Motor parameter cross-reasoning identification method based on verification driving and related device

Abstract

This invention discloses a verification-driven method and related apparatus for interleaved inference identification of motor parameters, belonging to the field of motor control technology. The method first acquires the motor's operating state parameters, then dynamically selects and executes interleaved actions from an extended action space using a strategy function. When the interleaved action is an inference action, the inferred value of the motor's target parameter is obtained; when the interleaved action is a verification action, the inferred value of the motor's target parameter is verified and feedback is generated. Based on the feedback, the strategy function is updated, and the above inference and verification actions are repeated until a preset termination condition is met, resulting in an estimated value of the motor's target parameter. Finally, the estimated value of the motor's target parameter and a reliability index are output. This method embeds verification into the inference chain, achieving a closed loop of "inference-verification-feedback-adjustment." Parameter tuning is performed through a motor factual knowledge base to prevent error accumulation, and verification is dynamically scheduled through a strategy function to balance verification intensity and inference efficiency, adapting to industrial multi-condition and small-data scenarios.

Description

Technical Field

[0001] This invention belongs to the field of motor control technology, specifically relating to a method and related device for motor parameter interleaving reasoning and identification based on verification-driven approach. Background Technology

[0002] Accurate identification of motor parameters is crucial for achieving efficient control and energy efficiency optimization. Complex and ever-changing industrial conditions demand not only high precision in the identification algorithm but also strong robustness and the ability to detect and handle abnormal states. Traditional parameter identification methods heavily rely on physical test benches, calculating parameters by injecting specific excitation signals and collecting response signals. This method is costly, cumbersome, and difficult to apply to already deployed and operating motors.

[0003] With the development of industrial intelligence, data-driven reasoning-based parameter identification models have shown great potential. These models can directly estimate parameters from motor operating data, potentially eliminating reliance on dedicated testing equipment. However, they place extremely high demands on the reliability of the reasoning process. While existing AI (Artificial Intelligence) reasoning frameworks can achieve thought chain reasoning using language models, they lack mechanisms for real-time, rigid verification of each step of the reasoning process in rigorous industrial physical systems. The reasoning process is like a "blind box," unable to guarantee its conformity to physical common sense during runtime, thus introducing process-level reliability risks.

[0004] Traditional approaches have significant limitations: they either place verification after the fact, making it difficult to correct accumulated errors during the reasoning process in a timely manner; or they rely on engineers to write a large number of hard-coded rules, leading to system rigidity and maintenance difficulties. Other solutions attempt to improve reliability through complex and redundant architectures, but this often exacerbates system complexity and unpredictability. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a verification-driven method and related device for interleaved inference and identification of motor parameters, so as to solve the problem that industrial AI vertical models accumulate intermediate errors and have low reliability in the final output due to opaque decision-making and inability to verify in real time during the inference process.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] The first aspect of this invention discloses a method for motor parameter interleaving reasoning and identification based on verification-driven methods, comprising:

[0008] Obtain motor operating status parameters;

[0009] Based on the motor operating state parameters, an inference state is constructed. The strategy function dynamically selects an interleaved action from the extended action space containing inference and verification actions according to the current inference state. If the selected interleaved action is an inference action, the inference action is executed according to the current inference state and the core parameters of the strategy function to obtain updated inference values ​​of the motor target parameters, and the inference state is updated based on the updated inference values ​​of the motor target parameters. If the selected interleaved action is a verification action, the verification action is executed by calling the built-in motor factual knowledge base through the verifier index module in the strategy function based on the latest inference values ​​of the motor target parameters to generate a verification state. Feedback is given to the strategy function according to the verification state, and the parameters of the strategy function are updated based on the feedback.

[0010] When the preset termination condition is met, the final estimated value of the motor target parameters is output, and the credibility index is calculated and output based on the strategy function and historical data of the verification status.

[0011] Preferably, the reasoning actions include parameter iterative optimization, working condition adaptation rule adjustment, and neural network model weight parameter adjustment and update; the verification actions include physical constraint verification, numerical stability verification, data integrity verification, and constraint violation detection.

[0012] Preferably, the expression of the strategy function is:

[0013]

[0014] in, ( ) is the policy function. For alternating movements, In the state of reasoning, These are the core parameters of the policy function. This is the parameter set for the validator index module. This is the core module of the strategy function; The output of the verifier index module represents the dynamic constraint strength of the motor factual knowledge base on the policy function decision, with a value range of 0 to 1.

[0015] Preferably, the reasoning state includes motor operating state parameters, historical reasoning results, and historical verification status.

[0016] Preferably, the verification status includes pass, failure, and warning; if the verification status is pass, the inference status is updated and the next round of inference is performed; if the verification status is failure, the process is backtracked to the last pass inference status and the step size of the inference action is shortened; if the verification status is warning, the verification frequency is adjusted.

[0017] If the inferred value of the motor target parameter is within the first threshold range of the motor factual knowledge base, the verification status is passed; if the inferred value of the motor target parameter is outside the first threshold range of the motor factual knowledge base but within the tolerance range or only triggers non-critical constraints, the verification status is a warning; if the verification status is neither passed nor a warning, it is a failure.

[0018] Preferably, if the proportion of the number of failures recorded in the verification state to the number of verification actions executed exceeds a set second threshold, the policy function is updated by the near-end policy optimization method, and the dynamic constraint strength of the output of the verifier index module in the policy function is updated.

[0019] Preferably, the credibility index includes verification pass rate and inference confidence; the verification pass rate is the proportion of verification actions performed when the verification status is passed.

[0020] The formula for calculating the inference confidence level is as follows:

[0021]

[0022] in, For the confidence level of the inference, Expectation operator To determine the dynamic constraint strength of the verifier index module, For the policy function, For optimal alternating movements, In the state of reasoning, These are the core parameters of the policy function. This is the parameter set for the validator index module.

[0023] A second aspect of the present invention discloses a motor parameter interleaving reasoning identification device based on verification-driven methods, comprising:

[0024] The parameter input module is used to acquire motor operating status parameters;

[0025] The inference execution module is used to construct an inference state based on the motor operating state parameters. The strategy function dynamically selects an interleaved action from an extended action space containing inference and verification actions based on the current inference state. If the selected interleaved action is a inference action, the inference action is executed based on the current inference state and the core parameters of the strategy function to obtain updated inference values ​​for the motor target parameters, and the inference state is updated based on these updated values. If the selected interleaved action is a verification action, the verification action is executed based on the latest inference values ​​for the motor target parameters by calling the built-in motor factual knowledge base through the verifier index module in the strategy function to generate a verification state. Feedback is provided to the strategy function based on the verification state, and the parameters of the strategy function are updated based on the feedback.

[0026] The result output module, when the preset termination condition is met, is used to output the final estimated value of the motor target parameters, and calculate and output the credibility index based on the strategy function and historical data of the verification status.

[0027] A third aspect of the present invention discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the verification-driven motor parameter interleaving reasoning identification method described in any of the preceding claims.

[0028] The fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the verification-driven motor parameter interleaving reasoning identification method as described in any of the preceding claims.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] This invention discloses a verification-driven method for interleaved inference and identification of motor parameters. It constructs a collaborative technical solution of verification-driven and interleaved inference, designing a lightweight, embedded, and dynamic verification mechanism and deeply integrating it with the inference process. This eliminates the need for complex and redundant designs. By guiding the dynamic interleaving of inference and verification actions through a strategy function, it effectively solves the problem of reliable identification of motor parameters in industrial multi-condition and small-data scenarios. This invention directly embeds verification actions into the inference chain, embedding a verifier index module containing a motor factual knowledge base within the strategy function, and integrating it with the inference strategy. The strategy function dynamically schedules the interleaved execution of inference and verification actions, forming a "reasoning-verification" interleaved closed loop. This allows the model to perform self-checks simultaneously during inference, enabling it to conduct self-verification and real-time adjustments based on the motor factual knowledge base at each step of the inference process. This effectively prevents the accumulation of erroneous inferences that violate physical laws from the source, achieving real-time reliable assurance of the inference process. Meanwhile, this invention achieves a dynamic balance between verification intensity and inference efficiency through dynamic scheduling of strategy functions, providing a highly reliable inference solution for motor parameter identification tasks that is behaviorally knowable, state-checkable, and error-correctable. This promotes the transformation of motor parameter identification from "black box inference" to "white box inference," and achieves the credibility of identification results through controllable inference process, fundamentally improving the operational reliability and practical acceptability of industrial AI vertical model inference engines in industrial scenarios.

[0031] Corresponding to the above method, this invention also discloses a verification-driven motor parameter interleaved reasoning identification device. This device achieves a real-time closed loop of "reasoning-verification-feedback-adjustment" by setting up a parameter input module, a reasoning execution module, and a result output module. This device embeds trust assurance into every step of the reasoning process, ensuring that intermediate inference results and final outputs satisfy physical constraints and engineering common sense, thereby shifting the parameter identification problem from a difficult-to-verify area to an easily verifiable area. Addressing the issues of opaque decision-making and difficulty in real-time verification during the reasoning process of industrial AI vertical models, this invention provides a new, controllable, and real-time reliable reasoning paradigm for the trustworthy deployment of industrial AI vertical models, expanding the traditional single reasoning action space into a joint space that integrates verification actions. Attached Figure Description

[0032] Figure 1 This is a flowchart of a motor parameter interleaving reasoning and identification method based on verification-driven method according to the present invention.

[0033] Figure 2 This is a flowchart illustrating the construction of the extended action space in this invention;

[0034] Figure 3 This is a flowchart illustrating the composition of the strategy function and the module for calling the extended action space in this invention.

[0035] Figure 4 This is a flowchart of the execution strategy function of the present invention;

[0036] Figure 5 This is a graph showing the change in the verification pass rate of the present invention;

[0037] Figure 6 This is a comparison chart of inference confidence in small data scenarios according to the present invention. Detailed Implementation

[0038] The present invention will now be described in further detail with reference to the accompanying drawings:

[0039] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0040] This invention constructs a verification-driven "inference-verification" interleaved architecture, expanding the action space into two categories: inference actions and verification actions. It integrates the core module of the policy function and the verifier index module, and dynamically interleaves the execution of inference and verification actions. Relying on the policy function (neural network), it dynamically schedules the two types of actions, and combines the built-in motor factual knowledge base as a dictionary (structured constraints) to output discrete verification states of pass, failure, or warning. At the same time, based on the self-optimization strategy of PPO (Proximal Policy Optimization), a real-time closed loop of inference and verification is formed, realizing collaborative inference of inference and verification actions, ensuring that each step of the inference process has reliable guarantees, and ultimately achieving reliable identification of motor target parameters. Moreover, the specific implementation steps are closely connected, ensuring the overall technical solution's effectiveness.

[0041] See Figure 1 The first aspect of this invention discloses a method for identifying motor parameters based on verification-driven interleaved reasoning, comprising the following steps:

[0042] S1, obtain motor operating status parameters;

[0043] S2, based on the motor operating state parameters, an inference state is constructed. The strategy function dynamically selects an interleaved action from the extended action space containing inference and verification actions according to the current inference state. If the selected interleaved action is a inference action, the inference action is executed according to the current inference state and the core parameters of the strategy function to obtain updated inference values ​​of the motor target parameters, and the inference state is updated based on the updated inference values ​​of the motor target parameters. If the selected interleaved action is a verification action, the verification action is executed by calling the built-in motor factual knowledge base through the verifier index module in the strategy function based on the latest inference values ​​of the motor target parameters to generate a verification state. Feedback is given to the strategy function according to the verification state, and the parameters of the strategy function are updated based on the feedback.

[0044] S3, when the preset termination condition is met, output the final estimated value of the motor target parameters, and calculate and output the credibility index based on the strategy function and historical data of the verification status.

[0045] In this step, the motor can be a single in-service motor or an out-of-service motor. The motor's operating status parameters are obtained through online real-time acquisition or offline historical data retrieval. Specifically, the motor's operating status parameters can be obtained through online real-time acquisition without a test bench, offline historical data retrieval without a test bench, or operating data acquisition in a test bench environment. In online scenarios, the motor's operating status parameters are periodically acquired through embedded sensors, with the sampling frequency matching the motor control cycle. In offline scenarios, the motor's operating status parameters are retrieved from the time-series dataset of the motor's historical operation, which must meet the requirements of timestamp synchronization and data integrity. Specifically, the acquired or retrieved motor operating status parameters include one or more of the following: speed, three-phase stator voltage, three-phase stator current, bus voltage, and temperature signal. This set of motor operating status parameters constitutes the initial input basis for the subsequent reasoning process, and its sampling frequency and timestamp are aligned to ensure that the operating status parameters of each motor are synchronized under a unified time reference.

[0046] In S2, the inferred value of the motor target parameter refers to the current estimate of the key electrical or magnetic circuit parameters to be identified, such as stator resistance. R s d-axis inductance L d q-axis inductance L q and flux linkage ψ f One or more of the following; In S2, inference and verification actions are executed alternately, explicitly fusing and alternating the originally separate operations of solving for motor target parameters and verifying results at the action space level; where: the inference action is a forward computation behavior aimed at optimizing motor target parameters, with the goal of improving the estimation accuracy of motor target parameters based on the current verification state; the verification action is a reverse verification behavior aimed at physical consistency and numerical rationality, with the goal of determining whether the current inferred value of motor target parameters is within the threshold range based on the constraints of the built-in motor factual knowledge base, that is, verifying the physical rationality and numerical stability of the inferred value of motor target parameters; the two do not constitute a master-slave relationship, but are executed alternately under the unified scheduling of the strategy function, forming a closed-loop logical chain of "inference-verification-feedback-adjustment". The strategy function is the core decision module for realizing the above-mentioned alternating scheduling. Its input is the current inference state and its output is the next action instruction. Its decision basis not only comes from the output of the data-driven model, but also deeply integrates the knowledge of the structured motor domain, thereby ensuring that the action selection conforms to both statistical laws and physical constraints.

[0047] The method of this invention first acquires the motor operating state parameters, then dynamically selects and executes interleaved actions from the extended action space via a strategy function. When the interleaved action is a reasoning action, the inferred value of the motor target parameter is obtained; when it is a verification action, the inferred value of the motor target parameter is verified and feedback is generated. Based on the feedback, the strategy function is updated, and the above reasoning and verification actions are repeated until the termination condition is met, resulting in an estimated value of the motor target parameter. The final output is the inferred value of the motor target parameter, along with a credibility index including the verification pass rate and the reasoning confidence. This invention deeply embeds the verification action into the reasoning chain, forming a strategy function through the coupling of the strategy core module and the verifier index module, realizing an interleaved closed loop of "reasoning-verification-feedback-adjustment" to ensure the credibility of the reasoning process. It can be routinely verified and dynamically adjusted based on the motor factual knowledge base to avoid the accumulation of erroneous reasoning. The method has a simple structure, dynamically scheduling the verification action through the strategy function, optimizing the balance between verification intensity and reasoning efficiency as needed, adapting to multiple industrial operating conditions and small data scenarios, and providing a highly reliable reasoning solution for motor parameter identification.

[0048] See Figure 2 In some embodiments of the present invention, before performing interleaved identification, an extended action space is formed by fusing the inference action space and the verification action space, thus expanding the traditional single inference action space A into an extended action space that incorporates the verification action space V. Specifically The inference action space A stores inference actions, including parameter iterative optimization, working condition adaptation rule adjustment, and neural network model weight parameter adjustment and update. The verification action space V stores verification actions, including physical constraint verification, numerical stability verification, data integrity verification, and constraint violation detection. The process specifically includes the following steps:

[0049] S201, Define a verification-driven inference environment, the inference environment including an inference action space A and a verification action space V.

[0050] Here, the inference action space A is a set of inference actions for the target parameters of the motor, including actions such as parameter iterative optimization, adjustment of operating condition adaptation rules, and adjustment and updating of neural network model weight parameters; specific parameter iterative optimizations include stator resistance. R s Gradient descent iteration of motor target parameters such as inductance, adjustment of operating condition adaptation rules such as adjustment of motor target parameter adaptation coefficients at different speeds, and adjustment and updating of neural network model weight parameters refer to the adjustment of the strategy function.

[0051] Wherein, the verification action space V is a set of actions specifically used for credibility verification, and its mathematical definition is as follows (1):

[0052] (1)

[0053] in, For physical constraint verification, physics The meaning of "physics" refers to physical laws, specifically the physical laws such as the motor voltage balance equation, which are used to verify the physical rationality of the inferred values ​​of the motor's target parameters, such as the deviation verification between theoretical current and actual observed current.

[0054] For numerical stability verification, numerical The meaning is numerical, specifically, it refers to detecting the fluctuation range of the motor target parameter inference value iteration process to ensure numerical convergence.

[0055] For data integrity verification, data The meaning of "data" is to verify whether the inferred values ​​of the target parameters of the motor are missing or abnormal.

[0056] To curb violations of testing, violation The meaning is violation, which is to verify whether the inferred values ​​of the motor target parameters conform to the typical range constraints in industrial scenarios.

[0057] S202 merges the reasoning action space and the verification action space to form an extended action space. And satisfy This extended action space serves as the action output range of the policy function, enabling dynamic scheduling of inference and verification actions.

[0058] In some embodiments of the present invention, the expression of the strategy function is:

[0059] (2)

[0060] in, For the policy function, The actions are alternating, and s represents the reasoning state. These are the core parameters of the policy function. For the validator index module, there is a set of parameters, where idx "Index" is an abbreviation for index. As the core module of the strategy function, core The meaning of "core"; The output of the validator index module represents the dynamic constraint strength of the motor factual knowledge base on the policy function decision, with a value ranging from 0 to 1. The policy function integrates the validator index module of the motor factual knowledge base and is deeply integrated with the inference strategy, thus constructing the validator index module and the core policy function module. The integrated architecture guides the dynamic interleaving of inference and verification in the A' space through policy functions.

[0061] Further, see Figure 3 The core components of the policy function are disclosed; the policy function is a neural network, where... The alternating action is either a reasoning action or a verification action; s is the reasoning state, which includes motor operating state parameters, historical reasoning results and historical verification states. Specifically, the motor operating state parameters are one or more of the following: speed, three-phase stator voltage, three-phase stator current, bus voltage and temperature signal. The core parameter of the policy function is the weight matrix and bias term set of the neural network, which is used to control the decision logic of the inference action. This is the parameter set of the verifier index module, associated with constraint information from the motor factual knowledge base. The core module of the strategy function represents the decision function that outputs interleaved actions a′ based solely on the current reasoning state s without considering the constraints of the motor factual knowledge base. It is the core of pure data-driven reasoning decision-making. The output of the verifier index module represents the dynamic constraint strength of the motor factual knowledge base on the policy function decision. The value range is 0~1, realizing the adaptive adjustment of common sense constraints. For example, when the inferred value of the motor target parameter deviates from the physical range, a high-weighted strengthening constraint is output.

[0062] Furthermore, the historical inference results in the inference state refer to the serialized record of the aforementioned motor target parameter inference values ​​at several historical moments. This record is used to document the evolution trend of the motor target parameters, identify oscillation or divergent behavior, and support the adjustment of working condition adaptation rules and dynamic step size control. The historical verification state refers to the verification state generated by the previous verification action, including pass, failure, and warning. It can also be used to synchronously feed back verification thresholds, deviations, and whether constraint violations are triggered.

[0063] See Figure 4 In a specific example, the strategy function works as follows:

[0064] First, obtain the motor operating state parameters and generate the inference state s.

[0065] Then, the policy function is calculated based on formula (2), where the core module of the policy function is... Output action decision value, The dynamic constraint strength between 0 and 1 is output, and the two are coupled to obtain the final action selection probability, thus expanding the action space. Dynamic selection of interleaved actions .

[0066] If the interleaved action is a reasoning action: call the core parameter of the policy function. Generate specific reasoning actions (such as parameter iteration optimization and working condition adaptation rule adjustment), and obtain the inferred values ​​of the target parameters of the motor after execution.

[0067] If the interleaved action is a verification action: the built-in motor factual knowledge library is retrieved through the verifier index module; the strategy function has selected the specific verification type (such as physical constraint verification, constraint violation detection) from the extended action space A', and then infers the type of the current motor target parameter value (such as stator resistance). R s d-axis inductance L d It retrieves the threshold or range of the corresponding parameter from the motor factual knowledge base, performs multi-dimensional verification, and generates and returns the verification status (pass, failure, warning).

[0068] The strategy function receives the verification status as feedback: if the verification status is passed, the inference status s is updated, including: adding the currently verified motor target parameter inference value to the historical inference result record, updating the verification status history, retaining the motor operating status parameters, and then continuing the inference of the next round of motor target parameter inference values; if the verification status is failed, it is backtracked to the most recent passed inference status and the inference step size is shortened; if the verification status is a warning, only the verification frequency is adjusted, and the parameter iteration direction is optimized to avoid error accumulation.

[0069] Repeat the above-mentioned interleaved closed loop of "reasoning-verification-feedback-adjustment" until the preset termination condition is met, and output the estimated value of the target parameters of the motor and the credibility index calculated based on the verification pass rate and reasoning confidence.

[0070] Furthermore, the parameter set of the validator index module The built-in motor factual knowledge library is categorized into three levels: physical constraint thresholds, typical parameter ranges, and operating condition adaptation rules. The first level, the physical constraint threshold library, contains the fundamental physical equations and constraint thresholds of motors, used to compare the deviation between the inferred values ​​of the motor's target parameters and the theoretical values ​​derived from the physical equations (such as the current value calculated by the voltage balance equation) and whether the deviation is within a preset threshold. The second level, the typical parameter range library, contains parameter value ranges categorized by motor type (such as permanent magnet synchronous motors and induction motors), used to check whether the inferred values ​​of the motor's target parameters fall within the industrially reasonable range for the corresponding type of motor. The third level, the operating condition adaptation rule library… The core of the matching rule library is the parameter adaptation logic under different operating conditions (such as speed, load, and temperature), used to verify whether the inferred values ​​of the motor target parameters match the dynamic adaptation requirements of the current operating condition. This three-level classification storage is not only highly structured, reusable, and easy to maintain, but also reduces redundant calculations in the inference process through precise layered judgment. It also provides clear logical basis for subsequent verification status output, ensuring that valid verification results are only generated when parameters fully comply with physical constraints, typical parameter ranges, and operating condition adaptation rule requirements; invalid results are generated when parameters violate these rules; and results requiring adjustment are generated for non-critical adaptation deviations. It should be understood that this motor factual knowledge library and extended action space... While not the same concept, they have a precise matching relationship: the extended action space defines the "type of verification action," and the motor factual knowledge base provides the "verification basis for verification action." Specifically, physical constraint verification depends on the physical constraint threshold library, constraint violation detection depends on the typical range of parameters library, and numerical stability verification depends on the working condition adaptation rule library. This ensures that each type of verification action is supported by a clear motor factual knowledge base, avoiding verification without a basis.

[0071] The verifier index module ensures the structured and callable nature of common-sense constraints, for example, physical constraint thresholds such as stator resistance. R s Common range, d-axis inductance L d and q-axis inductance L q The typical ratio, obtained through the parameter set of the validator index module. A neural network that embeds a common-sense database of motor facts into a policy function.

[0072] In some embodiments of the present invention, the verification status includes pass, failure, and warning; if the execution result is pass, the inference status is updated and the next round of inference is performed; if the execution result is failure, the process is backtracked to the last pass inference status and the step size of the inference action is shortened; if the execution result is warning, the verification frequency is adjusted and the parameter iteration direction is optimized to avoid error accumulation.

[0073] If the inferred value of the motor target parameter is within the first threshold range of the motor factual knowledge base, the verification status is passed; if the inferred value of the motor target parameter is outside the first threshold range of the motor factual knowledge base but within the tolerance range or only triggers non-critical constraints (such as slight deviation in numerical accuracy or missing minor data integrity items), the verification status is a warning; if the verification status is not passed or a warning, it is a failure.

[0074] In a specific example of the present invention, a real-time closed-loop process of "reasoning-verification-feedback-adjustment" is disclosed, including the following steps:

[0075] S211 (Inference): The policy function expands the action space based on the current inference state s. Select reasoning action (belonging to) The subset output is used to initialize or update the inferred values ​​of the motor target parameters; the action selection range of the policy function is always the expanded action space. By dynamically determining whether to output reasoning or verification actions, the "reasoning-verification" process is executed alternately.

[0076] S212 (Verification): The strategy function outputs a verification action. This verification action calls the motor factual knowledge base through the verifier index module to perform multi-dimensional verification on the current motor target parameter inference value, and outputs three verification states: pass, failure, and warning, as inference feedback constraints of the strategy function.

[0077] S213 (Feedback): The policy function accepts the verification status as feedback and dynamically adjusts its subsequent decision logic accordingly.

[0078] If the verification status is passed, maintain the current inference step size, verification frequency, and verification strength, and continue to the next round of inference.

[0079] If the verification status fails in a single verification action, the error recovery mechanism is triggered, which backtracks to the most recent inference node, adjusts the step size of the inference action to half of the original step size, increases the verification strength, and increases the verification frequency.

[0080] If the verification status is a warning, fine-tune the verification frequency and optimize the parameter iteration direction to avoid error accumulation.

[0081] Among them, the step size of the inference action refers to the update range of the inference value of the motor target parameter inference (such as the learning rate of gradient descent), which affects the inference speed; the verification frequency refers to how many rounds of inference actions are needed to perform a complete verification, which affects the balance between verification accuracy and efficiency.

[0082] S214 (Adjustment), Iterative Loop: Based on the adjusted inference state, repeat steps S211 to S213 to form a real-time closed loop. In this loop, because each round of interleaved execution verifies the inference value of the motor target parameter through the motor factual knowledge base, erroneous inferences are promptly backtracked and corrected. The inference value of the motor target parameter gradually converges to a physically reasonable range, and the strategy function is continuously optimized with verification feedback. Therefore, the verification pass rate gradually increases with the number of inference steps and approaches the preset target threshold, thereby realizing the continuous enhancement of the credibility of the inference process by the verification-driven mechanism.

[0083] In some embodiments of the present invention, if the proportion of the number of failures recorded in the verification status to the number of verification actions executed exceeds a set second threshold... If a systematic bias is found in the decision logic of the policy function, the core parameters of the policy function are updated using the PPO method. The strength amplification factor *m* of the verifier index module's output is adjusted synchronously to strengthen the common sense constraint. It should be noted that the recorded number of failures is the number of failures recorded over N consecutive periods.

[0084] Specifically, the verification failure rate is defined as the proportion of the number of failures in the verification state to the number of verification actions executed, and is calculated using the following formula:

[0085] (3)

[0086] in, To verify the failure rate, fail The meaning of failure To verify the number of passes, pass The meaning of "passing" To verify the number of times the action was executed, val The meaning of validation. This is a preset threshold.

[0087] When the above conditions are met, the following policy update will be performed:

[0088] First, update the strength amplification factor m of the verifier index module's output (using the following formula) to strengthen the constraint strength of the motor factual knowledge base and prevent the policy function from continuing to output reasoning actions that violate common sense:

[0089] (4)

[0090] Where m is the intensity amplification factor, used to adjust... The output amplitude needs to be clarified. The value is a dynamic constraint strength ranging from 0 to 1. It is coupled with the core module of the policy function through formula (2) to modulate the output distribution of the policy function: the larger the value, the higher the selection weight of the verification action in the extended action space, forming a "strong verification-cautious reasoning" mode; the smaller the value, the more the policy function depends on data-driven, forming a "fast reasoning-light verification" mode. This mechanism indirectly constrains the reasoning process through the dynamic scheduling of verification frequency.

[0091] In this process, the core parameters of the policy function are updated using the PPO method. This optimizes the decision-making logic of inference actions, enabling the policy function to learn to avoid actions that are prone to failure, thereby improving the accuracy of inference. During the PPO method update process, the parameter set of the validator index module is frozen. To avoid violating the common sense constraint mechanism, the process optimizes reasoning and decision-making without changing the common sense constraint rules.

[0092] Through the above process, the policy function outputs an interleaved action sequence. k is the number of interleaved actions, which are executed in a loop according to the logic of "reasoning-verification-feedback-adjustment". The verification results are fed back to the strategy function in real time, and the output of the next round of actions is dynamically adjusted.

[0093] Specifically, the policy function is updated using the PPO method, with the validation results as constraints. The objective function of the PPO method is... for:

[0094] (5)

[0095] in: These are the learnable parameters of the policy function. To sample the expected action based on the old policy function, For the interleaved action sequence sampled from the old policy function, This refers to the old policy function, i.e., the policy function before the update. "Old" means "old". min( The function is used to find the minimum value; the Chinese meaning of "min" is "small". Output the decision values ​​for the interleaved actions for the current policy function. The decision values ​​of the old policy function before the update all implicitly depend on the inference state s and the core parameters of the policy function. and the parameter set of the validator index module , The advantage function is based on the validation results. val for validation The abbreviation for "verification" in Chinese means "verification". clip ( ) is the clipping function. clip The Chinese meaning is truncation, core association verification pass rate. With verification state value estimation, it is used to quantify the optimization direction of the policy function under verification constraints. This is the cutting factor. This is used to limit the update magnitude of the policy function and ensure training stability.

[0096] It should be noted that the two update strategies described above are two complementary error handling mechanisms of this invention, each operating at a different time scale:

[0097] (1) Immediate error recovery at the micro level (action execution layer): In a single verification action, if the verification status is failed (as described in S213), the error recovery mechanism is immediately triggered, backtracking to the most recently passed inference node, adjusting the step size of the inference action to half of the original step size, and increasing the verification strength. This mechanism is designed for single inference errors, ensuring that local errors do not accumulate, and belongs to the fault tolerance at the action execution level.

[0098] (2) Macro-level strategy optimization (strategy learning layer): When the proportion of recorded failures to the number of verification actions exceeds the set second threshold, it is determined that there is a systematic bias in the decision-making logic of the strategy function. At this time, the core parameters of the strategy function are updated through the PPO method. And simultaneously adjust the strength amplification factor m of the verifier index module to strengthen the common sense constraint strength.

[0099] In some embodiments of the present invention, the loop is terminated and the current motor target parameter inference value is output as the motor target parameter estimate value when at least one of the following termination conditions is met: the motor target parameter inference value converges, the verification pass rate reaches a preset target threshold, or the preset maximum number of iterations is reached.

[0100] The convergence of the motor target parameter inference value terminates when the change in the motor target parameter inference value is less than the preset accuracy threshold during a preset number of iterations (e.g., 10 times).

[0101] The verification pass rate reaches the preset target threshold, which means that the verification pass rate reaches or exceeds the preset target threshold (such as 90%).

[0102] Reaching the preset maximum number of iterations means that the number of interleaved executions reaches the preset maximum number of iterations limit (e.g., the limit is 100 times).

[0103] If any termination condition is not met, the next round of S211-S213 will continue.

[0104] In some embodiments of the present invention, the credibility metrics include verification pass rate and inference confidence.

[0105] Among them, the verification pass rate The calculation formula is:

[0106] (6)

[0107] in, To verify the pass rate, rate The meaning is proportion. To verify the number of passes, To verify the number of times the action was executed.

[0108] The formula for calculating the inference confidence level is as follows:

[0109] (7)

[0110] in, The term "confidence" refers to the core confidence level. Expectation operator The output of the validator index module, For the policy function, To determine the optimal interleaved actions, the reliability of the estimated motor target parameters output by the quantization policy function is determined, where s represents the inference state. These are the core parameters of the policy function. This is the parameter set for the validator index module.

[0111] In a specific example

[0112] The estimated target parameters of the motor are as follows: ,in For stator resistance, For d-axis inductance, It is the q-axis inductance. For flux linkage, the Chinese meaning of "final" is "final".

[0113] A second aspect of the present invention discloses a motor parameter interleaving reasoning identification device based on verification-driven methods, comprising:

[0114] The parameter input module is used to acquire motor operating status parameters;

[0115] The inference execution module is used to construct an inference state based on the motor operating state parameters. The strategy function dynamically selects an interleaved action from an extended action space containing inference and verification actions based on the current inference state. If the selected interleaved action is a inference action, the inference action is executed based on the current inference state and the core parameters of the strategy function to obtain updated inference values ​​for the motor target parameters, and the inference state is updated based on these updated values. If the selected interleaved action is a verification action, the verification action is executed based on the latest inference values ​​for the motor target parameters by calling the built-in motor factual knowledge base through the verifier index module in the strategy function to generate a verification state. Feedback is provided to the strategy function based on the verification state, and the parameters of the strategy function are updated based on the feedback.

[0116] The result output module, when the preset termination condition is met, is used to output the final estimated value of the motor target parameters, and calculate and output the credibility index based on the strategy function and historical data of the verification status.

[0117] In some embodiments of the present invention, the inference execution module includes:

[0118] The inference environment construction module is used to define the inference action space A, the verification action space V, and the extended action space. ;

[0119] The integrated strategy index module is used to construct a strategy function containing a validator index module, and outputs an interleaved action sequence in the A′ space;

[0120] The interleaved inference execution module is used to execute actions according to the logical scheduling of "inference-verification-feedback-adjustment" to achieve dynamic interleaving;

[0121] The strategy optimization module is used to update strategy parameters based on the validation results using the PPO method, and to perform adaptive adjustments and backtracking.

[0122] The third aspect of this invention discloses a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used to implement a verification-driven motor parameter interleaving reasoning and identification method, including: S1, acquiring motor operation... S2, Based on the motor operating state parameters, construct an inference state. The strategy function dynamically selects an interleaved action from the extended action space containing inference and verification actions according to the current inference state. If the selected interleaved action is an inference action, execute the inference action according to the current inference state and the core parameters of the strategy function to obtain the updated inference value of the motor target parameter, and update the inference state based on the updated inference value of the motor target parameter. If the selected interleaved action is a verification action, execute the verification action by calling the built-in motor factual knowledge base through the verifier index module in the strategy function based on the latest inference value of the motor target parameter to generate a verification state. Feedback is given to the strategy function according to the verification state, and the parameters of the strategy function are updated based on the feedback. S3, When the preset termination condition is met, output the final estimated value of the motor target parameter, and calculate and output the credibility index based on the historical data of the strategy function and the verification state.

[0123] A fourth aspect of this invention discloses a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement a verification-driven motor parameter interleaved inference identification method in the above embodiments, including: S1, acquiring motor operating state parameters; S2, constructing an inference state based on the motor operating state parameters, and dynamically selecting an interleaved action from an extended action space containing inference and verification actions according to the current inference state; if the selected interleaved action is an inference action, the inference action is executed according to the current inference state and the core parameters of the strategy function to obtain an updated inference value of the motor target parameter, and the inference state is updated based on the updated inference value of the motor target parameter; if the selected interleaved action is a verification action, the verification action is executed by calling the built-in motor factual knowledge base through the verifier index module in the strategy function based on the latest inference value of the motor target parameter to generate a verification state; feedback is given to the strategy function according to the verification state, and the parameters of the strategy function are updated based on the feedback; S3, when a preset termination condition is met, the final estimated value of the motor target parameter is output, and a credibility index is calculated and output based on the historical data of the strategy function and the verification state.

[0124] The following description, in conjunction with specific embodiments, provides further details.

[0125] This embodiment takes industrial permanent magnet synchronous motors as the identification object and focuses on stator resistance in small data scenarios. d-axis inductance q-axis inductance and flux linkage For trusted identification, the first step is to construct a verification-driven inference environment: the inference action space A specifically includes the target parameters of the motor. , and The gradient descent iterative optimization action, and the target parameters of the motor under different speed conditions. , and The adaptation coefficient adjustment action; the verification action space V covers physical constraint verification (current deviation verification based on voltage balance equation), numerical stability verification (parameter iteration fluctuation amplitude detection), data integrity verification (verification of no missing running data), and constraint violation detection (parameter range compliance verification), and expands the action space to The action scheduler enables dynamic collaborative scheduling of inference actions and verification actions.

[0126] Based on the aforementioned reasoning environment, a policy-index integrated architecture is constructed: the policy function adopts a structure combining convolutional neural networks and fully connected networks, with inputs including real-time motor operating state parameters (speed, three-phase voltage, three-phase current), historical reasoning results, and historical verification states; the verifier index module incorporates a three-level structured motor factual knowledge base, where physical constraint thresholds include... Common range / Typical ratios, typical ranges of motor target parameters, and correlation of inductance variation trends at different speeds with operating condition adaptation rules; through the parameter set of the verifier index module. Embed the motor factual knowledge base into the strategy function, where When parameters deviate from the physical range, output a high weight (e.g., 0.9) to strengthen the constraint.

[0127] The interleaved inference execution process follows a dynamic loop of "inference-verification-feedback-adjustment": the policy function first outputs the inference action and initializes... , and This then triggers a verification action. Substituting the values ​​into the voltage balance equation, the deviation between the theoretical current and the actual observed current is calculated. If the deviation exceeds the first threshold, the verification state output fails. After receiving feedback, the strategy function will output the verifier index module's output. Adjust, revert to the initial state and reduce the step size of the reasoning action to half, re-output the reasoning action, and update. , and After performing the verification action again, if the deviation drops below the first threshold, the verification status outputs "pass". The performance change during this process is as follows: Figure 5 As shown in the figure, the verification pass rate changes over time. The horizontal axis represents the number of inference steps (0-100), and the vertical axis represents the verification pass rate. The curve represents the verification pass rate of this method. As the number of inference steps increases, the pass rate gradually increases from about 0.5 initially, and eventually approaches the target threshold of 0.9 (dashed line). This intuitively demonstrates the continuous strengthening effect of the verification-driven mechanism on the credibility of the inference process.

[0128] The policy function is updated using the PPO method, with the failure rate as the core constraint. During the update process, the parameter set of the validator index module is frozen. To maintain the stability of the constraints imposed by the motor factual knowledge base; the final output includes the estimated values ​​of the motor target parameters, the verification pass rate, and the inference confidence level, clarifying whether all estimated values ​​of the motor target parameters meet the accuracy requirements of the motor's physical characteristics and industrial control.

[0129] To address the small data limitations common in industrial scenarios, this embodiment compares the inference confidence performance of our method with that of traditional non-verification methods, such as... Figure 6 As shown in the figure, this graph compares the confidence levels under small data scenarios. The horizontal axis represents the amount of training data (10-100), and the vertical axis represents the inference confidence level. Black bars represent the proposed method (validation-driven), while uncolored bars represent the traditional method (no validation). The graph shows that when the training data amount is only 10, the inference confidence level of the proposed method reaches 0.82, while the traditional method is only 0.53. Even when the data amount increases to 50, the confidence level of the proposed method remains high at 0.92, while the traditional method only increases to 0.72. This result fully demonstrates that the proposed method can effectively maintain high reliability of the inference results under small data scenarios, adapting to the practical application needs of industrial motors lacking bench test data.

[0130] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for verifying and driving motor parameter interlaced reasoning identification based on, characterized in that, include: Obtain motor operating status parameters; Based on the motor operating state parameters, an inference state is constructed. The strategy function dynamically selects an interleaved action from the extended action space containing inference actions and verification actions according to the current inference state. If the selected interleaved action is an inference action, the inference action is executed according to the current inference state and the core parameters of the strategy function to obtain the updated motor target parameter inference value, and the inference state is updated based on the updated motor target parameter inference value. If the selected interleaving action is a verification action, then based on the latest motor target parameter inference value, the verification action is executed by calling the built-in motor factual knowledge library through the verifier index module in the strategy function to generate a verification state; based on the verification state, feedback is given to the strategy function, and the parameters of the strategy function are updated based on the feedback; The expression for the policy function is: in, ( ) is the policy function. For alternating movements, In the state of reasoning, This is the core parameter of the policy function. This is the parameter set for the validator index module. This is the core module of the strategy function; The output of the verifier index module represents the dynamic constraint strength of the motor factual knowledge base on the policy function decision, with a value range of 0 to 1; When the preset termination condition is met, the final estimated value of the motor target parameters is output, and the credibility index is calculated and output based on the strategy function and historical data of the verification status.

2. The method for motor parameter interleaving reasoning and identification based on verification-driven approach according to claim 1, characterized in that, The reasoning actions include parameter iterative optimization, working condition adaptation rule adjustment, and neural network model weight parameter adjustment and update; the verification actions include physical constraint verification, numerical stability verification, data integrity verification, and constraint violation detection.

3. The method for motor parameter interleaving reasoning and identification based on verification-driven approach according to claim 1, characterized in that, The reasoning status includes motor operating status parameters, historical reasoning results, and historical verification status.

4. The method for motor parameter interleaving reasoning and identification based on verification-driven approach according to claim 1, characterized in that, The verification status includes pass, failure, and warning; if the verification status is pass, the inference status is updated and the next round of inference is performed; if the verification status is failure, the process is backtracked to the last pass inference status and the step size of the inference action is shortened; if the verification status is warning, the verification frequency is adjusted. If the inferred value of the motor target parameter is within the first threshold range of the motor factual knowledge base, the verification status is passed; if the inferred value of the motor target parameter is outside the first threshold range of the motor factual knowledge base but within the tolerance range or only triggers non-critical constraints, the verification status is a warning; if the verification status is neither passed nor a warning, it is a failure.

5. The method for motor parameter interleaving reasoning and identification based on verification-driven approach according to claim 4, characterized in that, If the proportion of the number of failures recorded in the verification status to the number of verification actions executed exceeds a set second threshold, the policy function is updated using the near-end policy optimization method, and the dynamic constraint strength of the output of the verifier index module in the policy function is updated.

6. The method for motor parameter interleaving reasoning and identification based on verification-driven approach according to claim 4, characterized in that, The credibility metrics include verification pass rate and inference confidence; the verification pass rate is the proportion of verification actions executed when the verification status is passed. The formula for calculating the inference confidence level is as follows: in, For the confidence level of the inference, Expectation operator To determine the dynamic constraint strength of the verifier index module, For the policy function, For optimal alternating movements, In the state of reasoning, This is the core parameter of the policy function. This is the parameter set for the validator index module.

7. A motor parameter interleaving reasoning and identification device based on verification-driven methods, characterized in that, include: The parameter input module is used to acquire motor operating status parameters; The inference execution module is used to construct an inference state based on the motor operating state parameters. The strategy function dynamically selects an interleaved action to execute from the extended action space containing inference actions and verification actions according to the current inference state. If the selected interleaved action is an inference action, the inference action is executed according to the current inference state and the core parameters of the strategy function to obtain the updated motor target parameter inference value, and the inference state is updated based on the updated motor target parameter inference value. If the selected interleaving action is a verification action, then based on the latest motor target parameter inference value, the verification action is executed by calling the built-in motor factual knowledge library through the verifier index module in the strategy function to generate a verification state; based on the verification state, feedback is given to the strategy function, and the parameters of the strategy function are updated based on the feedback; The expression for the policy function is: in, ( ) is the policy function. For alternating movements, In the state of reasoning, This is the core parameter of the policy function. This is the parameter set for the validator index module. This is the core module of the strategy function; The output of the verifier index module represents the dynamic constraint strength of the motor factual knowledge base on the policy function decision, with a value range of 0 to 1; The result output module, when the preset termination condition is met, is used to output the final estimated value of the motor target parameters, and calculate and output the credibility index based on the strategy function and historical data of the verification status.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the verification-driven motor parameter interleaving reasoning identification method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the verification-driven motor parameter interleaving reasoning identification method as described in any one of claims 1 to 6.

Images