Photoelectric angle measurement device performance trend prediction method considering causal correlation

Abstract

The invention discloses a photoelectric angle measurement device performance trend prediction method considering causal correlation. The method comprises the following steps: constructing a performance trend prediction model; acquiring monitoring data of the preset index at a plurality of moments through combination of actual measurement and prediction; converting the monitoring data into a confidence distribution form according to the performance level; calculating a conditional mixing correlation coefficient according to the causal correlation of the subsystems; calculating the index weight and reliability of the preset index, converting the preset index into a confidence distribution form of the evidence, and determining an evidence fusion sequence; the final basic probability quality of each subsystem is calculated through evidence fusion, so that model parameters are preliminarily determined; and constructing an optimization objective function, and obtaining monitoring data of a preset index again to carry out model optimization training to obtain a trained photoelectric angle measurement device performance trend prediction model used for predicting the performance trend of the target photoelectric angle measurement device. According to the method, the causal correlation between the subsystems is incorporated into the performance trend prediction model, so that the prediction precision can be improved.

Description

Technical Field

[0001] This invention belongs to the field of health management of photoelectric angle measuring devices, and specifically relates to a method for predicting the performance trend of photoelectric angle measuring devices that considers causal correlation. Background Technology

[0002] Electronic theodolites, as reliable and precise measuring devices, are widely used in hydraulic engineering, resource exploration, and military fields. The photoelectric angle measuring device, as the core component of the electronic theodolite, is responsible for the measurement, calculation, and presentation of angles. Composed of complex optical, mechanical, and electronic structures, the photoelectric angle measuring device enables precise angle measurement and possesses strong anti-interference capabilities, allowing it to operate stably in various harsh environments such as high temperature, extreme cold, and strong interference.

[0003] In the operation and maintenance of photoelectric angle measuring devices, to ensure measurement accuracy and safe operation, professional personnel are required to regularly monitor and calibrate them. However, judging and predicting the device's performance status mainly relies on the experience and habits of the operators and maintenance personnel. These methods are costly and highly subjective. Therefore, scientifically and accurately predicting the performance trend of photoelectric angle measuring devices is a prerequisite for their reasonable and efficient maintenance and to ensure their safe and stable operation. Thus, predicting the performance trend of photoelectric angle measuring devices is of great significance.

[0004] Currently, the performance trend prediction methods for photoelectric angle measuring devices face two major challenges. First, photoelectric angle measuring devices contain numerous mechanical and electronic components, with complex working mechanisms and strong electromagnetic coupling in each subsystem. These factors make it difficult to establish a performance trend prediction model. Second, there is a causal correlation between the subsystems. This causal correlation means that the output signal of one subsystem is the input signal of another. This correlation causes the performance state of one subsystem to affect other subsystems as the signal propagates. For example, a decline in the performance state of subsystem 1 leads to a large error in its output signal. This erroneous signal is input to subsystem 2, which further propagates or even amplifies this deviation, affecting subsequent subsystems and ultimately impacting the overall performance state of the photoelectric angle measuring device. Therefore, incorporating the causal correlation between subsystems into the performance trend prediction model to calculate a more accurate performance trend is a key research issue. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method for predicting the performance trend of a photoelectric angle measuring device considering causal correlation. The technical problem to be solved by this invention is achieved through the following technical solution: A method for predicting the performance trend of a photoelectric angle measuring device considering causal correlation includes: S1, For the target photoelectric angle measuring device, a performance trend prediction model is constructed based on the evidence reasoning rules, which contains conditional mixed correlation coefficients and other parameter vectors as model parameters to be determined; S2, by combining actual measurement with prediction, the monitoring data of each preset index of the target photoelectric angle measuring device at multiple times are obtained; S3, based on the performance level of the target photoelectric angle measuring device, convert the monitoring data of each preset index into a confidence distribution form; S4, Calculate the conditional mixed correlation coefficient of each subsystem based on the causal correlation of the subsystems in the target photoelectric angle measuring device; S5, calculate the weight of each preset indicator based on the corresponding monitoring data; S6, calculate the reliability of each preset indicator, and based on the corresponding reliability, indicator weight, confidence distribution form of the monitoring data transformation and the performance level, transform each preset indicator into a confidence distribution form of evidence; S7, determine the order of evidence fusion based on the signal transmission order of the subsystem; S8. Based on the ER-rule method, and according to the index weights, the reliability, the evidence fusion order, the conditional mixture correlation coefficient, and the confidence distribution form of the evidence, the final basic probability quality of each subsystem is calculated through evidence fusion, thereby initially determining the model parameters. S9. Construct an optimized objective function and reacquire monitoring data of each preset index for model optimization training to obtain a trained photoelectric angle measuring device performance trend prediction model, which is used to predict the performance trend of the target photoelectric angle measuring device.

[0006] This invention provides a method for predicting the performance trend of a photoelectric angle measuring device that considers causal correlation. It establishes a new PTF (Performance Trend Forecasting) model and an ERr-CC (ER-rule considered casual correlation) model based on evidence reasoning rules. By employing the evidence reasoning rule (ER-rule), expert knowledge and monitoring data are effectively integrated to construct the performance trend prediction model. A conditional mixed correlation coefficient is incorporated as a discount factor into the evidence fusion process, effectively reducing the redundancy of health information caused by causal correlation between subsystems, thus optimizing the prediction model. By establishing an accurate mathematical model, this invention can effectively improve the accuracy of performance trend prediction for photoelectric angle measuring devices, and the prediction results have good traceability and verifiability, thereby ensuring the healthy and stable operation of the device. Attached Figure Description

[0007] Figure 1 A flowchart illustrating a method provided in an embodiment of the present invention; Figure 2 The measured data are the preset parameters of the target photoelectric angle measuring device in the experimental verification section of this invention. Figure 3 This section shows the historical performance status of the target photoelectric angle measuring device in the experimental verification part of this invention. Figure 4 This is the predicted result of the measurement standard deviation in the preset index of the target photoelectric angle measuring device in the experimental verification part of the present invention; Figure 5 This is a schematic diagram of the input-output relationship between the target photoelectric angle measuring device subsystems in the experimental verification section of this invention; Figure 6 This is the predicted performance trend of the target photoelectric angle measuring device in the experimental verification section of this invention. Detailed Implementation

[0008] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0009] To address the challenges of establishing mathematical models and the causal correlations between subsystems in predicting the performance trends of photoelectric angle measuring devices, this invention provides a method for predicting the performance trends of photoelectric angle measuring devices that considers causal correlations. Figure 1 As shown, the method may include the following steps S1 to S9: S1, For the target photoelectric angle measuring device, a performance trend prediction model is constructed based on the evidence reasoning rules, which contains conditional mixed correlation coefficients and other parameter vectors as model parameters to be determined; The target photoelectric angle measuring device in this embodiment of the invention refers to a photoelectric angle measuring device for which performance trend prediction is currently required.

[0010] The constructed performance trend prediction model is abbreviated as ERr-CC model, and is represented as follows: (1); in, This represents the output of the performance trend prediction model; Represents a functional relationship; This represents the measured data of each preset indicator at multiple times; This represents the conditional mixed correlation coefficient, which contains components corresponding to each preset indicator. This represents the remaining parameter vector.

[0011] In this embodiment of the invention, multiple preset indicators can be set for the target photoelectric angle measuring device. These preset indicators refer to test indicators that can reflect the performance status of the device, and are mainly selected based on product manuals, expert knowledge, etc. For example, they may include measurement standard deviation, temperature, etc. The preset indicators are used for performance evaluation of the target photoelectric angle measuring device.

[0012] The measured data of each preset indicator can be collected at multiple times using various acquisition methods, such as... That is The measured data of each preset indicator at multiple times, For example, it contains the first The measured data of a preset indicator at multiple times. Similarly, Conditional mixed correlation coefficient containing various preset indicators , This can also be called a discount factor for the correlation between indicators. There are other vectors composed of various parameters in this model; for simplicity, we will use them uniformly here. This represents the remaining parameter vector. and All of these need to be calculated in subsequent steps.

[0013] The performance trend prediction model built in S1 is an abstract model given in the form of parameters, which shows the architecture of the model. The specific data and calculation process will be given in subsequent steps.

[0014] S2, by combining actual measurement with prediction, the monitoring data of each preset index of the target photoelectric angle measuring device at multiple times are obtained; S2 to S8 present the preliminary determination process of the model parameters for the performance trend prediction model. The model parameters are derived based on the measured data of the preset indicators and the model calculation.

[0015] Specifically, S2 may include the following steps: S21, For the target photoelectric angle measuring device, acquire the measured data of each preset index at multiple times; For example, you can choose At each of the several time points, the measured data of each preset indicator are obtained.

[0016] S22, for each preset indicator, the measured data at multiple times are normalized to obtain the corresponding normalized measured data; To make the changing trends of the measured data of the preset indicators more obvious and facilitate prediction, the measured data of the preset indicators need to be normalized (i.e., standardized) before prediction. This is achieved using the following formula: . (2); in, This indicates a preset indicator at time [time]. Actual measured data; This represents the mean of all measured data for the preset indicator; This represents the standard deviation of all measured data for the preset index. This indicates that the preset indicator is at time [time]. Normalized measured data.

[0017] It is understandable that the measured data of each preset indicator at each time moment, after the above normalization process, will yield the corresponding normalized measured data.

[0018] S23, for each preset index, input the normalized measured data from multiple time points into a pre-trained LSTM (Long Short-Term Memory) network to obtain a preset number of predicted data points; where each step represents a time point; LSTM is a special type of recurrent neural network (RNN) that addresses the long-term dependency problem of traditional RNNs through gating mechanisms (input gate, forget gate, output gate), making it adept at handling long-term correlations in time-series data. It is widely used in natural language processing, speech recognition, and time series prediction. This invention can be implemented using existing LSTM network structures, but parameter settings are required. Specifically, the input layer dimension of the LSTM network depends on the number of features in the input data, the output dimension of the fully connected layers is determined by the number of features in the target output, and the number of hidden units in the LSTM layers needs to be adjusted according to the characteristics of the dataset. The detailed setup process is not described in detail here.

[0019] The LSTM network in this embodiment of the invention can be trained by pre-selecting and splitting the training set and test set using measured data from a target photoelectric angle measuring device at a preset index at a certain time.

[0020] During the training of the LSTM network, normalized measured data is input with alternating time steps, where data from the previous time step is used as input for the next time step. The alternating time step is one unit of time, as shown in the following formula: (3); Among them, for example This represents the normalized measured data of a preset index at time 1; The vector representing the first input; This represents the vector of the second input.

[0021] The specific training process will not be described in detail here.

[0022] A trained LSTM network can output predicted data at multiple time points based on normalized measured data of a preset index, according to a set number of steps. The processing can be represented by the following formula: (4); in, A vector representing the normalized measured data of the input preset index, which may contain normalized measured data at one time point or multiple time points; Represents a functional relationship; This represents a vector containing a preset number of predicted data points.

[0023] S24. For each preset indicator, the measured data at multiple times and the predicted data at a preset number of steps are combined as the monitoring data for that preset indicator.

[0024] For ease of understanding, the preset indicators are... real-time monitoring data express.

[0025] S3, based on the performance level of the target photoelectric angle measuring device, convert the monitoring data of each preset index into a confidence distribution form; S3 transforms the input information, which includes both quantitative and qualitative information. Let the preset indicators be... One, of which the first Preset indicators The evaluation framework is ,in Indicates the number of performance levels. exist The monitoring data at any time is The dimensions of monitoring data for different preset indicators may differ, so it is necessary to convert them into a confidence distribution form for standardization.

[0026] The confidence distribution of the monitoring data for each preset indicator is expressed as follows: (5); Among them, the preset indicators are: One, of which the first Preset indicators The evaluation framework is The performance levels of the target photoelectric angle measuring device are as follows: indivual; express Compared to the first Performance levels Confidence level; express At any moment Monitoring data; express No. Performance levels; express No. Reference values ​​for each performance level, .

[0027] S4, Calculate the conditional mixed correlation coefficient of each subsystem based on the causal correlation of the subsystems in the target photoelectric angle measuring device; The target photoelectric angle measuring device may contain multiple subsystems, and these subsystems are causally correlated. Causal correlation means that the output signal of one subsystem is the input signal of another subsystem. This causal correlation causes the performance state of one subsystem to affect other subsystems as the signal is transmitted. This embodiment of the invention sets a conditional mixed correlation coefficient in the performance trend prediction model to reduce the redundancy of health information caused by causal correlation.

[0028] Specifically, S4 includes the following steps: S41, Based on the observation data of the subsystem in the target photoelectric angle measuring device, determine the causal correlation of the subsystem, and then calculate the correlation coefficient between the subsystems; Assuming the output of subsystem A is the input of subsystem B, its observed data vector is represented as: ,in Indicates that subsystem A is in Observational data at any given time This represents the total number of observation times. Observational data refers to the numerical values ​​of the physical parameters output by the subsystem. The observation data vector for subsystem B is represented as: .

[0029] against and They exhibit a linear correlation, that is... and Satisfying the relation: (6); in, Let be an arbitrary constant, which can be determined through calculation. Then, the Pearson correlation coefficient can be used to calculate... and The correlation coefficient between them, i.e.: (7); in, and They represent The mean and standard deviation of all elements in the dataset; and They represent The mean and standard deviation of all elements in the dataset.

[0030] when and If the correlation between them is non-linear, then the correlation coefficient can be calculated using the empirical distance covariance: (8); in, express and Empirical distance covariance express and Empirical distance covariance express and The empirical distance covariance.

[0031] In summary, specifically through functions The method for calculating the correlation coefficient can be expressed as follows: (9); in, for and A general method for representing the correlation coefficient between them.

[0032] Therefore, the correlation coefficient between the two subsystems can be calculated using the above formula.

[0033] S42, construct a triangular correlation matrix based on all the calculated correlation coefficients; Without loss of generality, definition Subsystems The output observation data vector is ,in, Indicates the first Subsystem The output observation data vector. When The fusion order is At that time, we have the following triangular correlation matrix: (10); S43, For each subsystem, the overall correlation coefficient of the subsystem is calculated by summing the correlation coefficients of the corresponding rows in the triangular correlation matrix; Subsystem Overall correlation coefficient for: (11); Understandable It is the th in the triangular correlation matrix In the middle of the line Each element (i.e., the correlation coefficient) This sums up all the elements in this row.

[0034] S44. Based on the overall correlation coefficient of all subsystems, the conditional mixed correlation coefficient of each subsystem is calculated using the formula for calculating the conditional mixed correlation coefficient. Specifically, the formula for calculating the conditional mixed correlation coefficient is as follows: (12); in, Representation Subsystem The conditional mixed correlation coefficient, Therefore, the conditional mixed correlation coefficient of each subsystem can be calculated.

[0035] Conditional mixed correlation coefficient The value range is [0,1], with higher values ​​indicating a lower causal correlation between the subsystem and other subsystems. Specifically, Representation Subsystem Health information can be fully represented by other subsystems. Representative subsystem It is completely independent of other subsystems. Representative subsystem It has causal relationships with other subsystems.

[0036] S5, calculate the weight of each preset indicator based on the corresponding monitoring data; This invention uses the coefficient of variation (CVBW) method to calculate indicator weights. The coefficient of variation is a statistical indicator that reflects the degree of dispersion of data relative to the mean. The larger the coefficient of variation of a performance indicator, the greater its data volatility and the greater its impact on the output. Therefore, indicators with large coefficients of variation should be assigned larger weight values.

[0037] S5 includes the following steps: S51, for each preset indicator, calculate the mean and standard deviation of its monitoring data; Assuming preset indicators The monitoring data is ,make , They represent The mean and standard deviation of the monitoring data can be obtained from the following two formulas: (13); (14); S52, calculate the ratio of the standard deviation to the mean of the monitoring data calculated by the preset index, and obtain the ratio parameter corresponding to the preset index; S53, calculate the sum of the ratio parameters corresponding to all preset indicators to obtain the ratio parameters and values; S54. For each preset indicator, calculate the quotient of its ratio parameter and the sum of the ratio parameters to obtain the indicator weight of the preset indicator.

[0038] The calculation process for S52-S54 can be expressed as the following formula: (15); (16); in, Indicates preset indicators The weight of the indicators.

[0039] Therefore, through the above processing, a weight can be calculated for each preset indicator.

[0040] S6, calculate the reliability of each preset indicator, and based on the corresponding reliability, indicator weight, confidence distribution form of the monitoring data transformation and the performance level, transform each preset indicator into a confidence distribution form of evidence; Among them, the preset indicators serve as evidence, and the evidence has a reliability at each time point, including static reliability. and dynamic reliability Static reliability The specific values ​​are provided by experts. Dynamic reliability. The calculation is performed using a distance-based method; for details, please refer to relevant technical documentation, which will not be elaborated here. The calculations are then performed using various indicators at time... Pre-calculated disturbance coefficient (preset index) At any moment The disturbance coefficient can be used (For detailed calculation procedures, please refer to relevant technical documents.) Static reliability... and dynamic reliability Combining these factors can determine the reliability of the evidence. It can be abbreviated as Therefore, for each piece of evidence, a reliability can be obtained at each moment of the monitoring data. .

[0041] According to ER-rule, each preset indicator This can be transformed into a confidence distribution of evidence. The confidence distribution of evidence is expressed as follows: (17); Formula (17) refers to a single moment. For simplification, the moment is... Not shown. Indicates the first Preset indicators The confidence distribution of the transformed evidence; Indicates the first Performance levels; This indicates that the preset indicator is in the evidence The following was evaluated as a performance level The confidence level, and satisfying , ; Indicates preset indicators The weight of the indicators; Indicates preset indicators Reliability; This indicates the number of performance levels of the target photoelectric angle measuring device; , indicating the first Preset indicators The evaluation framework represents the set of possible evaluation results for a solution. This indicates global ignorance.

[0042] S7, determine the order of evidence fusion based on the signal transmission order of the subsystem; When evidence is correlated, it cannot be fused according to the commutative and associative laws. Therefore, determining the fusion order is crucial when fusing evidence that does not meet the independence requirement. This invention proposes a method for determining the fusion order of evidence based on signal transmission sequences.

[0043] Suppose a subsystem with causal correlation. The signal transmission relationship between them is as follows: The output is Input, The input is The output of the subsystem... signal transmission sequence for: . Representation Subsystem The signal transmission sequence number. Let express The order of fusion This means that this piece of evidence was first integrated. Therefore: (18); in, , This represents the total number of subsystems.

[0044] S8. Based on the ER-rule method, and according to the index weights, the reliability, the evidence fusion order, the conditional mixture correlation coefficient, and the confidence distribution form of the evidence, the final basic probability quality of each subsystem is calculated through evidence fusion, thereby initially determining the model parameters. S81, calculates the mixed weight factor of each preset indicator by fusing the corresponding indicator weights and reliability; To avoid confusion with the mixed weights of subsystems, the weights obtained by fusing evidence and reliability are called mixed weight factors. express: (19); in, Indicates the first Preset indicators The mixed weighting factor.

[0045] S82, based on the ER-rule method, and according to the confidence distribution of the monitoring data of the preset indicators, as well as the mixed weighting factors of the preset indicators, determines the evidence. The basic probability mass, where ; Among them, evidence The basic probability mass can be expressed as: (20); in, Denotes the power set, by It consists of all subsets; The evaluation framework representing the subsystem performance status is consistent with the evaluation framework for photoelectric angle measuring devices; and satisfies... ; Represents the empty set; It is a preset indicator. For the confidence distribution form of the monitoring data, please refer to formula (5). understand; Present evidence The basic probability mass.

[0046] S83, based on evidence The basic probability quality, the evidence fusion order, and the underlying indicators of the subsystem are used to perform evidence fusion to obtain the probability of the subsystem evidence fusion pointing to the performance level. Combined with the basic probability quality of the subsystem that has been calculated, the hybrid weight factor of the subsystem evidence fusion is calculated. First, based on the basic probabilistic mass generation method, we assume the subsystem... Composed of H underlying indicators, which are part of L preset indicators, the evidence fusion process can be represented by equations (21)-(26): (twenty one); (twenty two); (twenty three); (twenty four); (25); (26); in, After the H pieces of evidence are merged, they are assigned to the result. The unnormalized basic probability mass; Indicates the first The reliability of each underlying indicator; This indicates that the H-1 underlying metrics are merged and then allocated to... The basic probability mass; This represents the basic probability quality of the power set after the fusion of H-1 underlying indices; Indicates the first Each underlying metric is assigned to the result. The basic probability mass; This indicates that the H-1 underlying metrics are merged and then allocated to... The basic probability mass; Indicates the first Each underlying metric is assigned to the result. The basic probability mass; and This represents the possible evaluation results after the fusion of H-1 underlying indicators and the first... All possible evaluation results for each underlying indicator are included in the evaluation framework. ; and yes The two parts that were split off are used as intermediate variables. This is then assigned to the power set after evidence fusion. The unnormalized basic probability mass represents global ignorance. In formula (26) This represents the possible evaluation result after the fusion of H underlying indicators; This indicates that the H underlying metrics are merged and then allocated to... The unnormalized basic probability mass.

[0047] The fusion of H pieces of evidence yields the following result: The probability is denoted as It can be calculated using the following formula: (27); in, It is a subsystem Evidence fusion points to performance level The probability of.

[0048] Then, from the basic probability mass expression, we can see that the subsystem The basic probability mass can be expressed as: (28); in, and As calculated above, This indicates the subsystem after the fusion of H pieces of evidence. The mixed weighting factor.

[0049] Therefore, subsystem The hybrid weighting factor after evidence fusion can be calculated as follows: (29); Following the above method, the hybrid weight factor after evidence fusion for each subsystem can be calculated.

[0050] S84, after obtaining the mixed weight factor after the fusion of evidence from all subsystems, calculates the mixed weight after the fusion of the underlying indicators of each subsystem by normalization. Subsystem The mixed weighting factors are respectively Then the subsystem The mixed weights are: (30); S85 calculates the final basic probability quality of the subsystem based on the mixed weights and conditional mixed correlation coefficients after the fusion of the subsystem's underlying indicators.

[0051] Based on formula (20), the subsystem is obtained. The final basic probability mass expression is: (31); in, Indicates through subsystem The mixed weights are obtained by fusing the underlying metrics. Representation Subsystem The rating is graded Confidence level, . For subsystem The conditional mixed correlation coefficient, when Subsystem When independent, the basic probability quality degenerates into the ER-rule model, i.e., the independent evidence reasoning model.

[0052] Through the above processing, the final basic probability mass can be calculated for each subsystem.

[0053] Understandably, through S2~S8, the model parameters are initially solved, including the indicator weights of the preset indicators, reliability, conditional mixed correlation coefficients of the subsystems, and mixed weights obtained by fusing the underlying indicators of the subsystems.

[0054] S9. Construct an optimized objective function and reacquire monitoring data of each preset index for model optimization training to obtain a trained photoelectric angle measuring device performance trend prediction model, which is used to predict the performance trend of the target photoelectric angle measuring device.

[0055] In the performance monitoring of photoelectric angle measuring devices, factors such as disturbances and environmental changes inevitably affect the adaptability of model parameters, leading to a decrease in their adaptability. Therefore, it is necessary to train the model with multiple sets of data to optimize the model parameters. Mean Squared Error (MSE) is a commonly used standard for measuring model accuracy. MSE is calculated by measuring the actual output utility. and predicting output utility The accuracy of the ERr-CC model is measured by the average of the squared differences between the two values. The optimization objective function is: (32); in, Indicates mean square error; , These represent the mixed weights and conditional mixed correlation coefficients of each subsystem at a certain moment; This indicates that the target photoelectric angle measuring device was evaluated as having a certain performance level at a certain moment. The probability of; This represents all the times corresponding to the monitoring data; It represents the overall health level of the target photoelectric angle measuring device as actually calculated at a certain moment; The expected output utility of the target photoelectric angle measuring device at a certain moment is determined by technicians based on the applicable scenario, historical statistical data, and experience. The constraints of the objective function include: (33); (34); (35); (36); (37); in, Indicates the first The mixed weights of the subsystems Indicates the first Conditional mixed correlation coefficients of the subsystems , Indicates the number of subsystems.

[0056] The objective function can be solved using optimization functions in MATLAB, ultimately yielding the desired result. , , , .

[0057] It should be further explained that the calculation process for the overall health level of the target photoelectric angle measuring device includes: 1) At this moment, the final basic probability mass of each subsystem is fused to obtain the final performance evaluation result of the target photoelectric angle measuring device, which is expressed as: ; Using the calculation methods of formulas (21)-(26), the subsystem The final performance evaluation result of the target photoelectric angle measuring device can be obtained by fusing the fundamental probability mass, as shown in formula (38).

[0058] (38); In formula (38), express After the evidence from the subsystem is fused, it is assigned to the result. The unnormalized basic probability mass; Indicates assignment to the result The unnormalized basic probability mass, i.e. Subsystem evidence fusion allocation The unnormalized basic probability mass; This represents the various possible performance levels assessed for the target photoelectric angle measuring device, including... , here And in formula (26) same; This represents the evaluation framework for photoelectric angle measuring devices.

[0059] 2) Based on the final performance evaluation results of the target photoelectric angle measuring device and the utility of each performance level, the overall health level of the target photoelectric angle measuring device is obtained by summation calculation; According to utility theory, assuming performance levels The utility is , Therefore, the expected utility of the evaluation results, i.e., the overall health level of the target photoelectric angle measuring device, is... The calculation is as follows: (39); The summation of formula (39) is for conduct.

[0060] This refers to the overall health level of the target photoelectric angle measuring device as actually calculated. .

[0061] Understandably, by using S9, the model parameters are optimized and solved. At this point, the model parameters reach their optimal state, and thus the model parameters are determined, resulting in a trained performance trend prediction model.

[0062] Then, the performance trend of the target photoelectric angle measuring device can be predicted using the trained performance trend prediction model.

[0063] Specifically, measured data of preset indicators at new moments can be collected, normalized, and then input into the trained performance trend prediction model to obtain the performance evaluation result of the target photoelectric angle measuring device at a future moment. Then, by obtaining the performance evaluation results of the target photoelectric angle measuring device at multiple future moments, the performance trend prediction results of the preset indicators can be obtained. The performance trend prediction results of each preset indicator can be displayed by drawing a graph to achieve visualization.

[0064] The above describes the processing procedure of the photoelectric angle measuring device performance trend prediction method considering causal correlation provided in the embodiments of the present invention. It should be noted that the above steps in the embodiments of the present invention are not intended to be a strict limitation on the execution order. If there is no obvious order between some steps, the execution order can be appropriately adjusted, which will not be described in detail here.

[0065] To address the challenges of establishing mathematical models and the existence of causal correlations between subsystems in predicting the performance trend of photoelectric angle measuring devices, this invention provides a method for predicting the performance trend of photoelectric angle measuring devices that considers causal correlations. The specific process includes: First, to obtain the predicted performance indicators, an LSTM network is used to predict preset indicators. Second, to establish a performance trend prediction model for the target photoelectric angle measuring device, a conditional mixed correlation coefficient is used to calculate the causal correlation discount factor of the subsystems, and the indicator fusion order is given based on the signal transmission sequence. Then, an improved ER-rule method is used for evidence fusion to obtain the performance trend prediction result. Finally, using the root mean square error between the actual performance trend of the target photoelectric angle measuring device and the model prediction result as the performance indicator, the model parameters are optimized to obtain a trained performance trend prediction model for predicting the performance trend of the target photoelectric angle measuring device.

[0066] This invention establishes a novel PTF (Performance Trend Forecasting) model and an ERr-CC (ER-rule considered casual correlation) model based on evidence reasoning rules. By employing evidence reasoning rules (ER-rule), expert knowledge and monitoring data are effectively integrated to construct the performance trend prediction model. A conditional mixed correlation coefficient is incorporated as a discount factor into the evidence fusion process, effectively reducing the redundancy of health information caused by causal correlations between subsystems, thus optimizing the prediction model. By establishing an accurate mathematical model, this invention can effectively improve the accuracy of performance trend prediction for photoelectric angle measuring devices, and the prediction results have good traceability and verifiability, thereby ensuring the healthy and stable operation of the device.

[0067] To facilitate understanding of the execution process and related effects of the methods in the embodiments of the present invention, the experimental verification process is given below.

[0068] To verify the effectiveness of the ERr-CC model, this experiment conducted a case study using the photoelectric angle measuring device (OGD) in the most typical theodolite—the electronic theodolite. The main principle of the photoelectric angle measuring device (OGD) is as follows: the light signal generated by the rotation of the grating disk with the theodolite's aiming head is converted into an electrical signal using a photodiode. After filtering and amplification, the electrical signal is input into a microprocessor to calculate the rotation angle of the aiming head. This working principle makes the OGD widely used in various types of theodolites to achieve accurate angle measurement. The experimental verification mainly includes the following steps: Step 1: Setting up experimental conditions The experiment used a certain model of photoelectric angle measuring device as the experimental object. This device mainly consists of three subsystems: a photoelectric conversion circuit, a differential amplifier circuit, and a microprocessor, denoted as subsystems 1, 2, and 3, respectively. Based on expert knowledge and product specifications, six performance indicators were selected as preset indicators: measurement standard deviation, temperature, photodiode sensitivity, offset voltage, amplification factor, and offset error. The device used in the experiment had been manufactured for 18 months and was tested monthly in an indoor environment, resulting in 18 sets of measured data. The measured data are shown below. Figure 2 As shown.

[0069] Based on expert knowledge and product manuals, OGD performance status is divided into the following three levels: (40); Historical performance status of the equipment, such as Figure 3As shown in the diagram, green, yellow, and red represent performance statuses of Good, Middle, and Poor, respectively. Taking the data from the tenth month as an example: the probability of OGD being rated Good was 35.12%, and the probability of being rated Middle was 64.88%. The overall performance status of OGD showed a downward trend over time, which is consistent with the logic of device performance degradation.

[0070] Step Two: Performance Trend Prediction based on Figure 2 The measured data in the dataset are used to predict the measured data of preset indicators using the LSTM method. Figure 2 The first 14 months of the 18 months of measured data shown are used as training data for performance trend prediction, and the last 4 months are used as test data to predict seven preset indicators related to the performance status of the photoelectric angle measuring device. Taking the measurement standard deviation as an example, the prediction results are as follows: Figure 4 As shown.

[0071] In the figure, the green curve represents the measured data, and the yellow "X" represents the predicted data. It can be seen that the deviation between the predicted and measured data is small. To quantify the deviation, the Relative Root Mean Square Error (RRMSE) is used to measure the prediction accuracy. The formula for calculating RRMSE is: (41); (42); In formulas (39) and (40), Indicates the first Group of measured data, Indicates the first One predicted data point, This indicates the number of measured and predicted data points. This represents the average value of the measured data.

[0072] The relative root mean square error (RRMSE) of the predicted data for each preset indicator, calculated using formula (40), is shown below: Table 1 Relative Root Mean Square Error Table

[0073] As can be seen from Table 1, the relative root mean square error of the preset indicators is less than 0.15, indicating that the LSTM prediction effect is good.

[0074] Step 3: Model Parameter Setting and Calculation Based on GB / T 36537-2018 and expert knowledge, the reference values ​​for each preset indicator are as follows: Table 2 Reference Values ​​for Preset Indicators

[0075] Based on the reference values ​​of the preset indicators and formula (5), the monitoring data can be converted into confidence level form. For example, the standard deviation of the measurement at a certain moment. . ,therefore , , .

[0076] in, This indicates that a certain preset indicator is relative to the evaluation level. The confidence level.

[0077] The results, converted to confidence distributions, are input into the ERr-CC model to calculate the mixture weights of each subsystem. Due to the large amount of data, only a portion of the data is shown here: Table 3. Mixed weights of subsystems (partial)

[0078] Analyze the working mechanism of OGD and the actual output signals of each subsystem, such as Figure 5 As shown. It can be seen that, under ideal conditions, the output signal between subsystem 1 (photoelectric conversion circuit) and subsystem 2 (differential amplifier circuit) should satisfy: (43); in, This indicates the measured voltage, which is the output of subsystem 1. This represents the amplified voltage, which is the output of subsystem 2. It is a coefficient.

[0079] The output signals between subsystem 2 and subsystem 3 (microprocessor) should satisfy the following: (44); in, The output angle represents the output of subsystem 3. Therefore, subsystem 1 and subsystem 2 exhibit a linear correlation, while subsystem 2 and subsystem 3 exhibit a nonlinear correlation.

[0080] The conditional mixed correlation coefficients of subsystems 1, 2, and 3 were calculated using formulas (6)-(12) respectively, and the results are shown in the table below: Table 4. Conditional hybrid correlation coefficients of each subsystem

[0081] It is evident that subsystem 1 is independent of other subsystems, primarily because it is located at the beginning of the signal transmission sequence and is not affected by the performance status of other subsystems. Subsystems 2 and 3, located at the second and third positions of the signal transmission sequence respectively, show a gradually decreasing correlation coefficient with the overall system, while their dependence on the performance status of other subsystems gradually increases.

[0082] Based on the signal transmission sequence between subsystems, the fusion order of subsystems 1, 2, and 3 can be determined as follows: (45); Step 4: Model Parameter Optimization The optimized model parameters are shown in the table below, which further improves the model's prediction accuracy and running speed.

[0083] Table 5. Optimized model parameters (partial)

[0084] Based on the optimized ERr-CC model, the performance trend of the photoelectric angle measuring device is as follows: Figure 5 As shown. By Figure 5 It can be seen that the probability of OGD being rated as Good slowly decreases with the increase of usage time. For example, the probability of being rated as Good will drop to 36.16% in the tenth month. The model predicts that OGD will have a 30.14% probability of being rated as Poor, a 69.86% probability of being rated as Middle, and a 0% probability of being rated as Good in the twenty-first month.

[0085] It should be noted that, in the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for predicting the performance trend of a photoelectric angle measuring device considering causal correlation, characterized in that, include: S1, For the target photoelectric angle measuring device, a performance trend prediction model is constructed based on the evidence reasoning rules, which contains conditional mixed correlation coefficients and other parameter vectors as model parameters to be determined; S2, by combining actual measurement with prediction, the monitoring data of each preset index of the target photoelectric angle measuring device at multiple times are obtained; S3, based on the performance level of the target photoelectric angle measuring device, convert the monitoring data of each preset index into a confidence distribution form; S4, Calculate the conditional mixed correlation coefficient of each subsystem based on the causal correlation of the subsystems in the target photoelectric angle measuring device; S5, calculate the weight of each preset indicator based on the corresponding monitoring data; S6, calculate the reliability of each preset indicator, and based on the corresponding reliability, indicator weight, confidence distribution form of the monitoring data transformation and the performance level, transform each preset indicator into a confidence distribution form of evidence; S7, determine the order of evidence fusion based on the signal transmission order of the subsystem; S8. Based on the ER-rule method, and according to the index weights, the reliability, the evidence fusion order, the conditional mixture correlation coefficient, and the confidence distribution form of the evidence, the final basic probability quality of each subsystem is calculated through evidence fusion, thereby initially determining the model parameters. S9. Construct an optimized objective function and reacquire monitoring data of each preset index for model optimization training to obtain a trained photoelectric angle measuring device performance trend prediction model, which is used to predict the performance trend of the target photoelectric angle measuring device.

2. The method according to claim 1, characterized in that, The performance trend prediction model constructed in S1 is represented as follows: ； in, This represents the output of the performance trend prediction model; Represents a functional relationship; This represents the measured data of each preset indicator at multiple times; This represents the conditional mixed correlation coefficient, which contains components corresponding to each preset indicator. This represents the remaining parameter vector.

3. The method according to claim 1, characterized in that, S2 includes the following steps: S21, For the target photoelectric angle measuring device, acquire the measured data of each preset index at multiple times; S22, for each preset indicator, the measured data at multiple times are normalized to obtain the corresponding normalized measured data; S23, for each preset index, input the normalized measured data from multiple time points into the pre-trained LSTM network to obtain a preset number of predicted data points; where each step represents a time point; S24. For each preset indicator, the measured data at multiple times and the predicted data at a preset number of steps are combined as the monitoring data for that preset indicator.

4. The method according to claim 1, characterized in that, In S3, the confidence distribution of the monitoring data of each preset indicator is expressed as follows: ； Among them, the preset indicators are: One, of which the first Preset indicators The evaluation framework is The performance levels of the target photoelectric angle measuring device are as follows: indivual; express Compared to the first Performance levels Confidence level; express At any moment Monitoring data; express No. Performance levels; express No. Reference values ​​for each performance level, .

5. The method according to claim 4, characterized in that, S4 includes the following steps: S41, Based on the observation data of the subsystem in the target photoelectric angle measuring device, determine the causal correlation of the subsystem, and then calculate the correlation coefficient between the subsystems; S42, construct a triangular correlation matrix based on all the calculated correlation coefficients; S43, For each subsystem, the overall correlation coefficient of the subsystem is calculated by summing the correlation coefficients of the corresponding rows in the triangular correlation matrix; S44. Based on the overall correlation coefficient of all subsystems, the conditional mixed correlation coefficient of each subsystem is calculated using the formula for calculating the conditional mixed correlation coefficient. The conditional mixed correlation coefficient ranges from [0,1], and a higher value indicates a lower causal correlation between the subsystem and other subsystems.

6. The method according to claim 5, characterized in that, S5 includes the following steps: S51, for each preset indicator, calculate the mean and standard deviation of its monitoring data; S52, calculate the ratio of the standard deviation to the mean of the monitoring data calculated by the preset index, and obtain the ratio parameter corresponding to the preset index; S53, calculate the sum of the ratio parameters corresponding to all preset indicators to obtain the ratio parameters and values; S54. For each preset indicator, calculate the quotient of its ratio parameter and the sum of the ratio parameters to obtain the indicator weight of the preset indicator.

7. The method according to claim 6, characterized in that, In S6, the confidence distribution of the evidence is expressed as follows: ； in, Indicates the first Preset indicators The confidence distribution of the transformed evidence; Indicates the first Performance levels; This indicates that the preset indicator is in the evidence The following was evaluated as a performance level The confidence level, and satisfying , ; Indicates preset indicators The weight of the indicators; Indicates preset indicators Reliability; This indicates the number of performance levels of the target photoelectric angle measuring device; , indicating the first Preset indicators The evaluation framework This indicates global ignorance.

8. The method according to claim 7, characterized in that, S8 includes the following steps: S81, calculates the mixed weight factor of each preset indicator by fusing the corresponding indicator weights and reliability; S82, based on the ER-rule method, and according to the confidence distribution of the monitoring data of the preset indicators, as well as the mixed weighting factors of the preset indicators, determines the evidence. The basic probability mass of , where ; S83, based on evidence The basic probability quality, the evidence fusion order, and the underlying indicators of the subsystem are used to perform evidence fusion to obtain the probability of the subsystem evidence fusion pointing to the performance level. Combined with the basic probability quality of the subsystem that has been calculated, the hybrid weight factor of the subsystem evidence fusion is calculated. S84, after obtaining the mixed weight factor after the fusion of evidence from all subsystems, calculates the mixed weight after the fusion of the underlying indicators of each subsystem by normalization. S85 calculates the final basic probability quality of the subsystem based on the mixed weights and conditional mixed correlation coefficients after the fusion of the subsystem's underlying indicators.

9. The method according to claim 8, characterized in that, In S9, the optimization objective function is expressed as: ； in, Indicates mean square error; , These represent the mixed weights and conditional mixed correlation coefficients of each subsystem at a certain moment; This indicates that the target photoelectric angle measuring device was evaluated as having a certain performance level at a certain moment. The probability of; This represents all the times corresponding to the monitoring data; It represents the overall health level of the target photoelectric angle measuring device as actually calculated at a certain moment; The expected output utility of the target photoelectric angle measuring device at a certain moment is determined by technicians based on the applicable scenario, historical statistical data, and experience. The constraints of the objective function include: ; ; ; ; ,in, Indicates the first The mixed weights of the subsystems Indicates the first Conditional mixed correlation coefficients of the subsystems , Indicates the number of subsystems.

10. The method according to claim 9, characterized in that, The calculation process for the overall health level of the target photoelectric angle measuring device at any given time includes: At this moment, the final basic probability mass of each subsystem is fused to obtain the final performance evaluation result of the target photoelectric angle measuring device, which is expressed as: ； in, express After the evidence from each subsystem is fused, it is assigned to the result. The unnormalized basic probability mass; This represents the various possible performance levels assessed for the target photoelectric angle measuring device, including... ; This represents the evaluation framework for photoelectric angle measuring devices; Indicates assignment to the result The unnormalized basic probability mass; Based on the final performance evaluation results of the target photoelectric angle measuring device and the utility of each performance level, the overall health level of the target photoelectric angle measuring device is obtained by summation calculation; expressed as: ； in, Indicates performance level Its effectiveness.

Images