. The membership calculation is based on the fuzzy membership function (which defines the degree to which the input value belongs to a certain linguistic variable (i.e., membership ∈ [0, 1])). Each fuzzy linguistic variable is represented by a triangular membership function in the following form: ,in It's the starting point. is the peak point (the position where the membership degree is 1), It is the end point; is the current input value; Is the degree of membership of the input to a certain language value. Taking "flow rate is medium" as an example, let the triangular membership function parameter of "medium" be the starting point =30, peak value =50, end point =70, if the current flow rate (current representation in historical data) =44, then according to the formula, ; The trigonometric function parameters corresponding to "blocking" are: a=0.8, b=1.1, c=1.4, and the current input value is , falls in the interval (0.8,1.1), belongs to the rising edge, calculate: ; The triangular membership function parameters for "normal" temperature are a=36.0, b=37.0, c=38.0. The current input value is , falls in the interval (36.0,37.0), belongs to the rising edge, calculate .

[0095] Step S22: performing hypersurface support processing and asymmetric information entropy processing on the preliminary rule data to obtain support rule data and informative rule data respectively;

[0096] In one embodiment, the hypersurface support processing (used to represent the structural integrity of rules in the fuzzy space) is to construct a multi-dimensional hypersurface of the fuzzy input space, with the flow rate (Q), resistance (R), and temperature (T) as the three axes to form a three-dimensional fuzzy input domain; for each rule , define its support region as the volume of the fuzzy space in the input space where this rule can be activated; use the integral approximation method to estimate the support: , is the rule support, representing the total volume activated by the fuzzy rule in the three-dimensional fuzzy input space; the higher the support, the wider the coverage of this rule in the input space and the higher the probability of being activated, is the activation region of the rule , referring to the three-dimensional subspace in the fuzzy input space where all the premise conditions of this rule (such as Q = medium, R = blocked, T = normal) simultaneously meet a certain activation condition (such as activation degree ≥ threshold), is the fuzzy membership function value of the flow rate dimension, representing the current flow rate value in the rule under the "specified language label" (such as Q = medium), and the value ∈ [0, 1], is the fuzzy membership function value of the resistance dimension, representing the current resistance value in the rule under the "specified language label" (such as R = blocked), is the fuzzy membership function value of the temperature dimension, representing the current temperature value in the rule under the "specified language label" (such as T = normal), is the flow rate input variable, is the resistance input variable, is the temperature input variable. The larger the result, the more frequently this rule appears in historical samples and the wider its application range.

[0097] The asymmetric information entropy processing (used to evaluate the information gain of rule output) is to calculate the information entropy of the conclusion for each rule, such as , is the conclusion information entropy, representing the degree of uncertainty of various output labels in the current rule set. The larger the value, the more dispersed the output results are and the higher the fuzziness of the system prediction, is the output label, referring to the output language label of the fuzzy rule, such as "high adjustment", "fine adjustment", "no adjustment", etc., which is the "conclusion" part of the fuzzy rule, is the output label set, referring to the set of all possible fuzzy output language labels (such as A = {no adjustment, fine adjustment, medium adjustment, high adjustment}), is the probability distribution of the output label, which means label The relative frequency of occurrence in all rule conclusions can be obtained by statistically analyzing the preliminary rule set. The information entropy of the premise conditions under the rule condition is calculated as , for, for, for, Then, the information gain is calculated as . If the gain of a rule is high (e.g., >0.1), it indicates that the rule has a strong discriminative effect on the output results. Based on the above, two types of measurement data are obtained. For example, for rule number R_05, its support is 0.78 and its information gain is 0.12; for rule number R_09, its support is 0.52 and its information gain is 0.08.

[0098] Step S23: performing rule conflict correction on the preliminary rule data according to the support rule data and the informative rule data to obtain rule correction data;

[0099] In one embodiment, the system mainly identifies and handles two typical types of fuzzy rule conflict problems, including similar premises but conflicting outputs, that is, there are two or more rules whose input condition combinations are highly similar or exactly the same, but the corresponding output results are significantly different. For example, the input conditions "flow rate is medium, resistance is blocked" correspond to "adjustment intensity is high" and "adjustment intensity is fine-tuning" at the same time. Such inconsistent outputs will interfere with the reasoning consistency of the fuzzy system. The output fuzzy overlap is serious, that is, when multiple rules are activated, there is a significant overlapping area in their fuzzy output functions, which causes the synthesized fuzzy output graph to be distorted or the fuzzy boundary to be unclear, thereby reducing the resolution and interpretability of the adjustment output. In order to deal with the above conflict problems, the system first defines the rule conflict judgment criteria. Specifically, for any two rules and , if the similarity of the premise condition is higher than the set threshold (such as 0.9), and the membership function of the output fuzzy label has significant overlap (the overlap is greater than the threshold ), then the rule pair is considered to be in conflict. The premise similarity can be calculated using a set similarity metric (such as the Jaccard similarity coefficient), which measures the degree of overlap in the linguistic values ​​of the input variable sets of the two rules. The output membership function overlap measures the extent of overlap between the output functions of the two rules within the fuzzy domain, often estimated using the overlap integral or function intersection ratio. For rule pairs identified as conflicting, the system adopts one of the following conflict resolution strategies based on the rule's support and information gain values: 1. If a rule's support value is less than 0.5 or its information gain value is less than 0.05, indicating that the rule is weakly representative in the historical sample or insufficiently contributes to the output, the system marks the rule as "weak" and removes it. 2. If the conflicting rule pairs have comparable support but significant differences in information gain, the one with the higher information gain is retained to ensure that the rule with greater discriminative power for the adjustment output is retained. 3. If the conflict arises from the overlap of the output fuzzy functions themselves (for example, the peak regions of two "medium-tone" and "high-tone" functions overlap), the system redefines the output membership function, including shifting or compressing its domain, to achieve boundary de-overlapping of the output graph and refine the response. All valid rules after conflict identification and resolution are retained and included in the rule correction dataset R*.

[0100] Step S24: Encapsulate the rule correction data into a knowledge graph structure to obtain fuzzy rule data.

[0101] In one embodiment, each fuzzy rule is represented as a structured graph unit. This graph unit consists of multiple nodes and edges, and is used to represent the preconditions, output conclusions, and association relationships of fuzzy rules. Taking a fuzzy rule as an example, its logical form can be expressed as: "If the flow rate is medium, the resistance is blocked, and the temperature is normal, then the adjustment intensity is high." This rule is encapsulated as the following structured content in the graph. Construct nodes, including precondition nodes, which respectively represent fuzzy input variables and their linguistic values, such as "the flow rate is medium (Q_medium)", "the resistance is blocked (R_blocked)", "the temperature is normal (T_normal)"; an output result node, which represents the fuzzy output variable and its value, such as "the adjustment intensity is high (A_high)"; a rule identification node, which uniquely identifies this rule, such as "Rule_05". Construct edges, including constructing conditional edges, that is, connecting each precondition node to the rule identification node, indicating that this input is the activation condition of the rule. For example, "Q_medium" points to "Rule_05" through an edge of type "condition"; construct conclusion edges, that is, connecting the rule identification node to the output result node, indicating the inference output of this rule. For example, "Rule_*05*" points to "A_high" through an edge of type "conclusion"; for each rule node (such as Rule_05), store additional attribute information of this rule, including the activation threshold, that is, the minimum activation degree required for this rule to be activated in fuzzy inference; the support value, which represents the effective activation volume of the rule in the fuzzy input space; the information gain value, which represents the discrimination ability of the output of this rule; the domain of the output membership function, that is, the interval of the membership function corresponding to the output fuzzy variable. Through the above structured encapsulation, the system organizes all the corrected fuzzy rules into a visual, schedulable, and retrievable graph-based knowledge expression system, forming a rule graph data structure for the fuzzy inference engine, collectively referred to as fuzzy rule data.

[0102] Preferably, the hypersurface support processing is specifically as follows:

[0103] Construct a rule premise hypersurface based on the preliminary rule data to obtain rule premise hypersurface data;

[0104] In one embodiment, the system performs fuzzy hypersurface modeling on the premise part of each preliminary fuzzy rule to characterize its multidimensional distribution structure in the fuzzy input variable space. Define the fuzzy input space. Let the input variable set be X, including multiple fuzzy input variables, such as flow rate (Q), resistance (R) and temperature (T). Each input variable corresponds to a fuzzy language set, and a corresponding membership function is preset for each language value to describe the degree of belonging of the variable value under different semantic states. For example, the flow rate Q corresponds to language values ​​such as "low", "medium" and "high", and each language value is characterized by a membership function defined on the real number field. For each fuzzy rule, the system extracts the fuzzy language conditions of its premise part. For example, the premise of a certain rule is: "If the flow rate is medium, the resistance is blocked, and the temperature is normal", then its fuzzy hypersurface can be defined as the area in the three-dimensional input space that meets the following conditions: the flow rate of all points in the area has a membership greater than zero under the "medium" language value, the resistance has a membership greater than zero under the "blocked" language value, and the temperature has a membership greater than zero under the "normal" language value. The spatial area corresponding to the fuzzy hypersurface is the fuzzy premise area where the rule is activated. The system encapsulates the fuzzy premise hypersurface of each rule into a structured data format for subsequent rule comparison, reasoning calculation and graphical expression. The encapsulation content includes the rule number, the dimensions of the fuzzy input variables involved in the modeling, and the membership function expression or its discrete representation corresponding to each variable, which is used to describe the fuzzy activation interval of the rule in each dimension. For a rule , its premise is as follows: IF Q = medium AND R = blocked AND T = normal, then its fuzzy hypersurface , defined as: , this area represents the rules The activated fuzzy spatial region.

[0105] Fuzzy support data is obtained by calculating fuzzy support based on historical variable fuzzy data and rule premise hypersurface data;

[0106] In one embodiment, the system estimates the support of each fuzzy rule based on the historical fuzzy input sample data and the constructed rule premise hypersurface data to quantify the activation frequency and degree of the rule in the historical samples. Specifically, it performs sample input matching. For any j-th sample in the historical fuzzy sample set, let its input variable combination be {flow rate is , resistance is , temperature is }, the system calculates the membership function of the sample to a certain rule based on the corresponding membership function in the rule premise The activation level can be evaluated in one of the following two ways: one is to use the minimum membership method ,in is the activation degree of the sample pair rule, is the minimization function, is the flow rate under the "medium" linguistic value, is the resistance under the "blocked" linguistic value, is the temperature under the "normal" linguistic value, that is, taking the minimum value of the membership function values of the sample in the three dimensions of flow rate, resistance and temperature, representing the weakest factor in the matching degree between the sample and the rule; the second is to use the product membership method , is the activation degree of the sample pair rule, is the flow rate for a certain linguistic value (such as "medium"), is the resistance for a certain linguistic value (such as "blocked"), is the temperature for a certain linguistic value (such as "normal"), that is, taking the product of the three membership function values, representing the joint response strength of the sample to the overall premise conditions of the rule. Aggregate calculation of fuzzy support degree , is the fuzzy support degree of the rule, is the historical sample index, is the total number of historical samples, is the activation degree of the sample pair rule. Let the total number of historical samples be N, and the system averages the activation degrees of all sample pair rules to obtain the fuzzy support degree value of the rule. This support degree represents the average probability of the rule being effectively activated in all historical samples, that is, the mean value of the response degrees of all samples to the rule. To increase the comparability between different rules, the system normalizes the support degree values of all rules and maps them to the [0,1] interval.

[0107] Label the preliminary rule data according to the fuzzy support degree data to obtain the support degree rule data.

[0108] In one embodiment, based on the fuzzy support value of each rule obtained in the aforementioned steps, the system appends the value as an attribute annotation to the preliminary fuzzy rule set, forming a structured rule data set with support information, referred to as "support rule data". The annotation process includes the following key points: the system assigns a unique number to each rule and extracts its original premise combination and the corresponding fuzzy output label. The corresponding fuzzy support value is recorded as an additional field in the rule information to form a complete rule record unit. Each rule record includes at least the following field information, such as the rule number, premise combination (such as flow rate is "medium", resistance is "blocked", temperature is "normal"), fuzzy output label (such as "high profile" or "fine-tuning"), and support value. For example, rule R_05, whose preconditions are "medium" flow rate, "blocked" resistance, and "normal" temperature, and whose output label is "high-profile," has a corresponding fuzzy support value of 0.76. Rule R_09, whose preconditions are "medium" flow rate, "unobstructed" resistance, and "normal" temperature, and whose output label is "fine-tuning," has a support value of 0.48. The support value serves as an important indicator of the rule's applicability or credibility in real-world samples. If a rule's support value exceeds a preset threshold (for example, greater than or equal to 0.6), it indicates that the rule has a high activation frequency and representativeness in historical samples and is therefore preferred for retention. Conversely, if a rule's support value is significantly lower, it indicates insufficient sample coverage or overly narrow activation conditions. The system may consider merging, restructuring, or directly removing it.

[0109] Preferably, the asymmetric information entropy processing is specifically as follows:

[0110] Calculate the joint probability of premise and conclusion and the information entropy of conclusion for the preliminary rule data to obtain joint probability data and information entropy data of conclusion respectively;

[0111] In one embodiment, the joint distribution probability between the combination of the premise conditions of each rule and its conclusion (output label) is estimated. The historical fuzzy sample set is , each sample has a combination of preconditions , output label , statistics for each group Frequency of occurrence in the sample , calculate the joint probability , this value means that in all samples, the premise is and Conclusion Relative frequency of co-occurrence.

[0112] Evaluate the uncertainty of all conclusion labels (output fuzzy values). Based on the joint probability results, the system calculates each output label Specifically, for all premise combinations, Same output label The sum of the joint probabilities is: ; The system uses information entropy indicators for quantification, such as the information entropy calculation formula ,in, The entropy of the output label variable A is expressed as follows: a higher value indicates greater uncertainty. The marginal probabilities of the output labels are 0.10 for no adjustment, 0.35 for fine adjustment, 0.25 for medium adjustment, and 0.30 for high adjustment. Substituting this into the above formula yields an information entropy of approximately 1.89 for the output labels.

[0113] The premise information entropy is calculated based on the joint probability data and the conclusion information entropy data to obtain the premise information entropy data;

[0114] In one embodiment, the system calculates the conditional probability distribution corresponding to each output label based on the acquired premise-conclusion joint probability data and the marginal probability of the output label. For each output label (e.g., "no adjustment" or "high profile"), the relative probability of each premise-condition combination occurring under that label is calculated. The joint probability of a premise combination and a particular output label is divided by the marginal probability of that output label to obtain the conditional probability of that premise combination under that output label; the sum of all conditional probabilities is 1. The system performs an information entropy calculation operation on this conditional probability data. This operation processes each output label one by one, calculating the local information entropy formed by the probability distribution of all premise combinations under that label, and then performing a weighted sum to obtain an overall conditional information entropy value. A smaller value indicates a greater concentration of premise combinations under a particular output label and a more stable causal relationship in the system; conversely, a smaller value indicates a higher degree of premise uncertainty. For each output label, the information entropy of the premise combination distribution is calculated based on its conditional probability data. The local information entropy of all output labels is multiplied by the marginal probability of that label, and the final sum is calculated to obtain the overall conditional information entropy of the system. For example, the system output labels include "no adjustment," "fine adjustment," "medium adjustment," and "high adjustment." Each label category contains several premise combinations. After calculation, the system concludes that the overall conditional information entropy of this fuzzy rule system is 1.26. This value represents the average uncertainty of the premise distribution under all known output conditions.

[0115] Calculate information gain based on conclusion information entropy data and premise information entropy to obtain information gain data;

[0116] In one embodiment, the information gain is calculated as: ,in The ability of the information to interpret the conclusion as a prerequisite, that is, how much uncertainty is reduced, is the Shannon entropy of the conclusion variable, reflecting the original uncertainty of the conclusion itself, The information gain (IG) is the conditional entropy of the conclusion variable given the given premises, representing the residual uncertainty after the premises are given. A large information gain indicates that the premises have strong explanatory power and discriminative power for the conclusion variable, effectively constraining or predicting the conclusion, and the rule has high information value. A value close to zero indicates that the premises have little or no influence on the conclusion, the distribution of the conclusion does not change significantly under the premise constraints, and the rule is redundant or repetitive, lacking independent information contribution. The system calculates the corresponding information gain metric for each fuzzy rule, generating structured output data. Each record contains the rule number, conclusion entropy, conditional entropy, and the calculated information gain value. For example, for rule number "R_05," the conclusion variable entropy is 1.89, the conditional entropy is 1.26, and the calculated information gain is 0.63. This result indicates that the rule has moderate information discriminative power.

[0117] The preliminary rule data is labeled according to the information gain data to obtain the information rule data.

[0118] In one embodiment, the system uses an information gain scoring mechanism to rank each rule according to its information gain value. When the gain value is not less than 0.7, it is marked as a high information rule; when the gain value is between 0.3 and 0.7 (including the lower bound and excluding the upper bound), it is marked as a medium information rule; when the gain value is less than 0.3, it is marked as a low information rule. This classification strategy can effectively characterize the ability of rules to distinguish output results in fuzzy reasoning systems and provide a structured reference for rule importance assessment. In specific operations, the system adds informative annotation fields to each rule, including rule identifiers (such as Rule_05); corresponding gain values , in decimal form; informativeness level labels, used to indicate whether the rule is high, medium, or low information. The resulting annotation results are uniformly incorporated into the information rule dataset and can be written as additional attribute tags for rule nodes in the fuzzy rule knowledge graph for application scenarios such as graph reasoning path selection, rule fusion screening, and explanatory output.

[0119] Preferably, step S3 is specifically:

[0120] Step S31: Calculating the rule activation degree according to the fuzzy rule data to obtain the rule activation degree data;

[0121] In one embodiment, the current input sample is assumed to be a ternary parameter set, represented as ,in represents the flow rate input variable, is the current flow rate value; represents the resistance input variable, is the current resistance value; represents the temperature input variable, is the current temperature value. The system traverses each rule in the fuzzy rule data , calculate its activation degree. For example, the premise can be expressed as "if the flow rate is medium, the resistance is blocked, and the temperature is normal". Under this premise, the system obtains the membership value of the current input under the corresponding fuzzy language label from the membership function, which is recorded as, Indicates the current flow rate value membership under the “medium” language variable; Indicates the current resistance value Membership under the “blocking” language variable; Indicates the current temperature value Membership under "normal" language variables. Based on the above membership, the system uses one of the following two typical methods to calculate the rules Activation , such as the minimum method, which takes the smallest of the three membership values ​​as the activation ; Alternatively, use the product method, which is to multiply the three membership values This method is more suitable for continuous representations of the input space and can retain some matching information. The two methods above are selected based on the specific application scenario. In most continuous fuzzy control scenarios, the product method can achieve a smoother response. The system performs the above activation calculation on all fuzzy rules in sequence, outputting the activation value corresponding to each rule to form a rule activation dataset.

[0122] Step S32: performing language-type fuzzy reasoning based on the rule activation data to obtain fuzzy language variable data;

[0123] In one embodiment, the system determines the set of all fuzzy rules that are activated under the current input conditions, denoted as , where each rule ( The values ​​of are 1, 2, ... , which are the order items of the rules) have been calculated by calculating the corresponding activation , and the activation level is not lower than the set threshold. Each rule Corresponding to a fuzzy output language label, such as "fine-tuning", "medium-tuning" or "high-tuning". The system extracts the fuzzy membership function corresponding to each output label, which is recorded as ,in is the language label, and x is any value in the domain of the output variable. To reflect the clipping effect of the current activation strength on the output function, the system performs a lower limit clipping on the membership function. The clipping method is to clip each activation rule , whose output label is , the output membership function after clipping is recorded as , defined as follows , that is, based on the original membership function, the activation degree The upper limit is used and truncated to form a response function with a limited activation degree. This operation ensures that the intensity of the output response is strictly controlled by the activation degree, thereby maintaining the quantitative consistency between the system response and the input conditions. The system further classifies and integrates the pruned membership functions of all activation rules according to language labels to construct a linguistic fuzzy response set. Each response data includes the output language label (such as "high-key"); the activation value (such as 0.72); and the function value distribution of the pruned fuzzy membership function within the domain of the output variable (which can be represented as a vector or function graph).

[0124] Step S33: performing functional fuzzy reasoning on the fuzzy rule data to obtain functional fuzzy numerical data;

[0125] In one embodiment, all fuzzy rule sets activated under the current input conditions , the system has obtained the output membership function pruning result corresponding to each rule from the previous step, which is recorded as .in, The continuous variables in the domain of the output variable are normalized to the interval [0,1]. In order to integrate multiple clipped fuzzy output functions into a single composite control response function, the system uses the "maximum synthesis method" to perform function fusion. The specific method is to perform function fusion for any given value in the domain. , take the maximum value of the membership function clipped by all activation rules at this point as the value of the fusion output function at this point. It is expressed as ,in, is the index of the activation rule. A continuous fuzzy output contour curve is formed in the entire output definition domain, which is called the synthetic fuzzy control function graph. Its shape reflects the integrated effect of the output of each rule reasoning in the numerical dimension. The system approximates the above function in a discrete form. That is, the output variable definition domain x is divided into equal intervals (for example, every 0.1), and the corresponding x point is calculated. Function values ​​are then generated, forming a discrete fuzzy function representation structure. This functional fuzzy numerical data serves as the core input for the system output value calculation. Its structure has the following characteristics: it completely covers the normalized domain of the output variable; the function values ​​at each point satisfy the fusion principle after being clipped by fuzzy logic rules; and it can be directly fed into the defuzzification module to perform standard fuzzy control output calculations such as the centroid method.

[0126] Step S34: Perform double-layer interpretation and control encapsulation on the fuzzy language variable data and the functional fuzzy numerical data to obtain fuzzy inference data.

[0127] In one embodiment, the system integrates the following three types of information into a unified data structure: Rule activation data, which records the fuzzy rules activated under the current input conditions and their corresponding activation values. This part is used to present the rule matching status and reflect the scope and intensity of the system's response to the input. For example, the rule and are activated, with activation degrees of 0.72 and 0.65, respectively, representing the contribution of these two rules to reasoning under the current input. Linguistic fuzzy output data records the linguistic labels output by the activated rules and their trimmed membership function responses. This data is used for human-computer interaction displays, such as outputting fuzzy curves corresponding to labels like "high-profile" and "fine-tuning," supporting visual analysis and language interpretation. Each linguistic label is accompanied by a discrete representation of its fuzzy function, reflecting the membership values ​​at each point on the output domain. Functional fuzzy numerical output data records the composite fuzzy output function formed through the fusion of functional reasoning. This function is obtained by maximizing the trimmed values ​​of the output functions of all activated rules. It expresses the state of the fuzzy control response and can be directly used in defuzzification processes, such as centroid calculation, to derive numerical control results. The function is represented discretely, including sampling points on the output variable domain and the corresponding membership function values. By integrating and encapsulating these three types of data, the system forms a fuzzy reasoning data structure adapted for the "interpretation-control" dual channel. This structure can be used for numerical control processes within the system (such as defuzzification, feedback adjustment, etc.), as well as for scenarios such as user display, transparent interpretation of rules, and visualization of control behaviors outside the system.

[0128] Preferably, step S4 is specifically:

[0129] Step S41: performing interpretable centroid processing on the fuzzy reasoning data to obtain key reasoning path data;

[0130] In one embodiment, during the fuzzy rule inference phase, the system activates multiple rules based on the input conditions and fuses their corresponding output fuzzy sets into a continuous membership function graph. This membership function is defined within the domain of the output space (e.g., 0.0 to 1.0) and reflects the combined influence of different rules on the control output. The system discretizes this fused output function, sampling it across the domain using a fixed numerical step size (e.g., every 0.01), and calculates the numerical contribution of each sample point to the defuzzification process. This process primarily involves calculating the numerator (weighted integral term), multiplying the value of each sample point by its corresponding membership value. The sum of all products serves as the numerator, representing the weighted average of the output value. The denominator (normalization term) is calculated by summing the membership values ​​of all sample points, which serves as the denominator, reflecting the cumulative response strength of the overall membership function. The output centroid position is calculated by dividing the numerator by the denominator to obtain the "centroid position" of the control output, representing the expected solution for the fuzzy set. The system further performs a backtracking of the rule that plays a dominant role in shaping the final centroid output. The specific operation involves evaluating all participating fuzzy rules based on their membership activation and contribution to the fusion function. Rules are then sorted from highest to lowest in terms of contribution, and those whose cumulative contribution exceeds a preset threshold (e.g., 80%) are selected. For each selected rule, its preconditions (e.g., the value combination of the input linguistic variables), output label (e.g., "high adjustment," "medium adjustment," etc.), rule activation (i.e., the membership at which the rule is triggered), and its primary contribution interval in the fusion graph are recorded. For example, given the input conditions of "flow rate = medium," "resistance = blocked," and "temperature = normal," the system activates two fuzzy rules, R05 and R09, with corresponding output control values ​​of "high adjustment" and "medium adjustment," respectively. The centroid of the fused output function is located at 0.73. Based on the contribution calculation, the system finds that rule R05 contributes 61% to the output, and rule R09 contributes 27%. Therefore, the system identifies and records these two rules as the critical path for the current reasoning round.

[0131] Step S42: performing control output extraction on the key inference data to obtain control output data;

[0132] In one embodiment, the system calculates and extracts control output values ​​based on the obtained key reasoning path data and the fused membership function graph. The results of fuzzy reasoning are converted into clear control signals, and a traceable interpretation basis is established. The specific processing method is as follows: the system uses the centroid method as a standard fuzzy solution method. Based on the fused output membership function, it performs equidistant sampling within the output definition domain (0.00 to 1.00). Taking 100 equal divisions as an example, the system obtains the corresponding fuzzy membership value at each sampling point, forming a series of "sampling point-membership degree" data pairs. Based on this, the system performs the following numerical integration calculations: A weighted integral term (as the numerator) is calculated, and the value of each sampling point is multiplied by its corresponding membership value. The sum of all multiplication results is used to represent the expected position of the fuzzy function within the output interval; a membership integral term (as the denominator) is calculated, and the membership values ​​of all sampling points are accumulated to represent the overall response strength. The output control value is calculated by dividing the weighted integral term by the membership integral term to obtain the center of gravity of the fuzzy set, which is the numerical solution for the current control output. The system encapsulates the output control value as structured control data, which is used to drive the actual device control module. For example, this output value can serve as a control instruction for a pump control module or actuator, enabling flexible adjustment of the liquid flow rate. The system also retains information about key inference paths associated with this control output value, particularly the rule numbers with high contributions. This information is derived from the statistical results of the activated rule contributions in step S41. For example, under certain input conditions, the system's defuzzified control output value is 0.73, primarily determined by the contribution of the fused function graph formed by the rule paths P01 and P02. The system writes the output value 0.73 as the current control instruction into the control output data, simultaneously appending the rule paths P01 and P02 as interpretation basis. The resulting control output data not only contains the specific control value but also embeds information about the rule chain closely related to its generation logic. The center of mass position obtained in step S41 is primarily used for interpretability analysis of the inference path, while in step S42, it serves as the basis for the control signal output. This is further encapsulated as structured control data and used to drive specific control devices. The two are logically related but clearly functionally distinct.

[0133] Step S43: construct a control interpretation diagram based on the key reasoning path data and the control output data to obtain fuzzy control data.

[0134] In one embodiment, the system constructs a control interpretation graph (CEG) from the extracted key reasoning paths and the calculated control output values. The CEG is constructed using a graph structure, whose nodes include fuzzy input condition nodes, such as "Q_Medium," "R_Blocked," and "T_Normal"; activated rule nodes, such as "Rule_05" and "Rule_09"; fuzzy output label nodes, such as "A_High" and "A_Medium"; and control output nodes, such as "ControlValue_0.73." Semantic edge relationships are established between these nodes, with input condition nodes pointing to rule nodes (edge ​​type "Condition"); rule nodes pointing to output label nodes (edge ​​type "Conclusion"); and output label nodes pointing to control output nodes (edge ​​type "Contribution"). Each edge is assigned attributes, such as activation degree, contribution rate, and label coverage interval. The graph output is part of the fuzzy control data, used to display control paths and output sources. For example, in the CEG constructed by the system, nodes "Q_Medium" and "R_Blocked" are connected to "Rule_05" via edges. This node is then connected to "A_High" and points to the control result node "0.73." The edge annotations in the figure include "activation degree 0.72" and "contribution degree 61%".

[0135] Preferably, the interpretable centroid processing is specifically:

[0136] Perform tailoring and fusion based on fuzzy inference data to obtain fuzzy graphic data;

[0137] In one embodiment, the activated rule set is extracted. The fuzzy inference data contains all the fuzzy rule sets that are activated under the current input conditions, set as Each rule corresponds to an activation value, denoted as , the value range is [0,1], used to indicate the matching degree of the rule. Execute rule output extraction and trimming. For each activated rule , which corresponds to a fuzzy output set , which represents the membership function corresponding to the regulation behavior indicated by the conclusion of the rule. The fuzzy output set is pruned, that is, the original output membership function and activation degree are taken as smaller values ​​to obtain the pruned fuzzy output function. The pruned process is mathematically equivalent to performing the minimum operation on the original fuzzy set, that is, each output value The membership value at is defined as the sum of the original membership and activation The system synthesizes all the clipped fuzzy output functions and fuses them using the maximum rule, that is, for the same output value , selecting the maximum trimmed membership value of all rules at that point as the membership value of the output fuzzy graph. This operation essentially models the combined effect of multiple activated rules on the output. The fused fuzzy graph is discretized to construct a fuzzy graph data structure. This data structure is typically represented as a two-dimensional array, where one dimension is the sequence of output variable values ​​(e.g., equally spaced sampling points from 0 to 1) and the other dimension is the membership value corresponding to each sampling point.

[0138] Perform path tracing integration on the fuzzy graphic data to obtain the output center of gravity data;

[0139] In one embodiment, an integration window is defined, and the system performs an integration operation within the domain of the output variable. The value range of the output variable is from 0 to 1, that is, the output domain is a closed interval [0,1]. The fused membership function has been defined in this interval and is recorded as the output membership function. The centroid value calculation is performed, and the system uses the centroid method (also known as the centroid method of fuzzy sets) to determine the representative output value of the fuzzy graph. The output variable and its corresponding membership function value are weightedly integrated. The numerator represents the weighted position sum, and the denominator represents the total intensity of the fuzzy graph. The centroid value is obtained by dividing the two. Specifically, the centroid value is equal to the integral of the product of the output variable value and its membership function in the interval from 0 to 1, divided by the integral of the membership function itself. The system uses numerical integration approximation methods, such as trapezoidal integration or rectangular integration. The output variable interval is divided into several small intervals of equal width, and the center point of each interval is , is the center point order item, and the upper limit is the number of center points , and its corresponding membership degree is , and use this to construct the center of mass The approximate calculation formula of , by calculating all small intervals In the example, the weighted sum of the products of the variable values ​​and the membership degrees is divided by the sum of the membership degrees to obtain an approximate centroid output value. The system encapsulates this calculated centroid value as structured output data, which represents the control strength of the current fuzzy inference. This value is represented as a floating-point number and falls within a predefined output range. For example, a centroid value of 0.72 indicates that the system should output a control instruction or response value with a strength of 0.72 under the current input conditions.

[0140] Calculate the contribution interval based on the output center of gravity data to obtain the contribution interval data;

[0141] In one embodiment, the system defines a rule contribution function to measure the contribution of each rule in the fuzzy output graph. , its output trimming membership function and the final fuzzy output graph have a certain overlap in the output variable interval. The system estimates the contribution value of the rule by calculating the membership coverage in the overlapping area. The contribution is expressed as the integration of the product of the rule trimming output function and the output graph membership function in the output variable interval. The result represents the rule Weighted coverage of the output graph. Sort by contribution value of all rules and select the rules with the highest cumulative contribution value as the key rule set. Select the top 80% of the rules with the highest cumulative contribution as the main influencing factors. This ratio can be adjusted according to the actual system requirements. The system records the input variable interval corresponding to the prerequisite of each key rule to form the input trigger interval. For example, for the rule If the fuzzy preconditions are that the flow rate is within the "medium" linguistic value range (e.g., [0.4, 0.6]), the resistance is in a "blocked" state (e.g., [0.7, 1.0]), and the temperature is "normal" (e.g., [0.5, 0.9]), then this input condition corresponds to a three-dimensional input interval. The output trimming function of this rule is concentrated within a specific output range (e.g., [0.6, 0.9]). The system also records this output range and the contribution value calculated above. The system encapsulates this structured information into contribution interval data, which includes the rule number, input trigger interval, output response interval, and corresponding contribution value.

[0142] The reasoning paths of fuzzy reasoning data are aggregated according to the contribution interval data to obtain the key reasoning path data.

[0143] In one embodiment, the system performs path aggregation processing on the high-contribution rules involved in the fuzzy reasoning process based on the contribution interval data, aiming to extract the most representative reasoning paths to enhance the interpretability and reconstructability of the system decision-making process. By performing path aggregation logic processing, the system identifies all fuzzy rules with high contribution values ​​and constructs path segments based on the premise conditions of these rules. Each path segment consists of a set of fuzzy language values ​​of input variables, connected to the corresponding fuzzy output labels, forming a "conditional For example, if the premise of a rule is that the flow rate is "medium", the resistance is "blocked", and the temperature is "normal", and its output is "high", then the path segment is represented as: "Q=medium, R=blocked, T=normal A=high profile". Path structure encapsulation is performed, that is, the system encapsulates the premise conditions, activation values, output contribution values, and centroid values ​​of the fuzzy output corresponding to each path segment into a structured path vector. This path vector is used to represent a complete fuzzy reasoning link and its response strength and output value in the current sample. For example, the path vector contains the path number, the condition set (that is, the premise variables and their fuzzy values), the activation degree of the path (indicating the degree of match with the current input), the output contribution (indicating the degree of influence of the path on the output), and the output control value (such as the numerical result calculated by the centroid method). The system sorts all constructed path vectors in descending order according to their contribution, and selects the top K paths as key reasoning paths based on the set path number upper limit parameter K. This path set represents the most influential reasoning path under the current input conditions and is the main source of system execution results.

[0144] Preferably, the tailoring and fusion are specifically:

[0145] Calculate the activation degree based on the fuzzy reasoning data to obtain the rule activation degree data;

[0146] In one embodiment, the input fuzzy variable set is set to , each rule Definition format: IF Q = Medium AND R = Blocked AND T = Normal THENA = High, activation calculation method (using minimum method or product method), including minimum method (commonly used): , or, the product method (smoother): For example, if the membership values ​​of the current input sample under "Q is medium", "R is blocked", and "T is normal" are 0.9, 0.8, and 0.7 respectively, then if the minimum value method is used, the rule The activation degree is 0.7; if the product method is used, the activation degree is 0.9 × 0.8 × 0.7 = 0.504. The system associates the activation calculation result of each rule with its premise condition to form structured response data. The record content includes the rule number; the fuzzy language value corresponding to the premise condition; and the activation value for the current sample.

[0147] Output extraction is performed based on fuzzy reasoning data to obtain rule output data, wherein the output extraction is based on the activated rule set in the fuzzy reasoning data, and the fuzzy language variables and membership function expressions of each rule output are extracted;

[0148] In one embodiment, the system filters the set of fuzzy rules activated in the current reasoning round. Only rules with an activation degree greater than zero (i.e., an activation degree greater than zero) are considered. >0) to perform output extraction operations. Each fuzzy rule is represented by an "IF...THEN..." structure, and the output item is the fuzzy language variable specified in the THEN clause. For example, the rule The output item is "Adjustment degree equals high-profile," where "high-profile" is the output label of the rule. This label points to a fuzzy set that describes the degree of control the system should perform under this condition. The system extracts and models the membership function corresponding to the output label. Membership functions can be defined using triangular, trapezoidal, or Gaussian functions. Taking the fuzzy output variable "high-profile" as an example, its membership function can be expressed using a trapezoidal membership function. This function defines fuzzy membership within the numerical range [0.6, 1.0] and behaves as follows: below 0.6, the membership is zero; between 0.6 and 0.7, the membership increases linearly from zero to its maximum value; between 0.7 and 0.9, the membership remains at 1, indicating full membership; between 0.9 and 1.0, the membership gradually decreases linearly to zero; and beyond 1.0, the membership returns to zero. This definition clearly expresses the degree of adaptability of the "high-profile" control output within different numerical ranges. The output of each activation rule consists of the following elements: the rule number, which is used to identify the rule; the output label, which is the control language variable recommended by the rule; the type of output membership function (such as trapezoidal function); and the key numerical range defined by the function, for example, [0.6, 0.7, 0.9, 1.0] represents the starting, rising, plateau, and falling endpoints of the trapezoidal function.

[0149] Minimize the rule output data according to the rule activation data to obtain rule-cut data;

[0150] In one embodiment, it is assumed that the output fuzzy set of a fuzzy rule is represented by a membership function, which provides the membership degree corresponding to each value in the definition domain of the control output variable, and is recorded as the original output membership function The function reaches a maximum membership value of 1.0 within a certain range, indicating complete membership. At the same time, the activation degree of the rule in the current reasoning round is recorded as , whose value range is 0 to 1, is used to indicate the matching degree of the rule premise. Specifically, at each value point in the domain of the output variable Above, take the original membership and activation The smaller of is taken as the trimmed membership degree, that is, if Less than or equal to , then the membership degree of the point remains unchanged; if Greater than , then truncate it to This operation involves horizontally truncating the original membership function graph so that its maximum membership does not exceed the current rule's activation level, thereby forming a "clipped" output fuzzy set. For example, if the output fuzzy label of a rule is "high-profile," its original membership function exhibits a trapezoidal shape within the domain of the control output variable [0.6, 1.0], with a membership of 1.0 between 0.7 and 0.9. However, the activation of this rule under the current input is 0.72. After clipping, the function's membership in the original 1.0 interval is truncated to 0.72, while the membership values ​​in the remaining intervals remain the same or decrease, forming a new clipped membership function. After minimum clipping, each rule's output function is redefined as a "clipped output function," which has two core components: a discrete partitioning of the domain of the control output variable (e.g., equally spaced points between 0.6 and 1.0); and the clipped membership values ​​corresponding to each value point. Through this structure, the system can accurately represent the range of response strengths of each rule to the control output under the current input conditions.

[0151] The regular clipping data is aggregated and fused to obtain fuzzy graphic data.

[0152] In one embodiment, the system performs a point-by-point maximum fusion operation within the domain of the output variable.

[0153] Specifically, at each point within the output interval, the system compares the trimmed membership function values ​​of all activated rules and takes the maximum value as the membership value for that point. To achieve this, the system divides the range of the control output variable (for example, from 0.0 to 1.0) into a number of equally spaced sampling points. A common approach is to divide the range into intervals of 0.01, thus dividing the output interval into 101 sampling points. For each sampling point within the domain, the system performs the following processing: obtaining the trimmed membership value of each activated rule at that point; comparing the trimmed values ​​of all rules at that point and taking the maximum value as the fused membership value for that point; and combining the fused results at all points to form a fuzzy output function graph. After fusion is complete, the system forms a data structure consisting of a "sampling point sequence" and a "corresponding fused membership sequence," referred to as fuzzy graph data. This data can be viewed as a numerical mapping of the fuzzy control output within a continuous interval, with a clear functional meaning. The fused membership function can be plotted as a fuzzy output graph, which displays the fuzzy response intensity corresponding to each value of the control variable.

[0154] Preferably, the present application further provides a fuzzy algorithm-based intravenous catheter dynamic flow regulation and anti-blocking system, which is used to execute the fuzzy algorithm-based intravenous catheter dynamic flow regulation and anti-blocking method as described above. The fuzzy algorithm-based intravenous catheter dynamic flow regulation and anti-blocking system includes:

[0155] A fuzzy variable generation module is used to obtain historical parameter data of the intravenous catheter system and construct fuzzy input variables for the historical parameter data to obtain historical variable fuzzy data;

[0156] A fuzzy rule construction module is used to construct fuzzy rules based on historical variable fuzzy data to obtain fuzzy rule data;

[0157] A fuzzy reasoning engine module is used to perform fuzzy reasoning on fuzzy rule data to obtain fuzzy reasoning data;

[0158] The interpretable defuzzification module is used to perform interpretable defuzzification on fuzzy inference data to obtain fuzzy control data.

[0159] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes that fall within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0160] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for dynamic flow regulation and anti-blocking of intravenous catheter based on fuzzy algorithm, characterized in that: The following steps are involved: Step S1: acquiring historical parameter data of the intravenous catheter system, and constructing fuzzy input variables for the historical parameter data to obtain historical variable fuzzy data; Step S2: Matching the historical variable fuzzy data with the preset fuzzy rule template library to obtain preliminary rule data; performing hypersurface support processing and asymmetric information entropy processing on the preliminary rule data to obtain support rule data and informative rule data respectively; According to the support rule data and the information rule data, the preliminary rule data is corrected for rule conflicts to obtain rule correction data; the rule correction data is encapsulated in a knowledge graph structure to obtain fuzzy rule data; Step S3: performing fuzzy reasoning on the fuzzy rule data to obtain fuzzy reasoning data; Step S4: performing interpretable defuzzification on the fuzzy inference data to obtain fuzzy control data; The hypersurface support processing is specifically as follows: Constructing a rule premise hypersurface based on preliminary rule data to obtain rule premise hypersurface data; Fuzzy support data is obtained by calculating fuzzy support based on historical variable fuzzy data and rule premise hypersurface data; The preliminary rule data is labeled according to the fuzzy support data to obtain the support rule data; The specific processing of asymmetric information entropy is as follows: Calculate the joint probability of premise and conclusion and the information entropy of conclusion for the preliminary rule data to obtain joint probability data and information entropy data of conclusion respectively; The premise information entropy is calculated based on the joint probability data and the conclusion information entropy data to obtain the premise information entropy data; Calculate information gain based on conclusion information entropy data and premise information entropy to obtain information gain data; The preliminary rule data is labeled according to the information gain data to obtain the information rule data.

2. The method according to claim 1, characterized in that Step S1 is specifically as follows: Obtain historical parameter data of intravenous catheter system; Perform event sequence extraction on historical parameter data to obtain event sequence data; Perform parameter repair on event sequence data to obtain input parameter data; Perform input dimension screening based on input parameter data to obtain input screening data; Fuzzy linguistic variables are constructed for the input screening data to obtain historical variable fuzzy data.

3. The method according to claim 1, characterized in that Step S3 is specifically as follows: Calculate the rule activation degree according to the fuzzy rule data to obtain the rule activation degree data; Perform language-type fuzzy reasoning based on rule activation data to obtain fuzzy language variable data; Perform functional fuzzy reasoning on fuzzy rule data to obtain functional fuzzy numerical data; The fuzzy language variable data and the functional fuzzy numerical data are double-encapsulated by interpretation and control to obtain fuzzy reasoning data.

4. The method according to claim 1, wherein Step S4 is specifically as follows: Perform interpretable centroid processing on fuzzy reasoning data to obtain key reasoning path data; Perform control output extraction on key reasoning data to obtain control output data; A control interpretation diagram is constructed based on the key reasoning path data and control output data to obtain fuzzy control data.

5. The method according to claim 4, characterized in that The specific processing of interpretable centroid is as follows: Perform tailoring and fusion based on fuzzy inference data to obtain fuzzy graphic data; Perform path tracing integration on the fuzzy graphic data to obtain the output center of gravity data; Calculate the contribution interval based on the output center of gravity data to obtain the contribution interval data; The reasoning paths of fuzzy reasoning data are aggregated according to the contribution interval data to obtain the key reasoning path data.

6. The method according to claim 5, characterized in that The specific details of cutting and fusion are: Calculate the activation degree based on the fuzzy reasoning data to obtain the rule activation degree data; Output extraction is performed based on fuzzy reasoning data to obtain rule output data, wherein the output extraction is based on the activated rule set in the fuzzy reasoning data, and the fuzzy language variables and membership function expressions of each rule output are extracted; Minimize the rule output data according to the rule activation data to obtain rule-cut data; The regular clipping data is aggregated and fused to obtain fuzzy graphic data.

7. A fuzzy algorithm-based intravenous catheter dynamic flow regulation and anti-blocking system, characterized in that: Used to execute the method for dynamic flow regulation and anti-blocking of intravenous indwelling needles based on fuzzy algorithm as claimed in claim 1, the dynamic flow regulation and anti-blocking system for intravenous indwelling needles based on fuzzy algorithm comprises: A fuzzy variable generation module is used to obtain historical parameter data of the intravenous catheter system and construct fuzzy input variables for the historical parameter data to obtain historical variable fuzzy data; A fuzzy rule construction module is used to construct fuzzy rules based on historical variable fuzzy data to obtain fuzzy rule data; A fuzzy reasoning engine module is used to perform fuzzy reasoning on fuzzy rule data to obtain fuzzy reasoning data; The interpretable defuzzification module is used to perform interpretable defuzzification on fuzzy inference data to obtain fuzzy control data.

Images