Power supply state evaluation method in multi-mode working state

Abstract

The invention discloses a power supply state evaluation method in a multi-mode working state, and belongs to the technical field of complex system power supply management and reliability evaluation. The invention provides a system dynamic reliability evaluation and fault diagnosis method based on a classifier and a fuzzy dynamic Bayesian network in order to solve the problems of complexity and concurrent multi-fault probability increase in a multi-mode working mechanism. System reliability evaluation is realized through a fuzzy dynamic Bayesian network, concurrent fault positions are accurately positioned in combination with a fault diagnosis algorithm, and the system is assisted to isolate faults or start standby components, so that the system stability is enhanced and the mode switching process is optimized. Failure mechanism analysis is further carried out for main fault components, the residual life prediction model is optimized in combination with finite incremental learning and online data, and prediction precision and real-time performance are improved. Through cooperative control of software and hardware, the system realizes multi-mode power management and power consumption optimization, provides powerful support for reliability improvement and energy efficiency optimization of a complex measurement and control system, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the fields of power management and reliability assessment of complex systems, particularly to the dynamic reliability assessment and fault diagnosis technology of power supplies in multi-mode operating states, and belongs to the technical category of power management and optimization control. The present invention studies a dynamic reliability assessment method based on a classifier and a fuzzy algorithm, which can realize the real-time monitoring, fault diagnosis, and remaining life prediction of the power supply status in a multi-mode measurement and control system, and is widely applied to the power management, reliability improvement, and power consumption optimization scenarios of complex measurement and control systems. Background Art

[0002] 1. Analysis of the Failure Mechanism of Reliability Assessment Power Supplies

[0003] The components that constitute and cause power supply faults in a multi-mode power management system mainly include electrolytic capacitors, switching transistors, power diodes, and inductors. First, the physical structures of the various fault components of the DC-DC switching power supply in the multi-mode measurement, launch, and control system are analyzed and studied. Then, focusing on the influence of the stress conditions of the multi-mode measurement, launch, and control system on the degradation trends of the various components, the degradation characteristics and failure characterization parameters of the various components of the DC-DC switching power supply in the multi-mode measurement, launch, and control system are clarified. Finally, a feasibility analysis of life prediction is carried out for the DC-DC switching power supply based on its failure mechanism, and the overall failure characterization parameter of the DC-DC switching power supply is selected in combination with the actual engineering application of the multi-mode measurement, launch, and control system.

[0004] (1) Failure Mechanism of Aluminum Electrolytic Capacitors

[0005] Electrolytic capacitors, as energy storage and signal filtering components in the DC-DC switching power supply of a multi-mode measurement, launch, and control system, mainly include aluminum electrolytic capacitors and tantalum electrolytic capacitors according to their principles. Among them, the anode material of the tantalum electrolytic capacitor is metallic tantalum, and the working medium is a tantalum pentoxide (Ta2O2) film. In addition, the tantalum electrolytic capacitor can work stably under higher working electric field strength conditions and is easy to obtain a large capacitance, and its related characteristics ensure its advantages in applications such as high energy and power supply filtering. Aluminum electrolytic capacitors, with outstanding charge storage performance, strong use tolerance, and good economy, have become indispensable components in the application of electronic circuits such as the DC circuit of a multi-mode measurement, launch, and control system. Compared with tantalum electrolytic capacitors, although aluminum electrolytic capacitors are more widely used in the DC-DC switching power supply of a multi-mode measurement, launch, and control system, due to their high failure rate and susceptibility to temperature conditions, the following focuses on the analysis and research of their failure mechanism and the structure and equivalent circuit of aluminum electrolytic capacitors.

[0006] The main structure of the aluminum electrolytic capacitor includes two major parts: the positive and negative electrodes. The specific structure and its equivalent circuit are as shown in the appendix Figure 2 shown, where R ESR is the equivalent series resistance of the capacitor, and L ESis the equivalent inductance at the positive and negative electrode lead-out wires and the connection point, and C is the capacitance. The equivalent series resistance R of the capacitor ESR and the capacitor C can be expressed as:

[0007]

[0008] wherein, R0 is the equivalent series resistance of the two-pole metal foil and the lead-out wire end, R1 is the equivalent series resistance between the electrolyte and the separator paper; R2 is the electrolyte consumption resistance; C1 is the capacitance between the two-pole metals; C2 is the capacitance related to the loss of the electrolytic liquid; ω = 2πf, ω is the ripple current angular frequency, where f is the frequency. When the frequency increases, R ESR shows a downward trend. R1 is most affected by the change in the properties of the electrolyte. When the temperature of the electrolyte rises, the conductivity of the liquid also increases accordingly, which can be expressed by the following formula:

[0009]

[0010] T b = 27°C is the standard temperature value; E is the capacitance-temperature correlation coefficient; T e is the central temperature; R b is the resistance reference value under standard temperature conditions. From the above analysis, it can be deduced that when the temperature in the electrolytic liquid rises, R1 will show a decreasing trend. In addition, the equivalent series resistance R of the capacitor ESR has the following relationship with the volume of the electrolyte:

[0011]

[0012] where: V0 is the volume of the electrolyte of the aluminum electrolytic capacitor under the initial state conditions; R ESR0 is the equivalent series resistance value of the aluminum electrolytic capacitor under the initial state conditions; V is the volume of the electrolyte corresponding to the detection state conditions.

[0013] The performance of aluminum electrolytic capacitors mainly deteriorates and fails due to temperature stress. Temperature is generated on the one hand by factors such as pulse current, ripple current, and overvoltage, and on the other hand by environmental factors. Since the resistance value of the aluminum electrolytic capacitor R ESR is relatively large, when the pulse current and ripple current appear and act on the electrolytic capacitor for a long time, it will cause power loss and heat generation of the electrolytic capacitor; when the current exceeds the rated value, it will cause the element of the aluminum electrolytic capacitor to overheat and accelerate the failure of the aluminum electrolytic capacitor. If the input voltage of the DC-DC switching power supply in the multi-mode launch control system exceeds the maximum withstand voltage, the overvoltage will cause an increase in leakage current, which will also cause an increase in the core temperature of the aluminum electrolytic capacitor, accelerating the occurrence of performance degradation and failure. In addition, due to factors such as the increase in environmental temperature or the continuous accumulation of working time, the initial electrolyte temperature will rise with the environmental temperature, thereby promoting an increase in the conductivity of the electrolyte, RESR shows a downward trend; if the ambient temperature continues to rise, it will cause the electrolyte to evaporate, resulting in an increase in the viscosity of the electrolyte and a decrease in the volume V. At this time, the equivalent series resistance R of the capacitor ESR will show an increasing trend, and the capacitance value C will show a decreasing trend. During long-term storage or use, the evaporation of the electrolyte will also occur. If the ambient temperature is too low, R ESR will also show an increasing trend due to the condensation of the electrolyte.

[0014] Based on the above analysis, it can be determined that the characteristic phenomena of the failure of aluminum electrolytic capacitors are: R ESR will show an increasing trend, and the capacitance value C will show a decreasing trend.

[0015] (2) Failure mechanism of MOSFET

[0016] As a switching component of the DC-DC switching power supply in the multi-mode launch control system, the switching transistor mainly includes an Insulated Gate Bipolar Transistor (IGBT) and a Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET). Due to the special structural correlation between IGBT and MOSFET, there are great correlations between the two devices in terms of failure mechanism, main failure precursors, etc. MOSFET is more widely used in the DC-DC switching power supply of the multi-mode launch control system. Therefore, the failure mechanism of MOSFET devices will be analyzed and studied in detail below.

[0017] Appendix Figure 3 shows the N-channel and P-channel MOSFET devices and their equivalent circuit models, which are expressed as a switch S in series with the drain-source on-resistance R DS in series. The drain-source on-resistance (R DS ) mainly consists of the drain (D pole), source (S pole) and the resistance of the components between the two poles.

[0018] Excluding the faults caused by design defects, manufacturing negligence and other factors, the main influencing stresses for MOSFET failure are temperature stress and overvoltage stress. It can be clearly seen from the MOSFET structure diagram that MOSFET is composed of materials with different characteristics, including different densities, compositions, specific heat capacities, etc., which are overlaid on the substrate, resulting in different degrees of change or mismatch of temperature stress between different material layers. At the same time, MOSFET connects the chip and the substrate through leads, and these layer materials and connection points are also the on-resistance R DSComponents. When the performance of the MOSFET degrades, due to the mismatch of the coefficient of thermal expansion, the contact points will bear thermo-mechanical stress; with the accumulation of time, weaknesses are likely to form at these contact points. When there is a large short-term change or long-term continuous temperature stress on the junction temperature, these weaknesses will not be able to withstand the combined influence of internal and external stresses, resulting in solder fatigue, phenomena such as bond peeling and decreased adhesion, thus causing the complete failure of the MOSFET. In addition, when the external voltage is abnormal, the MOSFET is vulnerable to the influence of the gate-source drive voltage stress. When the gate-source drive voltage is greater than the maximum drive voltage, irreparable damage will occur in a short period of time. Since this phenomenon is not caused by the decay factor of the MOSFET itself, it is not considered in this project. Regarding the failure phenomenon of the MOSFET, relevant research has found that fault diagnosis and prediction can be carried out by monitoring three parameters: the on-resistance (R DS) , the gate threshold voltage, and the thermal resistance value. Among them, although the method of monitoring the gate threshold voltage is more sensitive than the method of monitoring the on-resistance at room temperature, the gate threshold voltage has a negative temperature coefficient, which is likely to make the measured value of the gate threshold voltage increased due to the performance attenuation of the MOSFET affected by the decrease in the measured value of the gate threshold voltage under the condition of rising temperature. Therefore, it is not applicable when the junction temperature is higher than room temperature; the thermal resistance measurement method cannot obtain real-time monitoring data in a short time. Especially for the new package DC-DC switching power supply module, limited by the existing measuring instruments, the feasibility of real-time measurement of this parameter is relatively low and the cost is high. Therefore, the on-resistance (R DS ) is used as the main failure characterization parameter of the MOSFET. Through the accelerated life experiment of the MOSFET, it is further found that the degradation trend of R DS over time follows an exponential curve, as shown in the following formula.

[0019] R DS (t) = αe βt + R0

[0020] where α and β are constant term coefficients, and R0 is the initial value of the on-resistance. According to the above research and analysis, the characterization phenomenon of the MOSFET failure is that the R DS value of the MOSFET will gradually increase.

[0021] The power diode mainly plays a rectifying role in the DC-DC switching power supply of the multi-mode launch control system. According to the characteristic parameters, power diodes can be divided into ordinary rectifying power type, fast recovery type and Schottky diodes. Since the internal circuit of the DC-DC switching power supply of the multi-mode launch control system mainly processes signals under high-frequency conditions, generally higher than 50HZ, it is required that the reverse recovery speed of the power diode is fast and the response time is short. Schottky diodes or fast recovery diodes are more widely used in the production and design of the DC-DC switching power supply of the multi-mode launch control system. Under high switching frequency conditions, the power diode and MOSFET are the main loss device sources of the DC-DC switching power supply of the multi-mode launch control system. Among them, the power diode loss is mainly affected by two factors: conduction voltage drop and reverse recovery time. Generally, the lower the conduction voltage and the shorter the reverse recovery time, the lower the loss of the corresponding power diode. Although the difference in recovery time between fast recovery diodes and Schottky diodes is small, the conduction voltage of fast recovery diodes is larger, and the rated current and voltage are also higher, which to a certain extent restricts the application of fast recovery diodes. Therefore, under the new development trend of current related products of the DC-DC switching power supply of the multi-mode launch control system, Schottky diodes are more suitable for applications under high-frequency and low-voltage conditions. Therefore, the following mainly analyzes and studies the failure mechanism of Schottky diodes.

[0022] (3) Failure mechanism of Schottky diodes

[0023] Different from the ordinary diode composed of P-type semiconductor and N-type semiconductor to form a P-N junction principle, the Schottky diode utilizes the metal-semiconductor structure and its equivalent circuit as shown. Among them, R Figure 4 is its junction resistance, C1 is its junction capacitance; L I is the electrode lead inductance; C Z is the silicon shell tube capacitance; R Z is its equivalent series resistance, and its specific value can be expressed as: Z R

[0024] R Z = R a + R su + R C

[0025] The Schottky diode in the circuit is mainly affected by current, voltage and temperature stress. When affected by the environment or the long-term high-frequency operation of the DC-DC switching power supply, the internal temperature of the Schottky diode will rise. Therefore, in this project, the temperature stress is mainly regarded as the main influencing stress for the degradation of the Schottky diode. The temperature stress mainly affects the two nodes of the anode metal and the N-epitaxial layer and the cathode metal and the N+ cathode layer, and becomes the main reason for the performance degradation of the Schottky diode.

[0026] According to the experimental data of the high and low temperature performance of a certain type of Schottky diode, it can be found that under the conditions of constant current and constant voltage and high temperature, the on-voltage gradually increases and the on-current gradually decreases, etc., showing state degradation. Among them, the above degradation state can be expressed as an increase in the on-resistance of the Schottky diode, and the failure degradation model is:

[0027] ΔR C =α(e βt -1)

[0028] Among them, ΔR C is the variable of the on-resistance of the Schottky diode, generally a positive increment, and R C =R Z +R t ; t is time; α and β are model constant parameters.

[0029] According to the above research and analysis, the characterization phenomenon of the failure of the Schottky diode is that the on-resistance R C of the Schottky diode shows an increasing trend.

[0030] Inductors mainly play roles such as energy storage and signal filtering in the DC-DC switching power supply of the multi-mode launch control system. During operation, the inductor coil continuously stores and releases energy, effectively filtering out the distorted signals in the signal. Inductors are generally classified according to the degree of self-closure of the inductor coil, and can also be classified according to the difference in packaging materials, and are correspondingly divided into ceramic inductors, ferrite inductors, and black magnetic crystal inductors, etc.

[0031] (4) Inductor failure mechanism

[0032] Inductors in the DC-DC switching power supply of the multi-mode launch control system mainly play roles of energy storage and filtering. The quality performance of inductors is generally determined by parameters such as inductance and maximum withstand voltage current value. Among them, the inductance is directly related to filtering and energy storage. The larger the inductance, the better the impedance performance for high-frequency signals, and high-frequency interference signals can be better suppressed. The larger the inductance, the more and denser the number of coil turns and their distribution are required. The basic structure and equivalent circuit of the inductor are respectively as Figure 5 shown.

[0033] Among them, R L is the equivalent loss resistance; C L is the equivalent distributed capacitance; L is the inductor. In low-frequency circuits, the inductor is equivalent to R L in series with L; for the high-frequency varying current of the DC-DC switching power supply of the multi-mode launch control system, the equivalent impedance of the inductor is:

[0034]

[0035] In the formula, R eis the equivalent resistance of the inductor, L e is the equivalent inductance.

[0036] The DC-DC switching power supply of the multi-mode launch control system generally operates in a high-frequency environment. The equivalent impedance of the inductor increases with the frequency. In addition, the performance degrades with the extension of working time, and the inductor coil will generate heat. The increase in heat causes the inductance of the inductor to decrease, and the energy storage and filtering efficiency decrease. In addition, under the conditions of overcurrent and overvoltage in the external circuit, the inductor coil is prone to aging and short circuit, which also leads to a decrease in inductance.

[0037] According to the above research and analysis, it is clear that the characterization phenomenon of the inductor failure is that the inductance of the inductor shows a downward trend and gradually decreases.

[0038] (5) Fault characterization phenomenon of the switching power supply

[0039] Table 1 Failure mechanism and characterization phenomenon of the main fault components of the DC-DC switching power supply

[0040]

[0041] Based on the above analysis of the failure mechanism of the relevant fault components of the DC-DC switching power supply of the multi-mode launch control system, the failure mechanism and characterization phenomenon of each main fault component are summarized as shown in Table 1.

[0042] According to the parameter degradation law of each main fault component of the DC-DC switching power supply of the multi-mode launch control system in the research results, it can be found that they are all of the gradual change type. Since the occurrence of the DC-DC switching power supply fault of the multi-mode launch control system is theoretically caused by the physical failure of a single or multiple components such as electrolytic capacitors and switching transistors. Therefore, it can be inferred that the change of the overall characteristic parameters of the DC-DC switching power supply should also be of the gradual change type theoretically.

[0043] At the same time, the main possible fault phenomena of the DC-DC switching power supply of the multi-mode launch control system are mainly as follows: (1) short circuit at the input end; (2) power supply interruption; (3) output voltage reduction; (4) output voltage increase; (5) output current increase. The external measured parameters of the packaged DC-DC switching power supply include four parameters: input voltage and current, output voltage and current. Under the condition of excluding the influence of external noise interference or uncontrollable factors, and without considering the associated faults such as overcurrent and overvoltage caused by the external circuit and device faults, online prediction and remaining life prediction can be carried out through the external measured parameters of the DC-DC switching power supply of the multi-mode launch control system and the data-driven prediction algorithm.

[0044] 2. Power supply remaining life prediction method based on finite incremental learning

[0045] (1) Design of the remaining life prediction model

[0046] The remaining useful life prediction model of the power supply module of the multi-mode launch, measurement, control and communication system is shown in the appendix Figure 6 as follows. During the training phase, the remaining useful life prediction model combines historical degradation data and online data to form a dataset, and randomly selects a training dataset and a test dataset from this dataset for model training and verification of the WE-OSELM algorithm that includes the IPW-OSELM algorithm and the error compensation algorithm respectively. Since in the absence of online data updates, the accuracy of multi-step prediction is directly affected by the training effect of the model. Therefore, how to quickly and effectively reduce the training model error and improve the model prediction ability is very important. The remaining useful life prediction model focuses on optimizing the number of hidden layer nodes of the IPW-OSELM algorithm through a limited incremental learning module to improve the model training ability and minimize the training error within a certain range. In the prediction phase, the error compensation algorithm and the IPW-OSELM algorithm perform multi-step prediction. When the predicted value meets the set failure threshold condition, the prediction is terminated and the remaining useful life is calculated.

[0047] The multi-step prediction method is shown in the appendix Figure 7 as follows. The multi-step prediction is based on the single-step prediction. The predicted value of the single-step prediction is sequentially used as the input value of the single-step prediction at the next moment and iterated cyclically. Among them, N is the number of input layer nodes, M is the multi-step prediction step length, X(i) is the true value at the i-th moment, and S(i) is the predicted value at the i-th moment.

[0048] (2) Limited incremental learning

[0049] In the online extreme learning machine and its derivative algorithms, the number of hidden layer nodes has always been an important parameter variable affecting the performance of the prediction algorithm. During the single-step prediction process of the WE-OSELM algorithm for three time series, selecting different numbers of hidden layer nodes in the IPW-OSELM algorithm model will lead to significant differences in the prediction result performance.

[0050] The flow chart of the limited incremental learning is shown in the appendix Figure 8 as follows. Assume that in the initial stage, the number of hidden layer nodes of the IPW-OSELM algorithm is L, and generally L can take a smaller value. At this time, the training error is ρ0. When the number of hidden layer nodes of the k-th training model is not greater than the maximum number of hidden layer nodes, the IPW-OSELM adds δL hidden layer nodes and randomly generates δL new hidden layer coefficients. However, when the number of hidden layer nodes of the model after adding hidden layer nodes L k >L max the incremental learning is stopped and the training model with the minimum training error is returned. Where L max is the set maximum number of hidden layer nodes.

[0051] After adding the number of hidden layer nodes at the k-th moment, the hidden layer output matrix of the IPW-OSELM is Wherein:

[0052]

[0053] ai and bi are respectively the output weight and bias parameter between the i-th hidden layer node and the input layer.

[0054] 3. Power supply remaining life prediction method based on online prediction model

[0055] In order to improve the real-time monitoring and life prediction capabilities of key components in dynamic reliability management, the power supply remaining life prediction method based on the online prediction model is further explored. As the core component of the multi-mode measurement and control system, the reliability of the DC-DC switching power supply in the system has a direct impact on the dynamic stability of the entire system. However, traditional prediction methods often have problems such as difficulty in capturing changes in data characteristics in real time, poor prediction accuracy, and insufficient algorithm stability. Therefore, this section proposes an online sequential extreme learning machine (OSELM) algorithm based on information perception weight and error compensation. By introducing a perception mechanism for time series data characteristics and an error correction mechanism, the online prediction of the power supply module life is realized to support the reliability management of the system in a dynamic environment.

[0056] The online sequential extreme learning machine algorithm based on information perception weight and error compensation mainly takes the time series data reflecting the performance changes of the research object as the prediction subject, perceives the characteristic changes of the prediction data through the multi-scale weight perception of the data, introduces the structure model of double-parallel single hidden layer feedforward neural networks (SLFNs), and further adds an error compensation algorithm module under the condition of effectively controlling the single prediction time, so as to effectively improve the timeliness and accuracy of online prediction.

[0057] (1) Design of time series online prediction model

[0058] The signals of the measurement and control system can be regarded as common time - series serial numbers. As an important prediction object in online prediction applications, the time series has attracted much attention because it can comprehensively and objectively reflect the performance of the prediction object, especially the change of performance over time. In addition, the online sequential extreme learning machine (OSELM) and its derivative algorithms inherit the theoretical basis of the extreme learning machine, transform the original static learning and training problems into dynamic online solutions of linear equations, and deduce the output weights through the recursive least - squares method to realize the online real - time update of the algorithm. Compared with other online prediction algorithms, the online sequential extreme learning machine and its derivative algorithms have the characteristics of fast prediction speed and high accuracy, but at the same time, there are also prominent problems such as large demand for training data sample size, poor timeliness, and insufficient algorithm stability.

[0059] Therefore, combining the technical requirements of the online prediction of the DC - DC switching power supply life in the multi - mode launch, measurement, control and communication system and the performance characteristics of OSELM and its derivative algorithms, a time - series online prediction model is proposed. The model is shown in the appendix Figure 9 as follows. The model mainly consists of two parts: the information perception weight online sequential extreme learning machine (IPW - OSELM) algorithm module and the error compensation algorithm module.

[0060] Among them, the IPW - OSELM algorithm module first relies on offline time - series data to complete the offline model training and determine the built - in parameters of the model; then it enters the online prediction stage, receives the online streaming time - series data and calculates the information perception weight of the current - moment data in real - time, and finally makes an online prediction according to actual needs. The error compensation algorithm module completes the algorithm model training by constructing the mapping relationship between the online streaming time - series data and the prediction error of the IPW - OSELM algorithm; inputs the online streaming time - series data in real - time, predicts the prediction error of the IPW - OSELM algorithm at the future moment, and corrects it.

[0061] The double - parallel SLFNs structure is shown in the appendix Figure 10 . Among them, the input data of both the IPW - OSELM algorithm and the error compensation algorithm are the data at the \(t\) - th moment \([(A t )]{A t =[x t ,x t-1 ,\(\cdots\),x t-m+1 ,A t \(\in R m}\), where \(x t \) is the online streaming time - series data at the \(t\) - th moment. In the IPW - OSELM algorithm, \(\omega a，b \) is the input weight between the \(a\) - th hidden - layer node and the \(b\) - th input - layer node, and \(\beta c，ais the bias parameter between the number of nodes in the a-th hidden layer and the number of nodes in the c-th output layer, S′ t+1 is the predicted value of the data at the (t + 1)-th moment in the online streaming time series of the IPW-OSELM algorithm. In the error compensation algorithm, ω′ a，b is the input weight between the a-th hidden layer node and the b-th input layer node, β′ c，a is the bias parameter between the number of nodes in the a-th hidden layer and the number of nodes in the c-th output layer, E′ t+1 is the predicted error value at the (t + 1)-th moment. S t+1 is the predicted value of the data at the (t + 1)-th moment in the online streaming time series of the WE-OSELM algorithm.

[0062] (2) Information perception weight design

[0063] The purpose of information perception weight design is mainly to improve the performance of the IPW-OSELM algorithm in aspects such as the sensitivity of online perception of the characteristics change of time series data and the prediction timeliness. Different from the analysis of only the data characteristics at the current moment and adjacent moments in algorithms such as T-OSELM, the information perception weight not only includes the comparison of the similarity of data characteristics between adjacent moments, but also adds the comparison of the similarity between the prediction point data and the regional data, and integrates the two parts for processing. The data structure of the online streaming time series is as shown in the appendix Figure 11 as follows.

[0064] Let [(A t )]{A t = [x t , x t-1 , K, x t-m+1 , A t ∈ R m} be the input data of the IPW-OSELM algorithm at the t-th moment, m be the number of input layer nodes x t , be the online streaming time series data at the t-th moment. The adjacent point data of the data [(x k )]{x k ∈ R} at the k-th moment is the data [(x k-1 )]{x k-1 ∈ R} at the (k - 1)-th moment; its adjacent region data is expressed as [(x k-d+1 , K, x k-1 )]{x t ∈ R, t = k - d + 1, K, k - 1}, d is the width of the regional data. Therefore, the similarity between the data at the k-th moment and the data at the (k - 1)-th moment, defined as the point similarity of the data at the k-th moment, is expressed as R pk = γ(x k , x k-1 ); the similarity of the input data at the k-th moment relative to its adjacent region data, defined as the regional similarity of the data at the k-th moment, is expressed as R γk= γ(x k , E(x k-d+1 , K, x k-1 ))。Among them, γ represents the similarity calculation function. In this project, the Euclidean distance similarity is adopted, as shown in the following formula.

[0065]

[0066] Under the condition of excluding the interference of factors such as noise, sudden unpredictable faults, and inherent design defects in the time series data, any fluctuations and changes in the time series data can be considered as effective changes reflecting the reliability of the research object. Therefore, there are four mutual relationships between the point similarity and the regional similarity respectively:

[0067] ① High point similarity and high regional similarity. This indicates that the data characteristics at the current moment have not changed significantly compared with the previous period. Therefore, the information perception weight value should be set low.

[0068] ② Low point similarity and low regional similarity. This indicates that the data characteristics at the current moment have changed significantly compared with the previous period, and there is a high possibility that the overall data characteristics will change significantly in the future. Therefore, the information perception weight value should be set high.

[0069] ③ High point similarity and low regional similarity. This indicates that the data characteristics have started to change since the adjacent moment. Therefore, the information perception weight should be set relatively high.

[0070] ④ Low point similarity and high regional similarity. This indicates that the data characteristics at the adjacent moment have fluctuated significantly. Therefore, the information perception weight should be set relatively low.

[0071] The mutual relationships among similarity, regional similarity, and information perception weight are shown in Table 2.

[0072] Table 2 Mutual relationships among point similarity, regional similarity, and information perception weight

[0073]

[0074] Based on the above analysis and combined with the design requirements, the information perception weight is designed as shown in Equation (2.69):

[0075]

[0076] Among them, Ω t is the information perception weight at the t-th moment; R pt is the point similarity of the data at the t-th moment; R rt is the regional similarity of the data at the t-th moment.

[0077] (3) Design of error compensation algorithm

[0078] Based on the theory of extreme learning machines, the online sequential extreme learning machine proposes an error correction mechanism that mainly updates the output weights of the hidden layer online. The core formula is shown as follows.

[0079]

[0080] Where: H k =[g(a1·x k +b1),..., g(a L ·x k +b L )] is the output matrix of the hidden layer at the k-th moment, x k is the input data at the k-th moment, [(a i , b i )]{a i ∈R, b i ∈R, i = 1,..., L} are the input weights and bias parameters between the input layer and the hidden layer respectively; L is the number of nodes in the network hidden layer; g(x) is the activation function;

[0081] The predicted value at the k-th moment is generally calculated by multiplying the output matrix H k of the hidden layer at the k-th moment by the output weight β k-1 of the hidden layer at the k - 1-th moment, and then calculating the difference with the true value T k at the k-th moment, so as to obtain the prediction error of the data at the k-th moment. Finally, the output weight k of the hidden layer at the k-th moment is calculated and used for the prediction of the data at the k + 1-th moment.

[0082] Analysis shows that the prediction error at the k-th moment in the online sequential extreme learning machine is expressed as ε k =T k -H k β k-1 , and this error is the prior error; correspondingly, the posterior error at the k-th moment is expressed as e k =T k -H k β k-1 . In the theory of extreme learning machines, the output weight β k of the hidden layer at the k-th moment obtained by deriving the cost function using the least squares method is based on the known posterior error e k , but generally the posterior error cannot be known in advance, so the prior error is correspondingly used as an approximation of the posterior error. This will inevitably lead to insufficient timeliness of the prediction results, especially when the data characteristics of time series data fluctuate or mutate within a limited range, and it is also likely to have an adverse impact on the stability of subsequent algorithms.

[0083] According to the above analysis, when there are fluctuations in the data characteristics or finite-range mutations in the time-series data, the prediction errors between adjacent moments of the prediction algorithm should also be a time series with gradual changes or finite mutations; in addition, the prediction model obtained through offline data training can be regarded as a certain fixed function, and there are the following relationships among the input data sequence, predicted data, real data, and the prediction model function during the online learning stage:

[0084] S a+N = f(x a , x a+1 ,..., x a+N-1 , x a+N-1 , T a+N-1 , S a+N-1 )

[0085] where f is the prediction model function; (x a , x a+1 ,..., x a+N-1 , x a+N-1 , T a+N-1 , S a+N-1 ) is the input data of the input layer nodes at the (a + N)th moment; N is the number of input layer nodes; T a+N-1 is the real data at the (a + N - 1)th moment, and there is x a+N-1 = T a+N-1 / S a+N which is the predicted data at the (a + N - 1)th moment. It can be considered that when the prediction model is trained, the prediction model function is fixed, and the predicted value is only related to the input data. Further analysis shows that under the condition that the prediction model function remains unchanged, the prediction error is only related to the input data.

[0086] Based on the above analysis and combined with the theory of the online sequential extreme learning machine algorithm, an error compensation algorithm is designed, and the flowchart is as shown in the appendix Figure 12 as follows.

[0087] Among them, the error prediction model needs to rely on the training data to complete the model training before it can perform error prediction. The training data consists of the time-series input data and the prediction error of the IPW-0SELM algorithm, expressed as [(A t , e t )]{A t ∈ R m , e t ∈ R, t = 1,..., N0}. The initial output weights of the hidden layer are obtained through training:

[0088]

[0089] where

[0090]

[0091] The output weight of the hidden layer at the t-th moment is updated to:

[0092]

[0093] where

[0094]

[0095] e t is the true prediction error of the IPW-OSLEM algorithm at the t-th moment, then the prediction error value of the IPW-OSLEM algorithm at the (t + 1)-th moment is:

[0096] e′ t+1 = H t+1 β′ t

[0097] Correspondingly, the predicted value of the time series data of the WE-OSELM algorithm at the (t + 1)-th moment is:

[0098] S t+1 = S′ t+1 + αe′ t+1

[0099] where α ∈ (0, 1] is the error compensation coefficient; S t+1 is the prediction of the time series data of the IPW-OSELM algorithm at the (t + 1)-th moment.

[0100] In summary, with the diversification of the functional requirements of complex measurement and control systems and the dynamic changes in the task execution environment, higher requirements are put forward for the reliability, real-time performance, and power consumption optimization of power management. In the existing technology, the fuzzy dynamic Bayesian network shows high flexibility and adaptability in system reliability assessment; the classifier-based fault diagnosis technology improves the system's fault handling ability through efficient concurrent fault location; the power supply life prediction method based on the online prediction model can combine historical data and real-time monitoring data to optimize the sustainability of system operation. However, these technologies still face many challenges in the complex environment of multi-mode mechanisms, such as the accurate diagnosis and isolation of concurrent faults, the accuracy of reliability assessment in complex dynamic environments, and the optimization and coordination of power consumption and performance during mode switching. Therefore, there is an urgent need for an innovative technology that can comprehensively utilize the above technologies in the multi-mode operating state to further improve the stability and energy efficiency of complex measurement and control systems. Summary of the Invention

[0101] The present invention provides a multi-mode power state assessment method, aiming to achieve significant power consumption reduction and system stability improvement in complex systems through precise power management and intelligent mode switching.

[0102] First, through the analysis of the failure mechanisms of key components in the power supply system and by combining with the fuzzy dynamic Bayesian network, the present invention proposes a dynamic reliability assessment method for the power supply system. This method can monitor the operating status of the power supply in real time, and evaluate the reliability of the power supply according to the failure mechanisms and parameter degradation laws, providing data support for subsequent remaining life prediction and fault diagnosis.

[0103] Secondly, the invention proposes a method for predicting the remaining life of the power supply based on finite incremental learning. By combining historical degradation data and real-time monitoring information, and continuously optimizing the prediction model, the accuracy and adaptability of the remaining life prediction are improved. On this basis, the present invention further designs a method for predicting the remaining life of the power supply based on an online prediction model, which can adjust the prediction model in real time, improve the real-time performance and accuracy of the prediction, and ensure that the system always maintains a high-efficiency and stable state during operation.

[0104] In addition, by combining fault diagnosis algorithms, the present invention can accurately identify multiple concurrent faults and perform adaptive adjustment on the system. When a fault occurs, the system can quickly activate standby components or perform mode switching, thereby effectively avoiding a decline in system performance and enhancing the robustness and stability of the system.

[0105] The multi-mode power management strategy of the present invention realizes the precise management and optimization of the power supply system through reasonable fault diagnosis and life prediction mechanisms. This technology can not only greatly improve the stability and reliability of the system in complex environments, but also provide a technical solution with broad application prospects for the field of power management.

[0106] Compared with the prior art, the present invention has the following advantages:

[0107] (1) By combining the power management strategy and dynamic reliability assessment method in multi-mode working states, the present invention can more accurately evaluate the operating status of the power supply in different working modes, thereby realizing more efficient power consumption management and system stability optimization.

[0108] (2) Adopting the system dynamic reliability assessment technology based on classifiers and fuzzy algorithms, and combining fault diagnosis algorithms to accurately locate concurrent faults, ensuring that the system can quickly identify and respond to faults in multiple working states, thereby effectively improving the fault tolerance and stability of the system.

[0109] (3) The power supply state assessment method proposed by the present invention combines finite incremental learning and an online prediction model, which not only optimizes the prediction accuracy of the remaining life of the power supply, but also improves the timeliness and accuracy of online prediction. By comprehensively considering historical degradation data and real-time monitoring data, it can evaluate the power supply state in real time and predict the remaining life, ensuring the efficient operation of the power management system in different working modes.

[0110] (4) In multi-mode power management, the present invention provides a method for predicting the remaining life of a power supply based on an online prediction model, which can dynamically adjust the power supply strategy during the operation of the power supply, reduce the system power consumption, and extend the service life of the power supply.

[0111] (5) The method of the present invention has good versatility and can be widely applied to various multi-mode systems. Especially in occasions with high key requirements such as high energy efficiency, long-term stable operation, and fault isolation, it can provide reliable power management and optimization solutions for other complex systems.

[0112] The following will describe the specific embodiments of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0113] Figure 1 is a system schematic diagram of the present invention.

[0114] Figure 2 is the structure of an aluminum electrolytic capacitor and its equivalent circuit diagram.

[0115] Figure 3 is the N-channel and P-channel MOSFET devices and their equivalent circuit diagrams.

[0116] Figure 4 is the structure of a Schottky diode and its equivalent circuit diagram.

[0117] Figure 5 is the basic structure and equivalent circuit diagram of an inductor.

[0118] Figure 6 is the remaining life prediction model.

[0119] Figure 7 is the multi-step prediction method.

[0120] Figure 8 is the flowchart of the finite increment learning.

[0121] Figure 9 is the time series online prediction model.

[0122] Figure 10 is the structure of a double-parallel single-hidden layer feedforward neural network.

[0123] Figure 11 is the online streaming time series data structure.

[0124] Figure 12 is the flowchart of the error compensation algorithm. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0125] In the following Figure 2 shown system prototype software interface, the system prototype can be made to pass through the following Figure 3The FPGA-based multi-mode power management module shown is used to Figure 4 the DC power supply module shown, as well as Figure 5 the host power supply module of the measurement and control system shown for power status assessment, and the assessment results will be presented on the software interface in real time.

[0126] As can be seen from the above steps and the drawings, the method of the present invention can effectively achieve the power status assessment and optimal control of the measurement and control system in the multi-mode working state, significantly reduce the system power consumption while ensuring the operation stability of the system in the complex dynamic environment, and is of great significance for real-time status monitoring and task adaptation. Through the system dynamic reliability assessment combining the classifier and the fuzzy algorithm, the present invention can evaluate the performance of the power supply in different working modes in real time, identify and handle faults in a timely manner, thereby effectively improving the fault tolerance ability of the system. Especially in the complex scenarios of multi-task switching, multi-fault diagnosis and energy efficiency optimization, the adaptive multi-mode control strategy can comprehensively consider the system status detection data, task requirements and fault diagnosis results to ensure the high efficiency and accuracy of power management and mode switching.

[0127] For abnormal situations such as sudden increase in power consumption, the present invention effectively avoids the decline of system performance and ensures the continuity of task execution through precise power regulation and intelligent mode switching. By using the power status assessment results, the power strategy can be dynamically adjusted to optimize power consumption management and improve system stability. However, for hardware sudden faults (such as severely damaged power supply modules) that cannot be compensated by mode switching or power regulation, the present invention does not cover their solutions.

[0128] Since the multi-mode control method of the present invention is based on a widely applicable adaptive adjustment mechanism and can adapt to different system requirements, its technology can be popularized and applied to the power management and energy efficiency optimization problems of other complex systems, and has good application prospects and popularization value.

Claims

1. A power status evaluation method and device for a measurement and control system based on adaptive multi-mode control, characterized in that, The system includes five parts: a status detection module, a task analysis module, a fault diagnosis module, a fuzzy dynamic Bayesian network evaluation module, and a power control module. Among them, the status detection module collects the operation status data of the system in real time. The task analysis module generates task load information according to the task requirements. The fault diagnosis module conducts fault analysis and location on the status data. The fuzzy dynamic Bayesian network evaluation module conducts dynamic reliability evaluation based on the system status data and the fault diagnosis results. The power control module adjusts the power status according to the evaluation results and the task requirements to achieve system power consumption optimization and stability improvement.

2. A power status evaluation method and device for a measurement and control system based on adaptive multi-mode control according to claim 1, characterized in that, The described status detection module includes multiple sensors, which can collect the operation data such as voltage, current, power, and temperature of the system in real time, providing basic data support for subsequent task analysis, fault diagnosis, and evaluation.

3. A power status evaluation method and device for a measurement and control system based on adaptive multi-mode control according to claim 2, wherein The described fuzzy dynamic Bayesian network evaluation module adopts fuzzy logic reasoning technology, combines historical data with real-time data, analyzes the dynamic reliability of the system, and generates a power status evaluation report to support power optimization and mode switching decisions.

4. A method and device for evaluating the power supply state of a measurement and control system based on adaptive multi-mode control according to claim 3, characterized in that, The described power control module dynamically adjusts the output voltage and current of the power supply according to the evaluation report and the task requirements, optimizes the power supply configuration, and ensures low power consumption and efficient operation in different operation modes.

5. A power state evaluation method for a measurement and control system based on adaptive multi-mode control, characterized in that The implementation includes the following steps: Step 1: The status detection module collects the information such as voltage, current, power, and temperature of the system in real time and transmits it to the task analysis module and the fault diagnosis module. Step 2: The task analysis module analyzes the requirements of the current task, and the fault diagnosis module monitors the system status, conducts fault location, and analyzes its impact. Step 3: The fuzzy dynamic Bayesian network evaluation module combines the status data and the fault diagnosis results to evaluate the dynamic reliability of the system, providing an evaluation basis for power optimization. Step 4: The power control module adjusts the output voltage and current of the power supply according to the evaluation results in Step 3 and the task requirements to complete power consumption optimization and mode switching.

6. The power state evaluation method according to claim 5, characterized in that, The system status data collected in Step 1 includes load characteristics, ambient temperature, and power input parameters, which are used to improve the accuracy of fault diagnosis and mode selection.

Images