A method and device for detecting availability of piezoelectric materials

Abstract

The present invention discloses a method and device for detecting the availability of piezoelectric materials, which belongs to the technical field of piezoelectric material detection, including obtaining a historical data time series of piezoelectric materials within a set period, then calculating a historical piezoelectric coefficient time series of the piezoelectric materials and inputting it into a trained improved ARIMA to obtain a future piezoelectric coefficient time series of the piezoelectric materials within a predicted period; calculating a life evaluation value of the piezoelectric materials according to the future piezoelectric coefficient time series, and judging whether the piezoelectric materials are available according to the life evaluation value. The present invention significantly improves the accuracy and consistency of data by eliminating errors in historical data, adopts an improved differential autoregressive moving average model to reduce the influence of ambient temperature on the piezoelectric coefficient, and improves the accuracy of prediction; judges whether the piezoelectric material is available in each stage according to the preset judgment rules for each stage; and comprehensively improves various aspects to achieve accurate judgment of whether the piezoelectric material is available.

Description

Technical Field

[0001] The invention relates to a method and a device for detecting the availability of a piezoelectric material, and belongs to the technical field of piezoelectric material detection. Background Art

[0002] The piezoelectric effect refers to the phenomenon that some materials will produce electric polarization when subjected to mechanical stress, thereby generating an electric field inside the material, or generating strain under the action of the electric field. This material that can realize the mutual conversion of mechanical energy and electrical energy is called a piezoelectric material. Piezoelectric materials are widely used in the field of sensors. The piezoelectric coefficient is an important parameter to measure the performance of piezoelectric materials. It determines the degree of response of the material when subjected to mechanical stress or electric field. The larger the piezoelectric coefficient, the better the piezoelectric performance of the material, and the higher the conversion efficiency of mechanical energy and electrical energy that can be achieved. Therefore, accurately predicting and measuring the piezoelectric coefficient of piezoelectric materials is of great significance for material selection, device design and performance optimization.

[0003] The piezoelectric coefficient of piezoelectric materials will gradually change over time. At present, with the continuous development of intelligent algorithms, the first-principles calculation method based on density functional theory is usually used to predict the piezoelectric coefficient. This type of calculation method is a calculation based on the field of quantum mechanics. By calculating the electronic structure and lattice parameters of piezoelectric materials, the theoretical value of the piezoelectric coefficient is derived. However, the calculation process of this method is extremely complex and requires high computing resources. In addition, the first-principles prediction is usually for specific material systems and structures, and may be difficult to apply to unknown complex material systems. There is currently a lack of an accurate judgment and detection method to determine whether piezoelectric materials can continue to be used. Summary of the invention

[0004] The object of the present invention is to provide a method and device for detecting the availability of a piezoelectric material, so as to accurately determine whether the piezoelectric material is available.

[0005] To achieve the above objectives, the present invention is implemented by adopting the following technical solutions:

[0006] In a first aspect, the present invention provides a method for detecting the availability of a piezoelectric material, comprising:

[0007] Obtain the historical data time series of the piezoelectric material within a set period, and obtain the historical data time series without abnormal values ​​after preprocessing;

[0008] Calculate the historical piezoelectric coefficient time series of the piezoelectric material based on the historical data time series without abnormal values;

[0009] Inputting the historical piezoelectric coefficient time series into the trained improved differential autoregressive moving average model to obtain the future piezoelectric coefficient time series of the piezoelectric material in the prediction period, wherein the improved differential autoregressive moving average model is obtained by improving the differential autoregressive moving average model based on the ambient temperature;

[0010] The life evaluation value of the piezoelectric material is calculated according to the future piezoelectric coefficient time series, and whether the piezoelectric material is usable is determined according to the life evaluation value.

[0011] Furthermore, the historical data includes force, charge and ambient temperature;

[0012] The historical data time series includes charge time series, ambient temperature time series and force time series:

[0013] The charge time series is obtained by the following method: all charges are arranged in the order of sampling time from earliest to latest, so as to obtain the charge time series; the ambient temperature time series is obtained by the following method: all ambient temperatures are arranged in the order of sampling time from earliest to latest, so as to obtain the ambient temperature time series; the force time series is obtained by the following method: all forces are arranged in the order of sampling time from earliest to latest, so as to obtain the force time series.

[0014] Furthermore, the preprocessing includes eliminating gross errors and reducing systematic errors;

[0015] The elimination of gross errors includes:

[0016] The comprehensive error of each charge in the charge time series is calculated using the following formula:

[0017] ;

[0018] Among them, i represents the time sequence number of the charge, is a weight function that reflects the importance ratio of the local weighted error to the global weighted error. represents the local weighted error of the charge on the i-th day in the historical data, represents the global weighted error of the charge on the i-th day in the historical data, represents the comprehensive error of the charge on the i-th day in the historical data;

[0019] If the comprehensive error of the charge amount on a certain day is greater than the set first threshold, the charge amount on that day is a gross error. The charge amount on that day is eliminated, and the corresponding ambient temperature and force on that day are also gross errors. The ambient temperature and force on that day are eliminated until the comprehensive errors of all charge amounts are less than or equal to the set first threshold.

[0020] The local weighted error is calculated by the following method:

[0021] The charge in the charge time series is divided into a plurality of small windows containing the same amount of charge, the local true value of the charge is calculated according to the median of the small windows, and the local weighted error is calculated according to the local true value;

[0022] The local true value of the charge is calculated according to the median of the small window, using the following formula:

[0023] ;

[0024] ;

[0025] ;

[0026] in, Indicates the charge y used to reflect the i-th day in the historical data i and the median y of the small window m The weighted function of the degree of deviation, σ is the standard deviation of all charges in the small window, represents the local true value of the charge on the i-th day in the historical data obtained by fitting the charge on the j-th day in the historical data, represents the charge on the jth day in the historical data, exp() represents the natural exponential function, W m represents the mth small window, represents the local weighted error of the charge on the i-th day in the historical data;

[0027] The global weighted error is calculated by the following method:

[0028] ;

[0029] ;

[0030] ;

[0031] in, represents the global mean of the charge time series, represents the standard deviation of the charge time series, n represents the maximum number of days in the historical data, and y i is the charge on the i-th day;

[0032] The reduction of systematic errors includes:

[0033] The residual sequence of the charge time series is calculated, and the residual sequence of the charge time series is divided into the first residual sequence group and the second residual sequence group. The difference between the first residual sequence group and the second residual sequence group is calculated by the following formula:

[0034] ;

[0035] Wherein, Δ represents the difference between the first residual sequence group and the second residual sequence group, represents the bth residual error in the residual sequence, and Located in the first residual sequence group after division, represents the cth residual error in the residual sequence, and Located in the second residual sequence group after division;

[0036] in, is a constant, The value rule of is: when n is an even number, ξ=n / 2; when n is an odd number, ξ=(n+1) / 2;

[0037] If the absolute value of Δ is not equal to 0, then there is a systematic error in the charge sequence. If the absolute value of Δ is equal to 0, then there is no systematic error in the charge sequence, and the systematic error is reduced by the offset method.

[0038] Furthermore, the expression of the improved differential autoregressive moving average model is:

[0039] ;

[0040] Among them, p is the number of autoregressive terms in the improved difference autoregressive moving average model, q is the number of sliding average terms in the improved difference autoregressive moving average model, τ represents the τth day in the forecast period, is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, is the set constant, is the error term of the τth day in the forecast period, r ο is the autocorrelation coefficient, is the correlation coefficient of ambient temperature, is the correlation coefficient of the error term, is the value of the piezoelectric coefficient of the hysteresis k order, is the error term of lag k, T is the ambient temperature, and k represents the lag order;

[0041] The ambient temperature is characterized by the following formula:

[0042] ;

[0043] Where T is the ambient temperature, τ represents the τth day in the period to be predicted, and W is the set period. Indicates the maximum value in the ambient temperature time series.

[0044] Furthermore, the number of autoregressive terms in the improved autoregressive moving average model and the number of moving average terms in the improved autoregressive moving average model are determined by the following method:

[0045] The autocorrelation function ACF is calculated by the following formula:

[0046] ;

[0047] The partial autocorrelation function PACF is calculated using the following formula:

[0048] ;

[0049] in, represents the predicted value of the piezoelectric coefficient on the τth day in the prediction period under the influence of all independent variables from lag 1 to k-1 order, represents the predicted value of the piezoelectric coefficient on the τ-kth day in the prediction period under the influence of all independent variables from lag 1 to k-1 order, is the value of the piezoelectric coefficient on the τth day in the period to be predicted, is the value of the piezoelectric coefficient on the τ-kth day in the period to be predicted, Cov() represents the covariance, Var() represents the variance, and z represents the maximum value of τ;

[0050] If the autocorrelation function of a certain lag order exceeds the second threshold, and the autocorrelation functions corresponding to all lag orders after the lag order are zero, then the lag order is selected as the value of q; if the partial autocorrelation function of a certain lag order exceeds the second threshold, and the partial autocorrelation functions corresponding to all lag orders after the lag order are zero, then the lag order is selected as the value of p;

[0051] By using the above method, any one of several combinations of p and q is determined as the values ​​of p and q.

[0052] Furthermore, after obtaining several combinations of p and q, the best values ​​of p and q are selected by Bayesian optimization method:

[0053] The discrete values ​​of p and q are processed continuously so that they can be applied to Gaussian processes;

[0054] Taking p, q and T as input and L(p,q,T) as output, we construct the following Gaussian process model:

[0055] ;

[0056] Where x refers to p, q and T, m(x) is the mean function of x, k(x,x′) is the RBF kernel function of x and x′, x and x′ are two different inputs, f(x) represents the output of the Gaussian process, GP() represents the execution of the Gaussian process, and L(p,q,T) represents the set objective function;

[0057] The set acquisition function defines the next set of p, q and T values ​​that are most likely to minimize L(p,q,T) as a new point (p^*, q^*, T^*) according to the Gaussian process model, calculates the corresponding L(p,q,T) value at the new point (p^*, q^*, T^*) and uses the value as the output of the objective function, updates the new point (p^*, q^*, T^*) and the corresponding L(p,q,T) to the Gaussian process model, starts iteration, and continues to select the next set of p, q and T values ​​for iteration according to the updated Gaussian process model until the rate of change of the minimum value of the objective function after a preset number of iterations is less than the preset rate of change, and determines to use the p and q values ​​at this time as their respective optimal values;

[0058] The objective function is calculated by the following formula:

[0059] ;

[0060] ;

[0061] Among them, L(p,q,T) represents the set objective function, n represents the maximum number of days in the historical data, is the value of the piezoelectric coefficient on the τth day in the period to be predicted, N is the sample size of the charge sequence after eliminating the error, represents the time required to improve the training and prediction of the difference autoregressive moving average model, is the weight of the mean absolute percentage error MAPE, To calculate the time weight.

[0062] Furthermore, before inputting the historical piezoelectric coefficient time series into the trained improved difference autoregressive moving average model, the trained improved difference autoregressive moving average model is tested, specifically including:

[0063] The Q statistic of the trained improved difference autoregressive moving average model is calculated by the following formula:

[0064] ;

[0065] Among them, Q LB represents the Q statistic, is the sample capacity of the residual sequence of the predicted value of the piezoelectric coefficient in the predicted period, is the residual sequence of the predicted value of the piezoelectric coefficient The autocorrelation coefficient of order, represents the order of the autocorrelation coefficient, and k is the lag order;

[0066] The residual sequence of the piezoelectric coefficient prediction value The order autocorrelation coefficient is calculated using the following formula:

[0067] ;

[0068] in, is the residual between the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted and the true value of the piezoelectric coefficient, and z represents the maximum value of τ;

[0069] Q LB Obey the chi-square distribution with kpq degrees of freedom, p represents the number of autoregressive terms in the improved autoregressive moving average model, q represents the number of sliding average terms in the improved autoregressive moving average model, obtain the chi-square distribution table and find out the confidence probability value corresponding to the kpq degrees of freedom. If the confidence probability value is less than or equal to the set significance level, the trained improved autoregressive moving average model needs to be further optimized. If the confidence probability value is greater than the set significance level, the trained improved autoregressive moving average model passes the test;

[0070] If the trained improved autoregressive moving average model needs to be further optimized, the first threshold is reduced, and the value of the calculation time weight is further reduced to reduce the proportion of time, until the trained improved autoregressive moving average model passes the test;

[0071] Before inputting the historical piezoelectric coefficient time series into the trained improved differential autoregressive moving average model, the method further includes the steps of performing a unit root test on the historical piezoelectric coefficient time series and performing a differential operation, wherein the differential operation includes:

[0072] If the historical piezoelectric coefficient time series is a stationary series, the difference order is 0;

[0073] If the historical piezoelectric coefficient time series is a non-stationary series, the historical piezoelectric coefficient time series is subjected to ψ-order differences until the historical piezoelectric coefficient time series becomes stationary, where ψ represents the difference order.

[0074] Furthermore, the life evaluation value of the piezoelectric material is calculated based on the future piezoelectric coefficient time series, and is calculated by the following formula:

[0075] ;

[0076] in, Indicates the estimated lifespan of the piezoelectric material on that day. is the theoretical value of the piezoelectric coefficient of the piezoelectric material, is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, is the minimum acceptable level of piezoelectric coefficient;

[0077] in, The calculation formula is:

[0078] ;

[0079] in, is the average value of the historical piezoelectric coefficient, is the standard deviation of the historical piezoelectric coefficients, is the set safety factor.

[0080] Furthermore, judging whether the piezoelectric material is usable according to the life evaluation value includes:

[0081] like is greater than the set third threshold and less than 1, it indicates that the piezoelectric material is in the early stage; if is greater than the set fourth threshold and less than the set third threshold, it indicates that the piezoelectric material is in the middle stage; if is greater than the set fifth threshold and less than the set fourth threshold, it indicates that the piezoelectric material is in the late stage; if is less than the set fifth threshold and greater than the set sixth threshold, it indicates that the piezoelectric material is in an unusable stage, wherein the values ​​of the third threshold, the fourth threshold, the fifth threshold, and the sixth threshold decrease in sequence;

[0082] When the piezoelectric material is in the early stage, and the decay rate is greater than 1.5 within the consecutive set number of days, it indicates that the piezoelectric material is in the early stage of abnormal decay; when the piezoelectric material is in the middle stage, and the decay rate is greater than 1.5 within the consecutive set number of days, it indicates that the piezoelectric material is in the middle stage of abnormal decay; when the piezoelectric material is in the late stage, and the decay rate is greater than 2 within the consecutive set number of days, it indicates that the piezoelectric material is in the late stage of abnormal decay;

[0083] When the piezoelectric material has experienced one of the early attenuation abnormality stage and the mid-term attenuation abnormality stage, a maintenance instruction is issued to a set maintenance unit;

[0084] When the piezoelectric material has experienced the late attenuation abnormal stage, or the piezoelectric material has experienced both the early attenuation abnormal stage and the mid-term attenuation abnormal stage, or the piezoelectric material is in the unavailable stage, it indicates that the piezoelectric material meets the unavailable judgment condition, the piezoelectric material is unavailable, and an unavailable instruction is issued. In other cases, the piezoelectric material is available;

[0085] The calculation formula of the decay rate is:

[0086] ;

[0087] in, represents the decay rate of the τth day in the period to be predicted, represents the decay rate of the τ-1th day in the period to be predicted, represents the decay rate;

[0088] The calculation formula of the attenuation rate is:

[0089] ;

[0090] in, is the predicted value of the piezoelectric coefficient on the τ-1th day in the period to be predicted.

[0091] In a second aspect, the present invention provides a piezoelectric material availability detection device, which is used to implement the above-mentioned piezoelectric material availability detection method; the piezoelectric material availability detection device includes:

[0092] The historical data time series acquisition module is configured to: acquire the historical data time series of the piezoelectric material within a set period of time, and obtain the historical data time series without abnormal values ​​after preprocessing;

[0093] The historical piezoelectric coefficient time series calculation module is configured to: calculate the historical piezoelectric coefficient time series of the piezoelectric material according to the historical data time series without abnormal values;

[0094] The future piezoelectric coefficient time series prediction module is configured to: input the historical piezoelectric coefficient time series into the trained improved differential autoregressive moving average model to obtain the future piezoelectric coefficient time series of the piezoelectric material in the predicted period, wherein the improved differential autoregressive moving average model is obtained by improving the differential autoregressive moving average model based on the ambient temperature;

[0095] The availability assessment module is configured to: calculate the life assessment value of the piezoelectric material according to the future piezoelectric coefficient time series, and determine whether the piezoelectric material is available according to the life assessment value.

[0096] Beneficial Effects

[0097] The present invention provides a method and device for detecting the availability of piezoelectric materials. The method and device significantly improve the accuracy and consistency of data by eliminating errors in historical data. The improved differential autoregressive moving average model is adopted to take the ambient temperature into consideration, thereby greatly reducing the influence of the ambient temperature on the piezoelectric coefficient and improving the accuracy of prediction. A set of evaluation criteria is designed to determine whether the piezoelectric material is available. The above improvements are combined to achieve accurate judgment on whether the piezoelectric material is available.

[0098] The present invention can comprehensively identify abnormal points in the data through comprehensive error detection combined with dual analysis of local errors and global errors; taking the comprehensive error as the center, local errors and global errors are developed around it, from local data fitting to global distribution evaluation, the degree of abnormality of data points is comprehensively measured, and on this basis, gross errors are accurately eliminated; this method significantly improves the accuracy and consistency of data through multi-level error analysis and weighted synthesis, laying a reliable foundation for subsequent data processing and analysis. The Malikov criterion can also be used to eliminate systematic errors in the collected data, eliminating the problem that ARIMA (differential autoregressive moving average model) is easily disturbed by abnormal values. The obtained prediction results are relatively accurate, providing a basis for the accurate construction of subsequent models;

[0099] The ARIMA adopted in the present invention performs well in processing time series data, and can accurately predict the future trends and periodic changes of time series. When predicting the piezoelectric coefficient of piezoelectric materials, ARIMA can capture the law of material performance changes over time, thereby improving the accuracy of the prediction; currently, the calculation cost of first-principles prediction is relatively high, and professional computers and hardware support are required, which has certain limitations in real-time performance, while ARIMA calculation and prediction are relatively simple, and can give better prediction results when processing different data sets under the condition of sufficient data, which makes it highly adaptable and real-time. For piezoelectric materials, materials of different types and different preparation processes may have different piezoelectric coefficient change laws; ARIMA can flexibly adapt to these changes and provide reliable predictions;

[0100] The bivariate ARIMA used in the present invention takes into account that the ambient temperature factor has a greater impact on the piezoelectric coefficient, which ultimately reduces the accuracy of the model prediction. Here, the ambient temperature factor is introduced as an exogenous variable into the ARIMA model of the piezoelectric coefficient, which greatly reduces the impact of the ambient temperature on the piezoelectric coefficient and improves the accuracy of the model prediction.

[0101] The automatic threshold judgment used in the present invention can make the machine run automatically without the help of ACF images and PACF images (the ACF graph shows the relationship between the lag period and the autocorrelation coefficient, with the horizontal axis being the lag period and the vertical axis being the autocorrelation coefficient. By observing the ACF graph, it can be judged whether the sequence has obvious periodicity and whether it has a strong autocorrelation relationship; the PACF graph shows the relationship between the lag period and the partial autocorrelation coefficient, with the horizontal axis being the lag period and the vertical axis being the partial autocorrelation coefficient. Conventional technology estimates the values ​​of p and q by manually observing the ACF image and the PACF image), thus avoiding the subjectivity of manual visualization; the cutoff points of the ACF image and the PACF image are often based on an empirical rule and require manual judgment, while the automatic threshold judgment estimates the values ​​of p and q based on clear statistical rules, which is suitable for automatic processing of large-scale time series data;

[0102] In order to make the parameters adaptive, the present invention uses the Bayesian optimization method to update ARIMA in real time; in order to solve the problem that the p and q values ​​in ARIMA are discrete integers, while the Bayesian optimization requires a continuous space, it is proposed to perform continuous processing on p and q, and in order to balance the model complexity and error, the objective function is weighted, which not only improves the adaptability of the model parameters, but also can obtain more accurate p and q values, avoids human intervention, and greatly improves the efficiency of modeling;

[0103] In order to effectively determine when the piezoelectric material is unusable, the present invention introduces a phased mechanism as the core of the design. Through the life assessment values ​​of the early, middle, late and unusable stages, accurate management of the entire material life cycle is achieved, and the visual assessment of material performance degradation is enhanced, providing a reliable and flexible solution for material management in high-reliability scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 is a flow chart of a method for detecting the availability of a piezoelectric material corresponding to Embodiment 1 of the present invention;

[0105] Figure 2 It is a flow chart of a piezoelectric material availability detection method corresponding to Example 2 of the present invention. DETAILED DESCRIPTION

[0106] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the protection scope of the present invention.

[0107] Example 1

[0108] This embodiment provides a method for detecting the availability of a piezoelectric material, including:

[0109] Obtain the historical data time series of the piezoelectric material within a set period, and obtain the historical data time series without abnormal values ​​after preprocessing;

[0110] Calculate the historical piezoelectric coefficient time series of the piezoelectric material based on the historical data time series without abnormal values;

[0111] Inputting the historical piezoelectric coefficient time series into the trained improved differential autoregressive moving average model to obtain the future piezoelectric coefficient time series of the piezoelectric material in the prediction period, wherein the improved differential autoregressive moving average model is obtained by improving the differential autoregressive moving average model based on the ambient temperature;

[0112] The life evaluation value of the piezoelectric material is calculated according to the future piezoelectric coefficient time series, and whether the piezoelectric material is usable is determined according to the life evaluation value.

[0113] The present invention significantly improves the accuracy and consistency of data by eliminating errors in historical data, adopts an improved differential autoregressive moving average model to take ambient temperature into consideration, greatly reduces the influence of ambient temperature on the piezoelectric coefficient, and improves the accuracy of prediction; designs a complete set of evaluation and judgment criteria to determine whether the piezoelectric material is usable; and combines the improvements in the above aspects to achieve accurate judgment on whether the piezoelectric material is usable.

[0114] Example 2

[0115] This embodiment provides a method for detecting the availability of a piezoelectric material, including:

[0116] Step 1: Collect the charge, force and ambient temperature from the historical data of piezoelectric materials for the past three years.

[0117] Keep applying a force of 0.01N to the piezoelectric material every day. At 12 noon every day, collect the charge and ambient temperature (the two should correspond one to one, that is, the charge and ambient temperature at the same time) and put the collected data into the database.

[0118] The historical data time series includes the charge time series, the ambient temperature time series and the force time series:

[0119] The charge amount time series is obtained by the following method: all charges are arranged in the order of sampling time from the earliest to the latest, so as to obtain the charge amount time series; the ambient temperature time series is obtained by the following method: all ambient temperatures are arranged in the order of sampling time from the earliest to the latest, so as to obtain the ambient temperature time series; the force time series is obtained by the following method: all forces are arranged in the order of sampling time from the earliest to the latest, so as to obtain the force time series. In this embodiment, the force is always kept at 0.01N.

[0120] Step 2: preprocess the collected data. The preprocessing includes checking and eliminating gross errors and checking and reducing systematic errors to obtain a data sequence without outliers.

[0121] Step 2.1: Sort the sampled charges according to the sampling time from the earliest to the latest, and obtain the charge time series r={y 1 ,y 2 ,y 3 ,y 4 ...y i ...y n}, where y 1 Indicates the charge on the first day of the historical data, y 2 Indicates the charge on the second day in the historical data, y 3 represents the charge on the third day in the historical data, y 4 represents the charge on the 4th day in the historical data, y i represents the charge on the i-th day in the historical data, y n It represents the charge on the nth day in the historical data, and n represents the maximum value of i, that is, the maximum number of days in the historical data.

[0122] Step 2.2: After the charge time series is arranged in chronological order (from earliest to latest sampling time), the gross errors in the charge time series are checked and eliminated using the global weighted error and the local weighted error.

[0123] By combining the characteristics of local weighted error and global weighted error, the local deviation and global deviation of data points are comprehensively considered. To this end, the present invention designs a calculation formula for the comprehensive error:

[0124] ;

[0125] Where i represents the time sequence number of the charge (i.e. the i-th day in the historical data), is a weight function, which is used to reflect the importance ratio of the local weighted error to the global weighted error. Its optimal value can be determined by cross-validation. In this embodiment, The optimal value of is 0.03; represents the local weighted error of the charge on the i-th day in the historical data, represents the global weighted error of the charge on the i-th day in the historical data, Represents the comprehensive error of the charge amount on the i-th day in the historical data.

[0126] The combined error Compared with the first threshold β = 3 (refer to the Rydt criterion), if >β, then The corresponding charge y on the i-th day in the historical data i It is a gross error and needs to be eliminated (the corresponding ambient temperature is also eliminated). Through this method, the gross errors in the charge quantity sequence are checked and eliminated one by one, and this step is repeated until the comprehensive errors of all charge quantities are less than or equal to the first threshold, thereby ensuring the accuracy and consistency of the final data. The sample size of the charge quantity sequence after eliminating gross errors and systematic errors is recorded as N.

[0127] The calculation method of the local weighted error is as follows:

[0128] By fitting small range data, the deviation of the charge amount from the local true value is reflected.

[0129] First, the charge in the charge time series is grouped into small windows of 5 days each, and the mth small window is defined as W m (m = 1, 2, 3, …).

[0130] With small window W m the median y m As the center point, a local fitting model is constructed to fit the local true value: ;

[0131] Among them, the weighting function The expression is:

[0132] ;

[0133] Weighting function Used to reflect the charge y on the i-th day in the historical data i and the median y of the small window m The degree of deviation, σ is the standard deviation of all charges in the small window, represents the local true value of the charge on the i-th day in the historical data obtained by fitting the charge on the j-th day, Represents the charge amount on the jth day in the historical data.

[0134] According to the fitting results of the local fitting model Calculate the local weighted error in charge:

[0135] .

[0136] The calculation method of the global weighted error is as follows:

[0137] This is done by calculating the global mean and standard deviation of the charge time series.

[0138] First, calculate the global mean and standard deviation of the charge time series:

[0139] ;

[0140] ;

[0141] in, represents the global mean of the charge sequence, represents the standard deviation of the charge series, and n represents the maximum number of days in the historical data.

[0142] The calculation formula of the global weighted error is further obtained:

[0143] .

[0144] Step 2.3: After the gross errors are eliminated, the Malikov criterion is used to test and reduce the systematic errors of the charge time series.

[0145] Calculate the residual sequence of the charge time series, and divide the residual sequence into two groups (the first residual sequence group and the second residual sequence group). Sum the two groups respectively, and subtract the two groups to determine whether there is a systematic error by whether the absolute value of the difference is equal to 0. The specific implementation process is as follows:

[0146] Here, the systematic error of the charge time series is reduced. The residual sequence v can be obtained by combining the global mean of the charge time series and the value of the charge sequence. The specific calculation formula is: v=r- .

[0147] According to the residual sequence v obtained above, the first ξ residual errors in the residual sequence v are added to obtain the sum of the first residual errors, and the last (n-ξ) residual errors are added to obtain the sum of the second residual errors, where ξ is the number of residual errors in the preset first residual sequence group, and the value of ξ is as follows: when n is an even number, ξ=n / 2; when N is an odd number, ξ=(n+1) / 2.

[0148] Subtract the first residual sequence group from the second residual sequence group to get the difference between them:

[0149] ;

[0150] Wherein, Δ represents the difference between the first residual sequence group and the second residual sequence group, represents the bth residual error in the residual sequence, and Located in the first residual sequence group after division, represents the cth residual error in the residual sequence, and Located in the second residual sequence group after division.

[0151] If the absolute value of Δ is not equal to 0, then there is a linear system error in the charge time series. If the absolute value of Δ is equal to 0, then there is no linear system error in the charge time series. The systematic error can be reduced by the offset method. The specific steps are to measure each measurement value (i.e., charge) twice, and take the average of the two measurements as the new charge, so as to reduce it. When reducing a certain charge, the corresponding ambient temperature is also reduced in the same way.

[0152] Step 3, based on the force value and the charge time series r after gross error elimination and systematic error reduction, the historical piezoelectric coefficient time series of the piezoelectric material is obtained according to the piezoelectric effect.

[0153] The expression for the piezoelectric effect is as follows:

[0154] ;

[0155] in, represents the charge on the i-th day in the charge time series r, d i represents the piezoelectric coefficient of the i-th day in the historical data of the piezoelectric material, and f represents the value of the force applied to the piezoelectric material (a constant of 0.01 in this embodiment).

[0156] The historical piezoelectric coefficients obtained in each time period (the first day, the second day, ..., the nth day in the historical data) are sorted in chronological order (in order from the beginning to the end) into a historical piezoelectric coefficient time series d(t) = {d 1 ,d 2 ,d 3 ,……,d i ,……d n}, d i Represents the piezoelectric coefficient of the i-th day in the historical data.

[0157] Step 4: Perform ADF (Augmented Dickey-Fuller Test, unit root test) on the historical piezoelectric coefficient time series obtained in step 3 to confirm the parameter ψ, and build an improved differential autoregressive moving average model (abbreviated as improved ARIMA) in combination with the corresponding ambient temperature series, and use Bayesian optimization for parameter estimation.

[0158] Step 4.1: Perform a unit root test on the historical piezoelectric coefficient time series obtained in step 3 to establish the parameter ψ of the differential autoregressive moving average model, where ψ is the difference order.

[0159] After the unit root test is performed on the historical piezoelectric coefficient time series in step 3, if the historical piezoelectric coefficient time series is a stationary series, the parameter ψ is taken as 0. If the historical piezoelectric coefficient time series is a non-stationary series, it is necessary to perform ψ-order differences on the historical piezoelectric coefficient time series until the series is stable and stationary. The parameter ψ takes the difference order in the difference process, and the difference order indicates the number of difference operations required to make the historical piezoelectric coefficient time series stable.

[0160] Step 4.2: After the parameter ψ is confirmed, the improved autoregressive moving average model (improved ARIMA) with two variables including ambient temperature is established.

[0161] Considering that the ambient temperature factor is also an important environmental factor affecting the piezoelectric coefficient of piezoelectric materials, at high temperatures, the polarization of piezoelectric materials will weaken, thereby reducing the piezoelectric coefficient. At low temperatures, the lattice may shrink, resulting in unstable piezoelectric response. Therefore, the ambient temperature is taken into account in the improved ARIMA. The expression of the improved ARIMA considering the ambient temperature is:

[0162] ;

[0163] Among them, p is the number of autoregressive terms (AR) in the improved ARIMA, which indicates how many past observations are used to predict the current value, q is the number of moving average (MA) terms in the improved ARIMA, which indicates the number of past forecast errors included in the forecast equation, and τ represents the τth day in the forecast period. is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, is a constant, The value of is −1.00×10 -16 , is the error term of the τth day in the forecast period, r ο is the autocorrelation coefficient, r ο The value of is 0.8111, is the correlation coefficient of ambient temperature, The value is 1.00×10 -10 , is the correlation coefficient of the error term, The value of is 0.3914, is the value of the piezoelectric coefficient of the hysteresis k order, is the error term of lag k, T is the ambient temperature, and k is the lag order.

[0164] The input of the improved ARIMA is the historical piezoelectric coefficient time series (composed of the piezoelectric coefficients within the historical set period), and the output is the future piezoelectric coefficient time series (composed of the piezoelectric coefficients within the future set period).

[0165] The lag order indicates the number of observations traced back when building a time series model. A lag order of 0 represents the original time series data without displacement, a lag order of 1 represents the time series data shifted one position to the left, and so on for multiple lag terms (the lag order is a number greater than 1).

[0166] Considering that the ambient temperature T has a seasonal cycle, we can consider characterizing T through a sinusoidal characteristic:

[0167] ;

[0168] Where W is the period, Represents the maximum value in the ambient temperature time series. The corresponding sine value is calculated based on the known τ and period W and substituted into the improved ARIMA, while avoiding the introduction of too many parameters.

[0169] Step 4.3: After the model is built, it is necessary to confirm the values ​​of the parameters p and q in the model, use the threshold to automatically determine the values ​​of the parameters p and q in ARIMA, and determine the approximate range of the combination of the values ​​of the parameters p and q as the prior knowledge for Bayesian optimization.

[0170] Step 4.3 specifically includes:

[0171] (1) Calculate the autocorrelation function (ACF) and partial autocorrelation function (PACF):

[0172] ;

[0173] ;

[0174] in, represents the predicted value of the piezoelectric coefficient on the τth day in the prediction period under the influence of all independent variables from lag 1 to k-1 order, represents the predicted value of the piezoelectric coefficient on the τ-kth day in the prediction period under the influence of all independent variables from lag 1 to k-1 order, is the value of the piezoelectric coefficient on the τth day in the period to be predicted, is the value of the piezoelectric coefficient on the τ-kth day in the period to be predicted, Cov() represents the covariance, Var() represents the variance, and z is the maximum value of τ, which is the last day in the period to be predicted.

[0175] (2) Set a threshold and compare the autocorrelation function and partial autocorrelation function corresponding to different lag orders with the threshold to confirm the values ​​of p and q.

[0176] The second threshold Y is set at a significance level of 0.05:

[0177] ;

[0178] Where N is the sample size of the charge sequence after eliminating gross errors and systematic errors.

[0179] The specific judgment rule is: if the absolute value of the autocorrelation coefficient or the partial autocorrelation function is greater than the second threshold value Y, it can be considered that the autocorrelation of this lag order is significant.

[0180] For ACF (autocorrelation function): if the autocorrelation function of a certain lag order exceeds the second threshold value Y, and the autocorrelation functions corresponding to all lag orders after this lag order are zero, then this lag order can be selected as the value of q.

[0181] For PACF (partial autocorrelation function): if the partial autocorrelation function of a certain lag order exceeds the second threshold Υ, and the partial autocorrelation functions corresponding to all lag orders after this lag order are zero, then this lag order can be selected as the value of p.

[0182] The several combinations of p and q values ​​thus determined are used as prior knowledge (here we only estimate several suitable combinations of p and q values, but the optimal value is not reached. Bayesian optimization is performed based on the obtained combinations of p and q values ​​to estimate the optimal values ​​of p and q).

[0183] Step 4.4: Use Bayesian optimization and combine the prior knowledge from step 4.3 to estimate the optimal values ​​of p and q in the improved ARIMA.

[0184] Define the objective function of optimization. The objective function is usually used to measure the quality of fitting. The performance index of the direct optimization model (MAPE, Mean Absolute Percentage Error) can more efficiently find the appropriate parameter combination. In order to simplify the calculation complexity and take into account the calculation time, the objective function is weighted. The objective function is defined as L(p, q, T) in the present invention, and its expression is:

[0185] ;

[0186] ;

[0187] Where N is the sample size of the charge time series after eliminating gross errors and systematic errors, is the value of the piezoelectric coefficient on the τth day in the period to be predicted, is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, represents the time required for ARIMA training and prediction, is the weight of MAPE, The value of is 1. To calculate the time weight, since the sample size in the present invention is large, the calculation time will significantly affect the model optimization process, so it is hoped that As small as possible, but the error and calculation time must be balanced. It can be taken as 0.05 (here 0.05 is an empirical estimate, which will be adjusted based on the final model performance test).

[0188] The discrete values ​​of p and q (i.e., the several combinations determined in step 4.3) are made continuous so that they can be applied to the Gaussian process (a process in Bayesian optimization).

[0189] In order to apply the discrete values ​​of p and q to the Gaussian process, the present invention defines two continuous variables and , and define a continuous search space , , Represents the upper limit of the search space. The object of operation in Bayesian optimization is and , so as to achieve the purpose of continuity and obtain the ideal and Then, use a rounding function to round the continuous values and Mapping back to discrete parameters p and q:

[0190] ;

[0191] ;

[0192] Among them, [] represents the rounding function.

[0193] The core of Bayesian optimization is to model continuous variables through Gaussian processes. However, in this embodiment, the parameters p and q of ARIMA are discrete integers. If optimization is performed directly in discrete space, the efficiency is low and it is difficult to use Gaussian processes. Therefore, it is considered to use mapping to convert discrete parameters into continuous parameters.

[0194] Since the values ​​of p and q are discrete, this continuation process is mainly to overcome the defects of discrete search space and avoid the difficulties of searching directly in discrete space.

[0195] (3) The general trend of the objective function near the sampling points is estimated through the Gaussian process, and further sampling is performed using the acquisition function (EI) within the range determined by p and q (i.e., the several combinations determined in step 4.3, which is the prior knowledge mentioned above). After the sampling is completed, the general trend is estimated again. The more sampling points there are, the closer the estimated trend is to the true trend of the objective function, thereby determining the location where the minimum value of the objective function occurs.

[0196] Using the prior knowledge obtained in step 4.3, we construct a Gaussian process model (a model representing the operations performed by the Gaussian process) with p, q, and T as inputs and L(p, q, T) as output. The expression of the Gaussian process model is:

[0197] ;

[0198] Where x is the parameter space of L(p,q,T), i.e., the input of the Gaussian process model, x=(p,q,T), i.e., x is used to represent p, q and T in the Gaussian process, m(x) is the mean function of x (usually 0), k(x,x′) is the RBF kernel function (Radial Basis Function kernel) of x and x′, x and x′ are two different inputs (i.e., at least one of the two sets of p, q and T represented by x and x′ has a different value), f(x) represents the output of the Gaussian process, i.e., L(p,q,T), and GP() means executing the Gaussian process on the parameters in the brackets.

[0199] The acquisition function (EI, Expected Improvement) selects the next set of p, q, and T values ​​that are most likely to minimize L(p,q,T) according to the Gaussian process model, defines the next set of p, q, and T values ​​that are most likely to minimize L(p,q,T) as a new point (p^*, q^*, T^*), calculates the corresponding L(p,q,T) value at the selected new point (p^*, q^*, T^*), and uses the value as the output of the objective function. The new point (p^*, q^*, T^*) and the corresponding L(p,q,T) are updated to the Gaussian process model, and the iteration begins. According to the updated Gaussian process model, the next set of parameters (p, q, and T values) are selected for prediction until the minimum value of the objective function does not change significantly after 3 iterations (the rate of change is less than 1%), and the p and q values ​​at this time are determined as their respective optimal values.

[0200] Step 5: Use the LB (Ljung-Box) test to improve the performance of ARIMA.

[0201] The null hypothesis of the LB test (white noise test) is that the residual sequence of the piezoelectric coefficients in the predicted period is a white noise sequence, that is, the data at each time point in the residual sequence of the piezoelectric coefficients in the predicted period is independent and there is no correlation. A white noise sequence is a random, disordered time series with a mean of 0, a constant variance, and no correlation between data at different time points. By calculating the test statistic (in this embodiment, the Q statistic is calculated), it is determined whether the residual sequence is a white noise sequence.

[0202] According to the latest proposed better Q statistic model (whose output is defined as Q LB ), whose expression is:

[0203] ;

[0204] Among them, Q LB represents the Q statistic, is the sample capacity of the residual sequence of the predicted value of the piezoelectric coefficient in the predicted period, that is, the length of the residual sequence of the predicted value of the piezoelectric coefficient. is the residual sequence of the predicted value of the piezoelectric coefficient The autocorrelation coefficient of order, represents the order of the autocorrelation coefficient, and k is the lag order.

[0205] The expression is:

[0206] ;

[0207] in, It is the residual between the predicted value of the piezoelectric coefficient on the τth day in the prediction period and the true value of the piezoelectric coefficient.

[0208] Because the residual sequence of the original piezoelectric coefficient prediction value is a white noise sequence, under the original hypothesis, Q LB Approximately obey the chi-square distribution with kpq degrees of freedom, obtain the chi-square distribution table and find the corresponding P (confidence probability) value from it, and compare it with the significance level (the significance level is generally 0.05, depending on the specific requirements, and 0.05 is taken in this embodiment). If the P value is less than the significance level, there is sufficient reason to reject the original hypothesis (that is, the residual sequence is not a white noise sequence, indicating that the model has not fully captured the correlation in the data, and it is necessary to reduce the first threshold in step 2.2 To further reduce outliers, The value of also needs to be further reduced to reduce the proportion of time). If the P value is greater than the significance level, the original hypothesis cannot be rejected (that is, the residual sequence of the piezoelectric coefficient prediction value is a white noise sequence, indicating that the model has fully captured the correlation in the data). At this time, the performance of the improved ARIMA is better and the subsequent steps can be carried out.

[0209] Step 6, first calculate the life evaluation value of the piezoelectric material on the day, then divide the life of the piezoelectric material into four stages according to the calculated life evaluation value of the piezoelectric material on the day, and finally evaluate the availability of the piezoelectric material based on the four divided stages and the attenuation rate of the piezoelectric material.

[0210] (1) Calculate the life evaluation value of the piezoelectric material on that day:

[0211] This embodiment establishes an evaluation model based on the life of a piezoelectric material. A core expression of the evaluation model based on the life of a piezoelectric material is:

[0212] ;

[0213] in, It indicates the life evaluation value of the piezoelectric material on that day. The physical meaning of the life evaluation value of the piezoelectric material is the ratio of the remaining life of the piezoelectric material to the total life of the piezoelectric material. is the theoretical value of the piezoelectric coefficient of the piezoelectric material, is the predicted value of the piezoelectric coefficient at time τ, is the minimum acceptable level of piezoelectric coefficient.

[0214] The specific calculation formula is:

[0215] ;

[0216] in, is the average value of the historical piezoelectric coefficient, is the standard deviation of the historical piezoelectric coefficients, is the safety factor (ranging from 1.5 to 2. Here, the acceptable level is considered to be relatively strict, but misjudgment is not allowed, so the safety factor is 1.8).

[0217] (2) Based on the life evaluation value of the piezoelectric material on that day, the life of the piezoelectric material is divided into four stages: In order to ensure the accuracy of the judgment, the life cycle of the material is divided into four stages: the early stage, the middle stage, the late stage and the unusable stage. The piezoelectric material goes through the early stage, the middle stage, the late stage and the unusable stage in sequence.

[0218] The above four stages are The specific division method is as follows: is greater than 0.8 and less than 1, it indicates that the piezoelectric material is in the early stage. In this embodiment, 0.8 is the third threshold value. is greater than 0.6 and less than 0.8, it indicates that the piezoelectric material is in the middle stage. In this embodiment, 0.6 is the fourth threshold value. If greater than 0.45 and less than 0.6, it indicates that the piezoelectric material is in the late stage. In this embodiment, 0.45 is the fifth threshold. If it is less than 0.45 and greater than 0.3, it indicates that the piezoelectric material is in an unusable stage.

[0219] (3) Based on the four divided stages and the attenuation rate of the piezoelectric material, the usability of the piezoelectric material is evaluated.

[0220] The decay rate is calculated as:

[0221] ;

[0222] in, represents the decay rate of the τth day in the period to be predicted, represents the decay rate of the τ-1th day in the period to be predicted, Indicates the decay rate.

[0223] The calculation formula of the attenuation rate is:

[0224] ;

[0225] in, is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, is the predicted value of the piezoelectric coefficient on the τ-1th day in the period to be predicted, Represents the decay rate of the τth day in the period to be predicted.

[0226] When the piezoelectric material is in the early stage and the attenuation rate is greater than 1.5 for three consecutive days, it indicates that the piezoelectric material is in the early attenuation abnormal stage.

[0227] When the piezoelectric material is in the mid-term stage and the attenuation rate is greater than 1.5 for three consecutive days, it indicates that the piezoelectric material is in the mid-term attenuation abnormal stage.

[0228] When the piezoelectric material has gone through one of the two stages, the early attenuation abnormal stage and the mid-term attenuation abnormal stage, a maintenance instruction is issued to the set maintenance unit. After receiving the maintenance instruction, the maintenance unit repairs the piezoelectric material to extend the service life of the piezoelectric material. At this time, the piezoelectric material can also be used in occasions with smaller loads.

[0229] When the piezoelectric material is in the late stage and the attenuation rate is greater than 2 for 3 consecutive days, it indicates that the piezoelectric material is in the late attenuation abnormal stage.

[0230] When the piezoelectric material has experienced a late attenuation abnormality stage, or the piezoelectric material has experienced both an early attenuation abnormality and a mid-term attenuation abnormality stage, the piezoelectric material is unavailable and an unavailable instruction is issued.

[0231] In summary, piezoelectric materials are unusable in any of the following three situations:

[0232] First, piezoelectric materials are in an unusable stage;

[0233] Second, piezoelectric materials have experienced an abnormal late attenuation stage;

[0234] Third, piezoelectric materials experience both early attenuation anomaly and mid-term attenuation anomaly stages.

[0235] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0236] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0237] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0238] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0239] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting the availability of a piezoelectric material, characterized in that: include: Obtain the historical data time series of the piezoelectric material within a set period, and obtain the historical data time series without abnormal values ​​after preprocessing; Calculate the historical piezoelectric coefficient time series of the piezoelectric material based on the historical data time series without abnormal values; Inputting the historical piezoelectric coefficient time series into the trained improved differential autoregressive moving average model to obtain the future piezoelectric coefficient time series of the piezoelectric material in the prediction period, wherein the improved differential autoregressive moving average model is obtained by improving the differential autoregressive moving average model based on the ambient temperature; Calculate the life evaluation value of the piezoelectric material according to the future piezoelectric coefficient time series, and judge whether the piezoelectric material is usable according to the life evaluation value; The expression of the improved difference autoregressive moving average model is: ； Among them, p is the number of autoregressive terms in the improved difference autoregressive moving average model, q is the number of sliding average terms in the improved difference autoregressive moving average model, τ represents the τth day in the forecast period, is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, is the set constant, is the error term of the τth day in the forecast period, r ο is the autocorrelation coefficient, is the correlation coefficient of ambient temperature, is the correlation coefficient of the error term, is the value of the piezoelectric coefficient of the hysteresis k order, is the error term of lag k, T is the ambient temperature, and k represents the lag order; The ambient temperature is characterized by the following formula: ； Where T is the ambient temperature, τ represents the τth day in the period to be predicted, and W is the set period. Represents the maximum value in the ambient temperature time series; The step of judging whether the piezoelectric material is usable according to the life evaluation value includes: like is greater than the set third threshold and less than 1, represents the life evaluation value of the piezoelectric material on that day, indicating that the piezoelectric material is in the early stage; if is greater than the set fourth threshold and less than the set third threshold, it indicates that the piezoelectric material is in the middle stage; if is greater than the set fifth threshold and less than the set fourth threshold, it indicates that the piezoelectric material is in the late stage; if is less than the set fifth threshold and greater than the set sixth threshold, it indicates that the piezoelectric material is in an unusable stage, wherein the values ​​of the third threshold, the fourth threshold, the fifth threshold, and the sixth threshold decrease in sequence; When the piezoelectric material is in the early stage, and the decay rate is greater than 1.5 within the consecutive set number of days, it indicates that the piezoelectric material is in the early stage of abnormal decay; when the piezoelectric material is in the middle stage, and the decay rate is greater than 1.5 within the consecutive set number of days, it indicates that the piezoelectric material is in the middle stage of abnormal decay; when the piezoelectric material is in the late stage, and the decay rate is greater than 2 within the consecutive set number of days, it indicates that the piezoelectric material is in the late stage of abnormal decay; When the piezoelectric material has experienced one of the early attenuation abnormality stage and the mid-term attenuation abnormality stage, a maintenance instruction is issued to a set maintenance unit; When the piezoelectric material has experienced the late attenuation abnormal stage, or the piezoelectric material has experienced both the early attenuation abnormal stage and the mid-term attenuation abnormal stage, or the piezoelectric material is in the unavailable stage, it indicates that the piezoelectric material meets the unavailable judgment condition, the piezoelectric material is unavailable, and an unavailable instruction is issued. In other cases, the piezoelectric material is available; The calculation formula of the decay rate is: ； in, represents the decay rate of the τth day in the period to be predicted, represents the decay rate of the τ-1th day in the period to be predicted, represents the decay rate; The calculation formula of the attenuation rate is: ； in, is the predicted value of the piezoelectric coefficient on the τ-1th day in the period to be predicted.

2. The method for detecting the availability of piezoelectric materials according to claim 1, characterized in that: The historical data includes force, charge and ambient temperature; The historical data time series includes charge time series, ambient temperature time series and force time series: The charge time series is obtained by the following method: all charges are arranged in the order of sampling time from earliest to latest, so as to obtain the charge time series; the ambient temperature time series is obtained by the following method: all ambient temperatures are arranged in the order of sampling time from earliest to latest, so as to obtain the ambient temperature time series; the force time series is obtained by the following method: all forces are arranged in the order of sampling time from earliest to latest, so as to obtain the force time series.

3. The method for detecting the availability of piezoelectric materials according to claim 2, characterized in that: The preprocessing includes eliminating gross errors and reducing systematic errors; The elimination of gross errors includes: The comprehensive error of each charge in the charge time series is calculated using the following formula: ； Where i represents the time sequence number of the charge, is a weight function that reflects the importance ratio of the local weighted error to the global weighted error. represents the local weighted error of the charge on the i-th day in the historical data, represents the global weighted error of the charge on the i-th day in the historical data, represents the comprehensive error of the charge on the i-th day in the historical data; If the comprehensive error of the charge amount on a certain day is greater than the set first threshold, the charge amount on that day is a gross error. The charge amount on that day is eliminated, and the corresponding ambient temperature and force on that day are also gross errors. The ambient temperature and force on that day are eliminated until the comprehensive errors of all charge amounts are less than or equal to the set first threshold. The local weighted error is calculated by the following method: The charge in the charge time series is divided into a plurality of small windows containing the same amount of charge, the local true value of the charge is calculated according to the median of the small windows, and the local weighted error is calculated according to the local true value; The local true value of the charge is calculated according to the median of the small window, using the following formula: ； ； ； in, Indicates the charge y used to reflect the i-th day in the historical data i and the median y of the small window m The weighted function of the degree of deviation, σ is the standard deviation of all charges in the small window, represents the local true value of the charge on the i-th day in the historical data obtained by fitting the charge on the j-th day in the historical data, represents the charge on the jth day in the historical data, exp() represents the natural exponential function, W m represents the mth small window, represents the local weighted error of the charge on the i-th day in the historical data; The global weighted error is calculated by the following method: ； ； ； in, represents the global mean of the charge time series, represents the standard deviation of the charge time series, n represents the maximum number of days in the historical data, and y i is the charge on the i-th day; The reduction of systematic errors includes: The residual sequence of the charge time series is calculated, and the residual sequence of the charge time series is divided into the first residual sequence group and the second residual sequence group. The difference between the first residual sequence group and the second residual sequence group is calculated by the following formula: ； Wherein, Δ represents the difference between the first residual sequence group and the second residual sequence group, represents the bth residual error in the residual sequence, and Located in the first residual sequence group after division, represents the cth residual error in the residual sequence, and Located in the second residual sequence group after division; in, is a constant, The value rule of is: when n is an even number, ξ=n / 2; when n is an odd number, ξ=(n+1) / 2; If the absolute value of Δ is not equal to 0, there is a systematic error in the charge time series. If the absolute value of Δ is equal to 0, there is no systematic error in the charge time series. The systematic error is reduced by the offset method.

4. The method for detecting the availability of piezoelectric materials according to claim 1, characterized in that: The values ​​of the number of autoregressive terms in the improved autoregressive moving average model and the number of moving average terms in the improved autoregressive moving average model are determined by the following method: The autocorrelation function ACF is calculated by the following formula: ； The partial autocorrelation function PACF is calculated using the following formula: ； in, represents the predicted value of the piezoelectric coefficient on the τth day in the prediction period under the influence of all independent variables from lag 1 to k-1 order, represents the predicted value of the piezoelectric coefficient on the τ-kth day in the prediction period under the influence of all independent variables from lag 1 to k-1 order, is the value of the piezoelectric coefficient on the τth day in the period to be predicted, is the value of the piezoelectric coefficient on the τ-kth day in the period to be predicted, Cov() means finding the covariance, Var() means finding the variance, and z means the maximum value of τ; If the autocorrelation function of a certain lag order exceeds the second threshold, and the autocorrelation functions corresponding to all lag orders after the lag order are zero, then the lag order is selected as the value of q; if the partial autocorrelation function of a certain lag order exceeds the second threshold, and the partial autocorrelation functions corresponding to all lag orders after the lag order are zero, then the lag order is selected as the value of p; By using the above method, any one of several combinations of p and q is determined as the values ​​of p and q.

5. The method for detecting the availability of piezoelectric materials according to claim 4, characterized in that: After obtaining several combinations of p and q, the best values ​​of p and q are selected using the Bayesian optimization method: The discrete values ​​of p and q are processed continuously so that they can be applied to Gaussian processes; Taking p, q and T as input and L(p,q,T) as output, we construct the following Gaussian process model: ； Where x refers to p, q and T, m(x) is the mean function of x, k(x,x′) is the RBF kernel function of x and x′, x and x′ are two different inputs, f(x) represents the output of the Gaussian process, GP() represents the execution of the Gaussian process, and L(p,q,T) represents the set objective function; The set acquisition function defines the next set of p, q and T values ​​that are most likely to minimize L(p,q,T) as a new point (p^*, q^*, T^*) according to the Gaussian process model, calculates the corresponding L(p,q,T) value at the new point (p^*, q^*, T^*) and uses the value as the output of the objective function, updates the new point (p^*, q^*, T^*) and the corresponding L(p,q,T) to the Gaussian process model, starts iteration, and continues to select the next set of p, q and T values ​​for iteration according to the updated Gaussian process model until the rate of change of the minimum value of the objective function after a preset number of iterations is less than the preset rate of change, and determines to use the p and q values ​​at this time as their respective optimal values; The objective function is calculated by the following formula: ； ； Among them, L(p,q,T) represents the set objective function, n represents the maximum number of days in the historical data, is the value of the piezoelectric coefficient on the τth day in the period to be predicted, N is the sample size of the charge sequence after eliminating the error, represents the time required to improve the training and prediction of the difference autoregressive moving average model, is the weight of the mean absolute percentage error MAPE, To calculate the time weight.

6. The method for detecting the availability of piezoelectric materials according to claim 1, characterized in that: Before inputting the historical piezoelectric coefficient time series into the trained improved difference autoregressive moving average model, the trained improved difference autoregressive moving average model is tested, including: The Q statistic of the trained improved difference autoregressive moving average model is calculated by the following formula: ； Among them, Q LB represents the Q statistic, is the sample capacity of the residual sequence of the predicted value of the piezoelectric coefficient in the predicted period, is the residual sequence of the predicted value of the piezoelectric coefficient The autocorrelation coefficient of order, represents the order of the autocorrelation coefficient, and k is the lag order; The residual sequence of the piezoelectric coefficient prediction value The order autocorrelation coefficient is calculated using the following formula: ； in, is the residual between the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted and the true value of the piezoelectric coefficient, and z represents the maximum value of τ; Q LB Obey the chi-square distribution with kpq degrees of freedom, p represents the number of autoregressive terms in the improved autoregressive moving average model, q represents the number of sliding average terms in the improved autoregressive moving average model, obtain the chi-square distribution table and find out the confidence probability value corresponding to the kpq degrees of freedom. If the confidence probability value is less than or equal to the set significance level, the trained improved autoregressive moving average model needs to be further optimized. If the confidence probability value is greater than the set significance level, the trained improved autoregressive moving average model passes the test; If the trained improved autoregressive moving average model needs to be further optimized, the first threshold is reduced, and the value of the calculation time weight is further reduced to reduce the proportion of time, until the trained improved autoregressive moving average model passes the test; Before inputting the historical piezoelectric coefficient time series into the trained improved differential autoregressive moving average model, the method further includes the steps of performing a unit root test on the historical piezoelectric coefficient time series and performing a differential operation, wherein the differential operation includes: If the historical piezoelectric coefficient time series is a stationary series, the difference order is 0; If the historical piezoelectric coefficient time series is a non-stationary series, the historical piezoelectric coefficient time series is subjected to ψ-order differences until the historical piezoelectric coefficient time series becomes stationary, where ψ represents the difference order.

7. The method for detecting the availability of piezoelectric materials according to claim 1, characterized in that: The life evaluation value of the piezoelectric material is calculated based on the future piezoelectric coefficient time series, and is calculated by the following formula: ； in, Indicates the estimated lifespan of the piezoelectric material on that day. is the theoretical value of the piezoelectric coefficient of the piezoelectric material, is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, is the minimum acceptable level of piezoelectric coefficient; in, The calculation formula is: ； in, is the average value of the historical piezoelectric coefficient, is the standard deviation of the historical piezoelectric coefficients, is the set safety factor.

8. A piezoelectric material availability detection device, characterized in that: The piezoelectric material availability detection device is used to implement the piezoelectric material availability detection method according to claim 1; The piezoelectric material availability detection device comprises: The historical data time series acquisition module is configured to: acquire the historical data time series of the piezoelectric material within a set period of time, and obtain the historical data time series without abnormal values ​​after preprocessing; The historical piezoelectric coefficient time series calculation module is configured to: calculate the historical piezoelectric coefficient time series of the piezoelectric material according to the historical data time series without abnormal values; The future piezoelectric coefficient time series prediction module is configured to: input the historical piezoelectric coefficient time series into the trained improved differential autoregressive moving average model to obtain the future piezoelectric coefficient time series of the piezoelectric material in the predicted period, wherein the improved differential autoregressive moving average model is obtained by improving the differential autoregressive moving average model based on the ambient temperature; The availability assessment module is configured to: calculate a life assessment value of the piezoelectric material according to a future piezoelectric coefficient time series, and determine whether the piezoelectric material is available according to the life assessment value; The expression of the improved difference autoregressive moving average model is: ； Among them, p is the number of autoregressive terms in the improved difference autoregressive moving average model, q is the number of sliding average terms in the improved difference autoregressive moving average model, τ represents the τth day in the forecast period, is the predicted value of the piezoelectric coefficient on the τth day in the period to be predicted, is the set constant, is the error term of the τth day in the forecast period, r ο is the autocorrelation coefficient, is the correlation coefficient of ambient temperature, is the correlation coefficient of the error term, is the value of the piezoelectric coefficient of the hysteresis k order, is the error term of lag k, T is the ambient temperature, and k represents the lag order; The ambient temperature is characterized by the following formula: ； Where T is the ambient temperature, τ represents the τth day in the period to be predicted, and W is the set period. Represents the maximum value in the ambient temperature time series; The step of judging whether the piezoelectric material is usable according to the life evaluation value includes: like is greater than the set third threshold and less than 1, represents the life evaluation value of the piezoelectric material on that day, indicating that the piezoelectric material is in the early stage; if is greater than the set fourth threshold and less than the set third threshold, it indicates that the piezoelectric material is in the middle stage; if is greater than the set fifth threshold and less than the set fourth threshold, it indicates that the piezoelectric material is in the late stage; if is less than the set fifth threshold and greater than the set sixth threshold, it indicates that the piezoelectric material is in an unusable stage, wherein the values ​​of the third threshold, the fourth threshold, the fifth threshold, and the sixth threshold decrease in sequence; When the piezoelectric material is in the early stage, and the decay rate is greater than 1.5 within the consecutive set number of days, it indicates that the piezoelectric material is in the early stage of abnormal decay; when the piezoelectric material is in the middle stage, and the decay rate is greater than 1.5 within the consecutive set number of days, it indicates that the piezoelectric material is in the middle stage of abnormal decay; when the piezoelectric material is in the late stage, and the decay rate is greater than 2 within the consecutive set number of days, it indicates that the piezoelectric material is in the late stage of abnormal decay; When the piezoelectric material has experienced one of the early attenuation abnormality stage and the mid-term attenuation abnormality stage, a maintenance instruction is issued to a set maintenance unit; When the piezoelectric material has experienced the late attenuation abnormal stage, or the piezoelectric material has experienced both the early attenuation abnormal stage and the mid-term attenuation abnormal stage, or the piezoelectric material is in the unavailable stage, it indicates that the piezoelectric material meets the unavailable judgment condition, the piezoelectric material is unavailable, and an unavailable instruction is issued. In other cases, the piezoelectric material is available; The calculation formula of the decay rate is: ； in, represents the decay rate of the τth day in the period to be predicted, represents the decay rate of the τ-1th day in the period to be predicted, represents the decay rate; The calculation formula of the attenuation rate is: ； in, is the predicted value of the piezoelectric coefficient on the τ-1th day in the period to be predicted.

Images