Artificial intelligence model control system and control method

Abstract

The invention relates to the technical field of artificial intelligence control, and discloses an artificial intelligence model control system and method, and the system comprises an equipment data collection module which is used for collecting the measurement data of edge equipment, and constructing a virtual object data unit; the data preprocessing module is used for preprocessing the virtual object data unit and respectively writing the virtual object data unit into an instant message area and a simulated object memory area; the model reasoning module is used for constructing a first feature vector and inputting the first feature vector into a random forest model to obtain a model prediction result; the credibility calculation module is used for calculating a credibility weight according to a first preset function; the control execution module is used for correcting the model prediction result and generating a current control output value; and the model adjustment module is used for constructing control behavior change characteristics and adjusting the artificial intelligence model according to the first control strategy. The real-time performance, the self-adaptive capability and the abnormal data processing capability of the control system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence control, and particularly relates to an artificial intelligence model control system and a control method. Background Art

[0002] At present, in the field of industrial control and automation, traditional control systems mostly adopt classical control strategies, such as PID control. Although these control methods are simple and effective, they usually have some limitations. For example, PID control cannot effectively cope with system nonlinearity and large-scale changes, and its parameter adjustment is rather cumbersome, making it difficult to achieve automatic optimization under real-time conditions.

[0003] With the rapid development of artificial intelligence technology, more and more control systems have begun to introduce machine learning and deep learning methods in order to improve control accuracy and system stability. Currently, machine learning-based control systems have gradually been applied to some complex systems, such as autonomous driving, smart grids, and industrial robots. However, these systems still have some problems: 1. Real-time problem: Currently, machine learning-based control systems often require a long time for training and data processing, making it difficult to adapt to high-dynamic real-time control tasks; 2. Poor adaptive ability of the model: Traditional machine learning models are difficult to automatically adjust parameters according to environmental changes, resulting in a lag in the response of the control system when facing sudden changes; 3. Problem of handling abnormal data: Abnormal data or noise in the control system can cause incorrect model output, affecting the stability and reliability of the control system. Summary of the Invention

[0004] The present invention provides an artificial intelligence model control system and a control method to solve the technical problems of poor real-time performance, insufficient model adaptive ability, and improper handling of abnormal data in related technologies.

[0005] The present invention provides an artificial intelligence model control system, including:

[0006] An equipment data acquisition module, configured to collect measurement data of multiple edge devices and construct corresponding virtual object data units;

[0007] A data preprocessing module, configured to preprocess the virtual object data units and write them into the immediate information area and the object-like memory area respectively, where the immediate information area is used to provide input data for the artificial intelligence model at the current moment, and the object-like memory area is used to store historical state data and execution records;

[0008] A model inference module, configured to extract the preprocessed virtual object data units from the immediate information area, construct a first feature vector, and input the first feature vector into a random forest model for inference to obtain a model prediction result;

[0009] A credibility calculation module, which is used to obtain the feedback response value returned by an external device after the execution of the previous control operation, calculate the prediction error based on the feedback response value and the current model prediction result, and calculate the credibility weight output by the artificial intelligence model according to a first preset function, where the external device represents the target control object for executing the control instruction generated by the artificial intelligence model;

[0010] A control execution module, which is used to correct the model prediction result according to the credibility weight, perform weighted processing in combination with the feedback response value of the previous control operation, generate the current control output value, and then generate a control instruction, and send it to the external device through the communication module to execute the corresponding control operation;

[0011] A model adjustment module, which is used to construct the control behavior change characteristics according to the current control output value and the previous control output value, write the prediction error, credibility weight, feedback response value and control behavior change characteristics into the anthropomorphic memory area, and then adjust the artificial intelligence model according to the first control strategy.

[0012] Further, the measurement data includes: temperature, current, voltage and frequency, and the virtual object data unit includes: device identifier, timestamp, temperature, current, voltage and frequency.

[0013] Further, the steps of preprocessing include:

[0014] S201, for each parameter of the measurement data, based on the measurement values at the current time point and its previous two time points, use the linear regression method to predict the trend value of the parameter at the current time point;

[0015] S202, calculate the offset between the measurement value at the current time point and the trend value, and use the offset as the input feature of the corresponding parameter;

[0016] S203, regard the parameter whose offset exceeds the first preset threshold as an outlier, and use the trend value at the current time point for replacement processing;

[0017] S204, perform normalization processing on the processed measurement data using the maximum-minimum normalization method.

[0018] Further, the first feature vector is composed of the offsets of the parameters in the preprocessed virtual object data unit, and the parameters include: temperature, current, voltage and frequency, where the expression of the first feature vector is: , represents the offset of temperature, represents the offset of current, represents the offset of voltage, represents the offset of frequency.

[0019] Further, input the first feature vector into the random forest model for inference to obtain the model prediction result. The specific steps include:

[0020] S301: Input the first feature vector into multiple decision trees in the random forest model respectively. Each decision tree performs path traversal on the first feature vector based on the internal partitioning rule until it traverses to the leaf node and outputs a sub-prediction value defined by the leaf node. The internal partitioning rule is used to perform feature partitioning on non-leaf nodes, and the feature partitioning includes judging the branch direction of the traversal path according to the feature splitting threshold set on the non-leaf node.

[0021] S302: Perform result fusion on the sub-prediction values output by multiple decision trees, and use the weighted average method to generate the model prediction result.

[0022] Further, the prediction error is the absolute difference between the current model prediction result and the feedback response value. The first preset function is constructed by the prediction error and the average value of the prediction errors in the past n rounds. n represents the preset number of historical control operation rounds. The determination process of the first preset function is as follows:

[0023] Obtain the model prediction result in the current round t and the feedback response value returned by the external device, calculate the absolute difference between the two to obtain the prediction error in the current round.

[0024] Based on the historical control operation rounds, extract the n-round prediction error values from round t - n to round t, where n is the preset number of historical rounds. Calculate the arithmetic average of the prediction error values to obtain the average value of the prediction errors in the past n rounds; Add a very small value between 0 and 1 to the average value of the prediction errors in the past n rounds to prevent the divisor from being zero; Divide the prediction error in the current round by the adjusted average value to obtain a ratio; Raise the ratio to the power of the second weight coefficient to obtain the result of the power processing; Multiply the first weight coefficient by the result of the power processing to obtain the scaled processing value; Add the value 1 to the scaled processing value to obtain the sum value; Take the reciprocal of the sum value to obtain the credibility weight in the current round, which is used as the output result of the first preset function.

[0025] Further, correcting the model prediction result according to the credibility weight includes: According to the deviation between the current model prediction result and the feedback response value in the previous round, combined with the current credibility weight, perform weighted fusion processing to obtain the current corrected control output value.

[0026] Further, the control behavior change feature is obtained through the following steps:

[0027] Calculate the first difference between the corrected control output value in the t-th round and the corrected control output value in the (t - 1)-th round; perform a square operation on the first difference to obtain the first squared result;

[0028] Obtain the standard deviation of the control output value, and add a very small value between 0 and 1 to the standard deviation to get the denominator adjustment value; divide the first squared result by the denominator adjustment value to obtain the first ratio; process the first ratio to obtain the first sum value;

[0029] Perform a natural logarithm operation on the first sum value to obtain the first logarithmic result; calculate the second difference between the corrected control output value in the (t - 1)-th round and the corrected control output value in the (t - 2)-th round; calculate the absolute difference between the first difference and the second difference;

[0030] Multiply the preset weight coefficient by the absolute difference to obtain the product result; calculate the power of the natural constant e to the power of the product result to obtain the exponential operation result;

[0031] Process the exponential operation result to obtain the second sum value; multiply the first logarithmic result by the second sum value to obtain the final control behavior change characteristic value.

[0032] Further, the first control strategy includes:

[0033] When the duration for which the control behavior change characteristic exceeds the second preset threshold exceeds the first preset time period, switch the currently used artificial intelligence model to the pre-set spare model;

[0034] Based on the paired data stored in the anthropomorphic memory area, perform a local re-training operation on the decision tree in the current artificial intelligence model, where the paired data includes: the first feature vector, the model prediction result, and the feedback response value; the local re-training operation includes: adjusting the output value of the leaf node and modifying the feature splitting threshold of the non-leaf node.

[0035] The present invention provides an artificial intelligence model control method, including the following steps:

[0036] S401, collect measurement data of multiple edge devices and construct corresponding virtual object data units;

[0037] S402, preprocess the virtual object data units and write them into the immediate information area and the anthropomorphic memory area respectively, where the immediate information area is used to provide the input data of the artificial intelligence model at the current moment, and the anthropomorphic memory area is used to store historical state data and execution records;

[0038] S403. Extract the preprocessed virtual object data unit from the instant information area, construct the first feature vector, and input the first feature vector into the random forest model for inference to obtain the model prediction result;

[0039] S404. Obtain the feedback response value returned by the external device after the execution of the previous control operation, calculate the prediction error based on the feedback response value and the current model prediction result, and calculate the confidence weight output by the artificial intelligence model according to the first preset function, where the external device represents the target control object for executing the control instruction generated by the artificial intelligence model;

[0040] S405. Correct the model prediction result according to the confidence weight, and perform weighted processing in combination with the feedback response value of the previous control operation to generate the current control output value, and then generate a control instruction, which is sent to the external device through the communication module to execute the corresponding control operation;

[0041] S406. Construct the control behavior change feature according to the current control output value and the previous control output value, and write the prediction error, confidence weight, feedback response value and control behavior change feature into the anthropomorphic memory area, and then adjust the artificial intelligence model according to the first control strategy.

[0042] The beneficial effects of the present invention are as follows: By setting up the instant information area and the anthropomorphic memory area, the present invention processes real-time control data and historical data respectively, optimizes the data access speed, reduces I / O conflicts, and supports efficient model inference and control strategy adjustment; By combining real-time feedback and historical data, the present invention uses the confidence calculation module to dynamically adjust the model output to ensure that the model can be adaptively adjusted under different control environments, thereby improving the stability and reliability of the system; From data collection, preprocessing, model inference to control execution, the present invention adopts dynamic correction and model adaptive adjustment strategies to ensure control accuracy and system stability, and can effectively handle various complex control tasks. Brief Description of the Drawings

[0043] Figure 1 is a schematic diagram of the modules of an artificial intelligence model control system of the present invention. Detailed Embodiments

[0044] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the scope of protection of the content of this specification. Each example can omit, substitute or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0045] AsFigure 1 As shown in Figure 1 , an artificial intelligence model control system includes:

[0046] An equipment data acquisition module 101, configured to collect measurement data of multiple edge devices and construct corresponding virtual object data units;

[0047] A data preprocessing module 102, configured to preprocess the virtual object data units and write them into an instant information area and an object-like memory area respectively. Among them, the instant information area is used to provide input data of the artificial intelligence model at the current moment, and the object-like memory area is used to store historical state data and execution records;

[0048] A model inference module 103, configured to extract the preprocessed virtual object data units from the instant information area, construct a first feature vector, and input the first feature vector into a random forest model for inference to obtain a model prediction result;

[0049] A credibility calculation module 104, configured to obtain a feedback response value returned by an external device after the execution of the previous control operation, calculate a prediction error based on the feedback response value and the current model prediction result, and calculate a credibility weight output by the artificial intelligence model according to a first preset function. Among them, the external device refers to a target control object for executing control instructions generated by the artificial intelligence model;

[0050] A control execution module 105, configured to correct the model prediction result according to the credibility weight, perform weighted processing in combination with the feedback response value of the previous control operation, generate a current control output value, and then generate a control instruction, and send it to the external device through a communication module to execute corresponding control operations;

[0051] A model adjustment module 106, configured to construct a control behavior change feature according to the current control output value and the previous control output value, write the prediction error, credibility weight, feedback response value, and control behavior change feature into the object-like memory area, and then adjust the artificial intelligence model according to a first control strategy.

[0052] In an embodiment of the present invention, the measurement data includes: temperature, current, voltage, and frequency. The virtual object data unit refers to encapsulating measurement data into a standardized and reusable data structure with a single edge device as the granularity. The virtual object data unit includes: device identifier, timestamp, temperature, current, voltage, and frequency.

[0053] In an embodiment of the present invention, an edge device refers to a field data acquisition device used to collect measurement data such as temperature, current, voltage, and frequency, including: a temperature sensor, a current sensor, a voltage monitoring unit, and a frequency acquisition component; an external device refers to a target control object used to execute control instructions generated by an artificial intelligence model, including an electric actuator, a drive module, or other response devices, which can return status information or response values after executing the instructions.

[0054] The device data acquisition module establishes connections with each sensor through a standardized acquisition interface RS485 serial port, and polls the data of all edge devices according to a preset acquisition cycle; further, the data acquisition module performs field parsing and structured encapsulation on the measurement data collected by each edge device, uses the device identifier as the index primary key, writes the collected temperature, current, voltage, and frequency into the structured data format according to preset fields, and associates the current timestamp to construct a virtual object data unit, which serves as the standard interface structure for the subsequent input of the artificial intelligence model and supports the automatic generation of feature vectors and the traceability of historical data.

[0055] In an embodiment of the present invention, the steps of preprocessing include:

[0056] S201, for each parameter of the measurement data, based on the measurement values at the current time point and the previous two time points, use the linear regression method to predict the trend value of the parameter at the current time point; specifically, according to the measurement values at three time points, calculate the regression coefficients by the least squares method, construct a regression equation, and substitute the measurement values at the previous two time points into the regression equation to obtain the trend value at the current time point, which is used to reflect the expected development direction of the parameter and helps to judge whether the current state deviates.

[0057] S202, calculate the offset between the measurement value at the current time point and the trend value, and use this offset as the input feature of the corresponding parameter.

[0058] S203, regard the parameter whose offset exceeds the first preset threshold as an outlier, and use the trend value at the current time point for replacement processing to maintain the continuity and stability of the feature data and prevent the outlier from affecting the model prediction result.

[0059] S204, perform normalization processing on the processed measurement data using the maximum-minimum normalization method, standardize all parameters to the interval from 0 to 1. This method can effectively eliminate the numerical differences caused by different units and scales between different parameters, and improve the numerical stability, calculation accuracy, and convergence efficiency of the artificial intelligence model during the training and inference processes.

[0060] In an embodiment of the present invention, the preprocessed data is written into two types of storage areas respectively, including: an immediate information area and an object - like memory area. Among them, the immediate information area is implemented as a ring buffer resident in memory, indexed by device ID + timestamp, and only stores the data of the last 2 control operation cycles, which is used for high - speed reading in the model inference stage; the object - like memory area is implemented as a local Mysql database, and writes the complete virtual object data unit and control feedback records by the minute, storing the data of at least 50 control operation cycles. Through this dual - area structure, the present invention decouples real - time control and historical learning, significantly reduces I / O conflicts without increasing the memory peak value, and improves the convergence stability of the artificial intelligence model.

[0061] In an embodiment of the present invention, the first feature vector is composed of the offsets of the parameters in the preprocessed virtual object data unit. The parameters include: temperature, current, voltage, and frequency. Among them, the expression of the first feature vector is: , represents the offset of temperature, represents the offset of current, represents the offset of voltage, represents the offset of frequency; this first feature vector can accurately reflect the deviation degree of the current state of the device compared with the predicted trend, and improve the model's ability to identify device behavior fluctuations and control anomalies.

[0062] In an embodiment of the present invention, the first feature vector is input into a random forest model for inference to obtain the model prediction result. The specific steps include:

[0063] S301, input the first feature vector into multiple decision trees in the random forest model respectively. Each decision tree performs a path traversal on the first feature vector based on the internal division rule until it traverses to the leaf node and outputs a sub - prediction value defined by the leaf node. The internal division rule is used to perform feature division on non - leaf nodes, and the feature division includes judging the branch direction of the traversal path according to the feature splitting threshold set on the non - leaf node;

[0064] Specifically, the internal partitioning rule refers to the threshold judgment logic executed by the non-leaf nodes of each decision tree in the random forest model on specific dimensional parameters in the first feature vector, which is used to determine the path direction of the input data in the decision tree structure. Specifically, each non-leaf node sets a splitting condition, such as the frequency parameter ≤ 60 Hz. When the first feature vector meets this judgment condition, the system sends it to the left sub-branch of this node; otherwise, it enters the right sub-branch. This path traversal process continues to be executed in the tree structure until reaching the leaf node, and the leaf node outputs a corresponding predicted value as the sub-prediction result of this decision tree. The above partitioning rule and the output value of the leaf node are automatically generated by minimizing the variance during the model training stage, forming a decision logic structure for efficient classification or regression. The structure of each tree maintains independence, supporting subsequent operations such as sub-structure replacement, leaf node update, or splitting condition adjustment;

[0065] S302. Perform result fusion on the sub-predicted values output by multiple decision trees, and use the weighted average method to generate the model prediction result.

[0066] In an embodiment of the present invention, the feedback response value refers to the result data returned by an external device after receiving and executing a control instruction, including but not limited to: whether the control instruction is successfully executed, such as whether the switch is successfully switched, whether the device starts or stops as expected, etc.; after executing the control instruction, the status value of the external device, such as parameters such as temperature, current, and voltage returned by a sensor, as verification data for the control action effect; other auxiliary information generated after the device executes, such as delay time, fault status, etc.

[0067] In an embodiment of the present invention, the prediction error is the absolute difference between the current model prediction result and the feedback response value, and the first preset function is constructed by the prediction error and the average value of the prediction errors in the past n rounds, where n represents the preset number of historical control operation rounds. The determination process of the first preset function is as follows:

[0068] Obtain the model prediction result of the current round t and the feedback response value returned by the external device, calculate the absolute difference between the two to obtain the prediction error of the current round;

[0069] Extract the predicted error values for n rounds from round t - n to round t based on the historical control operation rounds, where n is the preset number of historical rounds, calculate the arithmetic mean of the predicted error values to obtain the average of the predicted errors in the past n rounds; add a minimum value between 0 and 1 to the average of the predicted errors in the past n rounds to obtain an adjusted average in case the divisor is zero; divide the predicted error of the current round by the adjusted average to obtain a ratio; raise the ratio to the power of the second weight coefficient to obtain a result of power processing; multiply the first weight coefficient by the result of power processing to obtain a scaled processing value; add the value 1 to the scaled processing value to obtain a sum value; take the reciprocal of the sum value to obtain the credibility weight of the current round, which is used as the output result of the first preset function.

[0070] When specifically applied, the calculation formula of the above first preset function can be implemented through the following formula, for example: ;

[0071] Where, represents the credibility weight, which is used to dynamically evaluate the credibility of the current artificial intelligence model's prediction result, represents the model prediction result, represents the feedback response value, represents the prediction error, represents a minimum value between 0 and 1, represents the average of the predicted errors from the (t - n)-th round to the t-th round in the past, where t represents the index of the control operation round, and represent the first weight coefficient and the second weight coefficient respectively.

[0072] When the current prediction error is significantly greater than the average of the historical prediction errors, the calculated credibility weight will rapidly decrease, reflecting the unreliability of the current model output; conversely, when the prediction error remains within the normal fluctuation range, the credibility weight remains at a high level, maintaining the dominant role of the model in generating control instructions.

[0073] In an embodiment of the present invention, correcting the model prediction result according to the credibility weight includes: performing weighted fusion processing based on the deviation between the current model prediction result and the feedback response value of the previous round, combined with the current credibility weight, to obtain the current corrected control output value; where the calculation formula for correcting the model prediction result according to the credibility weight is: ;

[0074] Where, represents the current control output value, that is, the corrected model prediction result, which is used to generate control instructions, represents the feedback response value of the (t - 1)-th round.

[0075] The control instructions generated by the control execution module are sent to an external device via the communication module. The communication module is a wired communication interface, such as RS485 or CAN bus, for reliably transmitting the control instructions to the target control object.

[0076] In an embodiment of the present invention, the control behavior change feature is obtained through the following steps:

[0077] Calculate the first difference between the control output value after the t-th round of correction and the control output value after the (t - 1)-th round of correction; perform a square operation on the first difference to obtain the first square result;

[0078] Obtain the standard deviation of the control output value, and add a very small value between 0 and 1 to the standard deviation to obtain the denominator adjustment value; divide the first square result by the denominator adjustment value to obtain the first ratio; process the first ratio to obtain the first sum value;

[0079] Perform a natural logarithm operation on the first sum value to obtain the first logarithm result; calculate the second difference between the control output value after the (t - 1)-th round of correction and the control output value after the (t - 2)-th round of correction; calculate the absolute value difference between the first difference and the second difference;

[0080] Multiply the preset weight coefficient by the absolute value difference to obtain the product result; calculate the power of the natural constant e to the power of the product result to obtain the exponential operation result;

[0081] Process the exponential operation result to obtain the second sum value; multiply the first logarithm result by the second sum value to obtain the final control behavior change feature value.

[0082] When specifically applied, the above control behavior change feature can be implemented through the following formula, for example:

[0083] The control behavior change feature is calculated through a non-linear composite function, and the calculation formula is: ;

[0084] Where represents the control behavior change feature, which is used to measure the fluctuation intensity and mutation degree of the control output value in the time dimension, represents the control output value of the (t - 1)-th round, represents the control output value of the (t - 2)-th round, represents the standard deviation of the control output value, represents the exponential adjustment coefficient, which is used to adjust the sensitivity of the output value of the current artificial intelligence model to the error change;

[0085] The control behavior change characteristic takes into account the change amplitude between the current control output value and the previous round of control output value, as well as the difference in the change rate between two consecutive rounds of control values. By combining these two parts nonlinearly, this embodiment can give differentiated responses when facing both slow-changing and sudden-changing behaviors. The higher the control behavior change characteristic, the more drastic the control behavior and the more unstable the system state, which is more likely to trigger model switching or local adjustment operations at the control strategy layer.

[0086] In one embodiment of the present invention, the first control strategy includes:

[0087] When the duration of the control behavior change characteristic exceeding the second preset threshold exceeds the first preset time period, the currently used artificial intelligence model is switched to a preset backup model; the backup model is a stable model version obtained through historical training, and its structure is consistent with the main model, and can be switched seamlessly;

[0088] Based on the paired data stored in the simulated memory area, a local retraining operation is performed on the decision tree in the current artificial intelligence model, and the paired data includes: a first eigenvector, a model prediction result, and a feedback response value.

[0089] In one embodiment of the present invention, the local retraining operation includes: adjusting the output value of the leaf node and modifying the feature splitting threshold of the non-leaf node;

[0090] Specifically, adjusting the output value of a leaf node means re-evaluating the target output value corresponding to a leaf node when the historical prediction error of the leaf node is large or the deviation is accumulated without changing the decision tree structure. This embodiment collects the first feature vector and its feedback response value that hits the leaf node based on the paired data recorded in the simulacrum memory area, calculates the feedback mean and updates the output result of the leaf node, thereby improving the prediction accuracy of the branch;

[0091] Modifying the feature splitting threshold of non-leaf nodes means readjusting the feature division conditions of certain non-leaf nodes while keeping the node structure of the tree unchanged. This operation is applicable to split nodes where judgment errors frequently occur. This embodiment re-evaluates the optimality of the current splitting condition under the current data distribution based on the first feature vector in the historical pairing data and its feedback response value, and replaces the original feature splitting threshold by minimizing the local error, thereby optimizing the judgment logic of the node and improving the accuracy of path selection.

[0092] In one embodiment of the present invention, there is also provided an artificial intelligence model control method, comprising the following steps:

[0093] S401, collecting measurement data of multiple edge devices and constructing corresponding virtual object data units;

[0094] S402. Preprocess the virtual object data unit and write it into the real-time information area and the object-like memory area respectively. The real-time information area is used to provide the input data of the artificial intelligence model at the current moment, and the object-like memory area is used to store historical state data and execution records.

[0095] S403. Extract the preprocessed virtual object data unit from the real-time information area, construct the first feature vector, and input the first feature vector into the random forest model for inference to obtain the model prediction result.

[0096] S404. Obtain the feedback response value returned by the external device after the execution of the previous control operation, calculate the prediction error based on the feedback response value and the current model prediction result, and calculate the confidence weight output by the artificial intelligence model according to the first preset function. The external device represents the target control object for executing the control instruction generated by the artificial intelligence model.

[0097] S405. Correct the model prediction result according to the confidence weight, and perform weighted processing in combination with the feedback response value of the previous control operation to generate the current control output value, and then generate a control instruction, which is sent to the external device through the communication module to execute the corresponding control operation.

[0098] S406. Construct the control behavior change feature according to the current control output value and the previous control output value, and write the prediction error, confidence weight, feedback response value and control behavior change feature into the object-like memory area, and then adjust the artificial intelligence model according to the first control strategy.

[0099] It should be noted that the setting of the interval and threshold size is for the convenience of comparison. The size of the threshold depends on the amount of sample data and the base quantity set by those skilled in the art for each group of sample data, as long as it does not affect the proportional relationship between the parameters and the quantified values. And the above formulas are all calculations that remove the dimension and take their numerical values. The formulas are all obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0100] The embodiments of the present invention have been described above, but the present invention is not limited to the above specific embodiments. The above specific embodiments are only illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.

Claims

1. An artificial intelligence model control system, characterized in that, Including: An equipment data acquisition module, configured to acquire measurement data of multiple edge devices and construct corresponding virtual object data units; A data preprocessing module, configured to preprocess the virtual object data units and write them into an immediate information area and an object-based memory area respectively, wherein the immediate information area is used to provide input data for the artificial intelligence model at the current moment, and the object-based memory area is used to store historical state data and execution records; A model inference module, configured to extract the preprocessed virtual object data units from the immediate information area, construct a first feature vector, and input the first feature vector into a random forest model for inference to obtain a model prediction result; A credibility calculation module, configured to obtain a feedback response value returned by an external device after the execution of the previous control operation, calculate a prediction error based on the feedback response value and the current model prediction result, and calculate a credibility weight output by the artificial intelligence model according to a first preset function, wherein the external device represents a target control object for executing a control instruction generated by the artificial intelligence model; A control execution module, configured to correct the model prediction result according to the credibility weight, perform weighted processing in combination with the feedback response value of the previous control operation, generate a current control output value, and further generate a control instruction, and send it to the external device through a communication module to execute a corresponding control operation; A model adjustment module, configured to construct a control behavior change feature according to the current control output value and the previous control output value, write the prediction error, credibility weight, feedback response value and control behavior change feature into the object-based memory area, and further adjust the artificial intelligence model according to a first control strategy.

2. The artificial intelligence model control system according to claim 1, wherein, The measurement data includes: temperature, current, voltage and frequency, and the virtual object data unit includes: device identifier, timestamp, temperature, current, voltage and frequency.

3. An artificial intelligence model control system according to claim 1, characterized in that, The steps of preprocessing include: S201. For each parameter of the measurement data, based on the measurement values at the current time point and its previous two time points, use a linear regression method to predict the trend value of the parameter at the current time point; S202. Calculate the offset between the measurement value at the current time point and the trend value, and use the offset as the input feature of the corresponding parameter; S203. Regard the parameter whose offset exceeds a first preset threshold as an outlier, and use the trend value at the current time point for replacement processing; S204. Use the maximum-minimum normalization method to normalize the processed measurement data.

4. An artificial intelligence model control system according to claim 2, characterized in that, The first feature vector is composed of the offsets of the parameters in the preprocessed virtual object data unit, and the parameters include: temperature, current, voltage and frequency.

5. An artificial intelligence model control system according to claim 1, characterized in that, Inputting the first feature vector into a random forest model for inference to obtain a model prediction result, the specific steps include: S301. Input the first feature vector into multiple decision trees in the random forest model respectively. Each decision tree performs a path traversal on the first feature vector based on an internal partitioning rule until it traverses to a leaf node and outputs a sub-prediction value defined by the leaf node. The internal partitioning rule is used to perform feature partitioning on non-leaf nodes, and the feature partitioning includes judging the branch direction of the traversal path according to a feature splitting threshold set on the non-leaf node; S302, perform result fusion on the sub-prediction values output by multiple decision trees, and use the weighted average method to generate the model prediction result.

6. The artificial intelligence model control system according to claim 1, wherein, The prediction error is the absolute difference between the current model prediction result and the feedback response value. The first preset function is constructed by the prediction error and the average value of the prediction errors in the past n rounds. n represents the preset number of historical control operation rounds. The determination process of the first preset function is as follows: Obtain the model prediction result in the current round t and the feedback response value returned by the external device, calculate the absolute difference between the two, and obtain the prediction error in the current round. Based on the historical control operation rounds, extract the prediction error values in n rounds from round t - n to round t. Here, n is the preset number of historical rounds. Calculate the arithmetic mean of the prediction error values to obtain the average value of the prediction errors in the past n rounds. Add a very small value between 0 and 1 to the average value of the prediction errors in the past n rounds to prevent the divisor from being zero, obtaining an adjusted average value. Divide the prediction error in the current round by the adjusted average value to obtain a ratio. Raise the ratio to the power of the second weight coefficient to obtain a power processing result. Multiply the first weight coefficient by the power processing result to obtain a scaling processing value. Add the value 1 to the scaling processing value to obtain a sum value. Take the reciprocal of the sum value to obtain the credibility weight in the current round, which is used as the output result of the first preset function.

7. An artificial intelligence model control system according to claim 6, wherein, Correcting the model prediction result according to the credibility weight includes: according to the deviation between the current model prediction result and the feedback response value in the previous round, combined with the current credibility weight, perform weighted fusion processing to obtain the currently corrected control output value.

8. An artificial intelligence model control system according to claim 7, characterized in that, The control behavior change characteristics are obtained through the following steps: Calculate the first difference between the corrected control output value in the t-th round and the corrected control output value in the (t - 1)-th round; perform a square operation on the first difference to obtain a first squared result. Obtain the standard deviation of the control output value, and add a very small value between 0 and 1 to the standard deviation to obtain a denominator adjustment value; divide the first squared result by the denominator adjustment value to obtain a first ratio. Process the first ratio to obtain a first sum value. Perform a natural logarithm operation on the first sum value to obtain a first logarithmic result. Calculate the second difference between the corrected control output value in the (t - 1)-th round and the corrected control output value in the (t - 2)-th round; calculate the absolute difference between the first difference and the second difference. Multiply the preset weight coefficient by the absolute difference to obtain a product result. Calculate the power of the product result of the natural constant e to obtain an exponential operation result. Process the exponential operation result to obtain a second sum value; multiply the first logarithmic result by the second sum value to obtain the final control behavior change characteristic value.

9. An artificial intelligence model control system according to claim 1, characterized in that, The first control strategy includes: When the duration for which the control behavior change characteristic exceeds the second preset threshold exceeds the first preset time period, switch the currently used artificial intelligence model to the pre-set backup model. Perform a local retraining operation on the decision tree in the current artificial intelligence model based on the paired data stored in the anthropomorphic memory area, where the paired data includes: a first feature vector, a model prediction result, and a feedback response value; the local retraining operation includes: adjusting the output value of the leaf node and modifying the feature splitting threshold of the non-leaf node.

10. An artificial intelligence model control method, characterized in that, An artificial intelligence model control system according to any one of claims 1-9, comprising the following steps: S401, collect measurement data of multiple edge devices and construct corresponding virtual object data units; S402, preprocess the virtual object data units and write them into the immediate information area and the anthropomorphic memory area respectively, where the immediate information area is used to provide input data for the artificial intelligence model at the current moment, and the anthropomorphic memory area is used to store historical state data and execution records; S403, extract the preprocessed virtual object data units from the immediate information area, construct a first feature vector, and input the first feature vector into a random forest model for inference to obtain a model prediction result; S404, obtain the feedback response value returned by the external device after the execution of the previous control operation, calculate the prediction error based on the feedback response value and the current model prediction result, and calculate the credibility weight output by the artificial intelligence model according to a first preset function, where the external device represents the target control object for executing the control instruction generated by the artificial intelligence model; S405, correct the model prediction result according to the credibility weight, perform weighted processing in combination with the feedback response value of the previous control operation, generate the current control output value, and then generate a control instruction, and send it to the external device through the communication module to execute the corresponding control operation; S406, construct a control behavior change feature according to the current control output value and the previous control output value, write the prediction error, credibility weight, feedback response value, and control behavior change feature into the anthropomorphic memory area, and then adjust the artificial intelligence model according to a first control strategy.

Images