Reactor operation state online monitoring method and system based on neural network

Abstract

The invention relates to the field of reactor state monitoring, in particular to a reactor operation state online monitoring method and system based on a neural network. The method comprises the following steps: acquiring multi-dimensional working parameters and working labels of a reactor at a target moment; calculating the volatility of the target moment in any dimension, and calculating the credibility of the working parameters according to the volatility of the target moment in each dimension; calculating the confusion degree between the working parameters and the working labels of any dimension; calculating a random inactivation parameter according to the credibility and the confusion degree of all dimensions; and training a preset network according to the random inactivation parameters, inputting real-time multi-dimensional working parameters of the reactor into the trained preset network, outputting a reactor working label, and completing state monitoring. According to the technical scheme, the precision of the monitoring result of the reactor state can be improved.

Description

Technical Field

[0001] The present invention relates to the field of reactor status monitoring, and more specifically, to a method and system for online monitoring of reactor operating status based on a neural network. Background Art

[0002] In many industrial processes, reactors are common and critical equipment, widely used in chemical, pharmaceutical, petroleum refining and other fields. The operating status of the reactor directly affects the efficiency, quality and safety of production. Therefore, how to monitor the operating status of the reactor in real time, detect potential problems in time and issue early warnings has become an important task in modern industrial production.

[0003] Traditional reactor monitoring methods usually rely on physical sensors to collect parameters such as temperature, pressure, flow, concentration, etc. in the reactor, and infer the operating status of the reactor through changes in these parameters. However, with the increase in variables in industrial processes, a single sensor is often unable to fully and accurately reflect the operating status of the reactor. The dynamic process in the reactor is often affected by multiple factors, including fluctuations in raw material quality, changes in catalyst activity, and slight deviations in operating conditions. These factors are often coupled with each other, resulting in complexity and uncertainty in sensor data. In addition, the sensors in the reactor are limited by the installation environment or equipment accuracy, and there will be problems with large data noise or data missing, resulting in inaccurate reactor status monitoring results. Summary of the invention

[0004] In order to solve the problem of inaccurate reactor status monitoring results, the present invention proposes a method and system for online monitoring of reactor operating status based on neural network.

[0005] In a first aspect, the present invention discloses an online monitoring method for the operation status of a reactor based on a neural network, comprising: obtaining multi-dimensional working parameters and working labels of the reactor at a target moment, the target moment being any sampling moment, and one multi-dimensional working parameter corresponding to one working label; constructing a window with the target moment as the center and a preset length as the window size, and for any dimension, taking the average difference between the working parameters at the target moment and the working parameters at the reference moment as the volatility of the target moment in any dimension, the reference moment being the sampling moment other than the target moment in the window, and calculating the credibility of the working parameters according to the volatility of the target moment in each dimension; calculating the degree of confusion between the working parameters and the working labels of any dimension; calculating the product of the mean credibility of all sampling moments and the mean degree of confusion of all dimensions, and using the result of the product through negative correlation mapping as a random deactivation parameter; training a preset network according to the random deactivation parameter, inputting the real-time multi-dimensional working parameters of the reactor into the trained preset network, outputting the reactor working label, and completing status monitoring.

[0006] By analyzing the volatility of real-time sampling data, the complexity of the model can be dynamically adjusted to effectively avoid overfitting and underfitting problems, improve the prediction accuracy and robustness of the reactor in practical applications, and ensure that the model can provide stable monitoring results in a changing operating environment.

[0007] Preferably, the missing working parameters are filled using the mean filling method, the outliers in the working parameters are identified and processed using the Z-score method, and the noise in the working parameters is reduced using the Kalman filter.

[0008] Preferably, the credibility satisfies the relationship: , Indicates the target time The credibility of the working parameters, represents the mean volatility of all dimensions at the target moment, Represents the total number of dimensions, Indicates the target time In Dimension The volatility of Represents an exponential function.

[0009] Evaluating credibility based on the volatility differences in multiple dimensions can effectively avoid misjudgment due to abnormal fluctuations in a single dimension, thereby improving the overall monitoring accuracy of the reactor status. This method helps to effectively distinguish between noise and real anomalies, ensuring that the operating status of the reactor is monitored more accurately and reliably.

[0010] Preferably, calculating the degree of disorder includes: calculating the normal disorder degree of any dimension when the working tag is working normally, and the normal disorder degree satisfies the relationship: , Indicates that when the work label is normal, the dimension The normal level of disorder, Representation Dimension The value of the working parameter is The probability of represents a logarithmic function; the abnormal disorder degree of any dimension is obtained when the work label is abnormal work in the same way as the calculation method of the normal disorder degree; the mean of the normal disorder degree and the abnormal disorder degree is taken as the disorder degree between the work parameter and the work label of any dimension.

[0011] It can effectively distinguish between normal fluctuations and abnormal changes, and in practical applications, it helps to accurately monitor the operating status of the reactor or system, identify potential faults or abnormalities in advance, and thus improve the safety and reliability of the system.

[0012] Preferably, the calculation of the degree of disorder also satisfies the relationship: , Indicates the degree of confusion. represents the variance of all operating parameters in any dimension, It represents the variance of all working parameters in any dimension when the working label is normal operation. It represents the variance of all working parameters in any dimension when the working label is abnormal working.

[0013] This calculation method can effectively reflect the relationship between working parameters and working tags, and optimize the monitoring and abnormal warning effects of the operating status of the reactor or equipment.

[0014] Preferably, the preset network is a BP neural network, which includes an input layer, a hidden layer and an output layer; the input layer is used to receive the input of multi-dimensional working parameters, and the hidden layer extracts features from the input information and inputs the extracted features into the output layer to output the reactor working label.

[0015] Preferably, the training process of the preset network includes: taking the multi-dimensional working parameters of the reactor in history as input information, and taking the true value of the working label of the reactor as the label, to obtain a set of training data; calculating the loss value of the preset network through the working label of the reactor and the label, setting the random inactivation parameters of the neurons in the preset network, and using the cross entropy loss function as the loss function; iteratively updating the network parameters of the preset network, and when the preset network reaches the set maximum number of training times or the network loss value is less than the set loss value, stopping the update to obtain a trained preset network.

[0016] In a second aspect, the present invention discloses an online monitoring system for the operation status of a reactor based on a neural network, comprising: a processor; and a memory, wherein the memory stores computer instructions, and when the computer instructions are executed by the processor, the system executes the above-mentioned online monitoring method for the operation status of a reactor based on a neural network.

[0017] Beneficial effects of the present invention: The present invention dynamically evaluates the working state of the reactor through the volatility of multi-dimensional working parameters, credibility assessment and calculation of the degree of chaos, and then outputs real-time working labels by training neural networks. Specifically, through windowed data sampling, missing value filling, outlier processing and noise filtering, high-quality data input is ensured; and through the calculation of the degree of chaos and credibility, it can not only accurately reflect the system performance under different working conditions, but also train a neural network with strong predictive ability based on historical data, thereby improving the accuracy of the reactor's operating status monitoring results. This method can promptly identify abnormal conditions of the reactor, prevent equipment failures, and improve the safety and operating efficiency of the reactor. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 It is a flow chart of a method for online monitoring of reactor operating status based on neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0020] It should be understood that when the terms "first", "second", etc. are used in the claims, descriptions, and drawings of the present invention, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the description and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their collections.

[0021] The present invention provides a method for online monitoring of reactor operation status based on neural network. Figure 1 As shown, the online monitoring method for the operation status of a reactor based on a neural network includes steps S1 to S5, which are described in detail below.

[0022] S1, obtain the multi-dimensional working parameters and working labels of the reactor at the target time.

[0023] In one embodiment, sensors are used to collect the working parameters of the reactor. For example, by installing a voltage sensor, the DC voltage and AC voltage of the reactor can be monitored, the current sensor is used to collect changes in DC current and AC current, and the temperature sensor records the temperature changes inside and outside the reactor in real time. These sensor data can provide basic information for determining the working state of the reactor. In combination with the actual working conditions and equipment operation specifications, technicians in this field analyze the collected working parameters, evaluate the operating state of the reactor, and assign working labels to each collected working parameter according to the set standards and thresholds, and the working labels include normal operation or abnormal operation.

[0024] The target time is any sampling time. The working parameter is multi-dimensional, and a multi-dimensional working parameter corresponds to a working label.

[0025] The missing working parameters are filled using the mean filling method, the outliers in the working parameters are identified and processed using the Z-score method, and the noise in the working parameters is reduced using the Kalman filter. The preprocessing of multi-dimensional working parameters is completed.

[0026] S2, build a window with the target moment as the center and the preset length as the window size. For any dimension, take the average difference between the working parameters at the target moment and the working parameters at the reference moment as the volatility of the target moment in any dimension, and calculate the credibility of the working parameters based on the volatility of the target moment in each dimension.

[0027] It should be noted that the credibility of the collected working parameters has a crucial impact on the performance of the preset network, because low-credibility working parameters may introduce noise, which in turn affects the learning effect and prediction accuracy of the preset network. Therefore, it is necessary to conduct a credibility assessment on each working parameter and process it according to the assessment results. When the credibility of a certain working parameter is low, the probability of neuron inactivation in the neural network can be increased during the training process, so that the preset network can maintain strong robustness when facing these noisy training data. In this way, even if there is a certain amount of noise or outliers in the training data, the preset network can still avoid overfitting and maintain good generalization performance, ensuring that it still has stable performance under incomplete or inaccurate data.

[0028] In one embodiment, for any dimension, the average difference between the operating parameters at the target time and the operating parameters at the reference time is used as the volatility of the target time in any dimension, and the volatility is expressed by the relationship: , Indicates the target time In Dimension The volatility of Indicates the target time In Dimension The working parameter value, Indicates the reference time In Dimension The working parameter value, Indicates the window size.

[0029] Credibility satisfies the relationship: , Indicates the target time The credibility of the working parameters, represents the mean volatility of all dimensions at the target moment, Represents the total number of dimensions, Indicates the target time In Dimension The volatility of Represents an exponential function.

[0030] When the working parameters of multiple dimensions at a certain sampling moment fluctuate synchronously, that is, when the volatility of each dimension is close to the mean volatility of all dimensions at that sampling moment, then the credibility is high. This is because if the volatility of multiple dimensions is close to the mean, it means that these fluctuations are more likely to be part of normal fluctuations rather than external noise or occasional interference, further indicating that the fluctuations at this moment may reflect the real change in the state of the reactor system, or even a sign of abnormality. On the contrary, if the difference between the volatility of each dimension and the volatility mean is large, it means that the fluctuations of each dimension do not show synchronization, and this volatility deviates far from the mean, and the credibility is low. This situation usually means that the fluctuations in some dimensions may be caused by noise or other irrelevant factors, and thus cannot accurately reflect the true state of the reactor system.

[0031] Evaluating credibility based on the volatility differences in multiple dimensions can effectively avoid misjudgment due to abnormal fluctuations in a single dimension, thereby improving the overall monitoring accuracy of the reactor status. This method helps to effectively distinguish between noise and real anomalies, ensuring that the operating status of the reactor is monitored more accurately and reliably.

[0032] It should be noted that the credibility of a certain sampling moment is not only related to the fluctuation amplitude of a single dimension at that sampling moment, but also closely related to the synchronization of fluctuations between multiple dimensions. If multiple parameters of the reactor (such as voltage, current, temperature, etc.) have similar and large fluctuations at the same sampling moment, this synchronization usually means that there is a common problem in the reactor system, rather than simple random noise.

[0033] S3, calculates the degree of confusion between the job parameters and job labels in any dimension.

[0034] In one embodiment, calculating the degree of disorder includes: calculating the normal disorder degree of any dimension when the working label is working normally, and the normal disorder degree satisfies the relationship: , Indicates that when the work label is normal, the dimension The normal level of disorder, Representation Dimension The value of the working parameter is The probability of Represents a logarithmic function.

[0035] The abnormal disorder degree of any dimension is obtained in the same way as the calculation method of the normal disorder degree when the work label is abnormal work; the average of the normal disorder degree and the abnormal disorder degree is used as the disorder degree between the work parameter and the work label of any dimension.

[0036] By introducing the calculation method of the degree of chaos, the matching degree between the work label and the working parameters of each dimension can be quantified from a probability perspective. Under normal working conditions, the normal degree of chaos of a dimension reflects the distribution uncertainty of the working parameter value, which is specifically calculated by calculating the probability distribution of each possible working parameter value of the dimension under normal working conditions and measuring its information entropy; similarly, under abnormal working conditions, the abnormal degree of chaos measures the uncertainty of the parameter distribution under abnormal working conditions. Taking the mean of these two degrees of chaos as the degree of chaos of the dimension helps to accurately identify the changes in the working state by comparing the fluctuation characteristics under normal and abnormal conditions.

[0037] During normal operation, the distribution of parameter values ​​should be relatively stable with a low degree of chaos, while during abnormal operation, the parameter distribution usually deviates from the normal state, resulting in a high degree of chaos. Therefore, with the help of this comparison based on the degree of chaos, it is possible to effectively distinguish between normal fluctuations and abnormal changes, and in practical applications, help accurately monitor the operating status of the reactor or system, identify potential faults or anomalies in advance, and thus improve the safety and reliability of the system.

[0038] In another embodiment, the degree of confusion further satisfies the relationship: , Indicates the degree of confusion. represents the variance of all operating parameters in any dimension, It represents the variance of all working parameters in any dimension when the working label is normal operation. It represents the variance of all working parameters in any dimension when the working label is abnormal working.

[0039] When the intra-class variance (i.e., the parameter variance under each working label) is small, and the inter-class variance (i.e., the variance difference under different working labels) is large, it indicates that the distribution of parameters is significantly different under normal and abnormal working conditions, and the system has a strong ability to distinguish between normal and abnormal working conditions. At this time, the degree of confusion is low, which means that the matching degree between parameters and labels is high, reflecting the recognition accuracy of system status.

[0040] This calculation method can effectively reflect the relationship between working parameters and working tags, and optimize the monitoring and abnormal warning effects of the operating status of the reactor or equipment.

[0041] S4, calculate the product of the mean credibility of all sampling moments and the mean chaos degree of all dimensions, and use the result of the product through negative correlation mapping as the random inactivation parameter.

[0042] It should be noted that random dropout in neural networks is mainly used to prevent overfitting of the model and improve the generalization ability of the model. When the credibility of the working parameters is high, it means that there is less noise in the working parameters, and the preset network can better learn the real rules of the data. At this time, the necessity of adding random dropout is lower, because the preset network can already fit the training data well and is not easy to overfit.

[0043] When the degree of confusion between the working parameters and the labels is large, it indicates that there is a strong nonlinear relationship between the labels and the working parameters. The preset network needs more complexity to capture these relationships. Therefore, there is no need to introduce too much random inactivation, because increasing inactivation may limit the expressive power of the model, making it difficult to fully capture the complex patterns of the data.

[0044] In general, random dropout is used to avoid overfitting of the model and classifying some incorrect samples as correct samples. The preset network works well on the test set, but the effect decreases in actual use.

[0045] S5, train the preset network according to the random deactivation parameters, input the real-time multi-dimensional working parameters of the reactor into the trained preset network, output the reactor working label, and complete the state monitoring.

[0046] In one embodiment, the network is assumed to be a BP neural network, which includes an input layer, a hidden layer and an output layer; the input layer is used to receive the input of multi-dimensional working parameters, and the hidden layer extracts features from the input information and then inputs the extracted features into the output layer to output the reactor working label.

[0047] The training process of the preset network includes: taking the multi-dimensional working parameters of the reactor in history as input information, and taking the true value of the reactor's working label as the label, to obtain a set of training data; calculating the loss value of the preset network through the reactor's working label and the label, setting the random inactivation parameters of the neurons in the preset network, and using the cross entropy loss function as the loss function; iteratively updating the network parameters of the preset network, and when the preset network reaches the set maximum number of training times or the network loss value is less than the set loss value, stop updating to obtain a trained preset network.

[0048] By inputting the real-time multi-dimensional working parameters of the reactor into the preset network, the reactor working label can be obtained, thereby obtaining the reactor status result and completing the status monitoring.

[0049] An embodiment of the present invention also discloses an online monitoring system for the operation status of a reactor based on a neural network, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an online monitoring method for the operation status of a reactor based on a neural network according to the present invention is implemented.

[0050] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, and their configuration and functions are known in the art, so they will not be described in detail here.

[0051] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module or both. Any such computer storage medium may be part of a device or accessible or connectable to a device.

[0052] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

[0053] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for online monitoring of reactor operation status based on neural network, characterized in that: include: Obtaining multi-dimensional working parameters and working labels of the reactor at a target time, where the target time is any sampling time, and one multi-dimensional working parameter corresponds to one working label; A window is constructed with the target time as the center and the preset length as the window size. For any dimension, the average difference between the working parameters at the target time and the working parameters at the reference time is taken as the volatility of the target time in any dimension. The reference time is the sampling time in the window except the target time. The credibility of the working parameters is calculated according to the volatility of the target time in each dimension. Calculate the degree of confusion between work parameters and work labels in any dimension; Calculate the product of the mean credibility of all sampling moments and the mean chaos degree of all dimensions, and use the result of the product through negative correlation mapping as the random inactivation parameter; The preset network is trained according to the random deactivation parameters, the real-time multi-dimensional working parameters of the reactor are input into the trained preset network, the reactor working label is output, and the state monitoring is completed.

2. The method for online monitoring of reactor operation status based on neural network according to claim 1, characterized in that: The mean imputation method is used to fill in the missing working parameters, the Z-score method is used to identify and process the outliers in the working parameters, and the Kalman filter is used to reduce the noise in the working parameters.

3. The method for online monitoring of reactor operation status based on neural network according to claim 1, characterized in that: The credibility satisfies the relationship: , Indicates the target time The credibility of the working parameters, represents the mean volatility of all dimensions at the target moment, Represents the total number of dimensions, Indicates the target time In Dimension The volatility of Represents an exponential function.

4. The method for online monitoring of reactor operation status based on neural network according to claim 1, characterized in that: Calculating the degree of disorder includes: When the work label is calculated as normal operation, the normal disorder degree of any dimension satisfies the relationship: , Indicates that when the work label is normal, the dimension The normal level of disorder, Representation Dimension The value of the working parameter is The probability of represents the logarithmic function; The abnormal disorder degree of any dimension is obtained in the same way as the calculation method of the normal disorder degree when the work label is abnormal work; The mean of the normal disorder degree and the abnormal disorder degree is taken as the disorder degree between the working parameter and the working label of any dimension.

5. The method for online monitoring of reactor operation status based on neural network according to claim 1, characterized in that: The calculation of the degree of disorder also satisfies the relationship: , Indicates the degree of confusion. represents the variance of all operating parameters in any dimension, It represents the variance of all working parameters in any dimension when the working label is normal operation. It represents the variance of all working parameters in any dimension when the working label is abnormal working.

6. The method for online monitoring of reactor operation status based on neural network according to claim 1, characterized in that: The preset network is a BP neural network, which includes an input layer, a hidden layer and an output layer; the input layer is used to receive the input of multi-dimensional working parameters, and the hidden layer extracts features from the input information and inputs the extracted features into the output layer to output the reactor working label.

7. The method for online monitoring of reactor operation status based on neural network according to claim 1, characterized in that: The training process of the preset network includes: The multi-dimensional working parameters of the reactor in history are used as input information, and the actual value of the working label of the reactor is used as a label to obtain a set of training data; The loss value of the preset network is calculated through the reactor working label and label, and the random inactivation parameters of the neurons in the preset network are set. The loss function uses the cross entropy loss function; Iteratively update the network parameters of the preset network. When the preset network reaches the set maximum number of training times or the network loss value is less than the set loss value, stop updating to obtain a trained preset network.

8. The online monitoring system of reactor operation status based on neural network is characterized by: include: Processor; and A memory storing computer instructions, wherein when the computer instructions are executed by a processor, the system executes the online monitoring method for the operation status of a reactor based on a neural network according to any one of claims 1 to 7.

Images