Health monitoring method and device for reaction kettle array and computer equipment

Abstract

The invention discloses a health monitoring method and device for a reaction kettle array and computer equipment, and relates to the field of automatic control of chemical equipment. The health monitoring method for the reaction kettle array comprises the following steps: sampling process data of the reaction kettle array in a production process; extracting the first time sequence feature data set, and inputting the first time sequence feature data set into a preset time sequence neural network model to obtain a dimension-reduced second time sequence feature data set; determining a target feature vector data set at the current moment based on the historical production data set and the second time sequence feature data set; determining a weighted statistical distance value of each reaction kettle in the reaction kettle array based on a weighted Mahalanobis distance of the target feature vector data set; obtaining a health degree score corresponding to the weighted statistical distance value based on a preset mapping relationship; and controlling each reaction kettle to execute corresponding work based on the health degree score. Through the method, property loss caused by abnormity of the reaction kettles is reduced, and the safety of the reaction kettle array is improved.

Description

Technical Field

[0001] This application relates to the field of automated control of chemical equipment, and in particular to methods, devices and computer equipment for health monitoring of reactor arrays. Background Technology

[0002] A reactor array is a system composed of several reactors used in the flexible manufacturing of high-end chemicals. During the operation of a reactor array, if even one reactor malfunctions, the entire batch of materials will be scrapped. Therefore, it is necessary to control each reactor to ensure timely detection of any abnormalities and reduce chemical production costs.

[0003] If process data from each reactor is collected via sensors and input into a deep neural network to predict its remaining lifespan, several challenges arise. First, training the deep neural network requires a massive amount of anomaly samples. However, the processing costs of high-end chemicals are high, making it difficult to collect sufficient anomaly samples. Second, even with a sufficient number of anomaly samples, the computational demands of the anomaly detection algorithm are high, and the computing power of the edge controllers installed in each reactor is limited, making real-time detection of anomalies in the reactor array impossible. These problems can lead to batch scrapping of products, reactor malfunctions and damage, and even safety accidents such as toxic gas leaks or runaway explosions within the reactors. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide a health monitoring method, apparatus, and computer equipment for reactor arrays, which can reduce property losses caused by reactor malfunctions and improve the safety of reactor arrays.

[0005] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, this application provides a health monitoring method for a reactor array, comprising: The process data of the sample reactor array during the production of the target sample is sampled; a first time-series feature dataset is extracted from the process data and input into a preset time-series neural network model to obtain a second time-series feature dataset after dimensionality reduction; based on the historical production dataset of the target sample and combined with the second time-series feature dataset, the target feature vector dataset at the current moment is determined. The target feature vector dataset is used to indicate the residual between the historical production dataset and the process data at the current moment; based on the weighted Mahalanobis distance of the target feature vector dataset, the weighted statistical distance value of each reactor in the reactor array is determined, and a health score corresponding to the weighted statistical distance value is obtained according to a preset mapping relationship. Based on the health score, the reactors in the reactor array are controlled to perform corresponding operations.

[0006] In some implementations, the weighted statistical distance value of each reactor in the reactor array is determined based on the weighted Mahalanobis distance of the target feature vector dataset. This includes: obtaining a reference feature dataset of historical production datasets and determining the inverse matrix of the covariance matrix of the reference feature dataset; generating an attention weight matrix corresponding to the reaction process at the current moment based on a pre-established process map, wherein the process map is established based on the historical production dataset and the attention weight matrix is ​​used to indicate the weight of process data in different reaction processes; and calculating the weighted statistical distance value based on the target feature vector dataset, the inverse matrix, and the attention weight matrix.

[0007] In some implementations, a weighted statistical distance value is calculated based on the target feature vector dataset, the inverse matrix, and the attention weight matrix, including: The weighted statistical distance values ​​satisfy the following relationship: ; In the formula, Indicates xxx, Represents the attention weight matrix. Represents the inverse matrix. This represents the target feature vector dataset.

[0008] In some implementations, each reactor is equipped with an edge controller, which is used to determine the weighted statistical distance value of the corresponding reactor. The dimension of the second time series feature dataset and the computing power of the edge controller satisfy the following conditions: when calculating the weighted Mahalanobis distance based on the target feature vector dataset, the floating-point operation of the inverse matrix is ​​less than the upper limit of the single-cycle computing power of the edge controller under the first preset ratio, and the data storage in a single time window is less than the working memory of the edge controller under the second preset ratio. Among them, the dimension of the target feature vector dataset is the same as the dimension of the second time series feature dataset.

[0009] In some implementations, health monitoring methods also include: Based on the health score, corresponding control is performed on each reactor; if the health score is within the first safety threshold range, the PID controller used to control the reactor maintains the current control state; if the health score is within the second safety threshold range, the reactor with the health score within the second safety threshold range is replaced by a spare reactor in the reactor array; if the health score is within the third safety threshold range, the transmission channel between the reactors is blocked; wherein, the first safety threshold range is greater than the second safety threshold range, and the second safety threshold range is greater than the third safety threshold range.

[0010] In some implementations, a reactor with a health score within a second safety threshold range is defined as a target reactor. Replacing a reactor with a health score within the second safety threshold range with a standby reactor in the reactor array includes: mapping the status indicators of the target reactor to the standby reactor, so that the status indicators of the standby reactor are consistent with the status indicators of the target reactor, wherein the status indicators include at least one of temperature, pressure, and rotational speed; obtaining the node characteristics of the target reactor based on process parameters, and mapping the node characteristics to a PID controller connected to the standby reactor to adjust the controller parameters of the PID controller connected to the standby reactor; the node characteristics are used to indicate at least one of reaction progress, material properties, and historical disturbances within the target reactor.

[0011] In some implementations, a reactor with a health score within a second safety threshold range is defined as a target reactor; replacing the reactor with a standby reactor in the reactor array with a standby reactor includes: transferring material from the target reactor to the standby reactor and determining a switching function that determines the material transfer progress over time; adjusting the first flow rate of material flowing out of the target reactor and the second flow rate of material flowing out of the standby reactor according to the switching function.

[0012] In some implementations, the switching function is the Sigmoid function, which satisfies the following relationship: ; In the formula, This represents the switching function, where e represents xxx, k represents xxx, and t and t0 represent the current time and start time of the material transfer, respectively. The first flow and the second flow satisfy the following relationship: ; In the formula, Indicates the first flow rate. Indicates the second flow rate. This indicates the total amount of materials transferred.

[0013] Secondly, this application provides a health monitoring device for a reactor array, comprising a sampling module and a data processing module. The sampling module is used to sample process data of the reactor array during the production of a target sample. The data processing module is used to extract a first time-series feature dataset from the process data and input the first time-series feature dataset into a preset time-series neural network model to obtain a second time-series feature dataset after dimensionality reduction. Based on the historical production dataset of the target sample and combined with the second time-series feature dataset, a target feature vector dataset for the current moment is determined. The target feature vector dataset is used to indicate the deviation between the historical production dataset and the process data at the current moment. Based on the weighted Mahalanobis distance of the target feature vector dataset, a weighted statistical distance value for each reactor in the reactor array is determined. Based on a preset mapping relationship, a health score corresponding to the weighted statistical distance value is obtained.

[0014] Thirdly, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it performs any of the above-mentioned health monitoring methods.

[0015] The health monitoring method for reactor arrays provided in this application samples process data during the production of target samples by the reactor array, and performs dimensionality reduction on the process data through a preset temporal neural network model to obtain a target feature vector dataset. The inverse of the covariance matrix of the dimensionality-reduced reference feature dataset is then calculated, and an attention weight matrix is ​​introduced to obtain a weighted statistical distance value. A health score is obtained through a preset mapping relationship, reducing the computational load when controlling the reactor array. This enables the edge controller to monitor reactor anomalies in real time and control each reactor to perform corresponding operations, improving the accuracy of reactor array control. Furthermore, it allows for timely detection of reactor anomalies, reducing property losses caused by reactor anomalies and improving the safety of the reactor array. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the health monitoring device for the reactor array in the embodiments of this application; Figure 2 This is a flowchart of the health monitoring method for the reactor array in the embodiments of this application; Figure 3 This is a flowchart illustrating the determination of health scores in an embodiment of this application; Figure 4 This is a flowchart illustrating the corresponding control measures implemented for each reactor based on a health score in an embodiment of this application. Figure 5 This is a first flow chart of replacing the target reactor with a spare reactor in an embodiment of this application; Figure 6This is a second flow chart in the embodiments of this application, in which a spare reactor is used to replace the target reactor; Figure 7 This is a schematic diagram of a computer device in an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in specific embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0018] It should be noted that the terms "first," "second," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. "A plurality" or "several" indicates at least two. "Comprising" or "including" and similar terms mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. "Connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0019] This application provides a health monitoring method for a reactor array 300. In one embodiment, this health monitoring method can be applied to, for example... Figure 1 In the health monitoring device 100 of the reactor array 300 shown, the health monitoring device 100 includes a sampling module 11 and a data processing module 12. The sampling module 11 can sample the process data of the reactor array 300 during the production of the target sample. The data processing module 12 can determine the health score of each reactor 31 in the reactor array 300 based on the process data and execute the method steps described in the following embodiments.

[0020] In another embodiment, the health monitoring method can also be applied to a computer device, which can be a terminal device or a server. It should be noted that the computer device is not limited to the terminal devices and servers mentioned above; any device with data processing capabilities can be used as the computer device in this embodiment. The computer device is communicatively connected to the health monitoring device 100. Exemplarily, the communication connection can be a wired connection using a data bus as the transmission medium or a wireless connection. The health monitoring device 100 can sample process data from the reactor array 300 during the production of the target sample and send the process data to the computer device, causing the computer device to execute the method steps described in the following embodiments.

[0021] In another embodiment, the health monitoring method is also applied in a health monitoring device 100 and a computer device, with some steps of the health monitoring method being performed by the health monitoring device 100 and some steps being performed by the computer device.

[0022] like Figure 2 As shown, this application provides a health monitoring method for a reactor array 300, which includes the following steps: Step S201: Sample the process data of the reactor array 300 during the production of the target sample.

[0023] The process data includes temperature, pressure, current, stirring speed, and material volume.

[0024] In some embodiments, the number of sampling modules 11 is greater than the number of reactors 31 in the reactor array 300, wherein some sampling modules 11 are disposed in reactors 31, and some sampling modules 11 may also be disposed in the transmission channel between two reactors 31. For a sampling module 11 in a reactor 31, it may include at least one of a temperature sensor, a pressure sensor, a current sensor, a speed sensor, and a weighing sensor. Specifically, the temperature sensor is used to detect the temperature inside the reactor 31, and includes, but is not limited to, thermocouple sensors and resistance temperature detectors (RTDs); the pressure sensor is used to detect the pressure inside the reactor 31, and includes, but is not limited to, diffused silicon pressure transmitters and explosion-proof pressure transmitters; the current sensor is used to detect the current in the reactor 31, and includes, but is not limited to, Hall effect current sensors and coil current sensors; the speed sensor is used to detect the stirring speed of the reactor 31, and includes, but is not limited to, Hall effect speed sensors and electrically inductive speed sensors; the weighing sensor is used to detect the material volume inside the reactor 31 through weight conversion, and includes, but is not limited to, shear beam weighing sensors and column weighing sensors.

[0025] Step S202: Extract the first time series feature dataset from the process data, and input the first time series feature dataset into the preset time series neural network model to obtain the second time series feature dataset after dimensionality reduction.

[0026] The first time series feature dataset is used to indicate information such as temperature, pressure, current, stirring speed, and material capacity in the reactor 31.

[0027] It should be noted that the first time series feature dataset is a T*M data set, where T is the time step and M is the number of channels of the data in the first time series feature dataset (such as temperature, pressure, current, stirring speed and material capacity, etc.). The number of channels is consistent with the number of sensors in the sampling module 11.

[0028] The pre-defined Temporal Convolutional Network (TCN) model can reduce the dimensionality of the first time series feature dataset through dilated causal convolution and residual connections. While reducing the data size, it removes redundant information and obtains a second time series feature dataset that is concise, structurally normalized, and has clear features, which is convenient for subsequent calculation and analysis.

[0029] For example, the first time series feature dataset with dimension T*M Input into TCN to obtain d-dimensional second time series feature data. , where d=32.

[0030] In some implementations, the data processing module 12 can extract the first time series feature dataset from the process data and input the first time series feature dataset into a preset time-series neural network model to obtain the second time series feature dataset after dimensionality reduction.

[0031] In some implementations, the data processing module 12 includes several edge controllers 121 (see...). Figure 1 Each edge controller 121 is located in a reactor 31 and is connected to the sampling module 11 of the reactor 31 to obtain the process data sampled by the sampling module 11.

[0032] Step S203: Based on the historical production dataset of the production target sample, and combined with the second time series feature dataset, determine the target feature vector dataset for the current moment.

[0033] The historical production dataset for the target sample consists of digitized records of the "best quality batch" from the historical production records of the target sample. The data types in this dataset are no less than those in the process data (temperature, pressure, current, stirring speed, material capacity, etc.). The "best quality batch" is the best target sample selected from the historical production records according to preset indicators, which include, but are not limited to, the target sample having the most stable performance and / or the highest corrosion resistance.

[0034] For example, when producing PVC resin through reactor array 300, the preset indicators are that the viscosity, impurity particle count, and volatile content of the product are within a preset range, and the differences between each batch of products are small.

[0035] In some implementations, the target feature vector dataset at the current time satisfies the following relationship:

[0036] in, This is the target feature vector dataset at the current moment, which indicates the residual between the historical production dataset and the current process data. This is the historical production dataset for producing the target sample.

[0037] In some implementations, the data processing module 12 can determine the target feature vector dataset at the current moment based on the historical production dataset and the second time series feature dataset of the production target sample.

[0038] Through the above steps, the target feature vector dataset at the current moment is determined based on the historical production dataset and the second time series feature dataset of the production target sample. General process fluctuations (such as normal temperature changes caused by reactions within reactor 31) are eliminated, while abnormal features are retained, thereby improving the accuracy of control over reactor array 300.

[0039] Step S204: Based on the weighted Mahalanobis distance of the target feature vector dataset, determine the weighted statistical distance value of each reactor 31 in the reactor array 300.

[0040] In some implementations, the data processing module 12 can determine the weighted statistical distance value of each reactor 31 in the reactor array 300 based on the weighted Mahalanobis distance of the target feature vector dataset.

[0041] like Figure 3 As shown, in some implementations, the weighted statistical distance value of each reactor 31 in the reactor array 300 is determined based on the weighted Mahalanobis distance of the target feature vector dataset, specifically including the following steps: Step S301: Obtain the reference feature dataset of the historical production dataset and determine the inverse matrix of the covariance matrix of the reference feature dataset.

[0042] It should be noted that the weighted Mahalanobis distance is more sensitive to high-dimensional data, making it more difficult to calculate when dealing with high-dimensional data, and the computational cost of calculating the inverse of the covariance matrix is ​​also large.

[0043] Edge controller 121 can acquire a reference feature dataset of historical production dataset to determine the inverse matrix of the covariance matrix of the reference feature dataset. Since edge controller 121 is located in reactor 31, it needs to meet the requirements of low power consumption and small size, so the computing power of edge controller 121 is relatively small.

[0044] In this implementation, the reference feature datasets of the historical production datasets are all dimensionality-reduced through feature extraction using a pre-defined temporal neural network model. Therefore, when calculating the covariance matrix of the reference feature datasets and solving for the covariance matrix, the computational load can be significantly reduced, and the computational efficiency can be improved. This ensures that the computing power of the edge controller 121 is sufficient to meet the computational power required to invert the covariance matrix of the reference feature datasets, achieving millisecond-level real-time computation. In some embodiments, each reactor 31 is equipped with an edge controller 121, which is used to determine the weighted statistical distance value of the corresponding reactor 31.

[0045] The dimensionality of the second time series feature dataset and the computing power of the edge controller satisfy the following conditions: When calculating the weighted Mahalanobis distance based on the target feature vector dataset, the floating-point operation of the inverse matrix is ​​less than the single-cycle computing power limit of the edge controller 121 under the first preset ratio, and the data storage within a single time window is less than the working memory of the edge controller 121 under the second preset ratio.

[0046] The dimension of the target feature vector dataset is the same as the dimension of the second time series feature dataset.

[0047] Specifically, the order of magnitude of the floating-point operations of the inverse matrix is ​​the cube of the dimension of the inverse matrix. The single-cycle computing power limit refers to the maximum computing power of the edge controller 121 within a real-time control cycle. The real-time control cycle is a fixed time interval during which the health monitoring device 100 periodically executes the complete process of "sampling-calculation-control output". A single time window includes several real-time control cycles.

[0048] For example, if the first preset ratio is 50%, the real-time control cycle is 10ms, and the single-cycle computing power limit of the edge controller 121 is 10^8 FLOPs, then the floating-point operation of the inverse matrix is ​​less than 10^7 FLOPs. If a single time window includes two real-time control cycles, then the single time window is 20ms. If the second preset ratio is 10%, and the total memory of the edge controller 121 is 1MB, then the data storage within a single time window is less than 0.1MB.

[0049] Step S302: Generate an attention weight matrix corresponding to the reaction process at the current moment based on the pre-established process map.

[0050] The process map is built based on historical production datasets, and the attention weight matrix is ​​a d×d diagonal weight matrix, which is used to indicate the weight of process data in different reaction processes and assign the weight to the corresponding features.

[0051] For example, the process includes a violent exothermic stage and a stirring and mixing stage. In the violent exothermic stage, the attention weight matrix can amplify the weights of the feature dimensions related to temperature; in the stirring and mixing stage, the attention weight matrix can amplify the weights related to current and stirring speed.

[0052] It should be noted that the difference between the weighted Mahalanobis distance and the Mahalanobis distance is the introduction of an attention weight matrix. By actively assigning weights to different features, the system monitors features of different importance in the reaction process at the current moment, thereby achieving dynamic control and improving the accuracy of controlling the reactor array 300 in the reaction process at different moments.

[0053] Step S303: Calculate the weighted statistical distance value based on the target feature vector dataset, the inverse matrix, and the attention weight matrix.

[0054] In some implementations, the weighted statistical distance values ​​satisfy the following relationship: ; In the formula, This represents the weighted statistical distance value. This represents the transpose of the target feature vector dataset. Represents the attention weight matrix. Represents the inverse matrix. This represents the target feature vector dataset.

[0055] It should be noted that the transpose of the target feature vector dataset and the target feature vector dataset can guarantee the matching of the computational dimensions.

[0056] Step S205: Obtain the health score corresponding to the weighted statistical distance value based on the preset mapping relationship.

[0057] In some implementations, the health score satisfies the following relationship:

[0058] In the formula, Indicates health score, This represents the normal operating threshold of reactor 31, i.e., the allowable deviation range when reactor 31 is operating normally, which is obtained from several weighted statistical distance values.

[0059] In some implementations, the edge controller 121 can obtain a health score corresponding to the weighted statistical distance value of the reactor 31 based on a preset mapping relationship, thereby realizing the control of the reactor 31.

[0060] Step S206: Based on the health score, control each reactor 31 in the reactor array 300 to perform the corresponding work.

[0061] The health monitoring method for the reactor array 300 provided in this application samples process data during the production of a target sample by the reactor array 300, and performs dimensionality reduction on the process data through a preset temporal neural network model to obtain a target feature vector dataset. The inverse matrix of the covariance matrix of the dimensionality-reduced reference feature dataset is calculated, and an attention weight matrix is ​​introduced to obtain a weighted statistical distance value. A health score is obtained through a preset mapping relationship, reducing the computational load when controlling the reactor array 300. This enables the edge controller 121 to monitor abnormalities in the reactor 31 in real time and control each reactor to perform corresponding operations, improving the accuracy of control over the reactor array 300. This allows for timely detection of abnormalities in the reactor 31, reducing property losses caused by reactor 31 abnormalities and improving the safety of the reactor array 300. Figure 4 As shown, in some embodiments, the health monitoring method further includes performing corresponding controls on each reactor 31 based on a health score, specifically including the following steps: Step S401: Obtain a health score.

[0062] In some implementations, each reactor 31 is equipped with a PID controller 32, which can change the status parameters of the reactor 31. These status parameters include at least one of temperature, pressure, and rotational speed.

[0063] In this implementation, the data processing module 12 also includes a main controller 122 (see...). Figure 1 The main controller 122 is connected to each edge controller 121 to obtain the health score obtained by each edge controller 121, and performs corresponding control on the corresponding reactor 31 through the PID controller 32 based on the health score.

[0064] Step S402: Determine if the health score is within the first safety threshold range. If yes, proceed to step S403; otherwise, proceed to step S404.

[0065] It should be noted that if the health score is within the first safety threshold range, it means that the reactor 31 is in a healthy state and there is no need to change the control strategy for the reactor 31.

[0066] Step S403: The PID controller 32 used to control the reactor 31 maintains the current control state.

[0067] Step S404: Determine if the health score is within the second safety threshold range. If yes, proceed to step S405; otherwise, proceed to step S406.

[0068] It should be noted that if the health score is within the second safety threshold range, it indicates that the reactor 31 is in a sub-healthy state. The reactor 31 has abnormalities such as scaling and wear. Although it can still maintain its function, the control strategy for the reactor 31 needs to be changed to avoid the risk of sudden failure in the future.

[0069] Step S405: Replace reactor 31, whose health score is within the second safety threshold range, with a spare reactor from reactor array 300.

[0070] like Figure 5 As shown, in some implementations, the reactor 31 with a health score within the second safety threshold range is defined as the target reactor, and the reactor 31 with a health score within the second safety threshold range is replaced by a spare reactor in the reactor array 300. Specifically, this includes the following steps: Step S501: Map the status indicators of the target reactor to the standby reactor, so that the status indicators of the standby reactor are consistent with those of the target reactor.

[0071] Step S502: Obtain the node characteristics of the target reactor according to the process parameters, and map the node characteristics to the PID controller 32 connected to the standby reactor, so as to adjust the controller parameters of the PID controller 32 connected to the standby reactor.

[0072] Among them, node features are used to indicate at least one of the following in the target reactor: reaction progress, material properties, and historical disturbances.

[0073] Through the above steps, the standby vessel is adjusted so that it can operate with the same state indicators and controller parameters of the PID controller 32 as the target reactor, facilitating non-destructive thermal switching.

[0074] like Figure 6 As shown, in some implementations, the reactor 31 with a health score within the second safety threshold range is defined as the target reactor, and the reactor 31 with a health score within the second safety threshold range is replaced by a spare reactor in the reactor array 300. Specifically, this includes the following steps: Step S601: Transfer the material in the target reactor to the standby reactor and determine the switching function for the material transfer progress over time.

[0075] Step S602: Adjust the first flow rate of the material flowing out of the target reactor and the second flow rate of the material flowing out of the standby reactor according to the switching function.

[0076] In some implementations, the switching function is the Sigmoid function, which satisfies the following relationship: ; In the formula, Let represent the switching function, e represent the natural constant, k represent the slope parameter that controls the steepness of the curve of the switching function, and t and t0 represent the current time and start time of the material transfer, respectively.

[0077] The first flow and the second flow satisfy the following relationship: ; In the formula, Indicates the first flow rate. Indicates the second flow rate. This indicates the total amount of materials transferred.

[0078] The sum of the first flow rate and the second flow rate always equals the total amount of material transferred, satisfying the material conservation constraint.

[0079] Step S406: Determine if the health score is within the third safety threshold range. If yes, proceed to step S407; otherwise, end the process.

[0080] It should be noted that if the health score is within the third safety threshold range, it indicates that the reactor 31 is in a faulty state and the operation of the reactor array 300 needs to be stopped immediately to avoid safety hazards.

[0081] Step S407: Block the transmission channels between each reactor 31.

[0082] In this embodiment of the application, the first security threshold range is greater than the second security threshold range, and the second security threshold range is greater than the third security threshold range.

[0083] For example, if If so, the reactor 31 is determined to be in a healthy state and its current state is maintained; if If the reactor 31 is in a sub-healthy state, it is determined that the reactor 31 is abnormal but can still maintain its function, triggering a non-destructive thermal switching and scheduling the backup reactor to intervene. If the reaction fails, the reactor 31 is determined to be in a faulty state, triggering a safety interlock to urgently halt the reaction and perform maintenance on the reactor 31. Non-destructive thermal switching refers to switching the reactor 31 to a standby reactor while ensuring continuous production; the safety interlock refers to quickly stopping the reactor 31.

[0084] Through the above steps, corresponding controls are implemented on each reactor 31 based on the health score, ensuring production efficiency while avoiding safety hazards.

[0085] like Figure 7As shown, this application also provides a computer device 200, which includes a memory 21 and a processor 22. The memory 21 stores a computer program. When the computer program is executed by the processor 22, the processor 22 performs the steps of the above-described health monitoring method for the reactor array.

[0086] Specifically, processor 22 may include a central processing unit, or an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement embodiments of the present invention.

[0087] In some implementations, memory 21 may include a large-capacity memory for data or instructions.

[0088] For example, the memory includes a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disk, a magneto-optical disk, a universal serial bus (USB) drive, or any combination of the above-mentioned memory.

[0089] For example, memory 21 may be located inside or outside the computer device 200.

[0090] In some possible implementations, the computer device 200 also includes a communication interface 23 and a bus 24. The processor 22, memory 21, and communication interface 23 are connected via the bus 24 and communicate with each other.

[0091] The communication interface 23 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0092] Bus 24 includes hardware and / or software that couples components of computer device 200 together. For example, and not to limit, bus 24 may include an accelerated graphics port or other graphics bus, an enhanced industry standard architecture bus, a front-side bus, a low pin count bus, a memory bus, or other suitable bus or any combination of the above buses.

[0093] This application also provides a computer-readable storage medium (not shown) storing a computer program that, when executed by a processor, implements the above-described health monitoring method for a reactor array.

[0094] The computer-readable storage medium includes, but is not limited to, electronic, magnetic, optical, infrared, or other physical storage devices or equipment, which may contain or store information such as executable instructions, data, etc. More specific examples of computer-readable storage media include electrical connections based on one or more wires, RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, storage drives (such as hard disk drives), SSDs (Solid State Disks), any type of storage disk (such as optical discs), or similar memory, or any suitable combination of the foregoing. This application also provides a computer-readable storage medium (not shown) storing a computer program that, when executed, implements the aforementioned health monitoring method for a reactor array.

[0095] Computer-readable storage media include, but are not limited to, electronic, magnetic, optical, infrared, or other physical storage devices or apparatuses that may contain or store information such as executable instructions, data, etc. More specific examples of computer-readable storage media include electrical connections based on one or more wires, RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, storage drives (such as hard disk drives), SSDs (Solid State Disks), any type of storage disk (such as optical discs), or similar memory, or any suitable combination of the foregoing.

[0096] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for health monitoring of a reactor array, characterized in that, The process data of the reactor array during the production of the target sample are sampled; Extract the first time series feature dataset from the process data, and input the first time series feature dataset into a preset time-series neural network model to obtain the second time series feature dataset after dimensionality reduction; Based on the historical production dataset of the target sample, the target feature vector dataset at the current moment is determined by combining the second time series feature dataset. The target feature vector dataset is used to indicate the residual between the historical production dataset and the process data at the current moment. Based on the weighted Mahalanobis distance of the target feature vector dataset, the weighted statistical distance value of each reactor in the reactor array is determined, and the health score corresponding to the weighted statistical distance value is obtained according to the preset mapping relationship. Based on the health score, the reactors in the reactor array are controlled to perform corresponding operations.

2. The health monitoring method according to claim 1, characterized in that, The determination of the weighted statistical distance value of each reactor in the reactor array based on the weighted Mahalanobis distance of the target feature vector dataset includes: Obtain a reference feature dataset of the historical production dataset, and determine the inverse matrix of the covariance matrix of the reference feature dataset; An attention weight matrix corresponding to the reaction process at the current moment is generated based on a pre-established process map, wherein the process map is established based on the historical production dataset, and the attention weight matrix is ​​used to indicate the weight of the process data in different reaction processes. The weighted statistical distance value is calculated based on the target feature vector dataset, the inverse matrix, and the attention weight matrix.

3. The health monitoring method according to claim 2, characterized in that, The step of calculating the weighted statistical distance value based on the target feature vector dataset, the inverse matrix, and the attention weight matrix includes: The weighted statistical distance values ​​satisfy the following relationship: ； In the formula, This represents the transpose of the target feature vector dataset. This represents the attention weight matrix. Denotes the inverse matrix, This refers to the target feature vector dataset.

4. The health monitoring method according to claim 2, characterized in that, Each of the aforementioned reactors is equipped with an edge controller, which is used to determine the weighted statistical distance value of the corresponding reactor; The dimensionality of the second time-series feature dataset and the computing power of the edge controller satisfy the following conditions: When calculating the weighted Mahalanobis distance based on the target feature vector dataset, the floating-point operation of the inverse matrix is ​​less than the single-cycle computing power limit of the edge controller under the first preset ratio, and the data storage within a single time window is less than the working memory of the edge controller under the second preset ratio; wherein, the dimension of the target feature vector dataset is the same as the dimension of the second time series feature dataset.

5. The health monitoring method according to claim 1, characterized in that, The health monitoring method also includes: Based on the health score, corresponding controls are performed on each of the reactors; If the health score is within the first safety threshold range, the PID controller used to control the reactor maintains the current control state. If the health score is within the second safety threshold range, the reactor with the health score within the second safety threshold range is replaced by a spare reactor in the reactor array; If the health score is within the third safety threshold range, the transmission channel between the reactors is blocked. Wherein, the first security threshold range is greater than the second security threshold range, and the second security threshold range is greater than the third security threshold range.

6. The health monitoring method according to claim 5, characterized in that, The reactor whose health score falls within the second safety threshold range is defined as the target reactor; The step of replacing the reactor whose health score is within the second safety threshold range with a spare reactor from the reactor array includes: The status indicators of the target reactor are mapped to the standby reactor, so that the status indicators of the standby reactor are consistent with the status indicators of the target reactor. The status indicators include at least one of temperature, pressure and rotation speed. The node characteristics of the target reactor are obtained based on the process parameters, and the node characteristics are mapped to the PID controller connected to the standby reactor to adjust the controller parameters of the PID controller connected to the standby reactor; the node characteristics are used to indicate at least one of the reaction progress, material characteristics, and historical disturbances in the target reactor.

7. The health monitoring method according to claim 5 or 6, characterized in that, The reactor whose health score falls within the second safety threshold range is defined as the target reactor; The step of replacing the reactor whose health score is within the second safety threshold range with a spare reactor from the reactor array includes: The material in the target reactor is transferred to the standby reactor, and a switching function is determined to show how the material transfer progress changes over time. The first flow rate of the material flowing out of the target reactor and the second flow rate of the material flowing out of the standby reactor are adjusted according to the switching function.

8. The health monitoring method according to claim 7, characterized in that, The switching function is a Sigmoid function, and the switching function satisfies the following relationship: ； In the formula, Let e ​​represent the natural constant, k represent the slope parameter that controls the steepness of the curve of the switching function, and t and t0 represent the current time and start time of the material transfer, respectively. The first flow rate and the second flow rate satisfy the following relationship: ； In the formula, This represents the first flow rate. This indicates the second flow rate. This indicates the total amount of materials transferred.

9. A health monitoring device for a reactor array, characterized in that, include: The sampling module is used to sample the process data of the reactor array during the production of the target sample; The data processing module is used to extract a first time series feature dataset from the process data and input the first time series feature dataset into a preset time-series neural network model to obtain a second time series feature dataset after dimensionality reduction. Based on the historical production dataset of the target sample, the target feature vector dataset at the current moment is determined by combining the second time series feature dataset. The target feature vector dataset is used to indicate the deviation between the historical production dataset and the process data at the current moment. Based on the weighted Mahalanobis distance of the target feature vector dataset, the weighted statistical distance value of each reactor in the reactor array is determined; A health score corresponding to the weighted statistical distance value is obtained based on a preset mapping relationship.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, performs the health monitoring method according to any one of claims 1-8.

Images