Fault classification diagnosis method based on set membership estimation ellipsoid domain internally tangent
Through the ellipsoidal incision method based on crew estimation, the problem that existing fault diagnosis methods rely on accurate mathematical models is solved, real-time hierarchical diagnosis of fault severity is achieved, and the accuracy and real-time nature of fault detection are improved.
Patent Information
- Application Number
- CN202411818762.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-11
- Publication Date
- 2025-03-25
AI Technical Summary
The existing fault diagnosis methods require accurate system mathematical models, which are large in calculation and low in real time, which is not conducive to the operation and maintenance of actual systems.
Using the ellipsoidal incision method based on crew estimation, the crew observer is designed to obtain the predicted estimate set and estimated set of system state amounts, judge the severity of the fault by intersecting the ellipsoid and the area of the incision circle, and determine the fault grading threshold using machine learning.
It realizes the judgment of whether a fault occurs, and can diagnose the severity of the fault in real time, improving the accuracy and real-time nature of fault detection.
Smart Images
Figure CN119690041A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of fault diagnosis of control systems, and in particular to a fault classification diagnosis method based on set membership estimation and ellipsoid domain inscription. Background Art
[0002] With the continuous promotion and use of green energy, photovoltaic power generation technology has been more widely used, and at the same time, the problem of fault detection has also received in-depth attention.
[0003] Fault detection is the process of using various status information of the system being diagnosed and various existing knowledge to conduct comprehensive information processing, and finally obtain a comprehensive evaluation of the system operation status and fault conditions. Through fault diagnosis technology, it is possible to monitor the system operation status, determine whether a fault has occurred, and determine the time, location, size and type of the fault.
[0004] For equipment, its dynamic characteristics are often reflected in the changes of output and input data. Therefore, the implementation evaluation of the equipment and real-time detection of fault information can be achieved by analyzing the large amount of operating data obtained by the online monitoring system.
[0005] There are many research results on the problem of system fault diagnosis, which are mainly divided into analytical model-based methods and data-driven methods. First, the fault diagnosis method based on analytical models compares the measurable information of the object to be diagnosed with the system prior information expressed by the model to generate residuals. Further analysis of the residuals can realize fault diagnosis: when there is no fault, the residual is equal to zero or approximately zero; when a fault occurs in the system, the residual should deviate significantly from zero. The prerequisite for fault detection using this method is to obtain an accurate mathematical model of the system, which has certain limitations for actual power grid systems.
[0006] As the scale of power grids continues to expand, data-driven methods are widely used, including principal component analysis, partial least squares, subspace identification, etc. The subspace identification method uses input and output data to obtain the spatial state model of the system, and generates a residual space vector through the predicted value and actual value obtained by the spatial state model to diagnose system faults.
[0007] In recent years, with the continuous development and application of set membership estimation methods, it has also provided certain research results in fault diagnosis. In actual systems, due to the existence of parameter uncertainty, disturbances and noise, it is not easy to obtain an accurate model. However, the use of set membership estimation methods for research has lower requirements on the model of disturbances, noise, etc. of the system. Existing research results obtain the prediction set and estimation set of the system state quantity by designing a set membership observer, and judge whether the two sets intersect to determine whether the system has a fault.
[0008] Therefore, when faced with the problem of equipment fault diagnosis in the control system, how to solve the problem that traditional detection methods mainly stay at the stage of judging whether a fault has occurred, resulting in no judgment on the specific extent of the system fault has become a technical problem that technical personnel in this field urgently need to solve. Summary of the invention
[0009] In view of the above-mentioned defects of the prior art, the present invention provides a fault classification diagnosis method based on set membership estimation of the inscribed ellipsoid domain, and its purpose is to solve the problem that the existing fault diagnosis methods usually require accurate mathematical models of the system, disturbance and noise, etc., and have large amount of calculation and low real-time performance, which is not conducive to the operation and maintenance of the actual system.
[0010] To achieve the above object, the present invention discloses a fault classification diagnosis method based on set membership estimation ellipsoid domain inscription, comprising the following steps:
[0011] Step 1: Design a set membership observer to obtain the predicted estimation set and the estimated set of the state quantity of the system to be diagnosed;
[0012] Step 2: using an ellipsoid to describe the outer boundaries of the predicted estimation set and the estimation set, and obtaining the intersection of the two sets through the intersection between the ellipsoid of the predicted estimation set and the ellipsoid of the estimation set;
[0013] Step 3, making an inscribed circle in the intersection of the ellipsoid of the prediction estimation set and the ellipsoid of the estimation set, and calculating the area of the inscribed circle;
[0014] Step 4: Use a large amount of data obtained from prior experiments to adopt machine learning methods to determine the number of fault classification judgments and specific thresholds.
[0015] Preferably, in step 1, the prediction estimator of the prediction estimation set is specifically as follows:
[0016]
[0017] in, represents the estimated value of the system output voltage data y(t); J represents the parameter of the prediction estimator to be determined;
[0018] Then the prediction set is:
[0019]
[0020] The measurement estimator of the estimation set is as follows:
[0021]
[0022] Wherein, L1, L2 and L3 are the parameters of the measurement estimator to be determined, y(t) is the actual output of the system when a fault occurs, and the fundamental wave data of the system output voltage when a fault occurs is assumed to be described as follows:
[0023] y(t)=L1x f (t)+L2x p (t)+L3y p (t)+g(t);
[0024] Where g(t) represents a system failure; f represents a future time, and p represents a past time, which respectively represent the input and output data of the system at the past time or the future time;
[0025] when
[0026] Then the estimation set is:
[0027]
[0028] Preferably, in step 4, by collecting a large amount of priori data results and combining with a machine learning algorithm, the number of classifications m of the faults of the system to be diagnosed is determined.
[0029] And the threshold values γ for classification judgment k (k=1,2,...m);
[0030] When the area of the inscribed circle is greater than or equal to γ1, the system to be diagnosed has no fault;
[0031] When the area of the inscribed circle is less than γ1 and greater than or equal to γ2, a minor fault occurs in the system to be diagnosed;
[0032] When the area of the inscribed circle is less than γ2 and greater than γ3, a medium fault occurs in the system to be diagnosed;
[0033] When the area of the inscribed circle is smaller than γ3 and larger than γ4, a serious fault occurs in the system to be diagnosed.
[0034] Beneficial effects of the present invention:
[0035] The present invention targets various types of faults that may occur at different times and locations. Compared with the traditional fault monitoring based on set membership estimation method, the present invention can not only diagnose whether a fault occurs, but also judge the severity of the fault.
[0036] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 Solar energy diagram.
[0038] Figure 2 A flow chart showing an embodiment of the present invention is shown.
[0039] Figure 3 A schematic diagram showing a state in which the ellipsoid of the prediction estimation set and the ellipsoid of the estimation set substantially coincide with each other in one embodiment of the present invention.
[0040] Figure 4 A schematic diagram showing a state in which the ellipsoid of the prediction estimation set and the ellipsoid of the estimation set are separated in one embodiment of the present invention.
[0041] Figure 5 A schematic diagram showing a state in which the ellipsoid of the prediction estimation set and the ellipsoid of the estimation set are severely separated in one embodiment of the present invention. DETAILED DESCRIPTION
[0042] Example
[0043] like Figure 1 As shown, the system is composed of a distributed DC power supply, an inverter, a filter, a local load RLC, etc. The present invention uses the output voltage values before and after the fault of the system to design a set membership filter to obtain the prediction set and the estimation set of the system state quantity, thereby realizing fault detection.
[0044] like Figures 2 to 5 As shown, the fault classification diagnosis method based on set membership estimation and ellipsoid domain inscription includes the following steps:
[0045] Step 1: Design a set membership observer to obtain the predicted estimation set and the estimated set of the state quantity of the system to be diagnosed;
[0046] Step 2: Use an ellipsoid to describe the outer boundaries of the prediction estimation set and the estimation set, and obtain the intersection of the two sets through the intersection between the ellipsoid of the prediction estimation set and the ellipsoid of the estimation set;
[0047] Step 3, draw an inscribed circle in the intersection of the ellipsoid of the prediction estimation set and the ellipsoid of the estimation set, and calculate the area of the inscribed circle;
[0048] Step 4: Use a large amount of data obtained from prior experiments to adopt machine learning methods to determine the number of fault classification judgments and specific thresholds.
[0049] The present invention uses conventional geometric figures to describe the outer boundaries of the state quantity prediction set and the estimation set obtained by the set membership estimation method, such as the ellipsoid used in this article. When the system is not faulty, the prediction set ellipsoid of the state quantity and the estimation set ellipsoid should completely overlap; when the system fails, the prediction set ellipsoid of the state quantity and the estimation set ellipsoid deviate and no longer completely overlap, and even completely separate in severe cases. The intersection of the two ellipsoids is analyzed. Since the intersecting part is an irregular shape, it is not convenient to directly calculate the intersection area. Therefore, an inscribed circle is taken for the intersection of the two, and the area of the inscribed circle is regarded as the intersection area.
[0050] After obtaining a large amount of intersection area data of the prediction set and the estimation set through prior experiments, the present invention clusters the data through a machine learning method to determine the number of fault diagnosis levels and the specific threshold size, so that real-time hierarchical diagnosis of system faults can be achieved in subsequent system operation.
[0051] like Figure 3 to Figure 4 As shown in the figure, when the system does not fail, the predicted estimation set of the state quantity of the system to be diagnosed basically coincides with the ellipsoid of the estimation set;
[0052] When a system failure occurs, the predicted estimation set of the state quantity of the system to be diagnosed gradually deviates and separates from the ellipsoid of the estimation set.
[0053] The present invention determines the severity of the system failure by analyzing the intersection area of two ellipsoids. Generally speaking, the shape of the intersection is an irregular figure, and the area is difficult to directly determine by a formula. Therefore, it is further proposed to make an inscribed circle of the intersection, and indirectly represent the area of the intersection by the area of the inscribed circle.
[0054] In this embodiment, the state information is first analyzed. Assuming that the input voltage fundamental wave data is x, the output voltage fundamental wave data is y, and the prediction time domain is t, if the system has input data {x0, x1, ..., x2} at time {0, 1, ..., 2t-1}, 2t-1}, and the output data {y0,y1,...,y t-1}, then the set membership filter is as follows:
[0055] In step 1, the prediction estimator of the prediction estimation set is as follows:
[0056]
[0057] in, represents the estimated value of the system output voltage data y(t); J represents the parameter of the prediction estimator to be determined;
[0058] Then the prediction set is:
[0059]
[0060] The measurement estimator of the estimation set is as follows:
[0061]
[0062] Wherein, L1, L2 and L3 are the parameters of the measurement estimator to be determined, y(t) is the actual output of the system when a fault occurs, and the fundamental wave data of the system output voltage when a fault occurs is assumed to be described as follows:
[0063] y(t)=L1x f (t)+L2x p (t)+L3y p (t)+g(t);
[0064] Where g(t) represents a system failure; f represents a future time, and p represents a past time, which respectively represent the input and output data of the system at the past time or the future time;
[0065] when
[0066] Then the estimation set is:
[0067]
[0068] In some embodiments, in step 4, by collecting a large amount of prior data results and combining with a machine learning algorithm, the number of classifications m of the faults of the system to be diagnosed is determined.
[0069] And the threshold values γ for classification judgment k (k=1,2,...m);
[0070] When the area of the inscribed circle is greater than or equal to γ1, the system to be diagnosed has no fault;
[0071] When the area of the inscribed circle is less than γ1 and greater than or equal to γ2, a minor fault occurs in the system to be diagnosed;
[0072] When the area of the inscribed circle is less than γ2 and greater than γ3, a medium fault occurs in the system to be diagnosed;
[0073] When the area of the inscribed circle is smaller than γ3 and larger than γ4, a serious fault occurs in the system to be diagnosed.
[0074] The algorithm flow of designing a set membership observer to obtain the prediction set and estimation set of the system output is as follows:
[0075] Input: The fundamental wave data of the input voltage of the photovoltaic grid-connected power generation system at the past time xp (t), voltage fundamental wave data x to be input at the future time f (t), the output voltage fundamental wave data y of the system at the past moment p (t);
[0076] Output: predicted value of output voltage y(t) of photovoltaic grid-connected power generation system Prediction set Y(t+1|t), estimated value Estimation set Y(t), the inscribed circle area of the intersection area S(t), the number of fault classification judgments m, and the specific threshold γ k (k=1,2,...m).
[0077] 1. Inputx p (t),x f (t),y p (t),sett=0;
[0078] 2. CalculateL1, L2, L3 by set-membership toolbox;
[0079] 3. Calculate the center of the ellipsoid Y(t+1|t)by(1), calculate the center of the ellipsoid Y(t)by(2);
[0080] 4. Calculate S(t);
[0081] 5. If t=t end ,then stop; otherwise t=t+1and go to step2;
[0082] 6. Calculate m and γ k (k=1,2,...m)by machine learning.
[0083] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A fault classification diagnosis method based on set membership estimation and ellipsoid domain inscription; characterized in that: The steps include: Step 1: Design a set membership observer to obtain the predicted estimation set and the estimated set of the state quantity of the system to be diagnosed; Step 2: using an ellipsoid to describe the outer boundaries of the predicted estimation set and the estimation set, and obtaining the intersection of the two sets through the intersection between the ellipsoid of the predicted estimation set and the ellipsoid of the estimation set; Step 3, making an inscribed circle in the intersection of the ellipsoid of the prediction estimation set and the ellipsoid of the estimation set, and calculating the area of the inscribed circle; Step 4: Use a large amount of data obtained from prior experiments to adopt machine learning methods to determine the number of fault classification judgments and specific thresholds.
2. The fault classification diagnosis method based on set membership estimation ellipsoid domain inscription according to claim 1 is characterized in that: In step 1, the prediction estimator of the prediction estimation set is as follows: in, represents the estimated value of the system output voltage data y(t); J represents the parameter of the prediction estimator to be determined; Then the prediction set is: The measurement estimator of the estimation set is as follows: Wherein, L1, L2 and L3 are the parameters of the measurement estimator to be determined, y(t) is the actual output of the system when a fault occurs, and the fundamental wave data of the system output voltage when a fault occurs is assumed to be described as follows: y(t)=L1x f (t)+L2x p (t)+L3y p (t)+g(t); Where g(t) represents a system failure; f represents a future time, and p represents a past time, which respectively represent the input and output data of the system at the past time or the future time; when Then the estimation set is:
3. The fault classification diagnosis method based on set membership estimation ellipsoid domain inscription according to claim 1 is characterized in that: In step 4, by collecting a large amount of prior data results and combining them with machine learning algorithms, the number of fault classifications m of the system to be diagnosed is determined. And the threshold values γ for classification judgment k (k=1,2,...m); When the area of the inscribed circle is greater than or equal to γ1, the system to be diagnosed has no fault; When the area of the inscribed circle is less than γ1 and greater than or equal to γ2, a minor fault occurs in the system to be diagnosed; When the area of the inscribed circle is less than γ2 and greater than γ3, a medium fault occurs in the system to be diagnosed; When the area of the inscribed circle is smaller than γ3 and larger than γ4, a serious fault occurs in the system to be diagnosed.