Multi-configuration refinery controller operation evaluation method

Abstract

The disclosed multi-structure refining controller operation evaluation method comprises the following steps: forming a standard model file; establishing an operation evaluation system and importing the standard model file, acquiring real-time operation data, and performing variable evaluation and mismatch degree evaluation according to a fixed cycle; and comprehensively evaluating single indexes in the variable evaluation and the mismatch degree evaluation, and performing comprehensive scoring after weight calculation. The multi-structure refining controller operation evaluation method can timely research and judge the controller model mismatch condition and the advanced control system performance through the model structure research of the controller and the real-time operation state evaluation of the controller, timely propose an operation strategy, ensure the efficient and good operation of the advanced control system, and realize the optimal control of the device.

Description

Technical Field

[0001] This invention belongs to the technical field of controller operation evaluation methods, specifically relating to a multi-structure refining controller operation evaluation method. Background Technology

[0002] In recent years, with increased research and support for advanced control technologies in the refining and chemical industry, the construction of Advanced Process Control (APC) systems has seen significant growth in both quality and quantity. However, advanced control technologies require a high level of comprehensive knowledge in areas such as processes, equipment, and control algorithms, resulting in a high technical threshold for implementation and maintenance. This high technical barrier leads to difficulties in timely and accurate mastery of the operational performance of APC systems by technical personnel at all levels; difficulty in effectively adjusting controller models and parameters after changes in plant processes or production conditions; and a lack of knowledge transfer when personnel leave their positions, significantly restricting the lifespan and operational efficiency of various APC systems. Therefore, an effective evaluation system is urgently needed for advanced control systems in refining and chemical plants to improve their operational performance and extend their service life.

[0003] Currently, no patents in the field of advanced control technology for refining and chemical plants have been found regarding controller operation evaluation. The patents related to advanced control systems retrieved all focus on the application of advanced control systems. Regardless of whether a controller is built using a specific method or multiple controllers with different structures, there is no evaluation of its operational performance. The lifespan and operational efficiency of various advanced control systems are significantly limited. Furthermore, the advanced control industry has numerous software vendors with varying model structures and algorithm technologies, making identification difficult, and there is currently no unified evaluation method or means in China. For example, the patent "Advanced Control for Maximizing Propylene Production from Petroleum Feed in a Demanding Fluidized Bed Catalytic Cracking Process" (CN101611118B) is an invention that optimizes propylene production by applying appropriate optimization to maximize the production of light distillate olefins, such as propylene, and provides early warning of potential performance degradation and equipment failure in FCC equipment. It provides optimized control for specific objects, thus optimizing production. However, this patent only innovates the application of advanced control technology and does not evaluate the operational performance of the advanced control system controller itself. Summary of the Invention

[0004] The purpose of this invention is to provide a method for evaluating the operation of multi-structure refining controllers, which solves the problem of the lack of existing methods for evaluating the operation performance of refining unit controllers.

[0005] The technical solution adopted in this invention is: a multi-structure refining controller operation evaluation method, comprising the following steps:

[0006] Step 1: Create a standard model file;

[0007] Step 2: Establish an operational evaluation system and import standard model files to obtain real-time operational data. Perform variable evaluation and mismatch evaluation at fixed intervals.

[0008] Step 3: The individual indicators in the comprehensive variable assessment and mismatch assessment are weighted and then comprehensively scored.

[0009] The invention is further characterized in that,

[0010] Step 1 involves two methods for generating a standard model file based on whether an original controller model exists: If an original controller model exists, the original controller model is obtained, the model structure is identified, and the model structure is converted according to the standard defined by the system for different structures to generate a standard model file; if an original controller model does not exist, a standard model file is generated by sequentially performing step identification and model fitting based on variable configuration information and historical operating data.

[0011] Step 2, after importing the standard model file, includes variable evaluation using Harris metrics and T-squared evaluation. 2 The Q statistic is used to assess the mismatch of the controller model.

[0012] The Harris index is expressed as:

[0013]

[0014] In equation (1), The variance of the system output error under the action of the minimum variance controller. f is the variance of the system output error under the actual controller action. i Let η be the impulse response function of the controller model, i = 0, 1, 2, ..., η Harris The value of η ranges from 0 to 1. When the output variance of the actual system is closer to the output variance of the system under minimum variance control, η Harris The closer it is to 1, the better the performance of the circuit.

[0015] The specific steps for variable evaluation include:

[0016] Step 2.1.1: Calculate the process time delay d and the process transfer function containing the process time delay d based on real-time operating data.

[0017] Step 2.1.2: Perturb the zero-mean white noise sequence a t Get the output y t To build a time series model, we combine the standard model transfer function G and the process transfer function. The estimated disturbance transfer function N is obtained;

[0018] Step 2.1.3: Using actual data, based on the process transfer function obtained in Step 2.1.1... And the minimum variance of the output of the estimation process, based on the interference transfer function N obtained in step 2.1.2.

[0019] Step 2.1.4: Using actual data, based on the process transfer function obtained in Step 2.1.1... Using the interference transfer function N obtained in step 2.1.2, calculate the actual output variance under the action of the standard model transfer function G.

[0020] Step 2.1.5: Use equation (1) to obtain the result from step 2.1.3. Compared with the result obtained in step 2.1.4 In comparison, the Harris index was obtained.

[0021] T 2 The statistic is expressed as:

[0022]

[0023] In equation (2), P T Let Λ be the transpose of the characteristic matrix, x be the system input, and Λ be the eigenvalue. 1, =diag(λ1,…,λ) k T represents the diagonal matrix formed by the first k eigenvalues; 2 The statistic is the squared Mahalanobis distance of the principal component vectors, which analyzes the distribution of the projection of the standardized original data x into the principal component space; T 2 The statistic follows an F-distribution, which is a two-degree-of-freedom statistical distribution. 2 The test threshold for the statistic is:

[0024]

[0025] In equation (3), F k,n-k；α Let f be an F-distribution with degrees of freedom k and nk and a confidence level α, and m be the dimension of the system input matrix;

[0026] The Q statistic is expressed as:

[0027] Q = ||e|| 2 (4)

[0028] When the sensor measurement data of the system is within the normal range, the Euclidean distance of its residual vector e is within a fixed range. If a sensor malfunction causes a change in the measurement data, the residual vector e will reflect the malfunction and cause its Q statistic to exceed the above fixed range. This fixed range is defined as the threshold Q of the Q statistic. a The threshold Q a The threshold Q is calculated based on the last nk feature values. a Represented as:

[0029]

[0030] In equation (5), k is the number of principal components in the principal component analysis model, λ i Let c be the i-th eigenvalue of the covariance matrix R. α For a standard normal distribution with a confidence level of α, the confidence limit is set. When α = 0.05, c α =1.645.

[0031] The specific steps for assessing controller mismatch include:

[0032] Step 2.2.1, Data Preparation: Collect historical input and output variable data online;

[0033] Step 2.2.2: Calculate T according to formula (2). 2 Statistic;

[0034] Step 2.2.3, Residual sequence calculation: Calculate the residual sequence e based on the principal component analysis model, input variable data, and output variable data; calculate the Q statistic based on the residual sequence e using equation (4);

[0035] Step 2.2.4: The system background collects data and calculates Q and T daily. 2 Statistics, forming the analysis of controller mismatch Q and T 2 Statistical indicator series;

[0036] Step 2.2.5: Perform filtering processing on the indicator sequence;

[0037] Step 2.2.6: Perform maintenance on the controller, and then adjust the controller mismatch detection standard based on the statistics of historical operating data.

[0038] The beneficial effects of the present invention are as follows: The multi-structure refining controller operation evaluation method of the present invention, through the study of the controller model structure and the evaluation of the controller's real-time operation status, can promptly judge the controller model mismatch and the performance of the advanced control system, propose operation strategies in a timely manner, ensure the efficient and good operation of the advanced control system, and achieve optimal control of the equipment. Attached Figure Description

[0039] Figure 1 This is a flowchart of the multi-structure refining controller operation evaluation method of the present invention;

[0040] Figure 2 It is a block diagram of single-input single-output feedback control. Detailed Implementation

[0041] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0042] Example 1

[0043] This invention provides a method for evaluating the operation of a multi-structure refining controller, such as... Figure 1 As shown, the original controller model is obtained, its structure is identified, and model conversion is performed according to the standards defined by the system for different structures, forming a standard model file. In the absence of an original controller model, a standard model file is formed based on variable configuration information and historical operating data through step identification and model fitting. The step response method involves providing a step test signal to the original controller model and analyzing the response sequence to complete the standardized conversion of the controller model for process inputs and outputs. In advanced control systems, controller models mainly include proportional, first-order, and second-order transfer function forms; step tests are used to determine the function parameters and the types of functions included.

[0044] The standard model file is imported into the operational evaluation system, which is used to evaluate the controller's performance. This involves acquiring real-time operational data and performing variable and mismatch assessments at fixed intervals.

[0045] Variable evaluation: The Harris index method is used for variable evaluation. This index reflects the comparison between the actual operating condition of the control system and the ideal condition, and reflects the potential capability of the controller.

[0046] The univariate Harris index is:

[0047]

[0048] in The variance of the system output error under the action of the minimum variance controller. f is the variance of the system output error under the actual controller action. i Let η be the impulse response function of the controller model, where i = 0, 1, 2, ..., . Harris The value of η ranges from 0 to 1. When the output variance of the actual system is closer to the output variance of the system under minimum variance control, η Harris The closer it is to 1, the better the performance of the circuit.

[0049] By utilizing minimum variance calculations, time series analysis of daily operational data can be used as a benchmark for evaluating control loop performance. Taking a single-input, single-output process under conventional control as an example, see... Figure 2 Where d is the process time delay, It is a time-delay-free process transfer function, N is the disturbance transfer function, and a t It is a zero-mean white noise sequence, and G is the standard model transfer function. N is represented by the Diophantine equation:

[0050]

[0051] That is, f i The first d terms of the N-impulse response.

[0052] Based on this benchmark, the variable evaluation steps are as follows:

[0053] 1. Calculate the process time delay d and the process transfer function containing the process time delay d based on real-time operating data.

[0054] 2. By perturbing the zero-mean white noise sequence a t Get the output y t To build a time series model, we combine the standard model transfer function G and the process transfer function. The estimated disturbance transfer function N is obtained. If the setpoint is not zero, then y t It should be the actual output minus the set value;

[0055] 3. Utilize actual data, including real-time and historical operating data, and combine it with the obtained process transfer function. Given the disturbance transfer function N, estimate the minimum variance of the process output.

[0056] 4. Utilize actual data, including real-time and historical operating data, and combine it with the obtained process transfer function. Given the interference transfer function N, calculate the actual output variance under the action of the standard model transfer function G.

[0057] 5. and In comparison, the Harris index was obtained to evaluate the performance of the control loop.

[0058] Mismatch assessment: using T 2 The basic principle of using SPE statistical methods for model mismatch detection and evaluation is to convert multidimensional data into unified one-dimensional data for comparison, and to determine whether the models match by judging the correlation between normal operation data and subsequent measurement data.

[0059] T 2 The statistic is also called Hoteelling's statistic. 2 The statistic is expressed as:

[0060]

[0061] In the above formula, P T Let Λ be the transpose of the characteristic matrix, x be the system input, and Λ be the eigenvalue. 1,k =diag(λ1,…,λ) k ) represents the diagonal matrix formed by the first k eigenvalues. T 2 The mathematical meaning of the statistic is the squared Mahalanobis distance of the principal component vectors, which analyzes the distribution of the projection of the standardized original data x into the principal component space. 2 The statistic follows an F-distribution. The F-distribution is a two-degree-of-freedom statistical distribution, T 2 The test threshold for the statistic is:

[0062]

[0063] In the above formula, F k,n-k；α Let f be an F-distribution with degrees of freedom k and nk and a confidence level α, and m be the dimension of the system input matrix.

[0064] The Q statistic, also known as the Squared Prediction Error (SPE statistic), is expressed as:

[0065] Q = ||e|| 2

[0066] The Q statistic represents the squared Euclidean distance of the residual vector e in the residual space, and is a statistic that transforms multidimensional data into one-dimensional data.

[0067] When the system's sensor measurement data is within the normal range, the Euclidean distance of its residual vector e should be within a fixed range. If a sensor malfunction causes a change in the measurement data, the residual vector e will reflect the malfunction and cause its Q-statistic to exceed the aforementioned fixed range. This range is defined as the threshold Q of the Q-statistic. a The threshold Q a It can be calculated from the last nk feature values. Threshold Q a Represented as:

[0068]

[0069] in In the above formula, k is the number of principal components in the principal component analysis model, and λ is... i Let c be the i-th eigenvalue of the covariance matrix R.α For a standard normal distribution with a confidence level of α, the confidence limit is set. When α = 0.05, c α =1.645.

[0070] The model mismatch assessment process is as follows:

[0071] 1. Data preparation: Online automatic collection of historical input and output variable data;

[0072] 2. According to T 2 Calculation of the statistical expression T 2 Statistic;

[0073] 3. Residual sequence calculation: Based on the principal component analysis model, input variable data, and output variable data, calculate the residual sequence e; based on the residual sequence e, calculate the Q statistic;

[0074] 4. The system automatically collects data in the background and calculates Q and T daily. 2 Statistics, forming the analytical model mismatch Q and T 2 Statistical indicator series;

[0075] 5. Filter the index sequence from the previous step to avoid false alarms and reduce the impact of disturbances or anomalies on model mismatch;

[0076] 6. Maintain the model and then adjust the model mismatch detection criteria according to the specific situation. It is generally recommended to select them from the statistics of historical running data.

[0077] Performance evaluation: For T 2 Individual indicators such as Q and Harris evaluation value are weighted by coefficients based on engineering experience and control objectives. The weight coefficients are verified and calibrated according to the actual situation of the controller, and a comprehensive score is given after weight calculation.

[0078] Example 2

[0079] After weighted calculation, a comprehensive score is given, which can be divided into four levels: excellent, good, average, and poor. A comprehensive score below 60 is judged as "poor," indicating that the controller has potential problems, and it is recommended to test or rebuild part of the controller model; a comprehensive score of 60 to 80 is judged as "average," and it is recommended to adjust the controller model parameters; a comprehensive score of 80 to 90 is judged as "good," indicating that the controller is basically operating normally; and a comprehensive score of 90 to 100 is judged as "excellent," indicating that the controller is operating well.

[0080] Example 3

[0081] After weighted calculation, a comprehensive score is given, and the results are displayed: the performance of the controller is evaluated and the results are displayed on the evaluation display interface of the operation evaluation system.

[0082] Example 4

[0083] After weighted calculation, a comprehensive score is generated, and the results are displayed on the evaluation display interface of the operation evaluation system. Furthermore, richer indicators such as utilization rate and stability rate can be combined to achieve multi-dimensional and multi-level evaluation and display of controllers for multi-level organizations, multiple devices, and multiple advanced control systems.

[0084] Through the above methods, the multi-structure refining controller operation evaluation method of the present invention can:

[0085] 1) Filling a Gap: Establishing an operational evaluation system for advanced controllers in multi-structure refining processes. The advanced control industry has numerous software vendors, with varying model structures and algorithm technologies, making identification difficult. Currently, there is no unified evaluation method or means in China. This invention defines model standards, develops conversion methods, and establishes a unified evaluation benchmark to achieve the evaluation and demonstration of the operational performance of advanced control systems with different controller structures.

[0086] 2) Algorithm Improvement: More complex algorithms and more comprehensive statistics. Existing mathematical methods for controller evaluation remain at a conventional level, such as total quantity statistics, difference statistics, and percentage statistics. This invention uses mathematical algorithms based on control theory, as well as data-driven fault diagnosis methods, to more comprehensively and accurately reflect the state of the statistical object.

[0087] 3) Operation and Maintenance Support: Solving operation and maintenance challenges and providing remote support. Through the application of advanced algorithms, the operating status and performance of advanced control systems can be reflected more timely and accurately, effectively identifying the trend and extent of model mismatch and controller performance changes, reducing the influence of experience factors, lowering the knowledge threshold, and providing strong support for the operation and maintenance of advanced control systems.

Claims

1. A method for evaluating the operation of a multi-structure refining controller, characterized in that, Includes the following steps: Step 1: Generate a standard model file. The method for generating a standard model file can be categorized based on whether an original controller model exists: If an original controller model exists, obtain it, identify the model structure, and perform model structure conversion according to the system definition standards for different structures to generate a standard model file; if an original controller model does not exist, generate a standard model file based on variable configuration information and historical operating data, sequentially through step identification and model fitting. Step 2: Establish an operational evaluation system and import the standard model file, obtain real-time operational data, and perform variable evaluation and mismatch evaluation at fixed intervals; the operational evaluation system after importing the standard model file includes variable evaluation using the Harris index, and evaluation using... and Statistics are used to assess the mismatch of the controller model; Step 3: The individual indicators in the comprehensive variable assessment and mismatch assessment are weighted and then comprehensively scored.

2. The multi-structure refining controller operation evaluation method as described in claim 1, characterized in that, The Harris metric is expressed as follows: （1） In equation (1), The variance of the system output error under the action of the minimum variance controller. This represents the variance of the system output error under the actual controller operation. Let be the impulse response function of the controller model, i = 0, 1, 2, ...; The value of is between 0 and 1. When the output variance of the actual system is closer to the output variance of the system under minimum variance control, The closer it is to 1, the better the performance of the circuit.

3. The multi-structure refining controller operation evaluation method as described in claim 2, characterized in that, The specific steps for evaluating the variables include: Step 2.1.1: Calculate the process time delay based on real-time operating data. and processes with time delays process transfer function ; Step 2.1.2: Perturb the zero-mean white noise sequence Get output To build a time series model, combining the standard model transfer function and process transfer function The estimated disturbance transfer function is obtained. ; Step 2.1.3: Using actual data, based on the process transfer function obtained in Step 2.1.1... and the interference transfer function obtained in step 2.1.2 The minimum variance of the estimation process output ; Step 2.1.4: Using actual data, based on the process transfer function obtained in Step 2.1.1... and the interference transfer function obtained in step 2.1.2 Calculate the standard model transfer function Actual output variance under action ; Step 2.1.5: Use equation (1) to obtain the result from step 2.1.

3. Compared with the result obtained in step 2.1.4 In comparison, the Harris index was obtained.

4. The multi-structure refining controller operation evaluation method as described in claim 1, characterized in that, The The statistic is expressed as: （2） In equation (2), It is the transpose of the characteristic matrix. x For system input, It indicates the previous A diagonal matrix composed of eigenvalues; The statistic is the squared Mahalanobis distance of the principal component vectors, which analyzes standardized raw data. The distribution of projections in the principal component space; Statistics follow distributed, The distribution is a two-degree-of-freedom statistical distribution. The test threshold for the statistic is: （3） In equation (3), For degrees of freedom and Confidence level is of distributed, m Input matrix dimensions to the system; The statistic is expressed as: （4） When the system's sensor measurement data is within the normal range, its residual vector The Euclidean distance is within a fixed range; if a sensor malfunction causes a change in the measurement data, the residual vector... It can reflect the fault and cause it. The statistic is greater than the above fixed range; this fixed range is defined as follows: Threshold of statistics The threshold According to the following The threshold is calculated from the eigenvalues. Represented as: （5） In equation (5), , , ; The number of principal components in the principal component analysis model. covariance matrix The 1 eigenvalue, For confidence level The standard normal distribution confidence limit, when hour, .

5. The multi-structure refining controller operation evaluation method as described in claim 4, characterized in that, The specific steps for evaluating the controller mismatch include: Step 2.2.1, Data Preparation: Collect historical input and output variable data online; Step 2.2.2: Calculate according to formula (2) Statistic; Step 2.2.3, Residual Sequence Calculation: Based on the principal component analysis model, input variable data, and output variable data, calculate the residual sequence. Based on the residual sequence Calculated using equation (4) Statistic; Step 2.2.4: The system backend collects data and calculates daily results. and Statistics, forming analysis of controller mismatch and Statistical indicator series; Step 2.2.5: Perform filtering processing on the indicator sequence; Step 2.2.6: Perform maintenance on the controller, and then adjust the controller mismatch detection standard based on the statistics of historical operating data.

Images