Electrical automation control system based on human assistance AI mechanism

Abstract

The invention relates to the field of electrical automation control, and particularly discloses an electrical automation control system based on a human assistance AI mechanism. The system comprises a manual intervention interface, an AI basic controller, a data storage and management module and a consensus perception and collaborative decision maker. The system is characterized in that the consensus perception and collaborative decision-making device can receive a qualitative advanced control intention input by an operator, analyze the qualitative advanced control intention into a control signal in combination with historical interaction data and a current working condition, and synchronously calculate a real-time man-machine consensus index; and dynamically fusing the AI basic control signal and the manual analysis signal according to the index to generate a final control instruction. In addition, the system extracts an efficient man-machine collaboration mode for multiplexing through a collaboration strategy evolution module, and optimizes consensus degree evaluation through a calibration loop. According to the method, online real-time flexible fusion and coevolution of historical experience and AI calculation are realized, and the adaptive control capability of a complex industrial system is improved.

Description

Technical Field

[0001] This invention relates to the field of electrical automation control, and more specifically, to an electrical automation control system based on an artificial intelligence (AI) mechanism. Background Technology

[0002] As industrial automation systems become increasingly complex and intelligent, fully automatic control modes that rely solely on preset algorithms face challenges in adaptability and robustness when dealing with nonlinear, time-varying, and unpredictable operating conditions. Against this backdrop, incorporating the experience and judgment of human operators and the decision-making capabilities of higher management is crucial. Therefore, constructing a hybrid enhanced intelligent control system that integrates human and machine collaboration can improve the overall control quality and reliability of complex industrial processes, and is of significant developmental importance.

[0003] Currently, in the field of electrical automation control, technical solutions integrating artificial intelligence and human intervention mainly present two modes. One mode involves setting up independent human supervision and override control outside the automated control loop, allowing operators to directly intervene and switch to manual mode or modify setpoints in abnormal situations. This approach results in a disconnect between human intervention and automatic operation; the switching of control often introduces process disturbances, and the operator's experiential knowledge cannot be continuously and structurally empowered by the automatic control system. The other mode utilizes artificial intelligence to learn and imitate historical human operation data offline, aiming to replicate human strategies. However, this approach is inherently static and cannot dynamically interact with the operator or align with their intentions in real-time operation. When faced with novel operating conditions not covered by the training data, the system cannot proactively seek or effectively integrate online guidance from the operator, limiting its adaptability.

[0004] Therefore, to address the above issues, an electrical automation control system based on an AI-assisted mechanism is proposed. However, it faces the shortcomings of existing technologies mentioned in the background: it is difficult to achieve a deep integration between the real-time computing power of artificial intelligence and the flexibility and experiential advantages of human cognition; and it lacks a mechanism that can analyze, quantify, and integrate advanced human control intentions online in real time, enabling the two to collaborate smoothly and evolve together within a unified framework. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the defect of the separation between human intervention and AI autonomous control in the existing electrical automation control system. Specifically, human intervention often leads to abrupt switching of control modes; the experience and intuition of human operators are difficult to be quantified and continuously absorbed by the system; and the system lacks adaptability when facing new working conditions or different operators.

[0006] To address the aforementioned issues, this invention provides an electrical automation control system based on an AI-assisted mechanism. This system constructs a two-layer fusion control architecture centered on a dynamically quantified human-machine consensus index. It transforms the operator's qualitative high-level control intentions into executable, trust-adjustable control signals, which are then smoothly integrated with AI-based basic control signals. This achieves deep collaboration and continuous co-evolution between human experience and artificial intelligence while ensuring system safety.

[0007] Furthermore, to achieve accurate understanding and trust assessment of high-level control intentions, the system employs parallel intention understanding and consensus assessment modules. The intention understanding module, based on a continuously accumulating historical case library, uses analogical reasoning to translate semantic operational instructions into specific control parameter adjustment suggestions. The consensus assessment module, through statistical analysis of operators' historical behavioral patterns and intervention effectiveness, dynamically calculates a consensus index reflecting the credibility of the current instruction. These two modules work together to provide the system with structured and quantifiable decision-making support, laying the foundation for human-machine intelligence integration.

[0008] Furthermore, to enhance the system's accuracy in understanding domain-specific expressions and its sensitivity to changes in operator state, the intent understanding module, when parsing semantics, not only relies on a general language model but also incorporates co-occurrence patterns of terms related to specific working conditions mined from historical data. The consensus evaluation module introduces a time decay factor when calculating intervention effectiveness, making the system more focused on the operator's recent performance. These mechanisms enhance the system's adaptability and personalization across different application scenarios and operators.

[0009] Furthermore, to achieve a smooth and disturbance-free transfer of control between humans and machines, the system employs a weight mapping function with smooth S-shaped characteristics to map the real-time calculated consensus index into fusion weights. This design enables the final control commands to change continuously and stably as the human-machine trust relationship changes, mathematically avoiding system oscillations caused by sudden changes in control power and ensuring the stability of the control process.

[0010] Furthermore, to ensure absolute security during the integration process, the system incorporates a dynamic security constraint processing step before adopting human intentions. This step not only performs static boundary checks but also conducts forward-looking predictions and online optimizations based on the dynamic model of the controlled object, ensuring that all instructions remain within a safe feasible domain. This mechanism constructs a reliable security boundary for the injection of human wisdom, enabling the system to more boldly integrate high-value human decisions and fully unleash the potential of human-machine collaboration.

[0011] Furthermore, to optimize the system's response performance in known high-performance scenarios, a fast-track mechanism has been established. When the current situation is detected to closely match historical high-performance cases, the system will directly invoke the validated successful strategy and assign it a high initial trust level. This significantly improves decision-making speed and control quality under typical operating conditions, enabling the immediate consolidation and efficient reuse of excellent operational experience.

[0012] Furthermore, to enable the system to learn from historical interactions and solidify successful collaboration patterns, a collaborative strategy evolution function has been introduced. This function abstracts successful human-machine control sequences into reusable strategy templates through offline analysis. During real-time operation, the system can proactively match and load relevant templates, thereby providing forward-looking guidance for collaborative interactions in the future, effectively shortening the trial-and-error process when facing complex tasks and improving collaborative efficiency.

[0013] Furthermore, to facilitate the transfer and sharing of excellent collaborative knowledge, the system can perform generalization analysis on various extracted strategy templates to extract universal guidance strategies decoupled from specific operators. When applied to new scenarios or new users, the system can provide a high-quality initial collaborative framework, shortening the "learning adaptation period" and solving the problem of system cold start.

[0014] Furthermore, to ensure that the system's internal evaluation criteria for collaborative states continuously align with real human perception, a human-machine consensus calibration and optimization loop has been constructed. This loop allows experienced users to calibrate the system's evaluation results and continuously optimize the internal evaluation model using this feedback data. This enables the system to possess online learning capabilities based on human feedback, achieving personalized adaptation and long-term evolution of the evaluation criteria.

[0015] The technical effects and advantages of this invention are as follows: Compared to the existing technologies where manual intervention and automatic control are disconnected and switch abruptly, this invention constructs a continuous human-machine collaborative signal space by introducing a dynamic fusion and decision-making center into the control architecture. This center parses the operator's qualitative high-level control intentions into underlying control parameter compensation signals in real time. Simultaneously, based on the analysis of the operator's historical behavior patterns and intervention effects, it dynamically calculates a consensus index characterizing the current level of human-machine collaboration, and uses this index to nonlinearly weight and fuse the output signal of the basic AI controller with the manual compensation signal. This mechanism allows control commands to dynamically balance the continuum of contributions from both sides according to the specific context, thereby achieving a transition from manual / automatic mode switching to a smooth, non-disruptive human-machine control weight shift. This ensures the continuity of basic automatic control while allowing human experience to be injected into the control loop in a flexible manner.

[0016] To address the problem in existing solutions where human experience and knowledge cannot be continuously absorbed and utilized by the system, this invention designs a collaborative strategy evolution module based on offline analysis of historical interaction data. This module automatically identifies and extracts human-machine collaborative control trajectories generated by specific operators under specific conditions that consistently lead to excellent performance by analyzing stored sequences of working conditions, intentions, consensus levels, and performance data. These trajectories are then abstracted into reusable strategy templates. When the system encounters similar working conditions again, it can proactively load the corresponding template and initialize collaborative parameters, thereby transforming past successful collaborative experiences into usable knowledge inherent to the system. This allows the system's control capabilities to continuously accumulate and evolve with ongoing use, shortening the learning time for new working conditions.

[0017] To address the limitation of offline learning systems in responding to real-time online guidance, this invention establishes a human-machine consensus calibration loop. This system allows authorized operators to evaluate and calibrate the consensus level of human-machine collaboration, calculated and displayed in real-time within the system, under typical operating conditions. This calibration data, serving as high-quality feedback samples, is used to periodically optimize the intrinsic parameters of the consensus level assessment model. Through this process, the system's learning of operator control styles and evaluation standards shifts from a single, historical data-driven approach to interactive learning incorporating proactive online feedback. This allows the system's judgments of human-machine collaboration to continuously align with and adapt to the individual operator's cognitive framework, improving the accuracy and personalization of human-machine interaction. Attached Figure Description

[0018] Figure 1 This is a flowchart of the internal processing of the consensus perception and collaborative decision-making system of the present invention.

[0019] Figure 2 This is a flowchart of the control signal fusion and security processing branch of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1 The following is in conjunction with the accompanying drawings. Figures 1 to 2 The present invention will be further described in detail below with reference to specific embodiments.

[0022] In a typical deployment, this system can be installed on an industrial control server or edge computing device. The manual intervention interface is connected to the operator station (HMI) via industrial Ethernet; the AI ​​basic controller and consensus perception and collaborative decision-making unit run as software modules on the server; the data storage and management module can use a real-time database or a time-series database; the system exchanges data with field PLCs, DCSs, and sensor networks through standard communication protocols such as OPCUA and ModbusTCP. This deployment method does not affect the implementation of the core software method of this invention.

[0023] The implementation process of this invention follows a periodic control cycle. Each cycle includes the entire process from data perception to instruction execution and data recording, with data learning and strategy optimization running in the background. The following description uses industrial process control as an application scenario to illustrate the complete workflow of the system.

[0024] S100: System Initialization and Data Preparation After the system is powered on and started, each functional module completes its initialization in sequence.

[0025] 1. The data storage and management module establishes or loads a historical interaction database. This database is used to store structured historical interaction records, each of which contains fields such as timestamp, operator ID, operating data, control commands, and performance indicators.

[0026] 2. The consensus perception and collaborative decision-making unit loads the initial configuration parameters of each computing unit within it.

[0027] These parameters include: the initial rule set of the intent case library, and the weight mapping function. coefficient (gain coefficient) and offset The initial value can be set to , Boundary parameters of the dynamic safety constraint processor (including the physical limits of the control quantities of each actuator) and process safety range And the maximum rate of change limit for key controlled variables, such as the maximum rate of temperature increase. ), and the initial calculation parameters of the consensus evaluation module (such as the balance coefficient between the consistency factor and the effectiveness factor). The initial value can be set to 0.5; time decay constant. The initial value can be set to 30 days.

[0028] 3. The manual intervention interface and the AI ​​basic controller respectively complete the self-test of the communication links with the host human-machine interface (HMI) and field sensors and actuators to confirm that the data transmission and reception channels are normal.

[0029] When the system runs for the first time, the historical interaction database is empty, and the intent case library can be pre-configured by domain experts with a small number of general rule cases, or it can be temporarily left empty. At this time, the system will mainly rely on the AI ​​basic controller for control, and the consensus evaluation module will run with conservative default parameters and gradually accumulate data during system operation to start the learning process.

[0030] S200: Real-time data acquisition and signal input The system operates at a fixed period (For example The real-time control loop executes every second. At the start of each cycle, the following operations are performed in parallel: S201: Acquires real-time operating status data of the controlled object. The sensor network transmits measured values ​​to the system via fieldbus, and these values ​​are organized into a current operating status data vector. Taking chemical reaction processes as an example, This may include: reactor internal temperature Temperature setpoint Reactant feed flow rate Stirring power Cooling water inlet pressure This vector is used to characterize the current system operating point.

[0031] S202: The AI ​​basic controller generates a reference control signal. The AI ​​basic controller uses... As input, it runs its internal algorithm model, calculates and outputs the first control signal vector. .

[0032] For example, a multivariate model predictive controller (MPC) might output adjustments to the setpoints for heater power and cooling water flow rate, constituting... This signal represents the system's decision based on preset control logic in the absence of human intervention.

[0033] S203: Receives qualitative high-level control intents from the operator. The operator inputs non-numerical control commands through dedicated interface elements on the HMI. These interface elements are designed to receive instructions in at least one of the following forms: Target Priority Instruction: Specify the relative importance order of multiple control targets (such as "product quality", "energy consumption", "production rate") at the current moment through a slider or weight allocator.

[0034] Control situation assessment instruction: Select one from a set of qualitative options (such as "control is too aggressive", "control is too conservative", "response is slow", "oscillation is obvious").

[0035] Expected behavior description instructions: Enter a short statement in the text box that describes the desired system behavior, such as "slowly increase the reaction temperature to the set point" or "focus on monitoring the pressure parameters".

[0036] After the human intervention interface captures and formats this input, it will... Send to the consensus awareness and collaborative decision-making unit.

[0037] S300: Historical Data Query and Context Building Consensus perception and collaborative decision-making unit receive and Immediately afterwards, a query is initiated to the data storage and management module. The query request includes the current operator's ID and... The data storage and management module, based on a predefined similarity metric (such as the Euclidean distance of feature vectors), retrieves data related to the operator from the historical interaction database, and whose historical working conditions are consistent with the operator's. Several similar records constitute a historical context dataset. This dataset provides data support for subsequent intent parsing and consensus assessment based on specific operator experience.

[0038] S400: Analysis and Conversion of Advanced Control Intent This step will define the qualitative intent of the operator's input. This is converted into specific control parameter adjustment suggestions. This function is performed by the intent understanding module within the consensus perception and collaborative decision-making unit.

[0039] Input: Current high-level control intent Current operating condition feature vector (from Extract relevant intent cases from historical context datasets.

[0040] Output: Original artificial intervention signal vector .

[0041] Furthermore, the intent understanding module maintains a dynamically updated library of intent cases.

[0042] Each case record in the database is a tuple: {Case ID, Working condition feature vector F, Intent text I, Effective control parameter adjustment vector ΔU, Performance improvement flag Flag}.

[0043] Where F is a characteristic that reflects the essence of the working condition, for example... ; It represents the difference between the final manual control signal implemented in this case scenario and the AI ​​control signal at the time; Flag indicates whether the overall system performance has improved after this intervention.

[0044] The intent understanding module performs the following parsing process: 1. Feature Extraction and Case Retrieval: From extract .calculate Features of all cases in the case library similarity (For example, using cosine similarity). Select the top results with the highest similarity. Cases (such as) () as a candidate case set.

[0045] 2. Semantic relevance calculation: Calculation With the intent text of each candidate case semantic relevance .

[0046] This calculation is not only based on a general word vector model, but also incorporates domain knowledge learned from the system's historical data. The system analyzes the co-occurrence patterns of specific word combinations under similar operating conditions in the historical data.

[0047] For example, if historical data shows that under heating conditions, the terms "mild heating" and "slow increase" are often used interchangeably and lead to similar successful control actions, then even if the words are different, their semantic relevance remains the same. The value will also be relatively high. In the specific calculation, It can be represented as a weighted sum of general semantic similarity and domain relevance based on co-occurrence statistics.

[0048] One approach is for the system to analyze a historical intent case library during the offline phase, construct a co-occurrence matrix of "control intent words - working context", quantify the association strength of specific word combinations under specific working conditions by calculating statistical quantities such as point mutual information, and use this strength as a domain relevance component.

[0049] When the intent case library is empty or has no matching cases, the intent understanding module outputs a zero vector or generates a conservative adjustment suggestion based on preset general rules. .

[0050] 3. Weighted generation of control signals: For each candidate case Calculate its overall weight The generated original artificial intervention signal The result is obtained by weighting the effective adjustments of these candidate cases: ; This formula ensures that the more similar the historical successful experience is to the current operating conditions and the closer the intended expression is to the current control recommendations, the greater the impact on the current control recommendations.

[0051] As a preferred implementation, a fast-track mechanism is introduced to improve the system's response efficiency and decision reliability under familiar operating conditions. When a candidate case is calculated above... Overall weight Exceeding the preset high threshold When the value is 0.85 (e.g., 0.85), the intent understanding module determines the current context and case. Highly compatible. In this case, the system will bypass the weighted average calculation and directly use the case study. Effective adjustment vector in As The output is then sent. Simultaneously, a signal is sent to the consensus evaluation module, indicating that a high initial consensus score should be assigned to this interaction. (e.g., 0.9).

[0052] S500: Dynamic Evaluation of Human-Machine Collaborative Consensus This step aims to quantitatively assess the system's level of trust in the current operator and their intentions. This function is performed by the consensus assessment module within the consensus awareness and collaborative decision-making unit, and is computed in parallel with the S400.

[0053] Input: Current high-level control intent Current operating condition feature vector The operator's behavior records in the historical context dataset.

[0054] Output: Real-time consensus metrics Its range is [0, 1].

[0055] Furthermore, the consensus assessment module calculates the consistency factor comprehensively. and effectiveness factors Determine .

[0056] 1. Calculate the consistency factor This factor measures current intent. Consistency with the operator's behavioral habits under similar historical working conditions.

[0057] Find the relevant data from historical data. Most similar For each historical time point, the main intent patterns corresponding to these points are statistically analyzed (e.g., the most frequently occurring intent keywords or types). Calculations are then performed. The semantic similarity with this main intent pattern, after normalization, is obtained as follows: .

[0058] 2. Calculate the effectiveness factor This factor measures the historical effectiveness of the operator's recent interventions.

[0059] From the operator's recent history (e.g., most recent) In each effective intervention, the actual values ​​of the system's key performance indicators after each intervention are extracted. Compared with the system's estimated performance index value under the condition of no intervention. Calculate the relative improvement rate for each intervention. .

[0060] To reflect the time-varying nature of operator skills or process conditions, a time decay weighted calculation is used. Let the current time be... , No. The intervention occurred at the time when Then its weight is ,in is the time constant.

[0061] time constant The value used to control the decay rate of historical data influence can be adjusted according to process stability requirements or the update cycle of operator skill assessments.

[0062] For example, for processes where operating conditions change slowly, A larger period (e.g., 60 days) can be set to smooth out fluctuations; for scenarios requiring rapid adaptation to changes in operators, A shorter period can be set (e.g., 7 days). This calculation ensures that the impact of near-term intervention effects on consensus is weighted more heavily than that of long-term effects.

[0063] 3. Comprehensive Calculation: Final Consensus Index Calculated by the following formula: ,in An adjustable balance coefficient ( ).

[0064] By adjusting The system can be configured to prioritize consistent operator behavior. (approaching 1) or the actual effect of its recent intervention ( (approaching 0), the calculation results are constrained to the interval [0, 1] by the amplitude limiting function.

[0065] S600: Dynamic Fusion and Safety Constraints of Control Signals This step receives , and They then perform fusion calculations to generate the final security control commands.

[0066] Input: First control signal Original artificial intervention signals Consensus index C.

[0067] Output: Final control command vector .

[0068] Furthermore, the signal fusion module within the consensus perception and collaborative decision-making unit first, based on... Value calculation fusion weight Through a pre-defined weight mapping function. The function is configured to: when When the value is low, Approaching ;when When the value is high, Approaching In the intermediate transition zone, Follow It increases monotonically and changes smoothly without abrupt changes.

[0069] One specific function form is the sigmoid function: , where parameters Control the steepness of the curve, Determine the center point of the curve (i.e. Time corresponding (Value), this design ensures the continuity of control over changes in consensus.

[0070] Furthermore, before integration, it is necessary to... Safety constraints are applied to ensure the feasibility of control commands. This function is handled by the dynamic safety constraint processor (function). The specific process is as follows: 1. Boundary constraint handling (static projection): Check each constraint individually. Each component Does it exceed the physical limits of the corresponding actuator? and process safety static range If it exceeds the limit, then correct (project) it to the valid range to obtain... After processing all components, the projection vector is obtained. .

[0071] 2. Dynamic Process Constraint Handling (Model Prediction Optimization): To prevent control commands from causing the controlled variable to change too rapidly or violating dynamic correlation constraints, prediction is performed based on a simplified dynamic model of the controlled object (this model can be obtained in advance through conventional engineering methods such as mechanism analysis, system identification, or historical data fitting). As an initial control sequence, the trajectory of key variables (such as temperature and pressure) is predicted by the model over several future steps.

[0072] If the predicted trajectory violates any dynamic safety constraints (such as...) Then, we need to solve a constrained optimization problem. The objective of this optimization problem is: Find a control sequence that is consistent with... The difference is minimized; the constraint is that, under this control sequence, the future trajectory predicted by the model satisfies all dynamic safety constraints. Solve this optimization problem, and take the first control variable of the solution sequence as the optimized safety control vector. .

[0073] If the optimization problem has no feasible solution, the dynamic safety constraint processor will activate backup strategies, such as prioritizing the protection of the most critical safety variables and generating a suboptimal but absolutely safe solution by relaxing secondary constraints or using the decay value of the safety control instructions from the previous cycle. .

[0074] 3. Smoothing Filter: To avoid sudden changes in commands between adjacent control cycles, the following filter is applied: A first-order low-pass filter is performed to obtain the final second control signal after safety constraint processing. .

[0075] The filtering formula is: ,in For filter coefficients ( ).

[0076] Finally, the signal fusion module performs a linear weighted fusion calculation: .

[0077] S700: Control Command Output and Execution Consensus perception and collaborative decision-making will The commands are transmitted to the corresponding actuator via fieldbus or hardwiring. The actuator drives the controlled object according to the received commands, completing the physical action of one control cycle.

[0078] S800: Interactive Data Recording and Storage At the end of the current control cycle, the data storage and management module creates a new historical interaction record and stores it in the database. This record is a structured data packet, typically containing the following fields: Timestamp, operator ID, operating status data Advanced control intent text First control signal Original artificial intervention signals fusion weight Consensus indicators Final control command .

[0079] In addition, after a period of time (such as 30 control cycles), the system will calculate the comprehensive performance indicators (such as average deviation, energy consumption, output, etc.) of this intervention cycle and compare them with the simulated performance of pure AI control during the same period to calculate the performance improvement rate. If the improvement rate exceeds the preset threshold, the performance improvement flag in the corresponding historical record will be set to valid, thereby enriching the quantity and quality of high-quality cases in the intent case library and providing a more reliable data foundation for intent parsing and strategy evolution.

[0080] S900: Offline Evolution and Optimization of Cooperative Strategies As a preferred implementation, to overcome the problems of insufficient utilization of historical success experience and slow system performance improvement when facing new scenarios or new operators in fusion control based on real-time consensus, the system introduces a collaborative strategy evolution module. This module starts when system computing resources are idle (such as at night) and performs offline analysis.

[0081] The workflow of this module is as follows: 1. Data preparation and session segmentation: Read the historical interaction database and segment the data into multiple independent "interaction sessions" according to operator ID and continuous control time period.

[0082] 2. Successful Session Identification and Pattern Extraction: For each session, "high-performance successful sessions" are identified based on their final comprehensive performance indicators. For each successful session, the operational characteristics of its initial stage are extracted. As a triggering condition, the sequence of intents input by the operator in this session is abstracted as an intent pattern sequence. Extract consensus metrics from the session. A curve that changes over time serves as a reference trajectory for consensus. .

[0083] 3. Cooperative strategy template generation: This involves generating the extracted triples... It is encapsulated as a collaborative strategy template and stored in a dedicated strategy template library. This template encapsulates a proven and efficient human-machine collaborative control mode.

[0084] 4. Online application of templates: During real-time control (after S300), the system will use the current operating condition characteristics... Triggering conditions of each template in the strategy template library The system performs matching. If the matching degree exceeds a threshold, the corresponding collaborative strategy template is loaded. In subsequent S400 and S500 iterations, the system can then... Provide biased guidance for intent analysis, and refer to... Initialize or compare the consensus assessment to accelerate and guide the system into a known efficient collaborative state.

[0085] If the matching degree between the current working condition and the trigger conditions of all templates does not reach the threshold, the system will not load any templates and will operate entirely according to the real-time dynamic fusion process described in S400 to S600. This is the default working mode.

[0086] Furthermore, to achieve knowledge transfer and reuse, the collaborative strategy evolution module also performs cluster analysis on multiple templates in the strategy template library. By analyzing the efficient templates generated by different operators in different scenarios, it extracts the universal control logic that is independent of the specific operator's identity, forming a general guidance strategy. When the system is first deployed in a new process unit, or used by a new operator, the general guidance strategy most similar to their process can be loaded as the initial collaborative framework, thereby significantly shortening the system's "learning adaptation period" and improving initial control performance.

[0087] S1000: Calibration of Human-Machine Consensus and System Adaptation As a preferred implementation, in order to enable the system to quantify the "human-machine collaboration state" (i.e.) The consensus assessment model aligns more closely with the subjective judgment of human experts or experienced operators, and a calibration and optimization loop is incorporated into the system. This loop aims to address potential initial biases or adaptability issues arising from environmental changes in the consensus assessment model.

[0088] The workflow of this circuit is as follows: 1. Calibration Data Acquisition: When the system is operating in a preset, stable calibration state, the HMI will display the consensus index calculated in real time by the system. Authorized advanced users (such as process engineers) who believe the displayed value does not match their experience-based judgment can input a more accurate consensus value through a dedicated calibration interface. .

[0089] 2. Calibration Sample Record: The system randomly records a calibration sample, whose data structure is: {calibration condition data} Original high-level control intent System calculated value User calibration value The calibration sample, along with its complete timestamp and user identifier, is stored in a special partition of the historical interaction database.

[0090] 3. Parameter Optimization and Model Update: Once the accumulated calibration samples reach a certain number (e.g., 50), the system initiates a parameter optimization routine in the background. The optimization process aims to minimize the system's calculated values. Compared with user calibration value The mean square error between them is the objective function, where The set of parameters to be optimized in the consensus evaluation module (which may include balance coefficients) Time decay constant (e.g., hybrid weights in semantic relevance calculation). Gradient descent and other optimization algorithms are used to... Iterative updates will be performed.

[0091] 4. Hot Model Update: After optimization convergence, the updated parameters are updated. The system performs an online hot update to the consensus evaluation module, replacing the original parameters. Through this closed-loop process, the system's quantitative standards for "human-machine consensus" can continuously align with the evaluation habits of specific user groups or application scenarios, achieving personalization and continuous evolution of the evaluation model.

[0092] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0093] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An electrical automation control system based on an artificial intelligence-assisted mechanism, characterized in that, include: Human intervention interface, AI basic controller, data storage and management module, and consensus perception and collaborative decision-making unit; The manual intervention interface is used to receive qualitative high-level control intentions input by the operator; The AI ​​basic controller is used to generate a first control signal based on the real-time status data of the controlled object; The data storage and management module is used to continuously record and update the operator's advanced control intentions, corresponding operating condition data and control results, forming a historical interaction database; The consensus perception and collaborative decision-making unit is communicatively connected to the manual intervention interface, the AI ​​basic controller, and the data storage and management module, respectively. The consensus-aware and collaborative decision-making unit is configured as follows: Receive the advanced control intent and current operating condition data; Calculate the real-time human-computer consensus index based on the historical interaction database and the current system status; Based on the current operating condition data and the historical interaction database, the advanced control intent is parsed into a second control signal; The fusion weights of the first control signal and the second control signal are dynamically determined based on the human-machine consensus index, and the two are weighted and fused based on the fusion weights to generate the final control command.

2. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 1, characterized in that, The consensus perception and collaborative decision-making unit includes an intent understanding module and a consensus evaluation module; The intent understanding module is configured as follows: Maintain and query an intent case library, which stores the mapping relationship between historical operating condition characteristics, historical advanced control intents and corresponding effective control parameter adjustment amounts; The intent understanding module generates the second control signal from the mapping relationship based on the matching degree between the current operating condition features and the historical operating condition features, as well as the semantic relevance between the current advanced control intent and the historical advanced control intent. The consensus assessment module is configured as follows: The human-machine consensus index is determined by comprehensively calculating the consistency factor and the effectiveness factor. The consistency factor reflects the degree of consistency between the current advanced control intent and the operator's mainstream control intent pattern under similar historical operating conditions. The effectiveness factor reflects the degree of statistical improvement in the key performance indicators of the controlled object resulting from the operator's recent historical intervention behavior.

3. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 2, characterized in that, The calculation of semantic relevance takes into account the co-occurrence relationship of the advanced control intent in the historical control target set recorded in the intent case library and its corresponding operating condition characteristics. When calculating the effectiveness factor, the consensus evaluation module assigns a weight that decays over time to the performance improvement associated with historical intervention events.

4. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 2, characterized in that, The consensus perception and collaborative decision-making unit also includes a signal fusion module; The signal fusion module is configured to map the human-machine consensus index to a fusion weight value between 0 and 1 through a preset weight mapping function. The weight mapping function makes the fusion weight value monotonically increase as the human-machine consensus index increases, and has a smooth S-shaped transition characteristic in the middle of its value range.

5. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 4, characterized in that, The signal fusion module is further configured to: perform feasibility correction and smoothing processing on the second control signal according to the real-time safe operation boundary of the controlled object, to obtain a second control signal subject to safety constraints; The final control command U is generated using the following formula: ,in, The first control signal. is the second control signal subject to safety constraints, and w is the fusion weight value.

6. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 2, characterized in that, The consensus perception and collaborative decision-making unit is further configured such that when the intent understanding module determines that the matching degree between the current situation and a certain high-performance historical case in the intent case library exceeds a preset threshold, it prioritizes generating the second control signal based on the historical case and sets the initial value of the human-computer consensus index of the current interaction to a benchmark value higher than that output by the conventional calculation process.

7. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 1, characterized in that, It also includes a collaborative strategy evolution module; The collaborative strategy evolution module is configured to: analyze the historical interaction database offline, identify and extract the successful control trajectory that starts from a specific initial working condition, passes through a specific interaction mode, and finally reaches the predetermined performance target, and abstract it into a reusable collaborative strategy template. In real-time control, if the system state matches the triggering condition of a certain collaborative strategy template, the consensus perception and collaborative decision-making unit loads the template to initialize its collaborative parameters.

8. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 7, characterized in that, The collaborative strategy evolution module is also configured to: perform generalization analysis on multiple extracted collaborative strategy templates, extract the general control logic that is decoupled from the specific operator's identity, and form an initial collaborative guidance strategy for new scenarios or new operators.

9. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 1, characterized in that, The system also includes calibration and optimization loops; The calibration and optimization loop is configured to: receive calibration inputs for the human-machine consensus index calculated internally by the system under preset calibration conditions, and record them as calibration samples; based on the accumulated calibration samples, periodically optimize the parameters involved in consensus calculation in the consensus perception and collaborative decision-making unit.

10. The electrical automation control system based on artificial intelligence (AI) mechanism according to claim 1, characterized in that, The qualitative high-level control intent includes at least one of the following: a priority ordering instruction for multiple control objectives, a qualitative evaluation instruction for control strategy tendencies, or a descriptive instruction for expected system behavior.

Images