Task success rate hybrid evaluation method combining simulation and proxy model

Abstract

The application relates to a task success rate hybrid evaluation method combining simulation and an agent model. The method comprises the following steps: performing simulation sampling on an agent unit to obtain maintenance time and downtime of the agent unit; performing data analysis on the maintenance time and the downtime of the agent unit by using a data analysis method to obtain a training sample of the agent unit, and designing three types of agent models corresponding to the agent unit according to the training sample; generating the maintenance time and the downtime of the agent unit according to the three types of agent models, and obtaining the maintenance time and the downtime by using a simulation method; judging whether a task to be evaluated is successful according to a logic relationship of a unit series connection structure, maintenance time or downtime of all units and task redundancy maintenance time or downtime of the task to be evaluated, and evaluating the task to be evaluated multiple times to obtain an evaluation result of a success rate of the task to be evaluated. The method can improve the convergence speed of task success rate evaluation.

Description

Technical Field

[0001] This application relates to the field of maintenance and support technology, and in particular to a method, apparatus, computer equipment, and storage medium for evaluating task success rate by combining simulation and proxy models. Background Technology

[0002] Task success rate refers to the probability that a system will complete a predetermined task under a given maintenance and support plan. It is a top-level indicator that comprehensively reflects the system's reliability, maintainability, and spare parts availability. Solving for the task success probability has important guiding significance for general quality characteristic analysis, and it is also an effective means to further improve reliability and implement testability, maintainability, and spare parts availability in engineering.

[0003] With the rapid development of science and technology, intelligent, electronic, and digital equipment is being continuously applied to various types of equipment, resulting in increasingly complex system scales. Unlike simple equipment, quantitatively assessing the mission success of such systems requires consideration of numerous and complex factors. These factors are not only related to the system's reliability and inherent capabilities but also to the mission's execution time, the logical relationships for mission success, and the mission's operational requirements. Typically, there is a lack of corresponding mathematical formulas to describe these factors.

[0004] Monte Carlo simulation is one of the most effective methods for evaluating the success rate of large and complex systems. However, this method has two shortcomings: slow convergence speed, requiring a large number of simulation iterations to obtain high accuracy, necessitating a trade-off between computational accuracy and time cost; and fixed component states, requiring a complete re-simulation process based on the entire system once component parameters are adjusted, lacking applicability in dynamic scenarios. Summary of the Invention

[0005] Based on this, it is necessary to provide a hybrid evaluation method, apparatus, computer equipment, and storage medium for task success rate that combines simulation and proxy models, which can improve the convergence speed of task success rate evaluation and address the aforementioned technical problems.

[0006] A hybrid evaluation method for task success rate combining simulation and proxy models, the method comprising:

[0007] To obtain the state changes of various components in a complex system;

[0008] The complex system is divided into multiple cascaded basic structural units based on the state change; each basic structural unit includes a simulation unit and multiple surrogate units.

[0009] Simulation sampling was performed on the agent unit to obtain the maintenance time and downtime of the agent unit under various states;

[0010] Data analysis methods were used to analyze the maintenance time and downtime of the agent unit to obtain training samples of the agent unit. Based on the training samples, three types of agent models were designed for the agent unit.

[0011] Based on the three types of proxy models trained by each unit, the maintenance time and downtime of the proxy unit are generated, and the simulation unit uses simulation methods to obtain the maintenance time and downtime.

[0012] The success of the task to be evaluated is determined based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task to be evaluated.

[0013] The task to be evaluated is evaluated multiple times based on the pre-set task time, allowable maintenance time percentage, allowable downtime percentage, and number of task evaluations, to obtain the evaluation result of the success rate of the task to be evaluated.

[0014] In one embodiment, the state change quantity is a discrete quantity of the component lifetime distribution parameters; the distribution parameters include: the shape parameter and scale parameter of the Weibull distribution and the MTBF of the exponential distribution.

[0015] In one embodiment, the process of obtaining the state changes of various components in a complex system includes:

[0016] Based on historical failure data, degradation data, and the initial parameters of a given life distribution model, determine the range of parameter variations within the component's life cycle;

[0017] By combining the task success rate with component parameters based on the range of parameter changes, the interval in which the task success rate changes significantly can be determined.

[0018] By taking a small interval from the interval, the continuous parameter is discretized, and the number of discretized parameters is the change in the state of the component.

[0019] In one embodiment, the complex system is divided into multiple cascaded basic structural units based on the amount of state change, including:

[0020] Components that do not change or change only slightly during their lifespan are classified as surrogate units, while components that undergo significant state changes during their lifespan are classified as simulation units.

[0021] In one embodiment, the states of the proxy unit and the simulation unit are represented by the lifetime distribution parameters of each component in the unit and the encoding of the system operating conditions, including:

[0022] Components of the same type connected in parallel have the same tendency to deteriorate over their life cycle, so they are represented by the same life distribution parameter.

[0023] Different operating conditions are numbered and used as unit state characteristics, and the distribution parameters of a type of component that does not participate in unit operation are set to 0.

[0024] In one embodiment, the three proxy models include a first proxy model, a second proxy model, and a third proxy model; data analysis methods are used to analyze the maintenance time and downtime of the proxy units to obtain training samples for the proxy units, including:

[0025] For any state in the unit state set, sample N times. If the component fails but the unit does not stop, the number of times is a, and the maintenance time data is A; if the component fails and the unit stops, the number of times is b, and the maintenance time data is B1, and the downtime data is B2. The number of times no component failure occurs is Nab. The normal probability P1, the unit failure probability P2, and the unit downtime probability P3 in a single simulation are obtained.

[0026] Based on the normal probability P1 of no fault occurring in a single simulation, the unit fault probability P2, and the unit downtime probability, the training samples of the first surrogate model are constructed and mapped as: unit state parameters → (P2, P3).

[0027] The empirical distribution function of the maintenance time data A is used as an approximation of the overall distribution function to construct the training samples of the second surrogate model, which are mapped as: (unit state parameters, empirical distribution function values) → maintenance time.

[0028] In one embodiment, two independent random values ​​u and w are generated that are uniformly distributed on (0,1); where w is a conditional distribution c. u The values ​​of (u,v) are determined; the value of v is obtained through the inverse function of the conditional distribution, i.e. (u,v) is a random value with a Copula function C(u,v);

[0029] Substituting into the inverse function of the marginal distribution function x = F -1 (u) and y=G -1 (v) and (x,y) are random variables that satisfy the coupling between the two.

[0030] Empirical distribution values ​​were sampled from maintenance time data B1 and downtime data B2 to obtain uniformly distributed data U and V.

[0031] Using the elements in U and V as the x-axis and y-axis coordinates respectively, the data distribution is obtained;

[0032] The inverse function of the fitted conditional distribution is obtained based on the data distribution. The training samples are mapped as (w,u)→v; then, the training samples of the third proxy model under a single state are constructed using the inverse functions u→x and v→y, and mapped as u→x and (w,u)→y.

[0033] In one embodiment, the first surrogate model takes the unit state as input and outputs the failure probability and the downtime probability; the second surrogate model takes the unit state as input and a random number uniformly distributed on (0,1) as input and outputs the maintenance time of the unit under the failure probability; the third surrogate model takes the unit state as input and two independent random numbers uniformly distributed on (0,1) as input and outputs the correlated maintenance time and downtime of the unit under the downtime probability.

[0034] In one embodiment, the maintenance time and downtime of the proxy unit are generated based on the three types of proxy models trained by each unit, including:

[0035] During a single simulation, for each surrogate unit, the first step is to determine the unit's fault and downtime status. The first surrogate model generates (P2, P3) and simultaneously generates a uniformly random value P on (0, 1). If P ∈ [0, P2), it indicates that the unit only experienced a fault in this simulation, and the second surrogate model generates the repair time. If P ∈ [P2, P2+P3), it indicates that the unit experienced a downtime in this simulation, and the third surrogate model generates the repair time and downtime. Otherwise, no fault occurred in this simulation, and the repair time and downtime are 0.

[0036] In one embodiment, determining whether the task to be evaluated is successful based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task to be evaluated includes:

[0037] Based on the logical relationship of the unit serial structure, the system maintenance time or downtime is equivalent to the sum of the maintenance time or downtime of each unit. Then, based on the task redundancy maintenance time or downtime of the task to be evaluated, if the system maintenance time or downtime is greater than the task redundancy maintenance time or downtime of the task to be evaluated, the task fails; otherwise, the task succeeds.

[0038] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0039] To obtain the state changes of various components in a complex system;

[0040] The complex system is divided into multiple cascaded basic structural units based on the state change; each basic structural unit includes a simulation unit and multiple surrogate units.

[0041] Simulation sampling was performed on the agent unit to obtain the maintenance time and downtime of the agent unit under various states;

[0042] Data analysis methods were used to analyze the maintenance time and downtime of the agent unit to obtain training samples of the agent unit. Based on the training samples, three types of agent models were designed for the agent unit.

[0043] Based on the three types of proxy models trained by each unit, the maintenance time and downtime of the proxy unit are generated, and the simulation unit uses simulation methods to obtain the maintenance time and downtime.

[0044] The success of the task to be evaluated is determined based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task to be evaluated.

[0045] The task to be evaluated is evaluated multiple times based on the pre-set task time, allowable maintenance time percentage, allowable downtime percentage, and number of task evaluations, to obtain the evaluation result of the success rate of the task to be evaluated.

[0046] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0047] To obtain the state changes of various components in a complex system;

[0048] The complex system is divided into multiple cascaded basic structural units based on the state change; each basic structural unit includes a simulation unit and multiple surrogate units.

[0049] Simulation sampling was performed on the agent unit to obtain the maintenance time and downtime of the agent unit under various states;

[0050] Data analysis methods were used to analyze the maintenance time and downtime of the agent unit to obtain training samples of the agent unit. Based on the training samples, three types of agent models were designed for the agent unit.

[0051] Based on the three types of proxy models trained by each unit, the maintenance time and downtime of the proxy unit are generated, and the simulation unit uses simulation methods to obtain the maintenance time and downtime.

[0052] The success of the task to be evaluated is determined based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task to be evaluated.

[0053] The task to be evaluated is evaluated multiple times based on the pre-set task time, allowable maintenance time percentage, allowable downtime percentage, and number of task evaluations, to obtain the evaluation result of the success rate of the task to be evaluated.

[0054] The above-mentioned method, apparatus, computer equipment, and storage medium for evaluating task success rate by combining simulation and proxy models achieve an approximate description of complex simulation processes that change with state by constructing a proxy model. It simulates and generates unit maintenance time and downtime for evaluating system task success rate. At the same time, while ensuring high accuracy, it reduces the scale of repetitive simulation caused by state changes from the entire system to local units, avoiding a large amount of redundant calculations. It is more suitable for rapid evaluation of dynamic scenarios and improves the convergence speed of task success rate evaluation. Attached Figure Description

[0055] Figure 1 This is an application scenario diagram of a hybrid evaluation method for task success rate that combines simulation and proxy models, as shown in one embodiment.

[0056] Figure 2 This is a block diagram of the reliability of a test system in one embodiment;

[0057] Figure 3 This is a schematic diagram illustrating the structural division of the proxy unit and simulation unit in one embodiment;

[0058] Figure 4 This is a schematic diagram illustrating an example of a state feature encoding method in another embodiment; Figure 4 (a) is a schematic diagram of the input state characteristics of the basic unit; Figure 4 (b) to (f) are schematic diagrams of the working conditions corresponding to numbers 0, 1, 2, 3, and 4, respectively;

[0059] Figure 5 This is a schematic diagram of the correlation distribution of maintenance time and downtime and sample collection in one embodiment;

[0060] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] In one embodiment, such as Figure 1 As shown, a hybrid evaluation method for task success rate combining simulation and proxy models is provided, including the following steps:

[0063] Step 102: Obtain the state changes of each component in the complex system; divide the complex system into multiple serial basic structural units based on the state changes; each basic structural unit includes a simulation unit and multiple proxy units.

[0064] The task success rate assessment is required for complex systems with multiple components. To decompose the system into multiple serial basic units, it is first necessary to define the state changes of the components. The component state is represented by the component life distribution parameters. Based on existing fault data, degradation data and the initial parameters of the given life distribution model, the range of parameter changes within the component life cycle can be determined. By combining the task success rate with the component parameters, the intervals that cause significant changes in the task success rate are taken as small intervals, and the continuous parameters are discretized. The number of discretized parameters is the state change of the component. Figure 2 As an example, the test system block diagram is used. The lifetime distribution parameters of the components in the diagram are discretized. The shape parameter of the Weibull distribution and the discrete spacing of the MTBF of the exponential distribution are as follows in different intervals:

[0065]

[0066] δ represents the state parameter value, and d represents the spacing. For the shape parameter m, equidistant dispersion is adopted, with a spacing d = 2.

[0067] The discrete state parameters of the components shown in the figure are listed in Table 1. In the table, θ represents MTBF, and μ and σ represent the mean and variance of the normal distribution, respectively.

[0068] Table 1 Component Distribution Types and Parameters

[0069]

[0070] The system structure is divided based on the changes in component states. Considering that real-world systems commonly contain unrepairable and state-stable components whose states do not change or change only slightly during their lifespan, these components with minimal changes are classified into surrogate units. However, multi-state components in engineering projects are prone to causing combinatorial explosion problems in the state space; therefore, they are classified as simulation units. Complex systems can be divided into six surrogate units and one simulation unit, such as... Figure 3 As shown, agent units 2, 3 and 5 are reusable units.

[0071] For a single agent unit, because time data needs to be collected in each state of the unit to train the model, a unique feature vector representing the unit's state is required. Figure 4 The system shown is used as an example to illustrate the encoding method for state characteristics. The lifetime distribution of the components in the figure is a δ-parameter lifetime distribution. The specific method consists of three steps: Since parallel components of the same type have the same state degradation trend within their lifetime, the same lifetime distribution parameter is used to represent their state, such as... Figure 4 As shown in (a), the basic unit input state features include (δ1, δ2, δ3, δ4, δ5); different operating conditions are numbered and used as unit state features, such as... Figure 4As shown in (b) to (f), this unit has 5 different operating conditions, numbered 0, 1, 2, 3, and 4 respectively. For the component type that does not participate in the unit's operation, changes in its parameters do not affect the output results; therefore, the distributed parameters of the corresponding component type are set to 0. Figure 4 (d) Under this condition, components of categories 4 and 5 do not participate in the work, and δ4 and δ5 are set to 0.

[0072] The relevant information for the six types of agency units obtained using the above method is shown in Table 2.

[0073] Table 2. Proxy Unit Feature Parameters, Sample Size, and Reusable Units

[0074]

[0075] Step 104: Simulate and sample the proxy unit to obtain the maintenance time and downtime of the proxy unit under each state; use data analysis methods to analyze the maintenance time and downtime of the proxy unit to obtain training samples for the proxy unit, and design three types of proxy models corresponding to the proxy unit based on the training samples. Data can be obtained by sampling N times for any state in the unit state set: the number of times the unit does not stop due to component failure (a times), and the maintenance time data A = {T}. m1 ,T m2 ,…,T ma}; Component failure and unit downtime count b times, maintenance time data B1 = {T m1 ,T m2 ,…,T mb}, Downtime data B2 = {T d1 ,T d2 ,…,T db}; and the number of times no component failure occurred (Nab times).

[0076] Training samples for the first proxy model:

[0077] From the above data, the probability can be obtained as follows:

[0078]

[0079] Where P1 represents the normal probability of no fault occurring in a single simulation, P2 represents the unit fault probability, and P3 represents the unit downtime probability. Clearly, the unit fault and downtime probabilities are statistical results given the unit state. Therefore, the training samples for constructing the first surrogate model can be obtained, mapped as: unit state parameters → (P2, P3).

[0080] Training samples for the second proxy model:

[0081] Under the fault probability, the second surrogate model is used to generate maintenance times that are identically distributed to data A. For common probability distribution sampling, the maintenance time can be obtained from a uniform distribution on [0,1] using an inverse function. In this step, the maintenance time is generated by approximating the empirical distribution function of data A as the overall distribution function and fitting the inverse function of the empirical distribution function through the surrogate model.

[0082] The empirical distribution function is estimated as follows:

[0083]

[0084] In the formula express:

[0085]

[0086] According to equations (3) and (4), the empirical distribution function at point t in data A is... This is the number of samples in A less than or equal to t divided by a, where t∈{min(A),min(A)+k,…,min(A)+(l-2)k,max(A)}, and k is the step size. Further, l unbiased and consistent function estimates can be obtained. Therefore, the training sample input and output of surrogate model 2 are: (unit state parameters, → Repair time. The second surrogate model inverse sampling method eliminates the limitation of the inverse function method, which requires an explicit inverse function, and is applicable to sample collection of any form of distribution function or empirical distribution function.

[0087] Training samples for the third-party agent model:

[0088] The third surrogate model under downtime probability is used to generate an array of maintenance time and downtime with the same distribution as data B1 and B2 and similar variable correlation.

[0089] For a random variable (x,y) with an unknown joint distribution function H(x,y), we first need to generate a pair of random variables (u,v) with a Copula function C(u,v) that are uniformly distributed on the interval (0,1). Then, we sample using the inverse function to obtain a random variable (x,y) that is correlated.

[0090] To obtain (u,v) with C(u,v), we need to define the conditional distribution when U = u:

[0091]

[0092] First, generate two independent random values ​​u and w that are uniformly distributed on (0,1) (where w is a conditional distribution c). u The values ​​of (u,v) are determined; then the value of v is obtained through the inverse function of the conditional distribution, i.e. The resulting (u,v) is then a random value with the Copula function C(u,v); finally, it is substituted into the inverse marginal distribution function x = F. -1 (u) and y=G -1 (v) and (x,y) are random variables that satisfy the coupling between the two.

[0093] Based on the generalization ability of neural networks, equation (5) is modified as follows:

[0094] c u (u i ,v)=P(V≤v|U=u i )≈P(V≤v|u i-1 ≤U≤u i+1 (6)

[0095] For data B1 (B1 = {T m1 ,T m2 ,…,T mb}) and B2(B2={T d1 ,T d2 ,…,T db}) Sampling of empirical distribution values ​​yields uniformly distributed data. and Using the elements in U and V as the x-axis and y-axis coordinates respectively, the data distribution image is obtained as follows: Figure 5 As shown, in u i c at the location u (u,v) can be represented as:

[0096]

[0097] Where, N i Indicates [u i-1 ,u i+1 The number of sample points within the interval, N ij This indicates that the value of v for sample points within the above interval is less than v. j The number of samples.

[0098] From this, we can obtain the inverse function of the fitted conditional distribution. The training samples are mapped as: (w,u)→v; and by the inverse functions u→x and v→y, the sample input and output mappings in a single state of the third proxy model are: u→x, (w,u)→y.

[0099] After obtaining the training samples of the surrogate model, the neural network structure needs to be designed. The surrogate model neural network of each surrogate unit is obtained based on the input and output feature dimensions, as shown in Table 3.

[0100] Table 3 Number of input and output neurons in the surrogate model

[0101]

[0102] The number of hidden layer neurons is adjusted based on empirical formulas and a trial-and-error approach. Following general neural network design principles, the transfer function of the hidden layer neurons is the tangent sigmoid function (tan-sigmoid). Since the output has been normalized to [0,1], the transfer function of the output layer neurons can be set as the logarithmic sigmoid function (log-sigmoid). Finally, the Levenberg-Marquardt algorithm is used for model training.

[0103] Step 106: Generate the maintenance time and downtime of the agent unit based on the three types of agent models trained by each unit. The simulation unit obtains the maintenance time and downtime using simulation methods.

[0104] The maintenance time and downtime of a unit are generated using three types of surrogate models trained for each unit. The specific process is as follows: During a single simulation, for a certain surrogate unit, it is first necessary to determine the fault downtime status of the unit. The first surrogate model generates (P2, P3) and simultaneously generates a uniformly random value P on (0, 1). If P ∈ [0, P2), it means that the unit only experienced a fault in this simulation, and the maintenance time is generated by the second surrogate model. If P ∈ [P2, P2+P3), it means that the unit experienced a downtime in this simulation, and the maintenance time and downtime are generated by the third surrogate model. Otherwise, no fault occurred in this simulation, and the maintenance time and downtime are 0.

[0105] In addition, the maintenance time and downtime of the simulation unit are collected through simulation methods.

[0106] Step 108: Determine whether the task to be evaluated is successful based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task to be evaluated.

[0107] Based on the logical relationship of the unit serial structure, the system maintenance time (downtime) is equivalent to the sum of the maintenance times (downtimes) of each unit. Then, given the task redundancy maintenance time (downtime), if the system maintenance time (downtime) is greater than the given task redundancy maintenance time (downtime), the task fails; otherwise, the task succeeds.

[0108] Step 110: Based on the pre-set task time, allowable percentage of maintenance time, allowable percentage of downtime, and number of task evaluations, evaluate the task to be evaluated multiple times to obtain the evaluation result of the success rate of the task to be evaluated.

[0109] In a task with a duration of 700 hours, a maintenance time allowance of 3%, and a downtime allowance of 0.3%, the evaluation accuracy and efficiency of the method were verified by comparing the simulation evaluation results and the mixed rapid evaluation results under 100 sets of random conditions. The accuracy verification data are shown in Table 4.

[0110] Table 4 Numerical accuracy verification

[0111]

[0112] The data shows that the model's predictions are very close to the target. The average error in task success rate is only 0.61 percentage points, with a maximum error of 2 percentage points, which is sufficient to meet engineering requirements. r is the correlation coefficient between the two; the closer r is to 1, the higher the model's accuracy. The table shows r is 0.9980, indicating high accuracy of the surrogate model. It is worth noting that in completing the aforementioned 100 comparative evaluation experiments, the numerical simulation took 5.9 hours, averaging 3.5 minutes per evaluation. In contrast, the hybrid evaluation algorithm only takes 0.69 hours, with each evaluation taking only 25 seconds. This demonstrates that compared to simulation methods, the hybrid evaluation algorithm is better suited to dynamic scenarios with changing parameters. The surrogate model reduces the simulation scale from the entire system to local simulation units, and its efficiency advantage is particularly significant in variable scenarios. This efficiency improvement is crucial for the rapid evaluation of task success rates in complex systems and the development of targeted maintenance solutions.

[0113] In the above-mentioned hybrid evaluation method for task success rate combining simulation and proxy model, this application realizes an approximate description of the complex simulation process that changes with state by constructing a proxy model. It simulates and generates unit maintenance time and downtime for evaluating the system's task success rate. At the same time, while ensuring high accuracy, it reduces the scale of repetitive simulation caused by state changes from the entire system to local units, avoiding a large amount of redundant calculations. It is more suitable for dynamic scenarios and can be evaluated quickly, thus improving the convergence speed of task success rate evaluation.

[0114] In one embodiment, the state change quantity is a discrete quantity of the component lifetime distribution parameters; the distribution parameters include: the shape parameter and scale parameter of the Weibull distribution and the MTBF of the exponential distribution.

[0115] In one embodiment, the process of obtaining the state changes of various components in a complex system includes:

[0116] Based on historical failure data, degradation data, and the initial parameters of a given life distribution model, determine the range of parameter variations within the component's life cycle;

[0117] By combining the task success rate with component parameters based on the range of parameter changes, the interval in which the task success rate changes significantly can be determined.

[0118] By taking a small interval from the interval, the continuous parameter is discretized, and the number of discretized parameters is the change in the state of the component.

[0119] In one embodiment, the complex system is divided into multiple cascaded basic structural units based on the amount of state change, including:

[0120] Components that do not change or change only slightly during their lifespan are classified as surrogate units, while components that undergo significant state changes during their lifespan are classified as simulation units.

[0121] In one embodiment, the states of the proxy unit and the simulation unit are represented by the lifetime distribution parameters of each component in the unit and the encoding of the system operating conditions, including:

[0122] Components of the same type connected in parallel have the same tendency to deteriorate over their life cycle, so they are represented by the same life distribution parameter.

[0123] Different operating conditions are numbered and used as unit state characteristics, and the distribution parameters of a type of component that does not participate in unit operation are set to 0.

[0124] In one embodiment, the three proxy models include a first proxy model, a second proxy model, and a third proxy model; data analysis methods are used to analyze the maintenance time and downtime of the proxy units to obtain training samples for the proxy units, including:

[0125] For any state in the unit state set, sample N times. If the component fails but the unit does not stop, the number of times is a, and the maintenance time data is A; if the component fails and the unit stops, the number of times is b, and the maintenance time data is B1, and the downtime data is B2. The number of times no component failure occurs is Nab. The normal probability P1, the unit failure probability P2, and the unit downtime probability P3 in a single simulation are obtained.

[0126] Based on the normal probability P1 of no fault occurring in a single simulation, the unit fault probability P2, and the unit downtime probability, the training samples of the first surrogate model are constructed and mapped as: unit state parameters → (P2, P3).

[0127] The empirical distribution function of the maintenance time data A is used as an approximation of the overall distribution function to construct the training samples of the second surrogate model, which are mapped as: (unit state parameters, empirical distribution function values) → maintenance time.

[0128] In one embodiment, two independent random values ​​u and w are generated that are uniformly distributed on (0,1); where w is a conditional distribution c. u The values ​​of (u,v) are determined; the value of v is obtained through the inverse function of the conditional distribution, i.e. (u,v) is a random value with a Copula function C(u,v);

[0129] Substituting into the inverse function of the marginal distribution function x = F -1 (u) and y=G -1 (v) and (x,y) are random variables that satisfy the coupling between the two.

[0130] Empirical distribution values ​​were sampled from maintenance time data B1 and downtime data B2 to obtain uniformly distributed data U and V.

[0131] Using the elements in U and V as the x-axis and y-axis coordinates respectively, the data distribution is obtained;

[0132] The inverse function of the fitted conditional distribution is obtained based on the data distribution. The training samples are mapped as (w,u)→v; then, the training samples of the third proxy model under a single state are constructed using the inverse functions u→x and v→y, and mapped as u→x and (w,u)→y.

[0133] In one embodiment, the first surrogate model takes the unit state as input and outputs the failure probability and the downtime probability; the second surrogate model takes the unit state as input and a random number uniformly distributed on (0,1) as input and outputs the maintenance time of the unit under the failure probability; the third surrogate model takes the unit state as input and two independent random numbers uniformly distributed on (0,1) as input and outputs the correlated maintenance time and downtime of the unit under the downtime probability.

[0134] In one embodiment, the maintenance time and downtime of the proxy unit are generated based on the three types of proxy models trained by each unit, including:

[0135] During a single simulation, for each surrogate unit, the first step is to determine the unit's fault and downtime status. The first surrogate model generates (P2, P3) and simultaneously generates a uniformly random value P on (0, 1). If P ∈ [0, P2), it indicates that the unit only experienced a fault in this simulation, and the second surrogate model generates the repair time. If P ∈ [P2, P2+P3), it indicates that the unit experienced a downtime in this simulation, and the third surrogate model generates the repair time and downtime. Otherwise, no fault occurred in this simulation, and the repair time and downtime are 0.

[0136] In one embodiment, determining whether the task to be evaluated is successful based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task to be evaluated includes:

[0137] Based on the logical relationship of the unit serial structure, the system maintenance time or downtime is equivalent to the sum of the maintenance time or downtime of each unit. Then, based on the task redundancy maintenance time or downtime of the task to be evaluated, if the system maintenance time or downtime is greater than the task redundancy maintenance time or downtime of the task to be evaluated, the task fails; otherwise, the task succeeds.

[0138] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0139] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a hybrid evaluation method for task success rate that combines simulation and proxy models. The display screen can be an LCD screen or an e-ink display screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0140] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0141] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described above.

[0142] In one embodiment, a computer storage medium is provided that stores a computer program, which, when executed by a processor, implements the steps of the method described above.

[0143] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0144] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0145] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A hybrid evaluation method for task success rate combining simulation and proxy models, characterized in that, The method includes: Obtain the state changes of each component in a complex system; discretize the component lifetime distribution parameters, and the number of discretized parameters is the state change of the component; The complex system is divided into multiple cascaded basic structural units based on the state change quantities; each basic structural unit includes a simulation unit and multiple proxy units. The agent unit is simulated and sampled to obtain the maintenance time and downtime of the agent unit under each state; Data analysis methods are used to analyze the maintenance time and downtime of the agent unit to obtain training samples of the agent unit. Based on the training samples, three types of agent models are designed for the agent unit. The three types of agent models include a first agent model, a second agent model, and a third agent model. Based on the three types of proxy models trained by each unit, the maintenance time and downtime of the proxy unit are generated, and the simulation unit uses simulation methods to obtain the maintenance time and downtime. The success of the task to be evaluated is determined based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task to be evaluated. The task to be evaluated is evaluated multiple times based on the pre-set task time, allowable maintenance time percentage, allowable downtime percentage, and number of task evaluations to obtain the evaluation result of the success rate of the task to be evaluated. The first surrogate model takes the unit state as input and outputs the failure probability and downtime probability. The second surrogate model takes the unit state as input and a random number uniformly distributed on (0,1) as input and outputs the maintenance time of the unit under the failure probability. The third surrogate model takes the unit state as input and two independent random numbers uniformly distributed on (0,1) as input and outputs the correlated maintenance time and downtime of the unit under the downtime probability.

2. The method according to claim 1, characterized in that, The component lifetime distribution parameters include: shape parameters and scale parameters of the Weibull distribution, and MTBF of the exponential distribution.

3. The method according to claim 1, characterized in that, The process of obtaining the state changes of various components in a complex system includes: Based on historical failure data, degradation data, and the initial parameters of a given life distribution model, determine the range of life distribution parameter variation within the component's life cycle. By combining the mission success rate with the component lifespan distribution parameters based on the range of variation of the lifespan distribution parameters, the interval in which the mission success rate changes significantly can be determined. By taking a small interval from the interval, the continuous lifespan distribution parameters are discretized, and the number of discretizations is the component state change.

4. The method according to any one of claims 1 to 3, characterized in that, Based on the state change quantities, the complex system is divided into multiple series-connected basic structural units, including: Components with small state changes during their lifespan are classified as surrogate units, while components with large state changes during their lifespan are classified as simulation units.

5. The method according to claim 4, characterized in that, The states of the proxy unit and simulation unit are represented by the lifetime distribution parameters of each component in the unit and the encoding of the system operating conditions, including: Components of the same type connected in parallel have the same tendency to deteriorate over their life cycle, so they are represented by the same life distribution parameter. Different operating conditions are numbered and used as unit state characteristics, and the life distribution parameter of a type of component that does not participate in unit operation is set to 0.

6. The method according to claim 5, characterized in that, Data analysis methods were used to analyze the maintenance time and downtime of the agent unit, resulting in training samples for the agent unit, including: Perform on any state in the unit state set N The number of times the unit remained operational due to component failure during the sampling was [number]. a Repair time data A The number of component failures and unit downtimes is [number missing]. b Repair time data B 1. Downtime data B 2. The number of times no component failure occurred was N - a - b This yields the normal probability that no fault occurred in a single simulation. Unit failure probability and unit downtime probability ; Based on the normal probability that no fault occurred in a single simulation Unit failure probability and unit downtime probability The training samples for constructing the first agent model are mapped as: unit state parameters. ; Repair time data A The empirical distribution function is used as an approximation of the overall distribution function to construct the training samples of the second surrogate model, which are mapped as: (unit state parameters, empirical distribution function values). Repair time.

7. The method according to claim 6, characterized in that, The method further includes: Generate two independent random values ​​uniformly distributed on (0,1). ;in w For conditional distribution The value of is obtained through the inverse function of the conditional distribution. v Value, that is The For having Copula function Random values; The Substitute the inverse function of the marginal distribution function and , ( x , y That is, a random variable that satisfies the coupling between the two; Repair time data B 1 and downtime data B 2. Conduct By sampling the distribution values, uniformly distributed data can be obtained. U and V ; by U and V The elements are respectively used as x shaft and y The axis coordinates are used to obtain the data distribution. The inverse function of the fitted conditional distribution is obtained based on the data distribution. The training samples are mapped as follows: Then by the inverse function The training samples of the third agent model in a single state are constructed and mapped as follows: 。 8. The method according to claim 1, characterized in that, Based on the three types of proxy models trained in each unit, the maintenance time and downtime of the proxy unit are generated, including: During a single simulation, for each proxy unit, the first step is to determine the unit's fault shutdown status, which is generated by the first proxy model. Simultaneously generate uniform random values ​​on (0,1). P ;like If this indicates that only a fault occurred in this unit during the simulation, then the repair time will be generated by the second proxy model; if If the simulation shows that the unit has stopped working, the maintenance time and downtime will be generated by the third proxy model. Otherwise, if no fault occurs in this simulation, the maintenance time and downtime will be 0.

9. The method according to claim 8, characterized in that, The success of the task under evaluation is determined based on the logical relationship of the unit serial structure, the maintenance time or downtime of all units, and the task redundancy maintenance time or downtime of the task under evaluation. This includes: Based on the logical relationship of the unit serial structure, the total system maintenance time is equivalent to the sum of the maintenance times of each unit, and the total system downtime is equivalent to the sum of the downtimes of each unit. Then, the total system maintenance time is compared with the task redundancy maintenance time of the task to be evaluated, and the total system downtime is compared with the task redundancy downtime of the task to be evaluated. If the total system maintenance time is greater than the task redundancy maintenance time, or the total system downtime is greater than the task redundancy downtime, the task fails; otherwise, the task succeeds.

Images