BUCK circuit state evaluation and prediction method
By combining RF model and BP model and using the Hippo optimization algorithm for weighted combination, the problem of insufficient fault prediction accuracy of BUCK circuit is solved, and higher fault prediction reliability and accuracy are achieved.
Patent Information
- Application Number
- CN202411807119.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-10
- Publication Date
- 2025-04-15
AI Technical Summary
The existing BUCK circuit fault prediction methods have insufficient accuracy, especially the overall circuit performance fault prediction has been studied, and the accuracy of the existing prediction methods needs to be further improved.
Combining the RF model and BP model, combining the advantages of the two models, the prediction results of the two models are weighted and combined through the hippo optimization algorithm to improve the accuracy of fault prediction.
Through this method, the reliability and accuracy of BUCK circuit fault prediction is improved, and it is suitable for real-time systems, and can more effectively evaluate and predict circuit state.
Smart Images
Figure CN119830831A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of BUCK circuits, and in particular to a method for evaluating and predicting a BUCK circuit state. Background Art
[0002] There are many state assessment methods for power electronic circuits, such as feature extraction, health indicators, and multi-sensor data fusion.
[0003] Feature extraction is used to extract output signals and describe the state of the circuit using statistical features (such as mean, variance, skewness, kurtosis), frequency domain features (such as power spectral density), time-frequency features (such as wavelet packet energy), etc.; health indicators define and calculate the health indicators of the circuit to quantify the state and performance of the circuit; multi-sensor data fusion combines data from multiple sensors to obtain more comprehensive circuit status information.
[0004] At the same time, various prediction algorithms have also emerged, such as time series, neural network, and support vector regression prediction, which are the most widely used. The prediction accuracy of these three methods has high requirements on the amount of data and the range of variation. The accuracy of a single prediction algorithm can no longer meet the needs of circuit fault prediction with the development of science and technology; at the same time, most of the current predictions of circuit faults are fault predictions for key components in the circuit, such as MOSFET, electrolytic capacitors, inductors, etc. The fault characteristic parameters are basically the degradation values of the selected components, and there is less research on the overall circuit performance fault prediction.
[0005] In addition, existing prediction methods, such as using particle swarm and quadratic convex programming optimization algorithms for prediction, still need to improve their prediction accuracy. Summary of the invention
[0006] In view of the shortcomings of the existing methods, the present invention combines the RF model and the BP model, integrates the advantages of the two models, and improves the reliability of fault prediction; the prediction results of the two models are weighted combined through the Hippo optimization algorithm to further improve the accuracy of the prediction.
[0007] The technical solution adopted by the present invention is: a BUCK circuit state evaluation and prediction method comprises the following steps:
[0008] Step 1: Collect the degradation parameter values and corresponding voltage values of several components in the BUCK circuit within a certain time range;
[0009] As a preferred implementation mode of the present invention, the degradation parameter value=initial value of the component parameter*(1-degradation rate)^a, wherein a is the number of degradation times.
[0010] Step 2: Perform PCA dimensionality reduction on the degradation parameter matrix;
[0011] As a preferred embodiment of the present invention, the PCA method comprises:
[0012] Step 21, calculating the mean and standard deviation of each degradation parameter;
[0013] Step 22: Generate new degradation parameters using the mean, standard deviation and degradation parameters;
[0014] Step 23, constructing a covariance matrix according to the new degradation parameters;
[0015] Step 24: Use the covariance matrix to construct a characteristic equation, solve the eigenvalue and eigenvector of the characteristic equation, and use the eigenvector corresponding to the maximum eigenvalue as a reference to obtain the comprehensive degradation parameter matrix after dimensionality reduction.
[0016] Step 3: Calculate the parameter time series of the circuit state using the DTW distance method;
[0017] As a preferred embodiment of the present invention, step three specifically includes:
[0018] Step 31: Use the eigenvector R and the voltage vector V to construct a filling accumulation matrix D, the formula is:
[0019]
[0020] Where, R0=r1, V0=v1; D[0,0]=dis(R0,V0); m, n≥1;
[0021] Step 32: In D, starting from the first data point in the upper right corner, search for the alignment path to the left, down, and lower left nodes, and use the smaller data point as the next node; then, output the result, and so on, until all data points are found.
[0022] Step 4: Use Chebyshev inequality to divide the parameter time series of the circuit state into thresholds and construct a circuit health state assessment standard;
[0023] As a preferred implementation of the present invention, the thresholds include: a health threshold, a warning threshold and a fault threshold.
[0024] As a preferred embodiment of the present invention, the threshold is the average value μ and standard deviation σ of the parameter time series of the calculated circuit state, and μ-1.5σ is defined as the health threshold; μ-σ is the warning threshold; μ-0.5σ is the fault threshold.
[0025] Step 5: Use the parameter time series of the circuit state to construct a data set, input it into the RF model and the BP model respectively, and use the Hippo optimization algorithm to weight the predicted value of the RF-BP model to obtain the optimal weight with the best fitness;
[0026] As a preferred implementation of the present invention, step five specifically includes:
[0027] Construct RF-BP model and output predicted value f t , the formula is:
[0028] f t =λ1f 1t +λ2f 2t λ1,λ2∈[-2,2] (8)
[0029] Assume the error sequence is:
[0030] e 1t =y t -f 1t ,e 2t =y t -f 2t (9)
[0031] The weights are determined according to the minimum root mean square principle of the combined prediction error. The combined prediction error is:
[0032] e t =y t -f t =y t -λ1f 1t -λ2f 2t (10)
[0033] The root mean square of the combined prediction is the fitness J, then:
[0034]
[0035] Among them, f 1t (t=1,2,…,n) and f 2t (t=1,2,…,n) are the predicted values of RF model and BP model, y t (t=1,2,…,n) are measured values, λ1, λ2 are weight coefficients of RF model and BP model.
[0036] As a preferred embodiment of the present invention, the Hippo optimization algorithm optimizes the weight coefficient including:
[0037] Set the number of search agents and initialize the agent positions:
[0038] Y k =ld+(ub-lb)×rand (12)
[0039] Among them, Y k represents all the columns of the kth row in the proxy matrix; ld and ub are the lower and upper bounds of the proxy value, and rand is a randomly generated two-dimensional array;
[0040] The agent's position is randomly selected and updated using the formula:
[0041]
[0042]
[0043] pd=lb+(ub-lb)×rand
[0044] Y k+1 =pd+rand×(Y k -pd) (15)
[0045] In the formula, is the optimal weight, pd generates a random "predator" position;
[0046] Ensure that the agent position is within the boundary, as shown in formula (16):
[0047] Y k =max(min(Y k ,ub),lb); (16)
[0048] Thus the optimal weight is obtained.
[0049] As a preferred embodiment of the present invention, a BUCK circuit state evaluation and prediction system includes: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement a BUCK circuit state evaluation and prediction method.
[0050] As a preferred embodiment of the present invention, a computer-readable medium storing a computer program code is characterized in that the computer program code implements a BUCK circuit state evaluation and prediction method when executed by a processor.
[0051] Beneficial effects of the present invention:
[0052] 1. The component degradation value extraction of the present invention is simple, easy to understand and integrate to calculate the comprehensive degradation value, and is suitable for real-time systems;
[0053] 2. Calculate the DTW distance based on the voltage parameters and the comprehensive degradation values of the circuit components. The calculation is simple and also applicable to real-time systems.
[0054] 3. The RF model is combined with the BP model, and the weight parameters of the two models are optimized using the Hippo algorithm, which significantly improves the prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 This is a flowchart of the BUCK circuit state evaluation and prediction method of the present invention;
[0056] Figure 2Optimize algorithm flow chart for Hippo;
[0057] Figure 3 This is the schematic diagram of the BUCK equivalent circuit;
[0058] Figure 4 It is the DTW distance graph of the output voltage and the comprehensive degradation value over time;
[0059] Figure 5 It is the status indicator of the BUCK circuit;
[0060] Figure 6 This is a graph comparing the weighted combination model using the Hippo optimization algorithm with a single prediction model. DETAILED DESCRIPTION
[0061] The present invention is further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore it only shows the components related to the present invention.
[0062] like Figure 1 As shown, a BUCK circuit state evaluation and prediction method includes the following steps:
[0063] Step 1: Collect the degradation parameter values and corresponding voltage values of several components in the BUCK circuit within a certain time range;
[0064] like Figure 3 For the BUCK circuit diagram, build a similink simulated BUCK circuit, simulate the components that affect the output of the analog circuit, including: MOSFET, inductor, capacitor and resistor, and perform equivalent processing; MOSFET is equivalent to a frequency switch and internal resistance R ON , the inductor is equivalent to the ideal inductor L and the resistor R L The capacitor is equivalent to an ideal capacitor C and a resistor ESR, and the diode D is equivalent to a frequency switch with the same frequency and opposite phase as the MOSFET and an internal resistor R F .
[0065] The components that affect the state of the BUCK circuit are mainly resistors R ON , resistor R F , inductance L, capacitance C, resistance R L There are 6 component variables, including resistor ESR; the degradation parameter values and corresponding voltage values of the 6 components at the same time interval are measured;
[0066] The degradation parameter value is the initial value of the component parameter * (1 ± degradation rate) ^ a, where a is the number of degradations, where the resistance is + when it is degraded, and the capacitance and inductance are - when they are degraded; for example, the degradation rate is 0.00004; when t = 0, the resistance R ONThe initial value is 0.025Ω; when t=1, the first degradation parameter value is 0.025*(1+0.00004), the second degradation parameter value is 0.025*(1+0.00004)^2, and so on.
[0067] Assume that time is T = {T1, T2, T3, ..., T j}, degradation parameter matrix X = {X ji}, the corresponding voltage matrix V = [v1,v2,v3,...v j ]; where i is a different component, j is the number of samples, j=1,2,3...n, i=1,2,3...6, and the degradation parameter matrix is the set of degradation parameters.
[0068] Step 2: Perform PCA dimensionality reduction on the degradation parameter matrix; that is, use the PCA method to fuse the component parameters of the degradation parameter matrix X to obtain the comprehensive degradation value of the components at each moment [r1, r2, r3, ...r j ];
[0069] The PCA method includes:
[0070] 1. Calculate the mean and standard deviation of each degradation parameter using the formula:
[0071]
[0072] The degradation parameters of each component are standardized using the Z-score normalization method; ensuring that the degradation parameter measurements of each component are converted to the same scale, that is, the mean is 0 and the standard deviation is 1.
[0073] 2. Generate new degradation parameter Z ji , the formula is:
[0074]
[0075] 3. By Z ji Construct the covariance matrix, the formula is:
[0076]
[0077] Among them, Z T is the transpose of the matrix Z, and j is the number of samples.
[0078] 4. Construct the characteristic equation, the formula is:
[0079]
[0080] Solve the eigenvalues and eigenvectors of the characteristic equation to find the maximum eigenvalue λ m The corresponding eigenvector Based on the benchmark, parameter fusion is performed to obtain the comprehensive degradation parameter matrix: Reduce the 6 variables of j samples at a certain moment to 1 variable.
[0081] Step 3: Use the DTW distance method to calculate the parameter time series of the circuit state; that is, use the DTW distance method to convert the comprehensive degradation parameter matrix Bind one-to-one with the output voltage;
[0082] The DTW distance method includes:
[0083] Step 31: The characteristic vector R and the output voltage vector V are essentially a time series data, that is, R = [r1, r2, r3, ... r j ],V=[v1,v2,v3,...v j ]; fill the cumulative matrix D with R from left to right as the x-axis data and V from bottom to top as the y-axis data; give the rule of dis(x,y)=|xy|, and the data filling rule is as follows:
[0084]
[0085] Among them, R0=r1, V0=v1; D[0,0]=dis(R0,V0); m, n≥1.
[0086] Step 32: In matrix D, starting from the first data point in the upper right corner, look for the registration path to the left, down, and lower left nodes, and use the smaller data point as the next node; then, output the result, and so on, until all data points are found; the DTW distance of the two time series of R and V is as follows:
[0087] DTW(R,V)=D p (m,n) (7)
[0088] In the formula, the subscript p represents a series of paths corresponding to the minimum distance in the sequence.
[0089] From this, we can get the parameter time series representing the circuit state D'=[D'1,D'2,D'3,.......,D' j ].
[0090] Step 4: Use Chebyshev inequality in probability theory to divide D', derive health threshold, warning threshold and fault threshold, and construct circuit health status assessment criteria;
[0091] Calculate the mean value μ and standard deviation σ of D', define [D'1, μ-1.5σ] as the healthy interval, μ-1.5σ as the healthy threshold; [μ-1.5σ, μ-σ] as the warning interval, μ-σ as the warning threshold, and [μ-σ, μ-0.5σ] as the fault interval; the fault interval means that the circuit state is very bad and may be scrapped at any time, and μ-0.5σ is the fault threshold.
[0092] Step 5: Divide the parameter time series D' data of the circuit state into test data and training data, input the test set data into RF (random forest) and BP network respectively, predict the DTW distance of the circuit state parameter at the next moment, and obtain two sets of prediction data f 1t (t=1,2,...n) and f 2t (t=1,2,...n);
[0093] Based on the predicted values and true values of the RF and BP network models, the Hippo optimization algorithm is used to perform a weighted combination of the predicted values of the two models; first find the optimal fitness J, and then use J to obtain the optimal weight, so that the relative error between the combined predicted value and the true value is as small as possible.
[0094] The solution to the optimal weight includes:
[0095] Let f 1t (t=1,2,…,n) and f 2t (t=1,2,…,n) are the predicted values of RF model and BP model, y t (t=1,2,…,n) is the measured value, f t (t=1,2,…,n) is the prediction of RF-BP model, and the RF-BP combined model is:
[0096] f t =λ1f 1t +λ2f 2t λ1,λ2∈[-2,2] (8)
[0097] Among them, λ1 and λ2 are the weight coefficients of the RF model and the BP model.
[0098] By determining the optimal weight coefficients λ1 and λ2, the optimal RF-BP model is achieved;
[0099] like Figure 2 , let the error sequence be:
[0100] e 1t =y t -f 1t ,e 2t =y t -f 2t (9)
[0101] The weights are determined according to the minimum root mean square principle of the combined prediction error. The combined prediction error is:
[0102] e t =y t -f t =y t -λ1f 1t -λ2f 2t (10)
[0103] The root mean square of the combined prediction is the fitness J, then:
[0104]
[0105] The optimal weights λ1 and λ2 are found through the Hippo optimization algorithm. Let J be the fitness function of the optimization problem, and the goal is to obtain the minimum root mean square error.
[0106] Set the number of search agents to 50, that is, 50 1×2 matrices, each agent represents a hippopotamus; each hippopotamus is a candidate solution, and when the minimum root mean square error is obtained, the optimal solution is found. The optimal solution is the position of the agent, and the agent at this position is used to calculate the fitness J; 50 hippos form a population, that is, a 50×2 matrix Y; the algorithm traverses each agent in turn, from the first to the 50th, and the agent position is initialized as follows:
[0107] Y k =ld+(ub-lb)×rand (12)
[0108] Among them, Y k represents all the columns of the k-th row in the agent matrix, that is, the current position of the k-th agent, ld = [-2, -2] and ub = [2, 2] are the lower and upper bounds of the agent value, and rand is a two-dimensional array randomly generated in the range of [0, 1]. During the iteration of the algorithm, the position of the agent is randomly selected and updated according to formulas (13)-(15):
[0109]
[0110]
[0111] pd=lb+(ub-lb)×rand
[0112] Y k+1 =pd+rand×(Y k -pd) (15)
[0113] In the formula, is the optimal weight, pd generates a random "predator" position, and the agent will explore new positions near its current position, simulating the behavior of hippos looking for new food sources in their territory. At the same time, it is necessary to ensure that the agent position is within the boundary, as shown in formula (16):
[0114] Y k =max(min(Y k ,ub),lb); (16)
[0115] The global optimal solution, that is, the best proxy position, is used as the optimal weights for weighting the predicted values of RF and BP at the next moment, so as to improve the reliability and accuracy of BUCK circuit fault prediction.
[0116] The predicted value is compared with the threshold value to evaluate the state of the BUCK circuit at the next moment, so as to achieve the purpose of timely warning and maintenance.
[0117] Experimental process:
[0118] In order to demonstrate the evaluation effect of DTW distance combined with Chebyshev inequality and the prediction performance of the RF-BP combined prediction model weighted by the Hippo optimization algorithm, an example is given here; Figure 3 is the equivalent schematic diagram of the BUCK circuit. The experimental object is a 20V to 10V BUCK circuit. In order to better simulate the actual circuit situation, the input voltage is set to 21.42V. The performance of the BUCK is closely related to the degradation of components. In order to verify the reliability of the method of the present invention, a degradation rate that conforms to the actual situation is set for each component to simulate the degradation of the component, so as to achieve the purpose of verifying the method of the present invention. The initial values of the components and the degradation rate settings are shown in Table 1.
[0119] Table 1 Component initial value and degradation rate data
[0120] Component name Initial Value Degradation rate Inductance L 130μH 0.005000 Capacitor C 250μF 0.009000 <![CDATA[R ON ]]> 0.025Ω 0.000040 <![CDATA[R F ]]> 0.03Ω 0.000055 <![CDATA[R L ]]> 0.72Ω 0.000990 ESR 0.11Ω 0.000080
[0121] In the simulation environment, 100 parameter degradation simulation experiments were conducted on the BUKC circuit, that is, on the basis of setting a reasonable parameter degradation rate of components, 100 degradations were performed, and the output voltage of each degradation was measured;
[0122] PCA fusion is performed on 6 groups of data of 6 components, each group has 100 data, and a total of 600 data are obtained to obtain the comprehensive degradation values of the components at 100 time points, which correspond to 100 output voltage values. The comprehensive degradation values of the components and the 100 output voltage values are bound one by one using the DTW distance method; Figure 4 As shown in Figure 2, the DTW distance changes over time and shows an upward trend.
[0123] According to Chebyshev's inequality, the health threshold is calculated to be 0.1751×104, and the corresponding output voltage is about 9.8V; the warning threshold is 0.3262×104, and the corresponding output voltage is about 8.6V; the fault threshold is 0.4773×104, and the corresponding output voltage is about 7.1V; Figure 5 It can be seen from the status indicator that as the DTW distance increases, the output voltage also decreases significantly, which is in line with the actual working conditions.
[0124] 85 time point DTW distance data were selected, 80% of which were used as training sets and 20 as test sets. The data were input into the random forest and BP neural network prediction model to predict the data from 86 to 95 time points, and the Hippo optimization algorithm was used for weighted combination. The root mean square error was used as the fitness to obtain the optimal weights λ1 and λ2, so that the combined prediction value was closest to the true value. Similarly, 95 data were selected and input into the random forest and BP neural network prediction model to predict the last 5 data. The prediction values of the two models were weighted combined using the λ1 and λ2 obtained above to obtain the final prediction value. Figure 6 It can be seen that the weighted prediction value based on the Hippo optimization algorithm is closer to the true value, and the prediction performance is significantly improved compared to the single model.
[0125] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.
Claims
1. A method for evaluating and predicting a BUCK circuit state, characterized in that: The following steps are involved: Step 1: Collect the degradation parameter values and corresponding voltage values of several components in the BUCK circuit within a certain time range; Step 2: Perform PCA dimensionality reduction on the degradation parameter matrix; Step 3: Calculate the parameter time series of the circuit state using the DTW distance method; Step 4: Use Chebyshev inequality to divide the parameter time series of the circuit state into thresholds and construct a circuit health state assessment standard; Step 5: Use the parameter time series of the circuit state to construct a data set, input it into the RF model and BP model respectively, use the Hippo optimization algorithm to weight the predicted value of the RF-BP model, and use the best fitness to obtain the optimal weight.
2. The BUCK circuit state evaluation and prediction method according to claim 1, characterized in that: Step three specifically includes: Step 31: Use the eigenvector R and the voltage vector V to construct a filling accumulation matrix D, the formula is: Where, R0=r1, V0=v1; D[0,0]=dis(R0,V0); m, n≥1; Step 32: In D, starting from the first data point in the upper right corner, search for the alignment path to the left, down, and lower left nodes, and use the smaller data point as the next node; then, output the result, and so on, until all data points are found.
3. The BUCK circuit state evaluation and prediction method according to claim 1, characterized in that: Step 5 specifically includes: Construct RF-BP model and output the predicted value f t , the formula is: f t =λ1f 1t +λ2f 2t λ1,λ2∈[-2,2] (8) Assume the error sequence is: e 1t =y t -f 1t ,e 2t =y t -f 2t (9) The weights are determined according to the minimum root mean square principle of the combined prediction error. The combined prediction error is: e t =y t -f t =y t -λ1f 1t -λ2f 2t (10) The root mean square of the combined prediction is the fitness J, then: Among them, f 1t (t=1,2,…,n) and f 2t (t=1,2,…,n) are the predicted values of RF model and BP model, y t (t=1,2,…,n) are measured values, λ1, λ2 are weight coefficients of RF model and BP model.
4. The BUCK circuit state evaluation and prediction method according to claim 3, characterized in that: The Hippo optimization algorithm optimizes the weight coefficients by: Set the number of search agents and initialize the agent positions: Y k =ld+(ub-lb)×rand (12) Among them, Y k represents all the columns of the kth row in the proxy matrix; ld and ub are the lower and upper bounds of the proxy value, and rand is a randomly generated two-dimensional array; The agent's position is randomly selected and updated using the formula: AND k+1 =And k +rand×(Best λ1 -AND k ) T (13) AND k+1 =And k +rand×(Best λ2 -AND k ) T (14) pd=lb+(ub-lb)×rand Y k+1 =pd+rand×(Y k -pd) (15) In the formula, Best λ1 ,Best λ2 is the optimal weight, pd generates a random "predator" position; Ensure that the agent position is within the boundary, as shown in formula (16): Y k =max(min(Y k ,ub),lb); (16) Thus the optimal weight is obtained.
5. The BUCK circuit state evaluation and prediction method according to claim 1, characterized in that: The PCA method includes: Step 21, calculating the mean and standard deviation of each degradation parameter; Step 22: Generate new degradation parameters using the mean, standard deviation and degradation parameters; Step 23, constructing a covariance matrix according to the new degradation parameters; Step 24: Use the covariance matrix to construct a characteristic equation, solve the eigenvalue and eigenvector of the characteristic equation, and use the eigenvector corresponding to the maximum eigenvalue as a reference to obtain the comprehensive degradation parameter matrix after dimensionality reduction.
6. The BUCK circuit state evaluation and prediction method according to claim 1, characterized in that: Degradation parameter value = initial value of component parameter * (1 ± degradation rate) ^ a, where a is the number of degradation times.
7. The BUCK circuit state evaluation and prediction method according to claim 1, characterized in that: The thresholds include: Health threshold, warning threshold, and failure threshold.
8. The BUCK circuit state evaluation and prediction method according to claim 1, characterized in that: The thresholds include: The threshold is the average value μ and standard deviation σ of the parameter time series of the calculated circuit state. μ-1.5σ is defined as the health threshold; μ-σ is the warning threshold; μ-0.5σ is the fault threshold.
9. BUCK circuit status assessment and prediction system, characterized in that: include: a memory for storing instructions executable by a processor; A processor, configured to execute instructions to implement the BUCK circuit state evaluation and prediction method according to any one of claims 1 to 8.
10. A computer readable medium storing computer program code, characterized in that: When the computer program code is executed by a processor, the method for evaluating and predicting the state of a BUCK circuit as claimed in any one of claims 1 to 8 is implemented.