Component performance prediction method, system and device based on eagle algorithm and medium

Abstract

The invention discloses a component performance prediction method, system and device based on a hawk algorithm and a medium, and the method comprises the steps: collecting component data, and carrying out the preprocessing of the component data; optimizing the neural network based on an improved hawk algorithm, including: initializing parameters, and determining a globally optimal solution based on the fitness value of each individual in the population; calculating a solution distance and an angle cosine value between the current solution and the globally optimal solution, and pre-updating the current solution in response to the condition that the solution distance and the angle cosine value meet a preset condition to obtain an initial solution; performing secondary updating on the initial solution to obtain a secondary solution; outputting an optimal hyper-parameter of the neural network in response to the condition that the secondary solution meets an iteration condition, and configuring the neural network based on the optimal hyper-parameter; and inputting the preprocessed component data into a neural network to obtain a minimum value of a loss function so as to predict the performance of the component. In the embodiment of the specification, the robustness and convergence speed of the algorithm are improved by introducing the spatial configuration and the secondary updating strategy.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a component performance prediction method, system, device and medium based on the Sky Eagle algorithm. Background Art

[0002] Metaheuristic algorithm is a computer-based intelligent method that aims to solve complex optimization problems through iterative search algorithms. Due to its flexibility, metaheuristic algorithm has been effectively used in the optimization of various complex problems.

[0003] Among them, the Aquila Optimizer (AO) is a new nature-inspired metaheuristic algorithm proposed by Abualigah et al. in 2021. It can be applied to the industrial field. By simulating different hunting methods to establish a mathematical model, it can obtain the optimal solution to the problem to be solved. In related technologies, traditional AO algorithms focus on exploring and utilizing the search space to obtain optimization results and have good optimization capabilities. However, in its actual application, the traditional AO algorithm also has the problems of heavy dependence on parameters, mechanized repeated capture strategies, and inability to dynamically determine the optimal strategy, which leads to the traditional AO algorithm's slow convergence speed, weak search ability, and weak robustness. In addition, due to the imbalance between its global search and local development capabilities, it is easy to fall into local optimality. Summary of the invention

[0004] The present application provides a component performance prediction method, system, device and medium based on the Sky Eagle algorithm, which can at least be used to solve the problems of slow convergence speed, weak search capability, low robustness and easy falling into local optimality generated by the traditional AO algorithm in practical application fields.

[0005] In a first aspect, the present application provides a component performance prediction method based on the Sky Eagle algorithm, comprising the following steps: collecting component data and preprocessing the component data; optimizing a neural network based on an improved Sky Eagle algorithm, including: initializing parameters and determining a global optimal solution based on the fitness value of each individual in the population; calculating a solution distance and an angle cosine value between a current solution and the global optimal solution, and in response to the solution distance and the angle cosine value satisfying preset conditions, pre-updating the current solution to obtain an initial solution; performing a secondary update on the initial solution to obtain a secondary solution; in response to the secondary solution satisfying an iteration condition, outputting the optimal hyperparameters of the neural network, and configuring the neural network based on the optimal hyperparameters; inputting the preprocessed component data into the neural network to obtain a minimum value of a loss function to predict component performance.

[0006] Some embodiments of the present application select the optimal capture strategy through spatial configuration according to the distance information and angle information of the prey, which is conducive to combining spatial advantages, accurately and efficiently judging the dynamic changes of the prey, and adaptively adjusting the capture strategy based on the position changes of the prey, further improving the convergence speed and robustness of the AO algorithm, thereby improving the computational efficiency of the algorithm, avoiding mechanical repetition of capture strategies, and causing waste of computing resources. Some embodiments of the present application obtain the optimal hyperparameters through the improved AO algorithm (Improved Aquila Optimizer, I-AO), configure the neural network based on the optimal hyperparameters, optimize the use parameters of the neural network, and further improve the working efficiency of the neural network, so that it can more accurately predict component performance.

[0007] In a second aspect, in some embodiments of the present application, a component performance prediction system based on the Sky Eagle algorithm is provided, including: a collection and processing module, configured to collect component data and pre-process the component data; an algorithm optimization module, configured to optimize the neural network based on the improved Sky Eagle algorithm, including: parameter initialization, and determining the global optimal solution based on the fitness value of each individual in the population; calculating the solution distance and the angle cosine value between the current solution and the global optimal solution, and in response to the solution distance and the angle cosine value satisfying preset conditions, pre-updating the current solution to obtain an initial solution; performing a secondary update on the initial solution to obtain a secondary solution; in response to the secondary solution satisfying an iteration condition, outputting the optimal hyperparameters of the neural network, and configuring the neural network based on the optimal hyperparameters; a result prediction module, configured to input the pre-processed component data into the neural network to obtain the minimum value of the loss function to predict component performance.

[0008] In some embodiments of the present application, a component performance prediction device based on the Sky Eagle algorithm is provided, comprising: a memory and at least one processor, wherein the at least one memory is used to store computer instructions; and the at least one processor is used to execute at least part of the computer instructions to implement a component performance prediction method based on the Sky Eagle algorithm.

[0009] In some embodiments of the present application, a computer-readable storage medium is provided, wherein the storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer runs a component performance prediction method based on the Sky Eagle algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0011] Figure 1 An exemplary flow chart of a component performance prediction method based on the Skyhawk algorithm disclosed in some embodiments of the present application; Figures 2 (a)-2 (d) are schematic diagrams of the capture strategy of the Sky Eagle algorithm disclosed in some embodiments of the present application; Figure 3 An exemplary flow chart of secondary update disclosed for some embodiments of the present application; Figure 4 A system module diagram of a component performance prediction system based on the Sky Eagle algorithm disclosed in some embodiments of the present application; Figure 5(a)-5(j) It is a schematic diagram of an iterative curve for testing the I-AO algorithm based on ten benchmark test functions disclosed in some embodiments of the present application; Figure 6 A schematic diagram of prediction results of I-AO-TCN based on bearing components disclosed in some embodiments of the present application; Figure 7 Schematic diagram of prediction results of TCN based on bearing components disclosed in some embodiments of the present application. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0013] In order to facilitate understanding of the component performance prediction method, system, device and medium based on the Sky Eagle algorithm provided in the embodiments of the present application, the relevant technologies are first introduced in conjunction with the application scenarios.

[0014] In the prior art, metaheuristic algorithms often aim to solve complex optimization problems through iterative search algorithms. Among them, the Eagle Algorithm establishes a mathematical model based on the hunting habits of the Eagle, and explores and utilizes the search space to obtain optimization results through its different capture strategies. However, since its capture strategy is relatively mechanical and fixed, and is heavily dependent on parameters, it is impossible to dynamically adjust the capture strategy according to the position changes of the prey. As a result, the traditional Eagle Algorithm often has problems such as slow convergence speed, weak search ability, and easy to fall into local optimality in actual applications.

[0015] In view of this, the present application provides a component performance prediction method, system, device and medium based on the Skyhawk algorithm. Figure 1-7 The technical solutions disclosed in each embodiment of the present application are described in detail.

[0016] Figure 1 This is an exemplary flow chart of a component performance prediction method based on the Skyhawk algorithm disclosed in some embodiments of the present application. In some embodiments, the process 100 may be executed by a component performance prediction system 400 based on the Skyhawk algorithm.

[0017] Step 110 , collecting component data and preprocessing the component data. In some embodiments, step 110 may be performed by the collection and processing module 410 .

[0018] Component data refers to data related to the physical properties of the component to be tested during the test process. Taking a bearing as an example, the component data of a bearing may include vibration acceleration data, torque data, temperature data, and pressure data.

[0019] In some embodiments, the component data may be obtained in a variety of ways, such as by using corresponding experimental equipment or by measuring and obtaining the component data through various sensors (temperature sensors, pressure sensors, etc.).

[0020] In some embodiments, preprocessing the component data includes normalizing the component data.

[0021] In some embodiments, when the amount of component data is very large, after the normalization process is completed, the processed component data may be merged, for example, by merging the component data through a root mean square (RMS) operation.

[0022] The normalized or merged component data is divided into data blocks by sliding window slicing based on the sampling duration. The sampling duration refers to the duration of collecting component data. For example, the sampling duration can be 1 second.

[0023] Normalization can facilitate data calculation, thereby quantifying the data of each component. Furthermore, by processing data through data merging and sliding window slicing, approximate data can be summarized to reduce the amount of data, thereby preventing excessive data from slowing down the calculation speed and consuming too many computing resources during subsequent use. In addition, through the above data processing steps, the component data can be serialized, which can more intuitively and clearly characterize the change trend of the component data.

[0024] Step 120, optimizing the neural network based on the improved Sky Eagle algorithm. In some embodiments, step 120 may be performed by the algorithm optimization module 420, including steps 121-124.

[0025] Step 121, parameter initialization, and determining the global optimal solution based on the fitness value of each individual in the population.

[0026] In some embodiments, parameter initialization refers to initializing algorithm parameters in the AO algorithm.

[0027] Among them, the algorithm parameters include the population size , the spatial dimension of candidate solutions The maximum number of iterations is , distance threshold , Angle threshold and the maximum random walk iterations .initialization Group position vector .in represents a random number between [0,1], and Respectively represent the problems to be solved The upper and lower bounds of the initialization number of iterations .

[0028] In some embodiments, Indicates Among the populations, The position of each individual (eagle). In this application, the position of the eagle refers to the hyperparameters of the neural network. The resulting space matrix represents all candidate solutions to the problem.

[0029] In some embodiments, the distance threshold , Angle threshold , maximum number of random walk iterations It can be obtained by setting based on experience by the staff in the field. For example, the distance threshold The value range of can be [1e-4, 1e-8], the maximum number of random walk iterations The value range of can be [1, 10], the angle threshold It can be 0.5.

[0030] In some embodiments, the fitness value of each individual in the population is calculated by a fitness function, and the individual with the largest fitness value is determined as the global optimal solution. .

[0031] Step 122, calculate the solution distance and angle cosine value between the current solution and the global optimal solution, and in response to the solution distance and angle cosine value satisfying preset conditions, pre-update the current solution to obtain an initial solution.

[0032] The solution distance refers to the distance between the current solution and the global optimal solution. The difference in fitness between the current solution and the global optimal solution. The cosine of the angle between .

[0033] In some embodiments, the preset condition can be determined in a variety of ways. For example, the preset condition can be determined based on the experience of the staff in the field.

[0034] In some embodiments, as shown in Figure 2, the pre-update includes: expanding the search update graph 2 (a), reducing the search update graph 2 (b), expanding the development update graph 2 (c) and reducing the development update graph 2 (d).

[0035] In some embodiments, the preset conditions include a plurality of different judgment conditions, and the aforementioned judgment conditions include the relationship between the solution distance and the distance threshold, and / or the relationship between the angle cosine value and the angle threshold.

[0036] In some embodiments, as shown in FIG. 2( a), in response to the current solution and the global optimal solution The solution distance d is greater than or equal to the distance threshold, the angle cosine value The absolute value of is greater than the angle threshold of 0.5, that is, the judgment condition is At this time, the eagle is looking for prey in a place far away from its relative position. Due to the long distance and small angle, the eagle will catch prey by soaring high in the sky and diving vertically. The current solution position can be expanded and updated based on formulas (1) and (2): (1) (2) in, Updates generated for expanded search Second solution, For the The optimal solution obtained before the iteration, For the The average of the current solutions of all individuals at the iteration.

[0037] In some embodiments, as shown in FIG2(b), in response to the current solution and the global optimal solution The solution distance d is greater than or equal to the distance threshold, the angle cosine value The absolute value of is less than or equal to the angle threshold 0.5, that is, the judgment condition is At this time, the eagle is searching for prey at a place far away from its relative position. Due to the relatively long distance and large angle, the eagle will hover above the target prey, preparing to land and attack, and perform contour flight of short gliding attack. The position of the current solution can be narrowed down and updated based on formula (3): (3) in, Update generated to narrow search Second solution, is the flight distribution function, It is At the iteration Randomly generated solutions in the range, and Optimized spiral trajectory during flight.

[0038] Flight distribution function It can be calculated by formula (4): (4) in, , and is a random number between 0 and 1, , It can be calculated by formula (5): (5) and It can be calculated by formula (6) and (7): (6) (7) , , , in, , , Can take values ​​in [1, 20], is an integer from 1 to Dim.

[0039] In some embodiments, as shown in FIG2(c), in response to the current solution and the global optimal solution The solution distance d is less than the distance threshold, and the angle cosine value The absolute value of is greater than the angle threshold of 0.5, that is, the judgment condition is At this time, when the eagle is looking for prey that is relatively close to its relative position, due to the close distance and large angle, the eagle will fly low and slowly descend to catch the prey. The current solution position can be expanded and updated based on formula (8): (8) in, Updated for expanded development Second solution, and They are all development adjustment parameters, and their values ​​are all 0.1. is the upper bound of the current solution, is the lower bound of the current solution.

[0040] In some embodiments, as shown in FIG2(d), in response to the current solution and the global optimal solution The solution distance d is less than the distance threshold, and the angle cosine value The absolute value of is less than or equal to the angle threshold 0.5, that is, the judgment condition is When the eagle searches for prey that is very close to its relative position, due to the short distance and small angle, the eagle will catch the prey while walking. The position of the current solution can be narrowed and updated based on formula (9): (9) (10) (11) (12) in, Update generated for minification development Second solution, is the quality function of the balanced search strategy, Refers to the various movements used by predators to track their prey during hunting. It presents decreasing values ​​from 2 to 0, representing the flight slope of the Skyhawk from the first position (1) to the last position (t). It can be calculated by formula (12): and It can be calculated using the above formulas (10) and (11).

[0041] In some embodiments, after pre-updating using the above four update strategies, the updated fitness value is calculated and the global optimal solution is updated. , and get the initial solution .

[0042] Some embodiments of the present application, by introducing solution distance and angle cosine value, execute corresponding update strategies according to the changes in the spatial position of the eagle during the hunting process, thereby greatly improving the adaptability of different update strategies, dynamically adjusting different update strategies, and improving the problems of low search accuracy and premature convergence caused by ignoring the individual search experience of the eagle in the original algorithm, thereby improving the robustness and convergence speed of the AO algorithm, and further optimizing the AO algorithm.

[0043] Step 123, perform a second update on the initial solution to obtain a secondary solution. For a detailed description of step 123, see Figure 3 The corresponding content.

[0044] Further, it is determined whether the quadratic solution satisfies the iteration condition. In response to the quadratic solution not satisfying the iteration condition, the process returns to step 121 .

[0045] Step 124, in response to the quadratic solution satisfying the iteration condition, outputting the optimal hyperparameters of the neural network, and configuring the neural network based on the optimal hyperparameters.

[0046] In some embodiments, the iteration condition may include: the number of iterations , or the quadratic solution and the global optimal solution The solution distance is 0, that is .

[0047] Hyperparameters refer to the parameters set before the neural network starts learning in machine learning, that is, hyperparameters are parameter data that cannot be obtained through training during the model training process. Among them, the optimal hyperparameters refer to the configuration parameters that can make the learning performance and effect of the neural network better.

[0048] Step 130 , input the preprocessed component data into the neural network to obtain the minimum value of the loss function to predict component performance. In some embodiments, step 130 may be performed by the result prediction module 430 .

[0049] Figure 3 An exemplary flow chart of a secondary update disclosed in some embodiments of the present application. Figure 3 As shown, the process 300 includes the following steps: In some embodiments, the process 300 may be executed by the algorithm optimization module 420 .

[0050] Step 310, obtaining an additional random factor.

[0051] In some embodiments, the additional random factor may be determined in a variety of ways, such as Math.random(), random class, etc.

[0052] In some embodiments, the additional random factor can be determined based on the random coefficient. For example, the additional random factor can be equal to the sum of the random coefficient and the rand() function. Exemplarily, when the random coefficient is 0.5, the value range of the additional random factor is [0.5, 1.5].

[0053] Step 320: Based on the additional random factor and the initial solution, a random walk solution of the initial solution is calculated.

[0054] The random walk solution refers to the latest solution obtained by updating the random walk update strategy.

[0055] In some embodiments, steps 310 and 320 may be performed once or multiple times. It can be calculated by formula (13).

[0056] (13) in, is the fitness value of the current solution, is the maximum fitness value among the candidate solutions.

[0057] By executing steps 310 and 320 based on the adaptive execution times, the fitness value of the candidate solution can be dynamically adjusted to further balance the exploration and development needs during the random walk update process. The larger the number, the more times it is executed. The larger the value, the better. This helps the random walk update avoid falling into the local optimum. However, in the specific implementation process, The larger it is, the more computational cost and computational time will be spent. The smaller it is, the faster the algorithm converges, but there is also the problem of falling into the local optimum.

[0058] In some embodiments, preferably, The value of can be 5.

[0059] In some embodiments, the random walk solution can be calculated based on formula ((14): (14) in, refers to the random walk solution, that is, the position of the latest solution obtained by random walk update, is the initial solution, is an additional random factor, SD is the difference between the current solution and the initial solution The standard deviation between is the upper bound of the current solution, is the lower bound of the current solution.

[0060] Step 330: Mix the local information and the global information to obtain mixed data.

[0061] Local information refers to the current position information of some candidate solutions, and global information refers to the current position information of all candidate solutions. Mixed data is the sum of local information and global information.

[0062] In some embodiments, the local information may be half of the global information.

[0063] Step 340, based on the mixed data and the random walk solution, a random correction solution is calculated.

[0064] The random correction solution refers to the latest solution obtained by updating the random correction update strategy.

[0065] In some embodiments, the random correction solution can be calculated based on formula (15) as follows:

[0066] in, is a random corrected solution, It is local information. is global information, is the average position of candidate solutions in the current population.

[0067] Step 350: construct dynamic weights based on the nonlinear function.

[0068] In some embodiments, dynamic weights It can be calculated based on formula (16) as follows: (16) Among them, t is the current iteration number and T is the maximum iteration number.

[0069] Step 360, based on the momentum factor, dynamic weight and random correction solution, a quadratic solution is calculated.

[0070] The momentum factor is a parameter used to improve search flexibility. In some embodiments, the momentum factor can be preset by a person skilled in the art. Exemplarily, the momentum factor can be 0.9.

[0071] In some embodiments, the secondary solution It can be calculated based on the following formula (17): (17) The Improved Aquila Optimizer (I-AO) introduces a random walk factor to make the step size more adaptive, thereby effectively solving the problems of long convergence time and high randomness of the traditional AO algorithm, further improving the stability and robustness of the algorithm. Random correction updates based on mixed local and global information can enhance the flexibility of the search, avoid falling into local optimality in the global scope, construct dynamic weights through nonlinear functions, introduce adaptive adjustment strategies, make weight changes smooth, in order to further develop and explore the better areas, and increase the diversity of the search by adding momentum factors, so that the AO algorithm can capture the optimal solution more accurately and efficiently in the global scope.

[0072] Figure 4 This is a system module diagram of a component performance prediction system based on the Skyhawk algorithm disclosed in some embodiments of this application. Figure 4 As shown, the system 400 for predicting component performance based on the Sky Eagle algorithm includes a collection and processing module 410, an algorithm optimization module 420, and a result prediction module 430. For detailed descriptions of the functions of each module, please refer to the specification Figure 1-3 Detailed description.

[0073] The acquisition and processing module 410 is a module for acquiring and processing data. In some embodiments, the acquisition and processing module 410 is configured to acquire component data and pre-process the component data.

[0074] The algorithm optimization module 420 is a module for optimizing the AO algorithm. In some embodiments, the algorithm optimization module 420 is configured to optimize the neural network based on the improved Sky Eagle algorithm, including: parameter initialization, and determining the global optimal solution based on the fitness value of each individual in the population; calculating the solution distance and angle cosine value between the current solution and the global optimal solution, and in response to the solution distance and angle cosine value satisfying the preset conditions, pre-updating the current solution to obtain the initial solution; performing a secondary update on the initial solution to obtain the secondary solution; in response to the secondary solution satisfying the iteration conditions, outputting the optimal hyperparameters of the neural network, and configuring the neural network based on the optimal hyperparameters.

[0075] In some embodiments, the algorithm optimization module 420 is configured to map the angle cosine value to a random number of [0, 1]; in response to the distance between the mapped angle cosine value and the solution satisfying a preset condition, pre-update the current solution.

[0076] In some embodiments, the algorithm optimization module 420 is configured to obtain additional random factors; based on the additional random factors and the initial solution, calculate a random walk solution of the initial solution; mix local information with global information to obtain mixed data; based on the mixed data and the random walk solution, calculate a random corrected solution; construct dynamic weights based on nonlinear functions; and calculate a quadratic solution based on the momentum factor, the dynamic weights, and the random corrected solution.

[0077] The result prediction module 430 is a module for predicting component performance. In some embodiments, the result prediction module 430 is configured to input the preprocessed component data into a neural network to obtain the minimum value of the loss function to predict component performance.

[0078] It should be understood that the system and its modules shown in FIG4 can be implemented in various ways, for example, by hardware, software, or a combination of software and hardware.

[0079] It should be noted that the above description of the system and its modules is only for convenience of description and is not intended to limit the present specification to the embodiments described above. It is understandable that, after understanding the principle of the system, those skilled in the art may arbitrarily combine the modules or form a subsystem to connect with other modules without departing from the principle.

[0080] The following is a specific example to illustrate the prediction result test process of the component performance prediction method based on the Aquila algorithm disclosed in this specification. This example is based on a bearing component, and the performance prediction results of the traditional Aquila Optimizer (AO), Particle Swarm Optimization (PSO), Puma Optimizer (PO), Genetic Algorithms (GA), Whale Optimization Algorithm (WOA) and the improved Aquila Optimizer (I-AO) in this specification are tested through ten benchmark test functions F1-F10.

[0081] This embodiment is based on the Windows 11 operating system, the computer hardware configuration is Intel (R) Core(TM) i9-13900HX CPU@2.20 GHz and 32GB memory, and is implemented using MATLAB 2023a.

[0082] Table 1 Basic information of the 10 benchmark functions of CEC2019 , First, initialize the parameters of the algorithm to be tested and set its population size Set to 30, the maximum number of iterations The remaining parameters are set to 500, and the detailed description is shown in Table 2.

[0083] Table 2 Parameter settings of I-AO algorithm and other algorithms , Each algorithm was run independently 30 times, and the average and standard deviation of each algorithm's calculation results were recorded. The smaller the average value of each algorithm's calculation results, the stronger its optimization ability is, and the smaller the standard deviation, the better the stability and adaptability of the algorithm.

[0084] Based on the calculation results of the above algorithms, they are tested based on the 10 benchmark test functions (F1-F10) of CEC2019. The test results are shown in Table 3: Table 3 Comparison of optimization results of various algorithms , As shown in Table 3, Average represents the average value, STD represents the standard deviation, and the black bold font represents the value with the smallest average value and standard deviation during the test of the corresponding test function. Based on the ten benchmark test functions, the calculation results of each algorithm are compared. Compared with other algorithms, I-AO performs well in most test functions. Compared with traditional AO algorithms and other optimization algorithms, it has stronger optimization ability, better stability, and stronger adaptability.

[0085] In addition to statistical indicators, the convergence speed of the above algorithms can be evaluated based on the convergence curve. Figure 5(a)-5(j) A schematic diagram of an iterative curve for testing the I-AO algorithm based on ten benchmark test functions disclosed in some embodiments of the present application.

[0086] like Figure 5(a)-5(j) As shown in the figure, the horizontal axis of the curve is the current iteration number, and the vertical axis is the optimal value. The descent speed of the iteration curve of the I-AO algorithm is faster than that of other optimization algorithms, indicating that the convergence speed of the I-AO algorithm is also faster, which further illustrates that the corresponding update strategy based on the spatial configuration selection is conducive to enhancing the ability of the algorithm to get rid of local extreme values ​​during the optimization process, which can accelerate the convergence speed of the algorithm, improve its calculation accuracy, and significantly improve the optimization performance of the AO algorithm.

[0087] According to Tables 1, 2, 3 and Figure 5(a)-5(j) It can be seen from the iteration curve in that the I-AO algorithm has certain optimization and improvement effects in computing power and optimization ability compared with traditional optimization algorithms.

[0088] Next, we take the prediction of the remaining service life of a bearing as an example to test the performance of the I-AO algorithm in practical applications.

[0089] A time convolution network (TCN) is selected, and the hyperparameters of the TCN are selected using the I-AO algorithm. In this embodiment, the hyperparameters selected by the I-AO algorithm are the learning rate, the number of hidden unit layers in the convolution layer, and the regularization rate.

[0090] In order to further verify that the TCN optimized based on the I-AO algorithm has more accurate prediction ability and the stability of its prediction results, two evaluation indicators, Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), are introduced. Among them, MAE and RMSE can be used to reflect the real error and large or uncommon errors.

[0091] Table 4 Comparison of bearing performance , Among them, I-AO-TCN refers to TCN optimized by I-AO algorithm, and Bearing1-1 and Bearing2-3 are bearing models. According to the data in the table, the MAE and RMSE error margins of I-AO-TCN are the smallest. This shows that the I-AO algorithm has a certain optimization and improvement effect on the prediction and construction performance compared with a single neural network.

[0092] Figure 6 A schematic diagram of prediction results of I-AO-TCN based on bearing components disclosed in some embodiments of the present application; Figure 7 Schematic diagram of prediction results of TCN based on bearing components disclosed in some embodiments of the present application. Figure 6 , 7 As shown, the vertical axis is the remaining useful life (RUL) and the horizontal axis is time. Compared with the I-AO-TCN, the prediction curve of the single TCN fluctuates more and the prediction accuracy is not as accurate as that of the I-AO-TCN.

[0093] Therefore, according to the above Table 4 and Figure 6 , 7 From the relevant data, it can be found that the hyperparameter values ​​of the neural network after the I-AO algorithm optimization can help the neural network obtain more accurate prediction results in practical applications, thereby better predicting component performance.

[0094] From the above description, we can know that the I-AO algorithm has a good optimization effect on neural networks and can significantly improve the performance of neural networks. Therefore, the component performance prediction method based on the Sky Eagle algorithm in this specification can also be applied to other application scenarios, including any complex problem that requires optimization through neural networks.

[0095] Some embodiments of the present specification provide a component performance prediction device based on the Sky Eagle algorithm, characterized in that it includes: a memory and at least one processor, wherein the at least one memory is used to store computer instructions; and the at least one processor is used to execute at least part of the computer instructions to implement a component performance prediction method based on the Sky Eagle algorithm.

[0096] Some embodiments of the present specification provide a computer-readable storage medium, characterized in that the storage medium stores computer instructions, and when a computer reads the computer instructions in the storage medium, the computer runs a component performance prediction method based on the Sky Eagle algorithm.

[0097] Exemplarily, the computer program may be divided into one or more modules / units, the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0098] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0099] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The aforementioned processor is the control center of the terminal device, and uses various interfaces and lines to connect the various parts of the entire terminal device. The memory may be used to store the computer program and / or module, and the processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the mobile terminal, etc. In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0100] Among them, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the aforementioned computer program includes computer program code, and the aforementioned computer program code can be in source code form, object code form, executable file or some intermediate form. The aforementioned computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0101] The above embodiments of the present application focus on describing the differences between the various embodiments. As long as the different optimization features between the various embodiments are not contradictory, they can be combined to form a better embodiment. Considering the simplicity of the text, they will not be repeated here.

[0102] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A component performance prediction method based on the Sky Eagle algorithm, characterized in that: include Follow these steps: Collecting component data and preprocessing the component data; Optimize neural network based on improved SkyEagle algorithm, including: Initialize parameters and determine the global optimal solution based on the fitness value of each individual in the population; Calculating a solution distance and an angle cosine value between a current solution and the global optimal solution, and in response to the solution distance and the angle cosine value satisfying a preset condition, pre-updating the current solution to obtain an initial solution; Performing a secondary update on the initial solution to obtain a secondary solution; In response to the quadratic solution satisfying an iteration condition, outputting an optimal hyperparameter of the neural network, and configuring the neural network based on the optimal hyperparameter; The preprocessed component data is input into the neural network to obtain the minimum value of the loss function to predict component performance.

2. The component performance prediction method based on the Sky Eagle algorithm according to claim 1 is characterized in that: Calculating the solution distance and the angle cosine value between the current solution and the global optimal solution, and in response to the solution distance and the angle cosine value satisfying a preset condition, pre-updating the current solution to obtain an initial solution includes: Map the angle cosine value to a random number in [0, 1]; In response to the mapped angle cosine value and the solution distance satisfying the preset condition, the current solution is pre-updated.

3. The component performance prediction method based on the Sky Eagle algorithm according to claim 2 is characterized in that: The preset conditions include at least one judgment condition. In response to satisfying different judgment conditions, the corresponding pre-update is executed. The pre-update includes expanding search update, narrowing search update, expanding development update and narrowing development update.

4. The component performance prediction method based on the Sky Eagle algorithm according to claim 3 is characterized in that: The judgment condition includes the magnitude relationship between the solution distance and the distance threshold, and / or the magnitude relationship between the angle cosine value and the angle threshold.

5. The component performance prediction method based on the Sky Eagle algorithm according to claim 1 is characterized in that: The initial solution is updated twice to obtain a secondary solution including: Get additional random factors; Based on the additional random factor and the initial solution, a random walk solution of the initial solution is calculated; Mix local information with global information to get mixed data; Based on the mixed data and the random walk solution, a random correction solution is calculated; Build dynamic weights based on nonlinear functions; The quadratic solution is calculated based on the momentum factor, the dynamic weight and the random correction solution.

6. The component performance prediction method based on the Sky Eagle algorithm according to claim 5 is characterized in that: The value range of the additional random factor is determined based on the random coefficient.

7. The component performance prediction method based on the Sky Eagle algorithm according to claim 5 is characterized in that: The value range of the momentum factor is [0,1].

8. A component performance prediction system based on the Sky Eagle algorithm, characterized in that: include: A collection and processing module is configured to collect component data and pre-process the component data; The algorithm optimization module is configured to optimize the neural network based on the improved Sky Eagle algorithm, including: Initialize parameters and determine the global optimal solution based on the fitness value of each individual in the population; Calculating a solution distance and an angle cosine value between a current solution and the global optimal solution, and in response to the solution distance and the angle cosine value satisfying a preset condition, pre-updating the current solution to obtain an initial solution; Performing a secondary update on the initial solution to obtain a secondary solution; In response to the quadratic solution satisfying an iteration condition, outputting an optimal hyperparameter of the neural network, and configuring the neural network based on the optimal hyperparameter; The result prediction module is configured to input the preprocessed component data into the neural network to obtain the minimum value of the loss function to predict component performance.

9. A component performance prediction device based on the Sky Eagle algorithm, characterized in that: include: memory and at least one processor, wherein The at least one memory is used to store computer instructions; The at least one processor is configured to execute at least part of the computer instructions to implement the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer runs the method described in claims 1-7.

Images