Fault prediction method of power module in sighting and aiming system based on IMTBO-LightGBM

Abstract

The method for predicting the fault of the power module of the sighting system based on IMTBO-LightGBM belongs to the technical field of power module fault diagnosis, and includes the following steps: step S01, using the grey correlation analysis algorithm to process and screen the collected raw data; step S02, introducing the Chebyshev chaos map and non-inertia weight factor into the mountaineering team algorithm MTBO; step S03, using the improved mountaineering team algorithm IMTBO to optimize the parameters of the lightweight gradient boosting machine LightGBM; step S04, using the fault diagnosis model IMTBO-LightGBM with optimal parameters to predict the fault. The present invention uses the improved mountaineering team optimization algorithm to optimize the fault diagnosis model IMTBO-LightGBM of the lightweight gradient boosting machine to predict the fault of the sighting system power module, thereby improving the accuracy of the fault prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of power module fault diagnosis, and in particular relates to a method for predicting power module faults in an observation and aiming system based on IMTBO-LightGBM. Background Art

[0002] The aiming and sighting system consists of multiple modules such as gyro sensors, signal processing units, actuators, operating interfaces and auxiliary components. Its main task is to achieve accurate identification and tracking of targets. By improving aiming efficiency and accuracy, the target locking time can be shortened and the hit rate can be improved. However, in the modern warfare environment, with the continuous improvement of the performance requirements for military equipment, the frequency of use of the aiming and sighting system has increased significantly, which has also led to a higher risk of failure of the power module. The power module provides the necessary power support for the aiming and sighting system to ensure that each functional unit can work stably. It not only needs to meet the strict requirements of different components for electrical characteristics such as voltage stability and current supply capacity, but also needs to withstand the influence of various adverse factors such as temperature changes, vibration and shock in a complex battlefield environment. Once the power module fails, it may cause the entire aiming and sighting system to lose its function and affect the combat effectiveness of the system. Therefore, it is particularly important to predict the real-time fault of the power module of the aiming and sighting system. By adopting advanced fault prediction, potential power problems such as overheating, voltage fluctuations or short circuits can be warned in advance. This can not only effectively improve the stability of the power module and the system it supports, but also reduce the risk of mission interruption due to sudden failures, and enhance the mobility and firepower output efficiency on the battlefield.

[0003] In the field of power module fault prediction, artificial intelligence algorithms have been widely used in recent years. Despite this, different algorithms still have certain limitations in practical applications. Taking support vector machine (SVM) as an example, although it performs well in many fields, its choice of kernel function is somewhat arbitrary, and it faces the problems of slow training speed and poor accuracy when processing large-scale data, which affects the reliability of the prediction results. Although expert systems can apply professional knowledge to fault prediction, the difficulty of knowledge acquisition and the limited knowledge base make expert systems less efficient when dealing with complex problems, and cannot guarantee the accuracy and real-time performance of each prediction. In addition, the fault tree analysis method requires a lot of calculations and cumbersome steps, and consumes a lot of computing resources. When dealing with complex faults, it often leads to slow operation and is difficult to meet the needs of rapid response. Summary of the invention

[0004] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a fault prediction method for the power module of the sighting and observation system based on IMTBO-LightGBM, uses an improved mountaineering team optimization algorithm to optimize the lightweight gradient boosting mechanism to construct a fault diagnosis model IMTBO-LightGBM, performs fault prediction on the power module of the sighting and observation system, and improves the accuracy of fault prediction.

[0005] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0006] The fault prediction method of the power module of the sighting system based on IMTBO-LightGBM includes the following steps:

[0007] Step S01, collecting the signal value of the power module pin of the sighting system as raw data, using the grey relational analysis algorithm GRA to process and filter the collected raw data, constructing a data set for fault diagnosis model input, and dividing it into a training data set and a test data set;

[0008] Step S02, by introducing Chebyshev chaos mapping into the initialization phase of the mountaineering team algorithm MTBO and introducing non-inertial weight factors into the disaster threat phase, an improved mountaineering team algorithm IMTBO is obtained;

[0009] Step S03, using the improved mountaineering team algorithm IMTBO to optimize the parameters of the lightweight gradient boosting machine LightGBM, and constructing a fault diagnosis model IMTBO-LightGBM;

[0010] Step S04, using the training data set in step S01 to train the fault diagnosis model IMTBO-LightGBM in step S03;

[0011] Step S05, using the test data in step S01 to test the fault diagnosis model IMTBO-LightGBM trained in step S04;

[0012] Step S06: Use the fault diagnosis model IMTBO-LightGBM with optimal parameters after testing to predict the fault of the power module of the observation and aiming system.

[0013] Furthermore, the step S02 includes introducing a Chebyshev chaotic map at the initialization position in the initialization phase of the mountaineering team algorithm, and obtaining an improved initialization position update formula as follows:

[0014] ;

[0015] Where: X i+1 For the i +1 position for a mountaineering team member;X i For the i The positions of the climbing team members; a To control the parameters;

[0016] Introducing non-inertial weight factors in the disaster threat phase ,

[0017] ;

[0018] Where: is the non-inertia weight factor; t is the current iteration number; is the maximum number of iterations;

[0019] The improved position update formula is as follows:

[0020] ;

[0021] Where: For the i +1 new position for mountaineering team members; For the i New positions for members of the climbing team.

[0022] Furthermore, the construction of the lightweight gradient boosting machine LightGBM in step S03 includes:

[0023] (1) Calculate the fitting residual at the tth iteration using the formula:

[0024] ;

[0025] Where: t is the number of iterations, m is the mth tree, y t To predict the label, is the model prediction value of the mth tree, is the fitting residual of the mth tree;

[0026] (2) Calculate the predicted update value using the formula:

[0027] ;

[0028] Where: is the predicted update value of the mth tree, represents the predicted value of the mth tree, Represents the learning rate set by the model, is the model prediction value of the mth tree, Represents the input at the tth iteration.

[0029] Furthermore, the construction of the IMTBO-LightGBM fault diagnosis model in step S03 includes:

[0030] 1) Initialize the control parameters in IMTBO;

[0031] 2) Using Chebyshev chaotic mapping to generate the initial population position;

[0032] 3) Evaluate the initial population fitness;

[0033] 4) The climbers climb collaboratively and use the following formula to update their position:

[0034] ;

[0035] In the formula, For the i new positions for members of the climbing team; For the i The positions of the climbing team members; To be the leader of the mountaineering team; is the position of other players guided by the previous players; rand is a random number between 0 and 1;

[0036] 5) Determine whether there is a new player with a lower fitness than the previous one. If so, swap the position of the new player with the previous player. If not, determine whether there is a new player with a lower fitness than the best player using the following formula:

[0037] ;

[0038] Where: For the i new positions for members of the climbing team; The position of the member whose fitness is smaller than that of this member after iteration; The position of the member with greater fitness than this member after iteration; is the minimum value of the difference between the maximum and minimum members of fitness after iteration;

[0039] 6) If there is a new player with lower fitness than the best player, swap the positions of the best player and the new player; if not, determine whether the termination condition is met;

[0040] 7) Determine whether the termination condition has been reached. If not, return to step 4);

[0041] 8) Assign the optimal parameter combination to the LightGBM model.

[0042] Furthermore, in step S03, the improved mountaineering team algorithm IMTBO is used to optimize the parameters of the lightweight gradient boosting machine LightGBM, and the optimized parameters include learning rate, maximum number of iterations, subsampling, maximum depth of decision tree, minimum number of sample splits, and minimum number of samples of leaf nodes.

[0043] The beneficial effects of the present invention are:

[0044] 1. The present invention uses the grey relational analysis algorithm GRA to process the collected data. By calculating the correlation between sample sequences, it can effectively identify the key factors affecting the target system and reveal their internal connections, reduce dependence on experience, and improve the accuracy and reliability of the evaluation results.

[0045] 2. The present invention optimizes the main parameters of the lightweight gradient boosting machine LightGBM through the improved mountaineering team optimization algorithm IMTBO, which makes up for the defect of blindness in parameter selection during the training process and improves the prediction accuracy of the classification prediction model.

[0046] 3. The present invention uses a lightweight gradient boosting machine LightGBM to show higher training speed and prediction accuracy. LightGBM can naturally handle missing values, reducing the burden of data preprocessing. It also supports automatic processing of category features, further improving the adaptability and flexibility of the model. LightGBM supports a variety of loss functions, enabling it to flexibly cope with a variety of tasks such as classification and regression. DETAILED DESCRIPTION

[0047] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods.

[0048] The present invention provides a method for predicting faults of a power module of an observation and aiming system based on IMTBO-LightGBM, comprising the following steps:

[0049] Step S01, collect the signal value of the power module pin of the sighting system as the original data, use the grey relational analysis algorithm GRA to process and filter the collected original data, build a data set for fault diagnosis model input, and divide it into a training data set and a test data set. Specifically, 80% of the data is used as the training data set and 20% of the data is used as the test data set.

[0050] Specifically, the grey correlation analysis algorithm GRA is used to perform correlation analysis and sorting on the collected raw data, thereby filtering out the pin data with greater correlation with the fault and removing the data with less impact on the system, reducing the redundancy between the data, simplifying the state feature information, and constructing a data set for the fault diagnosis model input. More specifically, the step S01 includes the following steps:

[0051] Step S101, determine the analysis sequence, define the main behavior or target sequence of the research object as the reference sequence:

[0052] ;

[0053] In the formula, n is an unknown integer; y For reference sequence Y The value at the corresponding point;

[0054] Defining the impact reference sequence Y The factor sequences of :

[0055] ;

[0056] In the formula, i =1,2,..., n ; n is a positive integer; x i To compare the sequence The value at the corresponding point.

[0057] Step S102: Perform data normalization processing according to the following formula to solve the problem that the original data units and orders of magnitude may be different.

[0058] ;

[0059] ;

[0060] Where: k is an integer between 1 and n; y ( k ) is the reference sequence Y In the k The value of a point; maximum ( y ) is the reference sequence Y In the k The maximum value of points; my ( y )Reference sequence Y In the k The minimum value of points; x i ( k ) is the comparison sequence In the k The value of the point; max( x i ) is the comparison sequence In the k The maximum value of points; my ( x i) is the comparison sequence In the k The minimum value of points; is the normalized reference sequence Y In the k The value of a point; To compare the sequences after normalization In the k The value of a point.

[0061] Step S103, calculating the correlation coefficient:

[0062] Calculate reference sequence Y Compare each sequence The difference sequence between:

[0063] ;

[0064] In the formula, For reference sequence Y Compare with sequence In between k The difference of points; is the normalized reference sequence Y In the k The value of a point; To compare the sequences after normalization In the k The value of a point.

[0065] Determine the grey relational coefficient:

[0066] ;

[0067] In the formula, k is an integer between 1 and n; For the i Comparison sequence With reference sequence Y In the k The correlation coefficient of each point; is the resolution coefficient; For reference sequence Y Compare with sequence In between k The optimal value of the spread; For reference sequence Y Compare with sequence In between k The minimum value of the spread; For reference sequence Y and compare sequences exist k The maximum difference between the data points.

[0068] Step S104, calculate the correlation: calculate the correlation between each subsequence and the reference sequence Y The average correlation of:

[0069] ;

[0070] In the formula, S i To compare the sequence With reference sequence Y The overall correlation between For the i Comparison sequence With reference sequence Y In the k The correlation coefficient of each point; k is an integer between 1 and n.

[0071] Step S105, analysis and decision: sorting is performed according to the calculated correlation degree. The greater the correlation degree, the greater the impact. Based on the correlation result, the data with a large correlation degree with the power module failure is retained, and the data with a small correlation degree is eliminated.

[0072] Step S02, by introducing Chebyshev chaos mapping into the initialization phase of the mountaineering team algorithm MTBO and introducing non-inertial weight factors into the disaster threat phase, an improved mountaineering team algorithm IMTBO is obtained.

[0073] The step S02 comprises the following steps:

[0074] Step S201, initialization phase: Initialize the positions of the climbing members before climbing, and take the top of the mountain as the optimal solution to the optimization problem. The initialization position is described as:

[0075] ;

[0076] Where: X i For the i The positions of the climbing team members; U j and L j For the optimization problem j The upper and lower limits of the dimension; rand is a random number between 0 and 1; N The population size of the mountaineering team members; D To optimize the problem dimension.

[0077] There is still the problem of uneven initialization in the initialization stage. Therefore, the Chebyshev chaotic map is introduced to improve the MTBO algorithm, which has the advantages of uniform traversal and fast convergence. The Chebyshev chaotic map can make the population initialized uniformly distributed, thereby improving the convergence of the algorithm. The Chebyshev chaotic map is introduced to the initialization position, and the improved initialization position update formula is as follows:

[0078] ;

[0079] Where: X i+1 For the i +1 position for a mountaineering team member; X i For the i The positions of the climbing team members; a is the control parameter.

[0080] Step S202, collaborative climbing phase: After each iteration, the positions of the team members are sorted from best to worst, and each team member is guided by the team leader and the previous team members. The positions of the team members in this phase are updated as follows:

[0081] ;

[0082] In the formula, For the i new positions for members of the climbing team; For the i The positions of the climbing team members; To be the leader of the mountaineering team; is the position of other players guided by the previous players; rand is a random number between 0 and 1.

[0083] Step S203, disaster threat: In the case of random disasters such as avalanches, the mountaineering team members save themselves by the following formula, that is, to avoid the algorithm falling into the local optimum and move towards the best team member position. The mathematical description of the team member position update in this stage is as follows:

[0084] ;

[0085] In the formula, The location of mountaineering team members in the event of random disasters such as avalanches; For the i new positions for members of the climbing team; For the i The positions of the climbing team members; rand is a random number between 0 and 1.

[0086] When the team members are facing the threat of disasters during the climbing process, they need to quickly search for the best team member's position and move to it to ensure their life safety. Therefore, by introducing the non-inertia weight factor, the search ability of the climbers in finding the best team members when facing the threat of disasters is improved, so that they can move closer to the best team member position faster. , the improved position update formula is as follows:

[0087] ;

[0088] ;

[0089] Where: is the non-inertia weight factor; For the i +1 new position for mountaineering team members; For the i New positions for members of the climbing team, t is the current iteration number; is the maximum number of iterations.

[0090] Step S204, coordinated defense: During the mountaineering process, when a disaster such as an avalanche occurs, the mountaineering team rescues the trapped members. The positions of the members in this stage are updated as follows:

[0091] ;

[0092] Where: is the average position of all members of the climbing team; For the i new positions for members of the climbing team; For the i The positions of the climbing team members; rand is a random number between 0 and 1.

[0093] Step S205, team member update: delete the members who died in the disaster from the mountaineering team members, and randomly generate new members to replace them. The replacement formula is as follows:

[0094] ;

[0095] Where: For the i new positions for members of the climbing team; The position of the member whose fitness is smaller than that of this member after iteration; The position of the member with greater fitness than this member after iteration; It is the minimum value of the difference between the maximum and minimum members of fitness after iteration.

[0096] Step S03: Use the improved mountaineering team algorithm IMTBO to optimize the parameters of the lightweight gradient boosting machine LightGBM and build a fault diagnosis model IMTBO-LightGBM.

[0097] LightGBM is an efficient lightweight gradient boosting machine algorithm suitable for large-scale data and high-dimensional feature data. The LightGBM algorithm is used to train the regression prediction model of fault data. By constructing a series of decision trees, each tree is trained according to the prediction residual of the previous tree. At each iteration, LightGBM improves computational efficiency and accuracy by calculating the gradient of the current prediction error and selecting the best split point to minimize the loss function. In the regression prediction task, the model continuously optimizes the prediction of each tree and finally obtains a model that can accurately predict continuous values. Building a lightweight gradient boosting machine LightGBM includes the following steps:

[0098] (1) Calculate the fitting residual at the tth iteration using the formula:

[0099] ;

[0100] Where: t is the number of iterations, m is the mth tree, y t To predict the label, is the model prediction value of the mth tree, is the fitted residual of the mth tree.

[0101] (2) Calculate the predicted update value using the formula:

[0102] ;

[0103] Where: is the predicted update value of the mth tree, represents the predicted value of the mth tree, Represents the learning rate set by the model, is the model prediction value of the mth tree, Represents the input at the tth iteration.

[0104] The LightGBM algorithm involves several key parameters in the modeling process, the most important of which include learning rate learning_rate, maximum number of iterations n_estimators, subsample, maximum depth of decision tree max_depth, minimum number of sample splits min_samples_split, and minimum number of samples of leaf nodes min_samples_leaf. The learning rate learning_rate controls the step size of each update. If it is set too large, the training process may be unstable or even divergent; if it is set too small, more iterations may be required to converge. The maximum number of iterations n_estimators interacts with the learning rate learning_rate. When the learning rate learning_rate is small, the number of iterations needs to be increased to avoid premature convergence. Subsample is used to control the proportion of samples involved in fitting in each iteration. Setting it to less than 1 can effectively prevent overfitting. The depth of the decision tree and the complexity of each tree are adjusted by controlling the maximum depth, the minimum number of sample splits min_samples_split, and the minimum number of samples of leaf nodes min_samples_leaf. Appropriate values ​​can balance the complexity of the model and prevent overfitting or underfitting. Therefore, the need to find the right combination of parameters plays an important role in the accuracy and performance of the model in making predictions.

[0105] The improved mountaineering team algorithm IMTBO is used to optimize the parameters of the lightweight gradient boosting machine LightGBM, and the IMTBO-LightGBM fault diagnosis model is constructed, including the following steps:

[0106] 1) Initialize the control parameters in IMTBO;

[0107] 2) Using Chebyshev chaotic mapping to generate the initial population position;

[0108] 3) Evaluate the initial population fitness;

[0109] 4) The climbers climb collaboratively and use the following formula to update their position:

[0110] ;

[0111] In the formula, For the i new positions for members of the climbing team; For the i The positions of the climbing team members; To be the leader of the mountaineering team; is the position of other players guided by the previous players; rand is a random number between 0 and 1;

[0112] 5) Determine whether there is a new player with a lower fitness than the previous one. If so, swap the position of the new player with the previous player. If not, determine whether there is a new player with a lower fitness than the best player using the following formula:

[0113] ;

[0114] Where: For the i new positions for members of the climbing team; The position of the member whose fitness is smaller than that of this member after iteration; The position of the member with greater fitness than this member after iteration; is the minimum value of the difference between the maximum and minimum members of fitness after iteration;

[0115] 6) If there is a new player with lower fitness than the best player, swap the positions of the best player and the new player; if not, determine whether the termination condition is met;

[0116] 7) Determine whether the termination condition has been reached. If not, return to step 4);

[0117] 8) Assign the optimal parameter combination to the LightGBM model.

[0118] Step S04, using the training data set in step S01 to train the fault diagnosis model IMTBO-LightGBM in step S03;

[0119] Step S05, using the test data in step S01 to test the fault diagnosis model IMTBO-LightGBM trained in step S04;

[0120] Step S06: Use the fault diagnosis model IMTBO-LightGBM with optimal parameters after testing to predict the fault of the power module of the observation and aiming system.

[0121] The present invention processes the collected data through the grey relational analysis algorithm GRA, and optimizes the main parameters of the lightweight gradient boosting machine LightGBM through the improved mountaineering team optimization algorithm IMTBO, which makes up for the defect of blindness in parameter selection during the training process and improves the prediction accuracy of the classification prediction model.

[0122] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. Alterations, modifications, substitutions and variations of the above embodiments by a person skilled in the art are all within the scope of the present invention.

Claims

1. The power module fault prediction method of the sighting system based on IMTBO-LightGBM is characterized by: The steps include: Step S01, collecting the signal value of the power module pin of the sighting system as raw data, using the grey relational analysis algorithm GRA to process and filter the collected raw data, constructing a data set for fault diagnosis model input, and dividing it into a training data set and a test data set; Step S02, by introducing Chebyshev chaos mapping into the initialization phase of the mountaineering team algorithm MTBO and introducing non-inertial weight factors into the disaster threat phase, an improved mountaineering team algorithm IMTBO is obtained; Step S03, using the improved mountaineering team algorithm IMTBO to optimize the parameters of the lightweight gradient boosting machine LightGBM, and constructing a fault diagnosis model IMTBO-LightGBM; Step S04, using the training data set in step S01 to train the fault diagnosis model IMTBO-LightGBM in step S03; Step S05, using the test data in step S01 to test the fault diagnosis model IMTBO-LightGBM trained in step S04; Step S06, using the fault diagnosis model IMTBO-LightGBM with optimal parameters after testing to predict the fault of the power module of the observation and aiming system; The step S02 includes introducing a Chebyshev chaotic map at the initialization position of the initialization phase of the mountaineering team algorithm, and obtaining an improved initialization position update formula as follows: ； Where: X i+1 For the i +1 position for a mountaineering team member; X i For the i The positions of the climbing team members; a To control the parameters; Introducing non-inertial weight factors in the disaster threat phase , ； Where: is the non-inertia weight factor; t is the current iteration number; is the maximum number of iterations; The improved position update formula is as follows: ； Where: For the i +1 new position for mountaineering team members; For the i New positions for members of the mountaineering team.

2. The method for predicting faults of power modules of an observation and aiming system based on IMTBO-LightGBM according to claim 1 is characterized in that: The construction of the lightweight gradient boosting machine LightGBM in step S03 includes: (1) Calculate the fitting residual at the tth iteration using the formula: ； Where: t is the number of iterations, m is the mth tree, y t To predict the label, is the model prediction value of the mth tree, is the fitting residual of the mth tree; (2) Calculate the predicted update value using the formula: ； Where: is the predicted update value of the mth tree, represents the predicted value of the mth tree, Represents the learning rate set by the model, is the model prediction value of the mth tree, Represents the input at the tth iteration.

3. The method for predicting faults of power modules of an observation and aiming system based on IMTBO-LightGBM according to claim 1 is characterized in that: The construction of the IMTBO-LightGBM fault diagnosis model in step S03 includes: 1) Initialize the control parameters in IMTBO; 2) Generate the initial population position using Chebyshev chaotic mapping; 3) Evaluate the initial population fitness; 4) The climbers climb collaboratively and use the following formula to update their position: ； In the formula, For the i new positions for members of the climbing team; For the i The positions of the climbing team members; To be the leader of the mountaineering team; is the position of other players guided by the previous players; rand is a random number between 0 and 1; 5) Determine whether there is a new player with a lower fitness than the previous one. If so, swap the position of the new player with the previous player. If not, determine whether there is a new player with a lower fitness than the best player using the following formula: ； Where: For the i new positions for members of the climbing team; The position of the member whose fitness is smaller than that of this member after iteration; The position of the member with greater fitness than this member after iteration; is the minimum value of the difference between the maximum and minimum members of fitness after iteration; 6) If there is a new player with lower fitness than the best player, swap the positions of the best player and the new player; if not, determine whether the termination condition is met; 7) Determine whether the termination condition has been reached. If not, return to step 4); 8) Assign the optimal parameter combination to the LightGBM model.

4. The method for predicting faults of power modules of an observation and aiming system based on IMTBO-LightGBM according to claim 1 is characterized in that: In step S03, the improved mountaineering team algorithm IMTBO is used to optimize the parameters of the lightweight gradient boosting machine LightGBM. The optimized parameters include learning rate, maximum number of iterations, subsampling, maximum depth of decision tree, minimum number of sample splits, and minimum number of samples for leaf nodes.

Images