A method and system for detecting failure of a sensor of a petroleum refining plant

Abstract

The application discloses a kind of failure detection method and system of petroleum refining plant sensor, belong to sensor detection technical field;The sensor to be detected of petroleum refining plant is represented using partial differential equation model, and the nonlinear observer of the sensor to be detected is constructed based on the partial differential equation model of the sensor to be detected;Its output is observed by observer, and failure detection is carried out based on the detection residual between observation value and corresponding actual output;The nonlinear observer constructed can cope with the multiple-input multiple-output situation with position interference, and it satisfies that detection residual is ultimately bounded under the condition that the sensor to be detected is healthy, it is easy to determine the preset threshold for failure judgment based on detection residual, and it is also unnecessary to reduce order partial differential equation model to ordinary differential equation for solving, avoid false alarm and miss report situation possibly caused by model reduction, can accurately carry out failure detection to petroleum refining plant sensor.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of sensor detection, and more particularly relates to a failure detection method and system for sensors of a petroleum refining device. BACKGROUND

[0002] In petroleum chemical production, sensors are used to measure and collect process parameters of each production process and upload data to the control center to ensure safe operation and efficient production of the device. The petroleum chemical production environment is usually exposed outdoors and has the characteristics of high temperature, dust and high electromagnetic interference, which can easily lead to sensor failure during long-term operation, such as increased measurement data error, inability to detect and report equipment abnormalities in a timely manner, and thus reduce the reliability and service life of the equipment. At the same time, due to cost constraints, the safety standards of sensors in petroleum chemical production are usually low, and are vulnerable to network attacks and thus lead to failure. Therefore, failure detection of sensors of a petroleum refining device can timely detect failed or about-to-fail sensors, and replace or secure them to reduce adverse effects, which is an important means to ensure the safe operation of a petroleum refining device.

[0003] A petroleum refining device is a typical hybrid system, which is characterized by a lumped parameter system (LPS) as a whole, and is characterized by a distributed parameter system (DPS) in some specific sensor sensing processes, such as heat conduction, mass transfer and fluid flow processes. The existing method usually represents the partial differential equation model of the DPS with a set of infinite ordinary differential equations. In the process of judging whether the sensor is failed, in order to adapt to the existing linear system observer, the partial differential equation model is reduced in order to derive the ordinary differential equation of its finite dimension for solving, which has representation error, and thus cannot accurately detect the failure of the sensor of the petroleum refining device. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the present application provides a failure detection method and system for sensors of a petroleum refining device to solve the technical problem that the prior art cannot accurately detect the failure of the sensor of the petroleum refining device.

[0005] To achieve the above purpose, in a first aspect, the present application provides a failure detection method for sensors of a petroleum refining device, comprising:

[0006] obtaining current detection data of a to-be-detected sensor in a petroleum refining device; wherein the current detection data comprises state data of the to-be-detected sensor at a current time and all times before the current time;

[0007] The partial differential equation model of the to-be-detected sensor is used to construct a nonlinear observer of the to-be-detected sensor, the current detection data is input into the observer for solving, and an observation value is obtained; a difference between the observation value and an actual output of the to-be-detected sensor is calculated, and a detection residual is obtained;

[0008] It is judged whether the detection residual is greater than a preset threshold value, if yes, it is determined that the to-be-detected sensor is invalid, otherwise, it is determined that the to-be-detected sensor is not invalid;

[0009] The equation of the observer is:

[0010]

[0011] The boundary condition thereof is:

[0012]

[0013] The observation value thereof is:

[0014]

[0015] Wherein, is a state estimation vector of the to-be-detected sensor at position x at time t; 0 is a Lipschitz continuous nonlinear vector function, used to describe the dynamic characteristics of the observed to-be-detected sensor; is an error feedback control term; P1(x, t) is a first gain matrix of the observer, used to adjust the feedback of the to-be-detected sensor to the state estimation error; is an actual output of the to-be-detected sensor at time t; is a full rank matrix; n is the state data dimension of the to-be-detected sensor; represents the derivative of the state at position 0; P 10 (t) is a second gain matrix of the to-be-detected sensor, used to adjust the boundary error feedback of the observer at the boundary; u(t) = [u1(x, t), u2(x, t), …, u n (x, t)] T ∈ R n represents a control input vector; u i (x, t) is the control input of the i th state data acting on the to-be-detected sensor at position x at time t.

[0016] Further preferably, the preset threshold value is αρ; wherein, ρ is a detection threshold value, and its expression is:

[0017]

[0018] Where 1.5≤α≤2.5; l is the spatial length of the sensor to be detected; This represents the upper limit of error for the partial differential equation model of the sensor to be detected. is the upper limit of the perturbation of the partial differential equation model of the sensor to be detected; c is a constant in the partial differential equation model of the sensor to be detected.

[0019] More preferably, α = 2.

[0020] More preferably, the partial differential equation model of the sensor to be detected is:

[0021]

[0022] Its boundary conditions and output are defined as follows:

[0023] v x (0,t)=Qv(0,t), v(l,t)=u(t)

[0024] y(t)=v(0,t)

[0025] Among them, v(x,t)=[v1(x,t),v2(x,t),…,v n (x,t)] T ∈[L2(0,l)] n This represents the state vector of the sensor to be detected; v i (x,t) represents the i-th state data at position x of the sensor to be detected at time t; i = 1, 2, ..., n; n is the dimension of the state data of the sensor to be detected; L2(0,l) is the quadratic integrable function space defined on (0,l); d(x,t) = [d1(x,t), d2(x,t), ..., d n [x,t)]∈R n d is the perturbation vector; i (x,t) represents the disturbance received by the i-th state at position x of the sensor to be detected at time t; u(t) = [u1(x,t), u2(x,t), ..., u n (x,t)] T ∈R n Represents the control input vector; u i (x,t) is the control input for the i-th state data at position x of the sensor to be detected at time t; It is a full-rank square array; y(t) is a Lipschitz continuous nonlinear vector function used to describe the dynamic characteristics of the actual sensor under test; y(t) is the actual output of the sensor under test at time t.

[0026] Further preferably, the constraint condition of the partial differential equation model of the to-be-detected sensor comprises: for all x and t≥0, the following condition is satisfied: wherein, is a preset constant.

[0027] Further preferably, the above obtaining the current detection data of the to-be-detected sensor in the petroleum refining device comprises: collecting the current detection data of the to-be-detected sensor from the petroleum refining device and performing preprocessing; wherein the preprocessing comprises: denoising operation and normalization operation.

[0028] In a second aspect, the present application provides a failure detection system of a sensor of a petroleum refining device, comprising: a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the failure detection method provided in the first aspect of the present application.

[0029] In a third aspect, the present application provides a petroleum refining device, comprising: a sensor and the failure detection system provided in the second aspect of the present application.

[0030] In a fourth aspect, the present application further provides a computer readable storage medium, which comprises a stored computer program, wherein when the computer program is run by a processor, the device where the storage medium is located is controlled to execute the failure detection method provided in the first aspect of the present application.

[0031] Overall, through the above technical solutions conceived by the present application, the following beneficial effects can be achieved:

[0032] 1. The present application provides a failure detection method of a sensor of a petroleum refining device, which represents the to-be-detected sensor of the petroleum refining device by using a partial differential equation model, and constructs a nonlinear observer of the to-be-detected sensor based on the partial differential equation model of the to-be-detected sensor; then the output of the observer is observed, and the detection residual between the observation value and the corresponding actual output is used for failure detection; the nonlinear observer constructed by the present application compensates for the disturbance at different positions through the gain matrix, that is, even if the sensor at different positions is disturbed, the position difference can also be responded through feedback gain, so as to keep the residual stable, which can cope with the multiple-input multiple-output situation with position disturbance, and the detection residual is ultimately bounded under the condition that the to-be-detected sensor is healthy, which is easy to determine the preset threshold for failure judgment based on the detection residual, and at the same time, it is also unnecessary to reduce the partial differential equation model to an ordinary differential equation for solving, avoiding the false alarm and missed alarm caused by model reduction, and the failure of the sensor of the petroleum refining device can be accurately detected.

[0033] 2. Further, the failure detection method provided by the present application determines the preset threshold as ar based on boundedness analysis of system error and stability requirement; wherein, The failure detection accuracy of detection can be further improved. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 A flowchart of the failure detection method of the petroleum refining device sensor provided by the embodiment of the present application is shown in the figure.

[0035] Figure 2 A detailed flowchart of the failure detection method of the petroleum refining device sensor provided by the embodiment of the present application is shown in the figure.

[0036] Figure 3 A process diagram of the construction of the partial differential equation model and the nonlinear observer of the sensor to be detected provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0037] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0038] In order to achieve the above-mentioned purpose, in a first aspect, as shown in the figure, Figure 1 the present application provides a failure detection method of a petroleum refining device sensor, comprising:

[0039] a. obtaining current detection data of a sensor to be detected in a petroleum refining device; wherein, the current detection data comprises: equipment information of the sensor to be detected, and state data at a current time and all times before the current time;

[0040] In an optional embodiment, as shown in the figure, Figure 2 the above-mentioned obtaining current detection data of a sensor to be detected in a petroleum refining device comprises: collecting current detection data of a sensor to be detected from a petroleum refining device and performing preprocessing; wherein, the preprocessing comprises: denoising operation and normalization operation.

[0041] Specifically, the equipment information of the petroleum refining device sensor and the state data at the current time and all times before the current time are collected and preprocessed, such as cleaning, normalization and smoothing, to ensure the reliability and consistency of the data, and the specific process is as follows:

[0042] Data collection: The equipment monitoring system or data acquisition system in the oil refining plant is usually connected to various sensors and equipment for real-time monitoring and recording of equipment operating status, temperature, pressure, flow, etc. When the sensor fails or abnormity, the monitoring system will record these data for subsequent analysis and processing. The collected data comes from the sensor monitoring system automatically recording failure events and generating corresponding logs or alarms, the local database or remote server where the monitoring system stores failure data, the regular maintenance table of data engineers or maintenance personnel, etc. The collected data includes normal operation data before failure, after failure and without failure, which reflects the failure factor data related to equipment properties, failure types, operating environment, etc.

[0043] In particular, according to the sensor to be detected, the sensor equipment information (mainly including equipment properties such as equipment ID, equipment type, etc.) and the state data (such as operating status, temperature, pressure, flow, etc.) at the current time and all times before the current time are collected, and the data sources for collection include sensor detection system, data recorder, data acquisition equipment, database and history record, SCADA system (supervisory control and data acquisition system), third-party data supplier, etc. In order to analyze the failure component data and influence component data, the above selected equipment for collecting data needs to have sufficient data amount to obtain the required statistical confidence, and the time scale is greater than a complete equipment life cycle.

[0044] Data preprocessing: In order to avoid the invalid data affecting the overall result, the data is preprocessed such as cleaning, normalization and smoothing, so as to delete some repeated and invalid data caused by specific problems, and the equipment classification type needs to be determined according to expert suggestions; the purpose of data cleaning and denoising is to check data quality, identify and process missing values, abnormal values and noises, the missing values can be filled by interpolation method, and the abnormal values and noises can be removed or corrected by filter or abnormal detection algorithm; data normalization aims to normalize the collected data of different sensors to eliminate the difference between different dimensions and orders of magnitude, common normalization methods include minimum-maximum normalization and standardization; data smoothing is to smooth the data by applying moving average and exponential weighted moving average method to reduce noise and volatility.

[0045] b. Constructing a nonlinear observer of the sensor to be detected based on the partial differential equation model of the sensor to be detected, inputting the current detection data into the observer of the sensor to be detected for solving to obtain an observation value; calculating the difference between the observation value and the actual output of the sensor to be detected to obtain a detection residual;

[0046] As Figure 3As shown, according to the physical characteristics of the petroleum refining device and the response principle of the sensor to be detected, a partial differential equation of the sensor to be detected is established to describe the dynamic behavior of the system and the sensor characteristics, and then a detection observer is designed to produce the estimated state and output of the established system; specifically including the following processes:

[0047] 1) Sensor model establishment: in the present application, the sensor to be detected in the petroleum refining device is represented by a nonlinear distributed parameter system (DPS), which can be described by a partial differential equation (PDE), so according to the physical characteristics of the petroleum refining device and the response principle of the sensor, a suitable PDE is derived to describe it;

[0048] In an alternative embodiment, the sensor to be detected is regarded as a multi-input multi-output (MIMO) nonlinear distributed parameter system (DPS), and its partial differential equation model (PDE) is:

[0049]

[0050] The boundary conditions and output definitions are:

[0051] v x (0,t)=Qv(0,t),v(l,t)=u(t)

[0052] y(t)=v(0,t)

[0053] Wherein, v(x,t)=[v1(x,t),v2(x,t),…,v n (x,t)] T ∈[L2(0,l)] n represents the state vector of the sensor to be detected; v i (x,t) is the i-th state data of the sensor to be detected at position x at time t; i=1,2,…,n; n is the state data dimension of the sensor to be detected; L2(0,l) is a quadratic integrable function space defined on (0,l); d(x,t)=[d1(x,t),d2(x,t),…,d n (x,t)] n is a disturbance vector; d i (x,t) is the disturbance on the i-th state of the sensor to be detected at position x at time t; u(t)=[u1(x,t),u2(x,t),…,u n (x,t)] T ∈R n represents a control input vector; u i(x, t) is the control input of the i-th state data of the to-be-detected sensor at position x at time t; is a full rank matrix; is a Lipschitz continuous nonlinear vector function, which is used to describe the dynamic characteristics of the actual to-be-detected sensor; is the actual output of the to-be-detected sensor at time t, which is used to design an observer and generate a detection residual.

[0054] In an optional embodiment, the constraint condition of the partial differential equation model of the sensor includes that for all x and t≥0: wherein, is a preset constant.

[0055] In an optional embodiment, the spatial length l can be selected as 1, the constant (i.e., the system diffusion coefficient) c in the partial differential equation model of the to-be-detected sensor is 1, and the upper bound of the disturbance is

[0056] 2) Design of an observer: the observer is used to monitor the health state of the nonlinear DPS;

[0057] The present application designs an adaptive observer to generate the estimated state and output of the to-be-detected sensor (DPS system), which shows that the detection residual is ultimately bounded (upper bound, UB) under the healthy condition. Instead of converting the DPS into an infinite set of ODEs to define the failure detection observer and boundary conditions, the observer equation is obtained based on the condition of the PDE representation:

[0058]

[0059] The boundary condition is:

[0060]

[0061] The observation value is:

[0062]

[0063] wherein, is the state estimation vector of the to-be-detected sensor at position x at time t; 0<x<l; l is the spatial length of the to-be-detected sensor; L2(0, l) is a quadratic integrable function space defined on (0, l); is the i-th state estimation data of the to-be-detected sensor at position x at time t; i=1, 2, …, n; n is the state data dimension of the to-be-detected sensor; c is a constant in the partial differential equation model of the to-be-detected sensor; is a Lipschitz continuous nonlinear vector function, which is used to describe the dynamic characteristics of the observed sensor to be detected; is an error feedback control term; is the actual output of the sensor to be detected at time t; is a full rank matrix; n is the state data dimension of the sensor to be detected; P1(x, t) is the first gain matrix of the observer, which is used to adjust the feedback of the state estimation error of the sensor to be detected; denotes the derivative of the state at position 0; P 10 (t) is the second gain matrix of the sensor to be detected, which is used to adjust the boundary error feedback of the observer at the boundary; u(t) = [u1(x, t), u2(x, t), …, u n (x, t)] T ∈ R n denotes the control input vector; u i (x, t) is the control input acting on the i-th state data at position x of the sensor to be detected at time t.

[0064] It should be noted that P1(x, t) is responsible for adjusting the feedback of the state estimation error of the system observer, so as to improve the adaptability of the observer in space and time. P 10 (t) is the feedback gain matrix for the boundary condition, which is mainly used to adjust the boundary error feedback of the observer at the boundary (such as at x = 0).

[0065] It should be noted that the above observer can be a nonlinear Luenberger observer, or other observers such as Kalman filter, which is not limited here.

[0066] It can be regarded as an error correction mechanism. Taking the above observer as a nonlinear Luenberger observer as an example, P1(x, t) directly affects the performance of the observer, including the convergence speed and system stability, and its design considers error feedback control, error convergence speed, matching of system dynamic characteristics, stability and robustness.

[0067] Specifically, the detection residual is which can also be used to correct the state estimation error caused by the initial condition By comparing the difference in observer dynamics (observer) and the difference in actual system dynamics (partial differential equation model) v x (0, t) = Qv(0, t), v(l, t) = u(t), y(t) = v(0, t); wherein the nonlinear vector function f(v(x, t), x) satisfies the following conditions:

[0068] 1) for x e [0, 1], t > 0, v(x, t) e L2(0, 1), f(v(x, t), x) is Lipschitz continuous in v, continuous in x, and continuous in t;

[0069] 2) f(v(x, t), x) should satisfy where Δv represents a small change in v(x, t); ε f (Δv, x) is the approximation error, which satisfies where || · || represents the Frobenius norm or Euclidean norm of a matrix;

[0070] Thus, the state estimation error dynamics equation and the boundary condition under the healthy condition are:

[0071]

[0072] where, from When A(x, t) becomes large and positive, it will make the system unstable; in order to eliminate the term A(x, t) that may make the system unstable, a suitable observer gain is selected, and the Volterra integral transformation is obtained: and the observer gains P1(x, t) and P 10 (t) are: P 10 (t) = L(0, 0, t) converts the observer error dynamics formula to:

[0073] L(x, σ, t) is the unique solution of the well-posed PDE defined by the following formula:

[0074]

[0075] b > 0 is an arbitrary scalar;

[0076] In the above formula, and is the kernel matrix of the inverse transform

[0077] The observer error dynamic equation The observer gain P1(x, t) is used to adjust the error e(t) between the observer state and the actual state, so as to ensure the boundedness and asymptotic convergence of the error dynamics. The constructed observer satisfies:​​​

[0078] 1. Lyapunov stability: P1(x, t) is used to control the feedback strength of the observer error to detect the residual error e(t) to adjust the convergence speed, so as to ensure the stability of the observer error, that is, the Lyapunov derivative is negative. Based on the gain matrix P1(x, t), the Lyapunov derivative of the error dynamic equation can be kept negative, so as to ensure that the error converges to 0 under the condition of no failure.

[0079] 2. Error boundedness: In the presence of nonlinear approximation error and disturbance, the observer can keep bounded, thereby avoiding infinite growth of error and ensuring the accuracy of the observer.

[0080] When the above observer is a Kalman filter, its gain matrix is usually denoted as K(x, t), which has a similar effect as P1(x, t), which will not be described here.

[0081] The observer compensates for disturbances at different positions through the gain matrix, that is, even if the sensor at different positions is disturbed, the position difference can be responded through the feedback gain, so as to keep the residual error stable. The gain matrix is a multi-dimensional state matrix, and the observer sets an independent estimation and feedback correction item for each output dimension, so as to adapt to the complex dynamics of the MIMO system and make it have the ability to track and feedback correct the multi-dimensional variables of the MIMO system.

[0082] c. Determine whether the detection residual error is greater than the preset threshold value, if yes, determine that the sensor to be detected fails, otherwise, determine that the sensor to be detected does not fail.

[0083] In an optional embodiment, the above-mentioned preset threshold value is αρ; wherein ρ is a detection threshold, and its expression is:

[0084]

[0085] Wherein, 1.5≤α≤2.5; l is the spatial length of the sensor to be detected; is the upper limit of the error of the partial differential equation model of the sensor to be detected; is the upper limit of the disturbance of the partial differential equation model of the sensor to be detected; c is a constant in the partial differential equation model of the sensor to be detected.

[0086] In an optional embodiment, when the sensor system is stationary, 1.5≤α≤2; when the disturbance is large, 2≤α≤2.5. Preferably, α=2.

[0087] It should be noted that the preset threshold value is determined based on the boundedness analysis and stability requirement of the system error. Lyapunov function V(t) is used to study the change of error with time and space, so as to obtain the boundedness of error. The derivative of Lyapunov function is The convergence tendency of the system state is expressed, and it is usually required to be negative definite in the non-failure state.

[0088] The determination step of the preset threshold value is as follows:

[0089] 1. Select Lyapunov function: define a Lyapunov function V(t) for analyzing the stability of the system, which represents the sum of squares of errors and integrates the partial derivatives of the spatial position:

[0090]

[0091] Where Ξ(x, t) is the spatial representation of the observer error, and is the observer error By the result of integral transformation, and its dynamic equation becomes stable under the selection of appropriate kernel function, which helps to maintain the boundedness of error in the non-failure case. x (t) is the partial derivative of Ξ(x, t) with respect to the control variable x.

[0092] 2. Derivation and application of partial integration: derive V(t) with respect to time, and apply partial integration and substitute the boundary conditions:

[0093]

[0094] In the partial differential equation of the distributed parameter system, c represents the diffusion coefficient, which is usually related to the spatial dynamic characteristics of the system. b is the damping or attenuation coefficient in the observer error dynamic equation, which is used to control the convergence speed of the error at time t.

[0095] 3. Apply Poincare inequality: combine the error upper bound expression, use Poincare inequality to relate the error and its first derivative, and get:

[0096]

[0097] 4. Obtain the threshold formula: combine all the results, and finally obtain the residual upper bound under the non-failure condition, i.e. the detection threshold ρ:

[0098]

[0099] The final determined preset threshold value is 2ρ, which is doubled to reduce the risk of false positives; when deriving the threshold value through Poincara and Agmon inequality, a conservative estimate will be obtained, which means it is often lower than the actual value, and more importantly, it is obtained by detecting the selected threshold value with historical data. The following is the proof process of sensor detectability:

[0100] When the sensor fails, the observer error dynamic changes to:

[0101]

[0102] For t≥t s , apply the above formula, we get Ξ(l,t)=0;

[0103] Solving the partial differential equation represented by the formula, we get

[0104] The resulting detection residual is:

[0105] Where,

[0106]

[0107] When

[0108] Holds and the selected detection threshold satisfies:

[0109]

[0110] At this time will result in:

[0111]

[0112] Therefore, it can be guaranteed that the sensor failure that meets the given condition is detected.

[0113] In a second aspect, the present application provides a sensor failure detection system for a petroleum refining device, comprising: a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the failure detection method provided in the first aspect of the present application.

[0114] The related technical solutions are the same as the failure detection method provided in the first aspect of the present application, and will not be repeated here.

[0115] In a third aspect, the present application provides a petroleum refining device, comprising: a sensor and the failure detection system provided in the second aspect of the present application.

[0116] The related technical solutions are the same as the failure detection system provided in the second aspect of the present application, and will not be repeated here.

[0117] ​In a fourth aspect, the present application provides a computer readable storage medium, which comprises a stored computer program, wherein the computer program, when executed by a processor, controls a device in which the computer readable storage medium is located to perform the failure detection method according to the first aspect of the present application.

[0118] The related technical solution is the same as the failure detection method according to the first aspect of the present application, and will not be described here.

[0119] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting failure of a sensor of a petroleum refining plant, characterized by, The method comprises the following steps: acquiring current detection data of a sensor to be detected in a petroleum refining device; the current detection data comprises state data of the sensor to be detected at a current time and all times before the current time; a nonlinear observer of the sensor to be detected is constructed based on a partial differential equation model of the sensor to be detected, the current detection data is input into the observer for solving to obtain an observation value, and a difference between the observation value and an actual output of the sensor to be detected is calculated to obtain a detection residual; it is judged whether the detection residual is greater than a preset threshold, if yes, it is determined that the sensor to be detected is invalid, and if not, it is determined that the sensor to be detected is not invalid; an equation of the observer is: a boundary condition thereof is: ， an observation value thereof is: wherein, is a state estimation vector at time ; ; is a spatial length of the sensor to be detected; is a constant in the partial differential equation model of the sensor to be detected; is a Lipschitz continuous nonlinear vector function for describing the dynamic characteristics of the observed sensor to be detected; is an error feedback control term; is a first gain matrix of the observer for adjusting the feedback of the sensor to be detected to the state estimation error; is t an actual output of the sensor to be detected at time ; n is a full rank matrix; is a state data dimension of the sensor to be detected; denotes a state derivative at position 0; is a second gain matrix of the sensor to be detected for adjusting the boundary error feedback of the observer at the boundary; denotes a control input vector; is a control input acting on the th state data of the sensor to be detected at position at time The preset threshold is ; wherein, is a detection threshold, and an expression thereof is: wherein, ; is a spatial length of the sensor to be detected; is an error upper limit of the partial differential equation model of the sensor to be detected; is a disturbance upper limit of the partial differential equation model of the sensor to be detected; is a constant in the partial differential equation model of the sensor to be detected.

2. The method for failure detection of a sensor of a petroleum refining plant according to claim 1, characterized by, =2。 3. The method of claim 1, wherein the method further comprises: the acquiring of the current detection data of the sensor to be detected in the petroleum refining device comprises: current detection data of the sensor to be detected is collected from the petroleum refining device and is preprocessed; wherein, the preprocessing comprises a denoising operation and a normalization operation.

4. The method of claim 1-3, wherein the partial differential equation model of the sensor to be detected is: a boundary condition and an output definition thereof are: ， in, This represents the state vector of the sensor to be detected; For time The position of the sensor to be detected The first One status data; ; n The dimension of the state data of the sensor to be detected; For definition in The space of quadratic integrable functions on; The perturbation vector; For time The position of the sensor to be detected The first The disturbances that affect the state; Represents the control input vector; For time The lower action is applied to the position of the sensor to be detected. The first Control input for each status data; It is a full-rank square array; It is a Lipschitz continuous nonlinear vector function used to describe the dynamic characteristics of a real sensor under test; yes t The actual output of the sensor to be detected at any given time.

5. The method for failure detection of a sensor of a petroleum refining plant according to claim 4, wherein The constraint condition of the partial differential equation model of the sensor to be detected comprises that all of and satisfy: ; wherein, is a preset constant.

6. A system for failure detection of sensors in a petroleum refining plant, characterized by, The method comprises the following steps: a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the failure detection method of the sensor of the petroleum refining device according to any one of claims 1-5.

7. A petroleum refining plant characterized by comprising: The method comprises the following steps: a sensor and the failure detection system according to claim 6.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein when the computer program is run by a processor, the device where the storage medium is located is controlled to execute the failure detection method of the sensor of the petroleum refining device according to any one of claims 1-5.

Images