Control method and system based on offline fuzzy judgment

Abstract

The invention discloses a control method and system based on offline fuzzy judgment, and relates to the technical field of electric digital data processing. Comprising the steps of collecting physical quantity data of a target scene through a data collection module; performing integration processing on the collected data according to a preset scene rule to generate prompt words; performing fuzzy decision operation on the cue words by using the off-line large model, and outputting control suggestions; and converting the control suggestions into specific instructions by utilizing semantic analysis and an instruction conversion rule base, executing the specific instructions, feeding back actual state data, and optimizing parameters according to the fed-back data. The problems that in the prior art, data leakage risks exist, multi-source data fusion is difficult, and analysis depends on manual parameters are solved, and the purposes that the control precision is high, the system adaptability is enhanced, stability and independence are improved, and expandability and compatibility are good are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of calculation, estimation or counting, and in particular to electronic digital data processing. Background Art

[0002] In the field of automated control, traditional control systems are widely used in numerous industries, including industrial production, smart home appliances, and transportation. These systems mostly execute control operations based on clear logical rules and simple threshold judgments. For example, in industrial temperature control systems, cooling is activated when the temperature exceeds an upper threshold, while heating is activated when the temperature falls below a lower threshold. In intelligent building lighting control, lights are fully turned on when the light intensity falls below a preset value and turned off when it exceeds. In intelligent traffic light control, the green light duration is extended when traffic volume reaches a threshold.

[0003] However, this all-or-nothing control approach presents numerous practical challenges. In industrial production, chemical reactions require control of parameters like temperature and pressure that go beyond simple on-off states. Chemical reaction rates and product quality can remain stable within a wide range of parameters. Traditional control approaches, however, require drastic adjustments when parameters deviate from set values, leading to significant fluctuations in the reaction process, impacting product quality stability, and increasing energy consumption and equipment wear. For example, in fine chemical production, temperature control requires extremely high precision. Temperature fluctuations with traditional control approaches can reduce product purity and increase defective rates. In intelligent building environmental control, traditional control methods fail to meet the demands for personalization and comfort. Different activity scenarios have varying requirements for environmental parameters like lighting and temperature, and people have varying degrees of sensitivity. Traditional lighting control methods lack the flexibility to adapt to actual needs. For example, conference rooms require varying light intensity in different scenarios (e.g., meetings, training sessions, and rest periods), making traditional control difficult to meet. In the intelligent transportation sector, traffic flow varies in complex ways, and traditional signal control methods can exacerbate congestion. Traffic demand in different directions and time periods is not a simple binary state. The control method of fixed time intervals or simple traffic flow detection cannot be flexibly optimized and adjusted according to real-time road conditions. For example, traffic congestion is more prominent during special periods such as morning and evening rush hours and holidays.

[0004] For example, the Chinese patent with publication number CN118643291A discloses a data analysis method and system based on an offline large model, and provides the following technical solutions: The present invention provides a data analysis method and system based on an offline large model, the method comprising: a data layer for data collection, data preprocessing and data persistent storage; a data analysis layer combining big data technology, using an artificial intelligence large language model to complete model construction and model training in an offline environment; and applying the trained model to perform data analysis on new data; a visualization layer for user interaction, displaying the result data to the user, and also supporting user custom configuration. The present invention continuously trains the large model with continuously provided data, performs refined data preprocessing and feature extraction, and integrates efficient algorithms, and the analysis results will become more and more accurate. It can also achieve physical isolation, and by enhancing the interpretability of the model, the analysis results are more intuitive and easy to understand. The learning efficiency of AI ecological technology is very high, and the accuracy of the data will continue to increase with the length of model learning, without over-reliance on manual labor. However, the above-mentioned data analysis method and system based on offline large models are limited to the control field in application scenarios, lack multi-source data processing capabilities, and do not have data security measures and model continuous optimization mechanisms, and cannot solve complex data fusion and privacy protection problems. Summary of the Invention

[0005] The present invention solves the problems of data leakage risk, difficulty in multi-source data fusion, and analysis reliance on manual parameters in the existing technology, and proposes a control method and system based on offline fuzzy judgment, which achieves the goals of high control accuracy, enhanced system adaptability, improved stability and independence, and good scalability and compatibility.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A control method based on offline fuzzy judgment includes the following steps: S1: Collect physical quantity data of the target scene through the data acquisition module; S2: Integrate and process the collected data according to the preset scenario rules to generate prompt words; S3: Use the offline large model to perform fuzzy decision operations on the prompt words and output control suggestions; S4: Use semantic analysis and the instruction conversion rule library to convert control suggestions into specific instructions and execute them, feedback the actual status data, and optimize parameters based on the feedback data.

[0007] A control system based on offline fuzzy judgment, comprising: The data processing and prompt word generation module normalizes and calculates weights on the collected data, and generates prompt words according to preset rules; The offline large model operation module receives prompt words from the data processing and prompt word generation module, performs fuzzy logic operations based on the Transformer architecture, and outputs control suggestions; The semantic translation module converts the control suggestions sent by the offline large model operation module into instructions that can be recognized by the device; The control execution module receives instructions from the semantic translation module, drives the control device to execute the instructions and form a closed-loop feedback.

[0008] A full-process control method from data acquisition to closed-loop optimization is disclosed, and system stability and dynamic optimization of control strategies are achieved through modular design and feedback mechanism.

[0009] Preferably, the step S2 specifically normalizes and calculates weights on the collected data based on preset scenario rules, eliminates the dimensional influence, and generates scenario prompt words containing key parameters, wherein the scenario prompt words include parameter values ​​and dimensions as well as scenario stage identification information.

[0010] Dimensional differences are effectively eliminated through data normalization and weight calculation, and structured prompt words are generated in combination with scenario rules, which improves the standardization of large model input data and scenario adaptability.

[0011] Preferably, step S3 specifically includes adopting an offline large model of the Transformer architecture, which, after receiving the prompt word, converts the characters or numbers in the prompt word into vector representations through an embedding layer, then captures the long-distance dependencies between data through a multi-head attention mechanism, performs nonlinear transformation and feature extraction in a multi-layer perceptron, and finally makes a multi-parameter comprehensive decision based on the membership function, and outputs a control recommendation based on fuzzy logic.

[0012] Through the multi-level feature extraction capabilities of the Transformer architecture and combined with the fuzzy logic decision-making mechanism, nonlinear relationship modeling of multi-parameter collaboration in complex control scenarios is achieved.

[0013] Preferably, the step S4 further includes, after conversion into a specific instruction, entering an anomaly detection module to detect whether there is a conflict, and if there is a conflict, regenerating the specific instruction through a conflict resolution strategy library.

[0014] Preferably, the step S4 specifically includes, for different types of control devices, performing word segmentation processing on the text according to its built-in corresponding semantic analysis, judging the type according to the control protocol rule library, and selecting the corresponding instruction conversion rule according to the type to convert it into a specific instruction.

[0015] Through the protocol rule library and dynamic instruction conversion mechanism, accurate mapping of abstract strategies to specific device instructions is achieved, significantly improving the system's compatibility with different types of execution terminals.

[0016] Preferably, in step S4, for the DC motor, the offline large model outputs a speed range and a tendency value, and according to the motor PWM protocol, the speed range and the tendency value are converted into duty cycle parameters of a pulse width modulation signal and relevant instructions are generated; for lighting control, the offline large model outputs a brightness range and a tendency value, and according to the I2C protocol of the lighting controller, they are converted into digital dimming signal instructions including a device address, a brightness control command, and a brightness value.

[0017] Through the specific conversion rules of the device protocol, the fuzzy control suggestions are converted into executable instructions, which not only retains the strategy flexibility but also meets the strict requirements of industrial control for instruction determinism.

[0018] Preferably, the step S4 also includes receiving control instructions from the semantic translation module, driving the corresponding control device to perform operations, and feeding back the execution results to the data acquisition module to form a closed-loop control, specifically including: the motor adjusting the speed according to the PWM signal, the light changing the brightness according to the dimming signal, the valve adjusting the opening according to the control instruction, and the feedback sensor on the control device feeding back the actual execution status data to the data acquisition module. The system optimizes the parameters according to the feedback data and adjusts the control strategy in real time.

[0019] A closed-loop control loop is formed through sensor feedback, which enables the system to have dynamic parameter adjustment capabilities and effectively deal with control deviations caused by environmental variable disturbances.

[0020] Preferably, the data processing and prompt word generation module is connected to the data acquisition module, which collects physical quantity data through a variety of sensors and converts them into electrical signals or digital signals. The data acquisition module includes temperature and pressure sensors, flow sensors, light sensors, and temperature and humidity sensors, and transmits data through wired or wireless communication.

[0021] Through the support of multimodal sensor fusion and heterogeneous communication methods, the system's data acquisition stability in complex industrial environments is enhanced, providing a comprehensive data foundation for subsequent processing.

[0022] Preferably, the semantic translation module has a built-in protocol rule library to support command conversion between motors, lights and valve devices; the control execution module includes motor and light controllers and valves, and forms a closed-loop control link through feedback sensors; the offline large model operation module deploys a small-bit version of the offline large model with a specific parameter scale that has been quantized.

[0023] Through the extended design of the protocol rule library, the system can quickly connect to equipment from different manufacturers, and at the same time use feedback sensors to verify control accuracy, forming a complete control quality assurance system.

[0024] Compared with the prior art, the present invention has the following beneficial effects.

[0025] 1. Leveraging the powerful computing and analytical capabilities of large models, this system deeply mines collected data and applies fuzzy judgments to deliver more precise control ranges and strategies. In one chemical reaction, traditional control methods resulted in a temperature fluctuation range of ±5°C and a defective rate of up to 15%. However, this system, through comprehensive analysis of multiple parameters, narrowed the temperature fluctuation range to ±1°C, reducing the defective rate to 5%, significantly improving product quality stability.

[0026] 2. Whether in industrial production, smart buildings, or intelligent transportation, this system integrates scenario-specific data and sets rules to generate effective prompts for large-scale model operations. In intelligent building environmental control, lighting, temperature, humidity, and other environmental parameters can be flexibly adjusted based on time periods, regional activities, and user preferences, meeting diverse environmental needs and improving user comfort and satisfaction. Compared to traditional fixed-mode control systems, this system can adapt to scenarios at least 3-5 times more complex, greatly expanding its application scope.

[0027] 3. This system avoids over-regulation through precise control and optimization strategies. In industrial production equipment and intelligent building HVAC systems, it can precisely control equipment operating status based on actual needs to achieve energy savings. Field testing has shown that, under the same production scale or building usage conditions, energy consumption can be reduced by 15%-25%, effectively improving energy efficiency and aligning with the concept of sustainable development.

[0028] 4. By using an offline, large-scale model with a small bitrate, the system eliminates network dependency, enhancing stability and independence. This system can operate stably even in scenarios with poor network coverage or extremely high security requirements, such as control systems within remote industrial plants and military facilities, ensuring that the control process is not disrupted by network factors, thereby improving system reliability and availability.

[0029] 5. For different types of control devices and emerging application scenarios, it is easy to adapt by simply adjusting the corresponding rules and protocols in the data processing and prompt word generation module and the semantic translation module, which facilitates widespread promotion and application in various fields and has high practical value and market potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of a control method based on offline fuzzy judgment of the present invention.

[0031] Figure 2 This is a hardware block diagram of a control system based on offline fuzzy judgment of the present invention. DETAILED DESCRIPTION

[0032] To make the objectives, technical solutions, and advantages of the present disclosure more apparent, embodiments of the present disclosure are described in further detail below with reference to the accompanying drawings. The proportions of the components herein are not drawn to scale, and the proportions and dimensions shown in the accompanying drawings are not intended to limit the essential technical solutions of the present disclosure. These embodiments do not describe all details in detail, nor do they limit the present disclosure to the specific embodiments described.

[0033] See also Figure 1-2 As shown, a control method based on offline fuzzy judgment includes the following steps: S1: Collect physical quantity data of the target scene through the data acquisition module; S2: Integrate and process the collected data according to the preset scenario rules to generate prompt words; S3: Use the offline large model to perform fuzzy decision operations on the prompt words and output control suggestions; S4: Use semantic analysis and the instruction conversion rule library to convert control suggestions into specific instructions and execute them, feedback the actual status data, and optimize parameters based on the feedback data.

[0034] A control system based on offline fuzzy judgment, comprising: The data processing and prompt word generation module normalizes and calculates weights on the collected data, and generates prompt words according to preset rules; The offline large model operation module receives prompt words from the data processing and prompt word generation module, performs fuzzy logic operations based on the Transformer architecture, and outputs control suggestions; The semantic translation module converts the control suggestions sent by the offline large model operation module into instructions that can be recognized by the device; The control execution module receives instructions from the semantic translation module, drives the control device to execute the instructions and form a closed-loop feedback.

[0035] To address the problem that traditional control systems produce overly absolute control results, the present invention makes the following innovative improvements: a small-bit version of an offline large model with 8-bit quantization and a parameter size of less than 1 billion is deployed in the controller. This version is based on the Transformer architecture, avoids dependence on real-time network connections, improves system independence and stability, and reduces hardware resource requirements.

[0036] In specific scenarios, the system normalizes process data based on principles such as chemical reaction kinetics and thermodynamics. It then integrates various data types, such as temperature, pressure, and flow, into prompt words according to the dynamic rule of "key parameters first, secondary parameters last, combined with reaction stage identifiers." For example, in a chemical production scenario, for a specific chemical reaction, the weights of parameters such as temperature, pressure, and flow are first determined based on its reaction kinetics equation. The data is then normalized to eliminate dimensional effects, and prompt words are generated according to the rules. In intelligent transportation scenarios, data such as traffic volume, speed, and signal light status are integrated to generate prompt words based on traffic flow theory and real-time road condition optimization principles.

[0037] The generated prompt words are fed into the large model for computation. Based on deep learning and massive data training, the large model converts the characters or numbers in the prompt words into vector representations through an embedding layer. A multi-head attention mechanism then captures long-range dependencies between data. A multi-layer perceptron performs nonlinear transformations and feature extraction. Finally, a membership function is used to make a comprehensive multi-parameter decision, outputting fuzzy logic-based control recommendations, such as the appropriate temperature range and equipment operating status trends.

[0038] After the large model is output, it is converted into specific instructions that the control device can recognize through the semantic translation module. This module uses a conversion algorithm based on "text segmentation → keyword extraction → protocol matching" to address the differences in interface protocols and instruction formats among different types of control devices, such as motors, lights, and valves. This module converts abstract control strategies or parameter ranges into specific operational instructions. For example, it generates duty cycle instructions for motor PWM protocols and digital dimming signal instructions for lighting controllers using the I2C protocol. These instructions are then encoded according to the device's communication protocol.

[0039] Through the close collaboration of these components—the data integration module generates scenario-based prompts, the offline large model outputs fuzzy control strategies based on these prompts, and the semantic translation module converts these strategies into device instructions—this solves the problem of absolute control results in traditional control systems, meets the actual control requirements for control ranges, and achieves flexible, precise, and efficient control. Compared to existing technologies, this invention is not a simple superposition of technical solutions. Instead, it organically combines various modules to form a complete chain of "data acquisition → feature fusion → fuzzy decision-making → instruction adaptation," achieving a qualitative technological advancement.

[0040] like Figure 1 In one embodiment shown, Figure 1This is a flow chart of a control method based on offline fuzzy judgment of the present invention. The entire process of the present invention is as follows: First, the physical quantity data of the target scene is collected through the data acquisition module; then, according to the preset scene rules, the data is normalized and weighted, and after eliminating the dimension effect, a scenario prompt word containing parameter value, dimension and scene stage identification information is generated. Then, an offline large model based on the Transformer architecture is used to convert the prompt word into a vector through the embedding layer. The long-distance dependency between the data is captured through the multi-head attention mechanism. Nonlinear transformation and feature extraction are performed in the multi-layer perceptron. Finally, a multi-parameter comprehensive decision is made based on the membership function, and fuzzy logic control suggestions are output. Subsequently, for different control devices, the text is segmented through semantic analysis, and the corresponding conversion rules are selected according to the control protocol rule library: for example, for DC motors, the speed range and tendency value are converted into the duty cycle parameters of the pulse width modulation signal according to the PWM protocol; for lighting control, a digital dimming signal instruction containing the device address, brightness control command and brightness value is generated according to the I2C protocol. Finally, the drive control device performs operations such as motor speed regulation, light brightness adjustment, and valve opening adjustment. At the same time, the actual status data is transmitted back to the data acquisition module through the feedback sensor on the device, forming a closed-loop control and optimizing the parameter adjustment strategy in real time.

[0041] In one embodiment, when the step of converting control suggestions into specific instructions using semantic analysis and a command conversion rule library proceeds, the system receives voice or image input—if it's voice, it's processed by the ASR speech recognition module; if it's an image, it's analyzed using OCR / visual technology. This processed information then enters the semantic translation module for intent understanding, and the large model generates preliminary instructions. This then enters the core process: the anomaly detection module checks for conflicts within the instructions. If a conflict is found, it invokes the conflict resolution strategy library to regenerate a compliant instruction. If there is no conflict, the instructions are directly sent to the command queue manager. The command queue manager then distributes the verified instructions in parallel, generating motor control instructions through the motor command conversion module (with an energy optimization algorithm) and lighting control instructions through the lighting command conversion module. Ultimately, the optimized motor instructions drive the motor devices, and the synchronously generated lighting instructions control the lighting system, completing a complete closed-loop process from multi-modal input to multi-device coordinated response. It's particularly important to note that during the anomaly detection phase after conversion into specific instructions, the system iterates through detection and conflict resolution until the instructions are fully compliant. This mechanism effectively ensures instruction security and execution reliability.

[0042] The present invention ensures that input information accurately reflects the actual scene status through data collection and scenario-based prompt word generation, and realizes fuzzy logic decision-making independent of the network with the help of a large offline model based on Transformer. The abstract strategy is converted into device instructions in combination with the protocol adaptation capability of the semantic translation module, forming a complete control chain of "data closed loop-fuzzy decision-instruction generation-execution feedback". While improving the control accuracy, the offline model quantization processing reduces hardware resource consumption and enhances system stability. Its multimodal input fusion and anomaly detection mechanism further expands the compatibility of system application scenarios and ensures the safety and reliability of multi-device collaborative control. The instruction conversion algorithm based on the protocol rule library effectively solves the adaptation problem of traditional control instructions and device interfaces, and ultimately achieves the comprehensive optimization effect of reducing energy consumption and improving the complexity of scenario adaptation.

[0043] like Figure 2 In one embodiment shown, Figure 2 This is a hardware block diagram of a control system based on offline fuzzy judgment of the present invention. A control system based on offline fuzzy judgment includes: The data processing and prompt word generation module normalizes and calculates weights on the collected data, and generates prompt words according to preset rules; The offline large model operation module receives prompt words from the data processing and prompt word generation module, performs fuzzy logic operations based on the Transformer architecture, and outputs control suggestions; The semantic translation module converts the control suggestions sent by the offline large model operation module into instructions that can be recognized by the device; The control execution module receives instructions from the semantic translation module, drives the control device to execute the instructions and form a closed-loop feedback.

[0044] The data processing and prompt word generation module is connected to the data acquisition module. The data acquisition module collects physical data through various sensors and converts them into electrical or digital signals. These sensors include temperature and pressure sensors, flow sensors, light sensors, and temperature and humidity sensors, and transmit data via wired or wireless communication. The semantic translation module has a built-in protocol rule library that supports command conversion for motors, lights, and valves. The control execution module includes motor and light controllers, as well as valves, and forms a closed-loop control link through feedback sensors. The offline large-scale model computation module deploys a quantized, small-bit version of the offline large-scale model at a specific parameter scale.

[0045] In one embodiment, in different application scenarios, the data acquisition module collects relevant process data through various professional sensors. In industrial production scenarios, thermocouple temperature sensors accurately measure temperature, strain gauge pressure sensors detect pressure, and electromagnetic flow meters monitor flow. In smart building scenarios, photoresistor light sensors sense light intensity, and temperature and humidity composite sensors measure temperature and humidity. These sensors convert physical quantities into electrical or digital signals and transmit the data to the data processing and prompt word generation module via wired (such as RS485 bus, CAN bus) or wireless (such as Wi-Fi, Bluetooth, ZigBee) communication methods.

[0046] After receiving the collected data, the data processing and prompt word generation module integrates and processes it according to the preset rules of the specific scenario. Taking the chemical production scenario as an example, the weight of each parameter is first determined based on the chemical reaction kinetics equation and thermodynamic characteristics. For example, for a certain endothermic reaction, the temperature parameter weight is calculated to be 0.6, the pressure parameter weight is 0.3, and the flow parameter weight is 0.1 according to its reaction kinetics equation. Then the temperature, pressure, flow and other data are normalized to eliminate the dimension effect. After that, the data is combined according to the rule of "key parameters first, secondary parameters later, combined with the reaction stage identification" to generate prompt words. For example, for a certain polymerization reaction, if the current temperature is 90°C, the pressure is 3 standard atmospheres, the flow rate is 8 cubic meters / hour, and it is in the third stage of the reaction, then the prompt word "90°C, 3atm, 8m 3 / h, the third stage".

[0047] The small-bit version of the offline large model deployed in the offline large model operation module is based on the Transformer architecture, possessing numerous parameters and a multi-layer neural network structure. After receiving a prompt word, the embedding layer converts the characters or numbers in the prompt word into vector representations. A multi-head attention mechanism then captures long-range dependencies between data points. A multi-layer perceptron performs nonlinear transformations and feature extraction. Finally, a multi-parameter comprehensive decision is made based on membership functions, outputting fuzzy logic-based control recommendations, such as the optimal temperature range and equipment operating status trends. For example, in chemical reaction temperature control, the large model, based on the input prompt word and a large amount of similar reaction data, outputs an optimal temperature range of 85-95°C, with a preference for 90°C in the current state.

[0048] The semantic translation module converts the abstract control suggestions output by the large model into specific instructions that the control device can recognize. For different types of control devices, corresponding semantic analysis and instruction conversion rule libraries are built in. Taking motor control as an example, the large model outputs the suggestion that "the speed should be between 1200-1500 rpm, tending to be 1300 rpm." The semantic translation module first performs word segmentation on the text, extracting keywords such as "speed," "1200-1500 rpm," and "1300 rpm." It then determines the motor type (such as DC motor, AC asynchronous motor, etc.) based on the motor control protocol rule library. For DC motors, according to the motor PWM protocol, the speed range and tendency value are converted into the duty cycle parameters of the PWM (pulse width modulation) signal, and the relevant formula is used to generate instructions such as "set the PWM duty cycle to 60%-75%, and the target duty cycle to 65%." For lighting control, if the large model outputs "the brightness should be adjusted to 40%-60%, tending to 50%", the semantic translation module parses the text and converts it into a digital dimming signal instruction according to the I2C protocol of the lighting controller, such as "I2C bus instruction: 0x01 (device address), 0x02 (brightness control command), 0x32 (50% brightness value)".

[0049] The control execution module receives control commands from the semantic translation module, drives the corresponding control devices to execute operations, and feeds the execution results back to the data acquisition module, forming a closed-loop control. The motor adjusts its speed based on the PWM signal, the light changes brightness based on the dimming signal, and the valve adjusts its opening based on the control command. Simultaneously, feedback sensors on the control devices (such as the motor speed encoder and the valve opening sensor) feed actual execution status data back to the data acquisition module, allowing the system to adjust the control strategy in real time. For example, in chemical production, after the motor speed is adjusted, the new flow rate data is collected by the data acquisition module and entered into the next control cycle, ensuring that the production process is always in an optimized control state.

[0050] The present invention uses a data processing and prompt word generation module to normalize and dynamically calculate the weights of physical quantity data collected by multi-source sensors, generate structured prompt words that identify the fusion scene stage, and accurately characterize complex working conditions; the offline large model operation module is based on a small-bit version of the Transformer architecture with quantization processing, and captures the nonlinear correlation between parameters through embedding layer vectorization and multi-head attention mechanism, and outputs fuzzy logic control suggestions in combination with the membership function, thereby achieving high-precision decision-making in an offline environment; the semantic translation module relies on the protocol rule library to convert abstract suggestions into device instructions, solving the problem of multi-device interface adaptation; the control execution module drives motors, lights, valves and other equipment to execute instructions, and forms a closed-loop control in combination with feedback from speed encoders, opening sensors, etc., ultimately reducing temperature fluctuations and defective rates in chemical production, achieving energy consumption reduction in intelligent buildings, and ensuring the system's stable operation and dynamic optimization capabilities in complex scenarios through wired / wireless communication and multi-sensor collaboration.

[0051] In summary, the present invention constructs a closed-loop intelligent control system based on an offline large model, deploys a quantized specific parameter scale Transformer architecture offline large model to realize network-independent operation, combines scenario data normalization processing and dynamic weight calculation to generate structured prompt words, and relies on the "text segmentation-keyword extraction-protocol matching" algorithm to convert the fuzzy logic decision output by the large model into a control instruction that can be recognized by the device. Finally, through the collaboration of multiple modules, a complete control chain of "data acquisition → feature fusion → fuzzy decision → instruction adaptation → closed-loop feedback" is formed, covering key technical links such as offline intelligent computing architecture design, scenario data integration rules and prompt word generation method, and multi-protocol adapted semantic translation mechanism, to achieve closed-loop optimization from complex scenario state perception to precise device control.

[0052] The present invention achieves breakthrough improvements in multiple dimensions and provides innovative solutions for automated control in complex scenarios. In terms of control accuracy, the system abandons the traditional black-and-white binary control mode and relies on the deep analysis and fuzzy reasoning capabilities of the offline large model to achieve collaborative analysis and dynamic decision-making of multi-source data. It can output precise control intervals and tendency strategies for scenarios such as chemical reactions and environmental regulation, significantly improving product quality stability and process control sophistication. In terms of system adaptability, through scenario-based data integration rules and dynamic weight distribution mechanisms, it can flexibly adapt to multiple fields such as industrial production, smart buildings, and smart transportation. It automatically generates structured prompt words based on real-time data features and scenario stage identifiers, effectively responding to the coupling relationship and dynamic change requirements of different scenario parameters, and greatly expanding the applicable boundaries of traditional control systems. In terms of energy utilization efficiency, the system uses refined control strategies and optimization algorithms to ensure the operating efficiency of equipment while avoiding ineffective energy loss. Especially in continuously running industrial production equipment and large-scale building environment control, its adaptive adjustment capabilities can significantly reduce energy consumption and provide technical support for green and sustainable development.

[0053] In terms of stability and independence, the system utilizes a quantized, compressed, and offline large-scale model architecture, completely eliminating reliance on network connectivity and ensuring stable operation even in extreme environments or high-security scenarios. A closed-loop feedback mechanism continuously optimizes control strategies, effectively resisting external interference and significantly improving the system's robustness and fault tolerance. Regarding scalability and compatibility, the system adopts a modular design concept. Through rule-based configuration in the data processing and prompt word generation modules and protocol adaptability in the semantic translation module, it can quickly integrate with new control devices and is compatible with diverse communication protocols, providing open interfaces for technological iteration and cross-domain applications. Its flexible expansion capabilities not only meet the upgrade needs of existing industrial equipment, but also leave ample room for the integration of future intelligent devices. Overall, this technical solution, through its innovative "data perception - intelligent decision-making - precise execution - closed-loop optimization" process, builds a next-generation intelligent control system with high precision, strong adaptability, low energy consumption, and high reliability. It provides the underlying technical support for the intelligent transformation of industry, the construction of smart cities, and the automation upgrade of key sectors, fully demonstrating the application potential of the deep integration of artificial intelligence technology and the real economy.

[0054] The present invention is not limited to the above-mentioned embodiments. Regardless of any changes in shape or material composition, any structural design provided by the present invention is a variation of the present invention and should be considered within the scope of protection of the present invention.

Claims

1. A control method based on offline fuzzy judgment, characterized in that: The following steps are involved: S1: Collect physical quantity data of the target scene through the data acquisition module; S2: Integrate and process the collected data according to the preset scenario rules to generate prompt words; S3: Use the offline large model to perform fuzzy decision operations on the prompt words and output control suggestions; S4: Use semantic analysis and the instruction conversion rule library to convert control suggestions into specific instructions and execute them, feedback the actual status data, and optimize parameters based on the feedback data.

2. The control method based on offline fuzzy judgment according to claim 1, characterized in that: The step S2 specifically normalizes and calculates weights on the collected data based on preset scenario rules, eliminates the dimensional influence, and generates scenario prompt words containing key parameters. The scenario prompt words include parameter values ​​and dimensions as well as scenario stage identification information.

3. A control method based on offline fuzzy judgment according to claim 1 or 2, characterized in that: The step S3 specifically includes adopting an offline large model with a Transformer architecture. After receiving the prompt word, the characters or numbers in the prompt word are converted into vector representations through an embedding layer, and then the long-distance dependencies between the data are captured through a multi-head attention mechanism. Nonlinear transformation and feature extraction are performed in a multi-layer perceptron. Finally, a multi-parameter comprehensive decision is made based on the membership function, and a control recommendation based on fuzzy logic is output.

4. The control method based on offline fuzzy judgment according to claim 1, characterized in that: The step S4 also includes, after conversion into a specific instruction, entering an anomaly detection module to detect whether there is a conflict, and if there is a conflict, regenerating the specific instruction through a conflict resolution strategy library.

5. The control method based on offline fuzzy judgment according to claim 4, characterized in that: The step S4 specifically includes, for different types of control devices, performing word segmentation processing on the text according to its built-in corresponding semantic analysis, judging the type according to the control protocol rule library, and selecting the corresponding instruction conversion rule according to the type to convert it into a specific instruction.

6. The control method based on offline fuzzy judgment according to claim 5, characterized in that: In step S4, for the DC motor, the offline large model outputs the speed range and tendency value, and according to the motor PWM protocol, the speed range and tendency value are converted into duty cycle parameters of the pulse width modulation signal and related instructions are generated; For lighting control, the offline large model outputs the brightness range and tendency value, which is converted into a digital dimming signal instruction including the device address, brightness control command and brightness value according to the I2C protocol of the lighting controller.

7. A control method based on offline fuzzy judgment according to claim 5 or 6, characterized in that: The step S4 also includes receiving control instructions from the semantic translation module, driving the corresponding control device to perform operations, and feeding back the execution results to the data acquisition module to form a closed-loop control. Specifically, the motor adjusts the speed according to the PWM signal, the light changes the brightness according to the dimming signal, and the valve adjusts the opening according to the control instruction. The feedback sensor on the control device feeds back the actual execution status data to the data acquisition module, and the system optimizes the parameters based on the feedback data and adjusts the control strategy in real time.

8. A control system based on offline fuzzy judgment, adopting a control method based on offline fuzzy judgment according to any one of claims 1 to 7, characterized in that: include: The data processing and prompt word generation module normalizes and calculates weights on the collected data, and generates prompt words according to preset rules; The offline large model operation module receives prompt words from the data processing and prompt word generation module, performs fuzzy logic operations based on the Transformer architecture, and outputs control suggestions; The semantic translation module converts the control suggestions sent by the offline large model operation module into instructions that can be recognized by the device; The control execution module receives instructions from the semantic translation module, drives the control device to execute the instructions and form a closed-loop feedback.

9. The control system based on offline fuzzy judgment according to claim 8, characterized in that: The data processing and prompt word generation module is connected to the data acquisition module, which collects physical quantity data through multiple sensors and converts them into electrical signals or digital signals. The data acquisition module includes temperature and pressure sensors, flow sensors, light sensors, and temperature and humidity sensors, and transmits data through wired or wireless communication.

10. The control system based on offline fuzzy judgment according to claim 9, characterized in that: The semantic translation module has a built-in protocol rule library that supports command conversion for motors, lights, and valves. The control execution module includes motor and light controllers as well as valves, and forms a closed-loop control link through feedback sensors. The offline large model operation module deploys a small-bit version of the offline large model with a specific parameter scale that has been quantized.

Images