A Method for Predicting the Remaining Life of High-Speed Bearings in Offshore Wind Turbines

Abstract

A method for predicting the remaining life of high-speed bearings of an offshore wind turbine, comprising the following steps: Step 1: Install N vibration sensors at several sampling points on the bearing seats of the gearbox high-speed shaft on both sides of the offshore wind turbine corresponding to the gearbox body and on the generator bearing seat, respectively, to collect the vibration signals of each sampling point, where N is a natural number greater than 2; then convert the vibration signals from analog values to digital signals and perform preprocessing; Step 2: Convert the vibration signals of the sampling points preprocessed in Step 1 to obtain the time-domain characteristics and frequency-domain characteristics of the vibration signals of each sampling point, and save these characteristics as the vibration basic feature set of the high-speed bearings; Step 3: Calculate the monotonicity measure for the basic feature set obtained in Step 2 respectively, and determine the features with higher monotonicity as the sensitive features of bearing degradation; Step 4: Calculate the feature degradation rate based on the sensitive features obtained in Step 3 and construct an increasing health index; Step 5: Through exponential kernel space conversion, convert the health index into a kernel space index, determine the kernel space index threshold, and further determine the failure threshold of the health index; Step 6: Predict the change trend of the health index at future moments based on the self-constrained state space estimator, compare the health index prediction result with the failure threshold, and determine the remaining life of the bearing. The present invention can obtain the accurate remaining life of the high-speed bearings of the offshore wind turbine.

Description

Technical Field

[0001] The present invention belongs to the field of predicting the remaining life of high-speed bearings of offshore wind turbines based on vibration monitoring and analysis. Specifically, a new method for constructing a health index of high-speed bearings with strict monotonicity is proposed, which can realize the prediction of the remaining life of high-speed bearings only by using online historical data. Background Art

[0002] As one of the main moving components in the drive train of offshore wind turbines, once a rolling bearing fails, it will affect the production efficiency of the equipment and even cause the entire unit to shut down for maintenance, resulting in unnecessary economic losses. The high-speed bearings in the drive train of offshore wind turbines are in a working environment of high speed, high temperature and variable load for a long time, with a high failure rate and great difficulty in operation and maintenance. Therefore, it is very necessary to carry out fault diagnosis and life prediction for the high-speed bearings in the drive train of offshore wind turbines to ensure the safe and stable operation of the equipment, which is also the requirement and development trend of the current industrialization process. Compared with fault diagnosis, the research significance of predicting the remaining life of high-speed bearings of offshore wind turbines is greater. The prediction result of the remaining life can be used to analyze the future operation state and failure moment of the equipment, consider whether to prepare equipment components in advance, and can provide a basis for the preventive maintenance decision-making of offshore wind turbines, so as to extend the service time of relevant mechanical equipment, improve production efficiency, and prevent the occurrence of major safety accidents to a certain extent.

[0003] In the field of predicting the remaining life of rolling bearings, the current prediction methods can be mainly divided into three categories: physical failure models-based, data-driven models-based, and the fusion models of both. The method based on physical failure models characterizes the degradation trend of mechanical equipment by establishing corresponding physical or mathematical models, which requires combining specific prior knowledge or defect degradation equations and optimizing the model parameters according to the observed data. The data-driven method predicts the degradation trend of mechanical equipment through the collected operation data. This method does not require corresponding expert experience knowledge, so it is more practical for studying complex mechanical systems. The fusion model method combines the physical failure model and the data-driven model. The main idea is to correct the given empirical degradation model through real-time monitoring data. However, due to the combination of the two methods, the constructed model structure is relatively complex and difficult to solve, so this method has less research. The model-based method can utilize relevant experience knowledge and combine the potential fault information in the observed data at the same time, so the predicted result of the remaining life is more persuasive. However, due to the increasing complexity of mechanical equipment and the interference of the environment, the performance degradation process has strong randomness. Therefore, it is particularly important to construct a reasonable health index. At the same time, many existing remaining life prediction methods only perform well on test bench data or training sets, and often need to be trained with full life data first, which has certain limitations in actual engineering applications.

[0004] One of the key points in predicting the remaining life of high-speed bearings lies in the construction of health indicators. Most of the existing methods for constructing health indicators have the following deficiencies: ① When constructing health indicators using methods such as neural networks and support vector machines, vibration data over the entire life cycle are required for training, and the training results are often only applicable to bearings of the same failure type. ② Existing methods mainly construct health indicators based on the differences between the states of characteristic signals at different times and the state at the initial time, ignoring the fact that the degradation of bearings is an irreversible process, and a good health indicator should have a strict monotonic trend. ③ Some methods can avoid using full-life data for training, but the constructed indicators have poor predictability and are difficult to accurately predict the remaining life.

[0005] The existing technology CN110909509A bearing life prediction method specifically adopts some technical means, including the following steps. Step 1: Combine SDAE, InfoGAN, and LSGAN to construct InfoLSGAN, automatically extract interpretable and robust features from noise data, and solve the problem of gradient disappearance. Step 2: Train InfoLSGAN using an AC-based training algorithm to reduce the training time and accelerate the convergence speed. Step 3: According to the trained InfoLSGAN, use a softmax classifier to predict the remaining life of the bearings in the wind power gearbox for test samples. This method can predict the remaining life of the bearings in the wind power gearbox in the case of small samples. However, this algorithm is too complex and cannot be processed in the case of large samples, and it can only be applied for experimental purposes. The existing technology "Remaining Life Prediction of High-Speed Bearings in the Drive Train of Wind Turbines" (author Huang Yike) discloses a bearing life prediction method, which utilizes the sensitive features of signals, then obtains the failure threshold, compares the sensitive features with the failure threshold, and on the basis of analyzing the advantages and disadvantages of using particle filtering to solve prediction problems, proposes using the prediction result of the Gaussian process regression method as the observation value, and at the same time optimizes the prediction result through particle filtering, thereby constructing a PF-GPR fusion prediction model; determines the specific form and construction process of the state equation of the prediction model, the type of kernel function, and the calculation method of the remaining life.

[0006] The prediction methods of the existing technology have the problem of inaccurate prediction. Summary of the Invention

[0007] The present invention constructs a health indicator with good predictability through the historical vibration signals of high-speed bearings, and realizes a relatively accurate prediction of the remaining life of high-speed bearings.

[0008] The present invention proposes a recursive method for constructing the health index of high-speed bearings, which not only satisfies the true degradation law of bearings but also has good predictability. At the same time, in the life prediction stage, a self-constrained state space estimator is proposed, and the self-constrained curve is used to provide pseudo-observations for the iterative update of the state space estimator at future times, improving the prediction accuracy and obtaining a more accurate remaining life prediction result.

[0009] The present invention can be realized through the following technical solutions:

[0010] A method for predicting the remaining life of high-speed bearings of an offshore wind turbine, comprising:

[0011] Step 1: Install N vibration sensors at several sampling points on the bearing seats of the gearbox high-speed shaft on both sides of the offshore wind turbine corresponding to the gearbox housing and on the generator bearing seat, respectively, to collect the vibration signals of each sampling point, where N is a natural number greater than 2. Subsequently, convert the vibration signals from analog values to digital signals and perform preprocessing.

[0012] Step 2: Convert the vibration signals of the sampling points preprocessed in Step 1 to obtain the time-domain characteristics and frequency-domain characteristics of the vibration signals of each sampling point, and save these characteristics as the vibration basic feature set of the high-speed bearings.

[0013] Step 3: Calculate the monotonicity measure for the basic feature set obtained in Step 2 respectively, and determine the sensitive features of bearing degradation for the features with higher monotonicity.

[0014] Step 4: Calculate the feature degradation rate based on the sensitive features obtained in Step 3 and construct an increasing health index.

[0015] Step 5: Through exponential kernel space conversion, convert the health index into a kernel space index, determine the kernel space index threshold, and further determine the failure threshold of the health index.

[0016] Step 6: Predict the change trend of the health index at future times based on the self-constrained state space estimator, compare the health index prediction result with the failure threshold, and determine the remaining life of the bearing.

[0017] Further, in Step 3, the expression of the monotonicity measure:

[0018]

[0019]

[0020] In the formula, Mon is the monotonicity measure, and F(t) is the feature value at time t.

[0021] Further, in Step 4, the expression of the feature degradation rate:

[0022]

[0023]

[0024] Wherein, F i (t) is the eigenvalue of the i-th sensitive feature at time t, D i (t) is the feature degradation rate of the i-th sensitive feature at time t, I is the number of sensitive features at time t, D(t) is the feature degradation rate of all sensitive features at time t, t ERT is the end time of running-in;

[0025] Expression for constructing the health index:

[0026]

[0027] Wherein, F B is one of the basic features, representing the basic health state of the system; H(t) is the health index at time t.

[0028] Furthermore, in step five, the expression for exponential kernel space transformation:

[0029]

[0030] Wherein, p is a hyperparameter, K si (t) is the kernel space index at time t, t FDT is the starting point of degradation.

[0031] Furthermore, in step five, the expressions for the kernel space index threshold and the kernel space index residual:

[0032]

[0033]

[0034] J(x,y) = |time span of x i <y| (5)

[0035] Wherein, t (l) end is the last moment of Ksi (l) , Ksit is the kernel space index threshold, and Ksie is the kernel space index residual.

[0036] Furthermore, in step six, the expression for the self-constraint algorithm:

[0037]

[0038] Wherein, g(.) is the constraint curve, w is the parameter of the constraint curve, t is the calculation time, t now is the current time, yt is the historical observation at time t, is the fitting parameter of the constraint curve, τ is the future interval, is at time (t now +τ).

[0039] Furthermore, the self-constraint curve is incorporated into the state space estimator, and the self-constraint curve that conforms to the development trend of the historical data is obtained by regression using historical data, providing pseudo-observations for the iterative update of the state space estimator at future times, improving the prediction accuracy, and obtaining a more accurate remaining life prediction result.

[0040] The beneficial effects of the present invention are as follows:

[0041] 1) The existing health index construction methods do not take into account that the bearing degradation process is an irreversible process, so the constructed indexes do not have strict monotonicity. In view of this problem, the present invention proposes a new characteristic monotonicity trend evaluation index, which satisfies the true degradation situation of the bearing, has good trend and monotonicity at the same time, and is easy to predict.

[0042] 2) One of the difficulties in constructing the health index is to eliminate the influence of the dimensional difference between different features on the index construction result without normalizing the data. The present invention effectively extracts the degradation information contained in different features based on the feature degradation rate; on this basis, considering that the bearing degradation is a continuous process, a recursive health index construction method is established, and the mean value of the root mean square of the vibration signal in the normal operation stage is used as the initial health index value to characterize the normal state of different bearings; to solve the problem of difficult determination of the failure threshold, an exponential kernel space conversion method is proposed, and the failure threshold of the health index is determined based on the failure threshold of the kernel space index, so as to realize the prediction of the remaining life of the bearing only using historical vibration data, which has great engineering practical significance.

[0043] 3) On the basis of constructing a bearing degradation health index with good trend, based on the analysis of the advantages and disadvantages of the existing state space estimators, a self-constraint state space estimator prediction model is proposed. The main idea is to use the self-constraint curve to provide pseudo-observations for the iterative update of the state space estimator at future times, improve the prediction accuracy, and obtain a more accurate remaining life prediction result. Description of the Drawings

[0044] Figure 1 Flow chart of the method for predicting the life of the high-speed bearing of an offshore wind turbine based on the construction of a health index by fusing degradation features and a self-constraint state space estimator;

[0045] Figure 2 Time domain diagrams of the full-life vibration data of bearings 1, 2, 3, and 4 of an offshore wind turbine;

[0046] Figure 3 Example of constructing health indicators for high-speed bearings of offshore wind turbines: Results of constructing health indicators from the full-life data of bearings No. 1, No. 2, No. 3, and No. 4;

[0047] Figure 4 Example of constructing kernel space indicators for high-speed bearings of offshore wind turbines: Results of constructing kernel space indicators from the full-life data of bearings No. 1, No. 2, No. 3, and No. 4;

[0048] Figure 5 Flowchart of the remaining life prediction model;

[0049] Figure 6 Example of the remaining life prediction results for high-speed bearings of offshore wind turbines: Remaining life prediction results of bearings No. 1, No. 2, No. 3, and No. 4.

[0050] Embodiments of the present invention

[0051] The present invention provides a method for predicting the remaining life of high-speed bearings of offshore wind turbines. The following further explains the present invention in combination with the accompanying drawings and specific embodiments:

[0052] Figure 1 For the complete process of predicting the remaining life of high-speed bearings of offshore wind turbines, the technical solution of the present invention is mainly divided into six steps. The calculation formulas involved in these six steps can be found in the section of the invention content:

[0053] (1) By installing four vibration sensors at both ends of the high-speed shaft of the gearbox and the generator of the doubly-fed offshore wind turbine, measuring and collecting the structural vibration signals of bearings at different positions, converting the signals into digital signals, and performing preprocessing, Figure 2 Shows the vibration signal waveform diagram collected from the high-speed bearings of the unit.

[0054] (2) At each moment, calculate and obtain the time-domain characteristics and frequency-domain characteristics of the vibration signals, and save them as the vibration basic feature set of the high-speed bearings. This initial feature set is required to contain as much degradation information of the bearings as possible for subsequent construction of health indicators.

[0055] (3) Since the characteristic signals that can represent the degradation state of the bearings are often different at different moments, and considering that the bearing degradation is an irreversible process, monotonicity is an important property for the construction of health indicators. A sensitive feature selection method based on monotonicity measurement is proposed. The corresponding calculation formulas can be found in the section of the invention content.

[0056] (4) To eliminate the adverse effects of different numbers and dimensions of sensitive features at different moments on the construction of health indicators, calculate the feature degradation rate using sensitive features and construct an increasing health indicator.

[0057] (5) Considering that in actual applications, the true remaining life of the bearing is unknown and the degradation laws of different bearings are different, the failure thresholds of the constructed health indicators are often different. To obtain a more reasonable threshold determination method, an exponential kernel space transformation method is proposed, and the failure threshold of the health indicator is determined based on the failure threshold of the kernel space indicator.

[0058] (6) In the prediction stage, the present invention proposes a self-constrained state space estimator prediction model, which uses historical data regression to obtain a self-constrained curve and provides pseudo-observations for the iterative update of the state space estimator at future times, improving the prediction accuracy and obtaining a more accurate remaining life prediction result. The specific structure of the prediction model is as Figure 5 shown.

[0059] (7) Based on the constructed health indicator and the failure threshold determination method, the remaining life of the high-speed bearing can be predicted at different times. Figure 6 shows the true remaining life of the case bearing at different times and the prediction results of the present invention. It can be found that the present invention can better realize the remaining life prediction of the high-speed bearing of the offshore wind turbine, and only uses the historical data of the unit for each prediction, which has engineering application value.

[0060] In the above steps, the basic feature set in step two includes the features shown in Tables 1 and 2:

[0061] Table 1 Time-domain features

[0062]

[0063] Table 2 Frequency-domain features

[0064]

[0065] The calculation method of the monotonicity measure Mon in step three is as follows:

[0066]

[0067]

[0068] In the formula, Mon is the monotonicity measure, and F(t) is the eigenvalue at time t. The denominator of the Mon calculation formula is the number of [t1, t2] combinations that satisfy the condition 1 ≤ t1 < t2 ≤ t, and the numerator is the number of combinations that satisfy F(t2) ≥ F(t1) in the [t1, t2] combinations. The larger the calculated value of Mon, the more combinations that satisfy F(t2) ≥ F(t1), which also means that the monotonic increasing characteristic of this feature is stronger. Using Mon to select features with stronger monotonicity and screening them as bearing degradation sensitive features. It is found that having monotonicity in a certain section and selecting features with monotonicity can achieve a more accurate prediction of the state.

[0069] The monotonicity trend index can characterize the consistency between the change trend of a certain characteristic signal and the bearing performance degradation state, and its value is limited between [0, 1]. The closer it is to 1, the better the monotonicity trend of the characteristic.

[0070] Among them, the expression of the characteristic degradation rate in step four:

[0071]

[0072]

[0073] In the formula, F i (t) is the characteristic value of the i-th sensitive characteristic at time t, D i (t) is the characteristic degradation rate of the i-th sensitive characteristic at time t, I is the number of sensitive characteristics at time t, D(t) is the characteristic degradation rate of all sensitive characteristics at time t, t ERT is the time point when the running-in period ends and the healthy operation period begins. The numerator of the calculation formula of D i (t) represents the product of the characteristic value of the i-th sensitive characteristic at time t and at t ERT time, and the denominator represents the sum of the characteristic values of the i-th sensitive characteristic from t ERT time to t time. The determination method of t ERT is as follows: The frequency band corresponding to the point with the largest amplitude in the power spectrum undergoes a switch, including sudden disappearance, gradual transfer, etc., and the disappearance point or the transfer completion point is used as the end point of the running-in period.

[0074] Among them, the expression for constructing the health index in step four is:

[0075]

[0076] In the formula, F B is one of the basic characteristics, representing the basic health state of the system; H(t) is the health index at time t. Since the root mean square is a statistic that effectively characterizes the bearing vibration energy, the root mean square is used as F B .

[0077] Among them, the expression of the exponential kernel space transformation in step five:

[0078]

[0079] In the formula, p is the kernel space hyperparameter, Ksi(t) is the kernel space index at time t, t FDT is the degradation starting point. The specific meaning of the hyperparameter p is to describe the degree of non-linear scaling transformation of the distance between H(t) and H(t FDT ) in the kernel space. The determination of t FDT adopts the same method as the end point t of the running-in period ERTSame determination method: The frequency band corresponding to the point with the largest amplitude in the power spectrum undergoes a switch, including sudden disappearance, gradual transfer, etc., and the disappearance point or the completion point of the transfer is used as the starting point of degradation.

[0080] The determination method of the hyperparameter p is as follows. Suppose the kernel space indexes of a group of L bearings operating under similar working conditions are expressed as Ksi_s = (Ksi (1) , …, Ksi (l) , …, Ksi (L) ), and the expressions for the kernel space index threshold and the kernel space index residual are:

[0081]

[0082]

[0083] J(x, y) = |time span of x i <y| (11)

[0084] In the formula, t (l) end is the last moment of Ksi (l) , Ksi_t is the kernel space index threshold, and Ksi_e is the kernel space index residual.

[0085] For different values of the hyperparameter p, the corresponding kernel space index Ksi_s can be calculated, and the corresponding kernel space index threshold Ksi_t and the kernel space index residual Ksi_e can be calculated. Determine the hyperparameter p corresponding to the minimum Ksi_e, then the kernel space index threshold Ksi_t can be determined, and the failure threshold of the health index H(t) can be deduced by the inverse calculation formula of the exponential kernel space conversion.

[0086] Among them, the state space expression in step six:

[0087] a t = a t-1 + n a,t-1

[0088] x t = x t-1 + a t

[0089] y t = x t + n y,t (12)

[0090] In the formula, a t and x t are the system states at time t, y t is the observation at time t, n a,t-1 is the state noise, n y,tis the observation noise. The state space can iteratively estimate the system state.

[0091] Among them, the self-constrained state space estimator in Step 6 introduces the self-constrained algorithm into the state space estimator, and the state space estimator adopts the Bayesian framework iterative algorithm - the particle filter algorithm. The expression of the self-constrained algorithm:

[0092]

[0093] In the formula, g(.) is the constraint curve, w is the parameter of the constraint curve, t is the calculation time, t now is the current time, y t is the historical observation at time t, is the fitting parameter of the constraint curve, τ is the future interval, is the pseudo-observation at time (t now +τ). In this research case, a first-degree polynomial curve is selected as the constraint curve. Substituting different time intervals τ can obtain the pseudo-observations for a future period of time.

[0094] First, use the known historical observation data to obtain the constraint curve through least squares linear regression, which is consistent with the development trend of the historical data of the constraint curve. Then, the constraint curve provides pseudo-observation data for the prediction stage. Finally, use the pseudo-observations to iteratively update the state space estimator to obtain the predicted value of the system state at future times.

[0095] Among them, the remaining useful life expression in Step 6:

[0096] RUL = inf{t: x(t + t now ) > FT | t now} (14)

[0097] In the formula, t now is the current time, FT is the failure threshold, and x(t) is the system state at time t.

[0098] The system state at future times is predicted by the self-constrained state space estimator, and the failure threshold is set by the exponential kernel space conversion method. By comparing whether the predicted value of the system state at future times exceeds the failure threshold, the predicted value of the remaining useful life can be obtained.

[0099] After obtaining the predicted value of the remaining useful life, according to the past actual situation and combined with the product situation, the life prediction value is associated with the remaining life duration to obtain a reference remaining life duration.

[0100] Figures 2 to 6 Two examples of the remaining life prediction of the high-speed bearings of offshore wind turbines are given. Figure 2Shows the original vibration data of the high-speed bearings used in the implementation cases, where the No. 1 and No. 2 bearings are the bearings on the motor side of the high-speed shaft of the gearbox; the No. 3 and No. 4 bearings are the bearings at the drive end of the generator.

[0101] Figure 3 Shows the implementation results of the health index construction method of the present invention. It can be seen that the health index not only has strict monotonicity, but also has an obvious change trend and is easy to predict.

[0102] Figure 4 Is the drawing result of the kernel space index curve. From it, the kernel space index thresholds of the No. 1 and No. 2 bearings can be determined to be 0.097, and the kernel space index thresholds of the No. 3 and No. 4 bearings can be determined to be 0.091.

[0103] Figure 5 Shows the specific structure of the prediction model, corresponding to Figure 1 the self-constrained state space model in

[0104] Figure 6 Is a comparison chart of the predicted remaining life and the actual remaining life of the high-speed bearings in the drive train of an offshore wind turbine based on the present invention. From it, it can be statistically obtained that when the actual remaining life of the bearing is 10 days, the predicted average error is 1 - 2 days; when the actual remaining life of the bearing is 30 days, the predicted average error is 1 - 4 days. The prediction results are good and can meet the actual application requirements.

[0105] The above content is a further detailed description of the present invention in combination with specific preferred technical solutions. It cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention belongs, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for predicting the remaining life of a high-speed bearing of an offshore wind turbine, characterized in that Including: Step 1: Install N vibration sensors at several sampling points on the bearing seats of the gearbox housing corresponding to the bearings on both sides of the high-speed shaft of the offshore wind turbine and on the generator bearing seat to collect the vibration signals at each sampling point, where N is a natural number greater than 2; then convert the vibration signals from analog values to digital signals and perform preprocessing; Step 2: Convert the signals preprocessed in Step 1 to obtain the time-domain features and frequency-domain features of the vibration signals at each sampling point, and save these features to obtain the vibration basic feature set of the high-speed bearings; Step 3: Calculate the monotonicity measure for the time-domain features and frequency-domain features in the basic feature set obtained in Step 2 respectively, and determine the time-domain features and frequency-domain features with higher monotonicity as the sensitive features of bearing degradation; Step 4: Calculate the feature degradation rate based on the sensitive features obtained in Step 3 and construct an increasing health index; The expression of the feature degradation rate is: where F i (t) is the eigenvalue of the i-th sensitive feature at time t, D i (t) is the feature degradation rate of the i-th sensitive feature at time t, I is the number of sensitive features at time t, D(t) is the feature degradation rate of all sensitive features at time t, and t ERT is the end time of running-in; The expression for constructing the health index: where F B is one of the basic features, representing the basic health status of the system; H(t) is the health index at time t; Step 5: Through exponential kernel space transformation, convert the health index into a kernel space index, determine the kernel space index threshold, and then determine the failure threshold of the health index; Step 6: Predict the change trend of the health index at future moments based on the self-constrained state space estimator, compare the health index prediction result with the failure threshold, and determine the remaining life of the bearing.

2. The method according to claim 1, wherein: In Step 3, the expression of the monotonicity measure: In the formula, Mon is the monotonicity measure, and F(t) is the feature value at time t.

3. The method according to claim 2, wherein: In Step 3, after calculating the monotonicity measure, determine the intervals of the time-domain features and frequency-domain features with higher monotonicity as the sensitive features of bearing degradation.

4. The method according to claim 1, wherein: In Step 5, the expression of the exponential kernel space transformation: where p is a hyperparameter and K si (t) is the kernel space index at time t, and t FDT is the starting point of degradation.

5. The method according to claim 1, wherein: In Step 5, the expressions of the kernel space index threshold and the kernel space index residual: J(x,y) = |time span of x i <y| (5) where t (l) end is the last moment of Ksi (l) , Ksi_t is the nuclear space index threshold, and Ksi_e is the nuclear space index residual.

6. The method according to claim 1, characterized in that: In Step 6, the expression of the self-constrained algorithm: where g(.) is the constraint curve, w is the parameter of the constraint curve, t is the calculation time, and t now is the current time, y t is the historical observation at time t, is the fitting parameter of the constraint curve, τ is the future interval, and is the pseudo-observation at time (t now + τ).

7. The method according to claim 1, characterized in that: Integrate the self-constrained curve into the state space estimator, and use historical data regression to obtain a self-constrained curve that conforms to the development trend of historical data, providing pseudo-observations for the iterative update of the state space estimator at future moments.

Images