Chemical production monitoring system based on cloud computing

Abstract

The invention relates to the technical field of chemical production monitoring, and discloses a cloud computing-based chemical production monitoring system, which is characterized in that an on-site data acquisition module is used for acquiring chemical production data in real time and carrying out filtering and denoising processing to obtain preprocessed data; the network transmission module is used for encrypting the transmitted data and reducing the data transmission quantity by adopting a data compression technology; the cloud computing platform module is used for dispersedly storing the data on a plurality of storage nodes by utilizing a distributed storage technology, classifying the data, analyzing the data in real time by adopting a streaming computing technology, detecting the abnormal condition of the data, and immediately triggering an alarm mechanism when the abnormal condition exceeds a preset safety threshold value; the monitoring application module is used for displaying various parameters and equipment operation states, and providing decision suggestions for production optimization in combination with a chemical production process and equipment; the production efficiency can be improved, the production cost can be reduced, the product quality can be improved, and the market competitiveness of enterprises can be enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical production monitoring, and in particular to a chemical production monitoring system based on cloud computing. Background Art

[0002] Chemical production processes are usually characterized by high temperature, high pressure, flammability, explosiveness, toxicity and harmfulness. The production process is complex and involves numerous equipment and process parameters. Therefore, real-time and accurate monitoring of chemical production processes is crucial. Traditional chemical production monitoring systems generally use local servers for data storage and processing. This approach has many limitations. On the one hand, the storage and computing capabilities of local servers are limited, making it difficult to cope with the massive data generated during the chemical production process. Especially when processing complex data analysis and prediction tasks, they are often inefficient and cannot meet the needs of real-time monitoring. On the other hand, traditional monitoring systems have poor scalability. When the production scale expands or the monitoring demand increases, the cost of upgrading and maintaining the system is high, and system performance bottlenecks are prone to occur. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a chemical production monitoring system based on cloud computing.

[0004] The present invention provides a chemical production monitoring system based on cloud computing, which includes: The field data acquisition module is used to collect chemical production data in real time through sensors at the chemical production site, and filter and denoise the collected data to obtain pre-processed data; The network transmission module is used to transmit the pre-processed data to the cloud computing platform module by combining industrial Ethernet and wireless network, encrypt the transmitted data, and use data compression technology to reduce the data transmission volume; The cloud computing platform module is used to receive data transmitted by the network transmission module, and uses distributed storage technology to disperse and store the data on multiple storage nodes. It classifies and processes the data according to the data type and application requirements, and uses streaming computing technology to perform real-time analysis to detect data anomalies. When the data exceeds the preset safety threshold, an alarm mechanism is immediately triggered; The monitoring application module is used to display various parameters of the chemical production process and the operating status of equipment in real time on a visual interface based on the processing results of the cloud computing platform module, and to provide decision-making suggestions for production optimization in combination with chemical production processes and equipment.

[0005] Optionally, in a first implementation of the present invention, the field data acquisition module includes: The conversion submodule is used for sensors at chemical production sites to collect production data in real time according to the frequency, convert the collected original analog signals into digital signals and output them to complete data collection; The filtering submodule is used to perform preliminary filtering on the collected original digital signal using the mean filtering method to remove the impulse noise and salt and pepper noise in the data to obtain the filtered data; The reconstruction submodule is used to input the filtered data into the denoising diffusion probability model, input the clean signal data reconstructed by the denoising diffusion probability model into the cyclic variational autoencoder, reconstruct the processed output data, and obtain the preprocessed data.

[0006] Optionally, in a second implementation of the present invention, the reconstruction submodule includes: The filtered data is divided into time periods of fixed length according to time. The data in each time period constitutes a sequence. Noise is gradually added to the data until the data reaches the set maximum noise level. Using the UNet model, starting from the final noisy state, the noise is gradually predicted and removed until the final reconstructed clean data is obtained. The clean data is arranged in chronological order to obtain time series data, and then input into the recurrent variational autoencoder. The encoder generates a hidden state sequence to capture the temporal dependency, calculates the parameters of the variational posterior distribution through the hidden state, samples the latent variables from the posterior distribution, inputs the latent variables into the decoder, initializes the decoder hidden state, generates a reconstructed sequence time step by time, and obtains the preprocessed data.

[0007] Optionally, in a third implementation of the present invention, the network transmission module includes: The capture submodule is used to divide the preprocessed data into blocks according to time windows. Each block extracts temporal features through the Transformer encoder and uses the self-attention mechanism to capture long-range dependencies in the data. The compression submodule is used to input the feature vector output by the Transformer into the INN and compress the data dimension through reversible transformation; The perturbation submodule is used to generate adversarial samples using quantum neural networks, perturb the compressed data, map the perturbed data into quantum state space, and enhance encryption through quantum superposition and entanglement; The encryption submodule is used by the sender and receiver to generate session keys through the quantum key distribution protocol, and to encrypt the compressed data using a symmetric encryption algorithm. The key is derived from the dynamically generated session key and the key matrix embedded in the INN, and the session key is encrypted using an asymmetric encryption algorithm.

[0008] Optionally, in a fourth implementation of the present invention, the compression submodule includes: The feature vector output by the Transformer encoder is used to generate a three-dimensional key matrix based on the current timestamp, device ID, and random number. The three-dimensional key matrix is ​​mapped to the same dimensional space as the feature vector through a hash function. The INN network is divided into multiple coupling layers. Each coupling layer includes a coupling layer, a transformation function, and a reversible transformation. After each coupling layer, the feature dimensions are rearranged through a channel shuffle operation, the feature selection mechanism is applied to discard redundant dimensions, and the attention mechanism is used to assign weights to different features. The feature vector after multi-layer coupling transformation is reshaped to obtain the compressed latent space representation, and the compression ratio is calculated. The check code is added to generate the final output vector.

[0009] Optionally, in a fifth implementation of the present invention, the perturbation submodule includes: Input the compressed data into the quantum neural network and generate a perturbation vector that maximizes the loss function by adjusting the parameters until the perturbation reaches a preset intensity. Normalize the perturbed data vector, use the normalized data elements as the amplitude of the quantum state, map them to the quantum register, apply the Hadamard gate operation, convert the quantum bit from the quantum bit initialization to the superposition state, and encode the data characteristics into the phase and amplitude of the quantum state through the parameterized rotation gate; Non-local correlations are created between quantum bits to generate highly entangled states, and quantum entanglement-enhanced encryption is performed through controlled rotation gates.

[0010] Optionally, in a sixth implementation of the present invention, the cloud computing platform module includes: The collection submodule is used to collect hardware indicators, network bandwidth and historical load data of storage nodes, and generate multi-dimensional features for each node, where the multi-dimensional features include at least processing power, storage capacity and reliability score; The allocation submodule abstracts storage nodes as vertices in a graph, with physical connections between nodes as edges. It applies a graph convolutional network to learn node embeddings, calculates similarity based on node embeddings, groups similar nodes into data shards, and allocates shards based on data type and access pattern. It monitors node load in real time and, when hotspots emerge, uses a graph convolutional network to predict load trends and dynamically adjust shard allocation. The classification submodule is used to automatically identify data types through data metadata and classify structured data, semi-structured data, and unstructured data; Set up submodules to deploy data stream processing tasks in Apache Flink, set parallelism and watermarks to process out-of-order data, and extract statistical features from time series data using sliding windows. The construction submodule is used to construct a causal graph. The TC-CNN model is used to capture the temporal dependencies and causal relationships between variables. The residual between the actual observation value and the TC-CNN model prediction value is calculated. When the residual exceeds the preset threshold and violates the causal graph relationship, it is marked as an anomaly.

[0011] Optionally, in a seventh implementation of the present invention, the process of constructing the causal graph includes: Filter out key variables from the data and organize them into structured multivariate time series data, where the key variables include at least temperature, pressure, flow rate and liquid level; A constraint-based causal discovery algorithm is used to perform causal relationship analysis on multivariate time series data. Taking the data within the time window as the unit, the conditional independence between variables is calculated, and irrelevant edges are gradually eliminated to construct a preliminary causal graph structure. The preliminary causal graph structure is then verified and corrected to obtain the final causal graph.

[0012] Optionally, in an eighth implementation of the present invention, the construction submodule includes: Construct a TC-CNN network and train it using multivariate time series data to obtain a TC-CNN model. The TC-CNN network uses causal convolutional layers at the bottom layer, adds an attention mechanism to the middle layer, and outputs the predicted value of each variable through a fully connected layer at the top layer. The trained TC-CNN model is deployed in a real-time computing environment and integrated with the data stream processing task. For the real-time incoming data, the model predicts the value of each variable at a fixed time step and outputs the predicted value at the future time point. For each time step, the difference between the actual observation value and the TC-CNN prediction value is calculated to obtain the residual sequence. Based on the constructed causal graph, the relationship between the variable with residual anomaly and its causal association variable is checked to see whether it meets the expectations. When the residual exceeds the preset threshold and violates the causal graph relationship, it is marked as an anomaly.

[0013] Optionally, in the ninth implementation of the present invention, the processing results output by the cloud computing platform module include at least real-time process parameters, equipment operation status data, anomaly detection results, and data analysis and prediction results in the chemical production process.

[0014] In the technical solution provided by the present invention, chemical production data is collected in real time through sensors at the chemical production site, and the collected data is filtered and denoised to obtain pre-processed data; the pre-processed data is transmitted to the cloud computing platform module by combining industrial Ethernet and wireless networks, the transmitted data is encrypted, and data compression technology is used to reduce the amount of data transmission; distributed storage technology is used to store data in multiple storage nodes, and the data is classified and processed according to the type of data and application requirements. Streaming computing technology is used for real-time analysis to detect data anomalies. When the preset safety threshold is exceeded, an alarm mechanism is immediately triggered; various parameters of the chemical production process and the operating status of the equipment are displayed in real time on the visual interface, and decision-making suggestions are provided for production optimization in combination with chemical production processes and equipment; the present invention uses the distributed storage and parallel computing technology of the cloud computing platform to quickly process the chemical production process. The massive data generated during the process meets the needs of real-time monitoring and data analysis, and improves the response speed and processing efficiency of the monitoring system; the cloud computing platform has good scalability and can flexibly increase computing resources and storage resources according to the expansion of chemical production scale and the increase in monitoring needs, without the need for large-scale hardware upgrades and modifications to the system, reducing the construction and maintenance costs of the system; through real-time monitoring and timely alarms, it can quickly discover safety hazards and equipment failures in the chemical production process, take timely measures to deal with them, avoid accidents, ensure the safe operation of chemical production, provide scientific decision-making support for production optimization, help improve production efficiency, reduce production costs, improve product quality, and enhance the market competitiveness of enterprises; the monitoring application layer supports multi-terminal access, so that monitoring personnel and managers can view production monitoring data through different devices anytime and anywhere, which is convenient for timely grasp of production conditions and remote management and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0016] Figure 1 A schematic diagram of the structure of a chemical production monitoring system based on cloud computing provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a field data acquisition module provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a network transmission module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar items and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that shown or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, product, or device.

[0018] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A schematic diagram of the structure of a chemical production monitoring system based on cloud computing provided by an embodiment of the present invention includes: The field data acquisition module is used to collect chemical production data in real time through sensors at the chemical production site, and filter and denoise the collected data to obtain pre-processed data; The network transmission module is used to transmit the pre-processed data to the cloud computing platform module by combining industrial Ethernet and wireless network, encrypt the transmitted data, and use data compression technology to reduce the data transmission volume; The cloud computing platform module is used to receive data transmitted by the network transmission module, and uses distributed storage technology to disperse and store the data on multiple storage nodes. It classifies and processes the data according to the data type and application requirements, and uses streaming computing technology to perform real-time analysis to detect data anomalies. When the data exceeds the preset safety threshold, an alarm mechanism is immediately triggered; The monitoring application module is used to display various parameters of the chemical production process and the operating status of equipment in real time on a visual interface based on the processing results of the cloud computing platform module, and to provide decision-making suggestions for production optimization in combination with chemical production processes and equipment.

[0019] In this embodiment, please refer to Figure 2 , the field data acquisition module includes: The conversion submodule is used for sensors at chemical production sites to collect production data in real time according to the frequency, convert the collected original analog signals into digital signals and output them to complete data collection; The filtering submodule is used to perform preliminary filtering on the collected original digital signal using the mean filtering method to remove the impulse noise and salt and pepper noise in the data to obtain the filtered data; The reconstruction submodule is used to input the filtered data into the denoising diffusion probability model, input the clean signal data reconstructed by the denoising diffusion probability model into the cyclic variational autoencoder, reconstruct the processed output data, and obtain the preprocessed data.

[0020] In this embodiment, the reconstruction submodule includes: The filtered data is divided into time periods of fixed length according to time. The data in each time period constitutes a sequence. Noise is gradually added to the data until the data reaches the set maximum noise level. Using the UNet model, starting from the final noisy state, the noise is gradually predicted and removed until the final reconstructed clean data is obtained. The clean data is arranged in chronological order to obtain time series data, and then input into the recurrent variational autoencoder. The encoder generates a hidden state sequence to capture the temporal dependency, calculates the parameters of the variational posterior distribution through the hidden state, samples the latent variables from the posterior distribution, inputs the latent variables into the decoder, initializes the decoder hidden state, generates a reconstructed sequence time step by time, and obtains the preprocessed data.

[0021] In this embodiment, please refer to Figure 3 , the network transmission module includes: The capture submodule is used to divide the preprocessed data into blocks according to time windows. Each block extracts temporal features through the Transformer encoder and uses the self-attention mechanism to capture long-range dependencies in the data. The compression submodule is used to input the feature vector output by the Transformer into the INN and compress the data dimension through reversible transformation; The perturbation submodule is used to generate adversarial samples using quantum neural networks, perturb the compressed data, map the perturbed data into quantum state space, and enhance encryption through quantum superposition and entanglement; The encryption submodule is used by the sender and receiver to generate session keys through the quantum key distribution protocol, and to encrypt the compressed data using a symmetric encryption algorithm. The key is derived from the dynamically generated session key and the key matrix embedded in the INN, and the session key is encrypted using an asymmetric encryption algorithm.

[0022] In this embodiment, the compression submodule includes: The feature vector output by the Transformer encoder is used to generate a three-dimensional key matrix based on the current timestamp, device ID, and random number. The three-dimensional key matrix is ​​mapped to the same dimensional space as the feature vector through a hash function. The INN network is divided into multiple coupling layers. Each coupling layer includes a coupling layer, a transformation function, and a reversible transformation. After each coupling layer, the feature dimensions are rearranged through a channel shuffle operation, the feature selection mechanism is applied to discard redundant dimensions, and the attention mechanism is used to assign weights to different features. The feature vector after multi-layer coupling transformation is reshaped to obtain the compressed latent space representation, and the compression ratio is calculated. The check code is added to generate the final output vector.

[0023] In this embodiment, the disturbance submodule includes: Input the compressed data into the quantum neural network and generate a perturbation vector that maximizes the loss function by adjusting the parameters until the perturbation reaches a preset intensity. Normalize the perturbed data vector, use the normalized data elements as the amplitude of the quantum state, map them to the quantum register, apply the Hadamard gate operation, convert the quantum bit from the quantum bit initialization to the superposition state, and encode the data characteristics into the phase and amplitude of the quantum state through the parameterized rotation gate; Non-local correlations are created between quantum bits to generate highly entangled states, and quantum entanglement-enhanced encryption is performed through controlled rotation gates.

[0024] In this embodiment, the cloud computing platform module includes: The collection submodule is used to collect hardware indicators, network bandwidth and historical load data of storage nodes, and generate multi-dimensional features for each node, where the multi-dimensional features include at least processing power, storage capacity and reliability score; The allocation submodule abstracts storage nodes as vertices in a graph, with physical connections between nodes as edges. It applies a graph convolutional network to learn node embeddings, calculates similarity based on node embeddings, groups similar nodes into data shards, and allocates shards based on data type and access pattern. It monitors node load in real time and, when hotspots emerge, uses a graph convolutional network to predict load trends and dynamically adjust shard allocation. The classification submodule is used to automatically identify data types through data metadata and classify structured data, semi-structured data, and unstructured data; Set up submodules to deploy data stream processing tasks in Apache Flink, set parallelism and watermarks to process out-of-order data, and extract statistical features from time series data using sliding windows. The construction submodule is used to construct a causal graph. The TC-CNN model is used to capture the temporal dependencies and causal relationships between variables. The residual between the actual observation value and the TC-CNN model prediction value is calculated. When the residual exceeds the preset threshold and violates the causal graph relationship, it is marked as an anomaly.

[0025] In this embodiment, the process of constructing the causal graph includes: Filter out key variables from the data and organize them into structured multivariate time series data, where the key variables include at least temperature, pressure, flow rate and liquid level; A constraint-based causal discovery algorithm is used to perform causal relationship analysis on multivariate time series data. Taking the data within the time window as the unit, the conditional independence between variables is calculated, and irrelevant edges are gradually eliminated to construct a preliminary causal graph structure. The preliminary causal graph structure is then verified and corrected to obtain the final causal graph.

[0026] In this embodiment, the construction submodule includes: Construct a TC-CNN network and train it using multivariate time series data to obtain a TC-CNN model. The TC-CNN network uses causal convolutional layers at the bottom layer, adds an attention mechanism to the middle layer, and outputs the predicted value of each variable through a fully connected layer at the top layer. The trained TC-CNN model is deployed in a real-time computing environment and integrated with the data stream processing task. For the real-time incoming data, the model predicts the value of each variable at a fixed time step and outputs the predicted value at the future time point. For each time step, the difference between the actual observation value and the TC-CNN prediction value is calculated to obtain the residual sequence. Based on the constructed causal graph, the relationship between the variable with residual anomaly and its causal association variable is checked to see whether it meets the expectations. When the residual exceeds the preset threshold and violates the causal graph relationship, it is marked as an anomaly.

[0027] In this embodiment, the processing results output by the cloud computing platform module include at least real-time process parameters in the chemical production process, equipment operation status data, anomaly detection results, and data analysis and prediction results.

[0028] In this embodiment, the field data acquisition module is responsible for the real-time collection and preliminary processing of raw data from the chemical production site, and is the source of system data; temperature sensors (such as thermocouples, RTDs): cover key points such as reactors, pipelines, towers, etc., with a sampling frequency of 1-10Hz. Pressure sensors (piezoresistive, capacitive): monitor pipeline pressure and container liquid level with an accuracy of 0.01%FS. Flow sensors (electromagnetic, ultrasonic): measure material flow rate and support bidirectional flow detection. Vibration and acoustic sensors: monitor the mechanical vibration frequency of equipment (such as pumps and compressors) in real time to detect early faults. Gas analyzers (infrared, laser): detect the concentration of toxic and harmful gases (such as H2S, CO), with a response time of <1 second; deploy industrial-grade embedded systems equipped with real-time operating systems; integrate programmable logic controllers (PLCs) and distributed control systems (DCS) to achieve local data caching and preprocessing.

[0029] In this embodiment, the transmission network is dynamically allocated according to the data type (such as real-time monitoring data, historical log) and priority (such as alarm information, general parameters); key production data (such as temperature and pressure thresholds) are preferentially transmitted through industrial Ethernet to ensure low latency; non-critical data (such as equipment status) are transmitted through 5G / 4G networks to balance bandwidth utilization; the network status (packet loss rate, delay) is monitored in real time, and the transmission rate is dynamically adjusted; when the network is congested, the sampling frequency of non-real-time data is reduced or data aggregation (such as average value and maximum value statistics) is performed; the pre-processed data is divided into blocks according to time windows (such as 10 seconds of data as a block), and the time series features of each block are extracted through the Transformer encoder; a self-attention machine is used to collect and analyze the data. The system captures long-distance dependencies in the data (such as the association between different sensor data); the feature vector output by the Transformer is input into the INN, and the data dimension is compressed through a series of reversible transformations; during the compression process, the dynamically generated key matrix (based on the current timestamp, device ID or random number) is embedded in the hidden space; the key matrix is ​​a necessary condition for decryption and is bound to the data for transmission; a quantum neural network (QNN) is used to generate adversarial samples to perturb the compressed data to increase the difficulty of decryption; the perturbed data is mapped to the quantum state space (such as quantum bits), and higher-strength encryption is achieved through quantum superposition and entanglement characteristics; each edge computing node (such as a local server in a factory workshop) is trained using local data Sparse autoencoders extract key features of the data; differential privacy technology is applied during the training process to protect the privacy of the original data by adding Gaussian noise, ensuring that the model parameters do not leak sensitive information; the edge nodes upload the trained model parameters (not the original data) to the cloud; the cloud server uses the federated averaging algorithm to aggregate the parameters of all nodes and update the global model; the updated model is sent to each edge node to optimize the local data compression and encryption strategy; the sender and receiver generate session keys through the quantum key distribution (QKD) protocol to ensure the unconditional security of key transmission; the session key is updated periodically (for example, every 5 minutes) and combined with the timestamp mechanism to prevent replay attacks; the compressed data is encrypted using a symmetric encryption algorithm (such as AES) Data is encrypted using a key derived from a dynamically generated session key and a key matrix embedded in the INN. Session keys are encrypted using an asymmetric encryption algorithm (such as RSA) to ensure the security of key exchange. The compression ratio is dynamically adjusted based on network bandwidth and data importance. For example, in high-bandwidth scenarios, a lower compression ratio is used to preserve more original information. In low-bandwidth or emergency scenarios, a higher compression ratio is used to prioritize timely data transmission. The optimal compression strategy is selected for different data types by combining lossless compression (such as LZ77 and Huffman coding) and lossy compression (such as INN-based latent space dimensionality reduction). Predictive coding is used for trend changes in time series data (such as a slow temperature increase), transmitting only the residual between the actual value and the predicted value.After receiving encrypted data, the cloud uses the matching key matrix and session key to decrypt it. Data integrity is ensured by verifying the data's hash value or digital signature, detecting tampering during transmission. The decrypted data is fed into a Transformer decoder, and the original data dimensions are restored using the reversible transform of the INN. The time series relationship of the data is reconstructed using an attention mechanism to restore complete production monitoring data. A global federated model is used to assess the quality of the received data and detect outliers or potential attacks. The recovered data is stored in a cloud database for subsequent analysis, visualization, or decision support. Quantum entanglement is used to generate unclonable keys to ensure encryption security. Quantum state measurements are used to detect eavesdropping attempts, triggering key updates and data retransmission upon detection. Encryption strength and compression strategies are dynamically adjusted based on data sensitivity (such as data involving core process parameters or personnel safety). For highly sensitive data, multi-layer encryption (such as simultaneous use of quantum state encoding and traditional encryption) and lower compression rates are enabled. Data characteristics (such as traffic patterns and data distribution) are monitored in real time during transmission to identify potential cyberattacks or data leaks. Once an anomaly is detected, the affected transmission channel is immediately disconnected, the emergency encryption mechanism is triggered, and security management personnel are notified.

[0030] In this embodiment, the INN network is divided into multiple coupling layers (CouplingLayer), each layer includes: Split operation: divide the input tensor into two parts x1 and x2. Transformation function: use a neural network (such as a convolutional network) to calculate the transformation parameters. Reversible transformation: apply affine transformation or additive transformation. The first layer of coupling operation: split the standardized feature vector x into x1 and x2, calculate the transformation parameters, apply the reversible transformation, and perform an XOR operation on the first part K1 of the key matrix K and y2. The second layer of coupling operation: merge y2' with x1, re-split it into y1 and y2'', calculate the transformation parameters, apply the reversible transformation, and perform an XOR operation on the second part K2 of the key matrix K and y1'. Through alternating processing through N coupling layers, the data dimension is gradually compressed and the key is embedded.

[0031] In this example, a highly available Flink cluster (at least three TaskManager nodes) is configured, along with the total memory and number of CPU cores. Incremental checkpointing is enabled, with a checkpoint interval (e.g., 30 seconds) and timeout set. Necessary dependencies are added, including Flink SQL, the Kafka Connector, and the HDFS Connector. Parameters for connecting to external systems (e.g., Kafka broker addresses and HDFS paths) are configured. A Flink KafkaConsumer is used to connect to the chemical production data topic, with a deserializer (e.g., JSON deserialization). The consumer group ID and starting consumption position (e.g., EARLIEST or TIMESTAMP) are set. Based on the number of data source partitions and cluster resources, the initial parallelism is set (e.g., allocating four slots to each TaskManager). A higher parallelism (e.g., parallelism = number of partitions) is used for I / O-intensive operations (e.g., Kafka consumption), while a lower parallelism is used for compute-intensive operations (e.g., complex aggregations). Event time semantics are implemented, using the data's native timestamp field. Configure a Timestamp Assigner to extract the timestamp field (such as timestamp or event_time) from the data. For bounded out-of-order data, use a BoundedOutOfOrdernessTimestampExtractor to generate a watermark, setting the maximum out-of-order time (such as 5 seconds). For unbounded out-of-order data, implement a custom watermark generator and dynamically adjust the watermark based on business logic (such as sensor data characteristics). Define a sliding window, setting the window size (such as 5 minutes) and the sliding step (such as 1 minute). Use SlidingEventTimeWindows.of(Time.minutes(5), Time.minutes(1)) to create a sliding window assigner. Apply a ProcessWindowFunction or AggregateFunction to extract statistical features: basic statistics: mean, variance, maximum, minimum, median. Trend features: linear regression slope (to determine upward / downward trends) and difference series. Volatility indicators: standard deviation and interquartile range (IQR).

[0032] Use AggregateFunction for incremental aggregation, gradually computing intermediate results as window data arrives, reducing memory usage. For complex features (such as moving averages), use Flink's ValueState or ListState to manage historical data. Configure custom triggers (such as EventTimeTrigger) to control when window computations are triggered. Set a window cleanup policy to release no longer needed window state to avoid memory leaks. Calculate dynamic thresholds for statistical features within a window (such as the 3σ principle based on historical data). Train anomaly detection models (such as Isolation Forest and PCA) using historical data from a sliding window. When a window statistic exceeds a threshold, mark it as a potential anomaly and append context (such as the window time range and the severity of the anomaly). Output anomaly data to a dedicated alert topic or write it to external storage (such as Elasticsearch). Write processed statistical features and anomaly information to a Kafka result topic. Configure a serializer (such as JSON serialization) and retry policy. For data that requires long-term storage, configure an HDFS or object storage sink and store it by time partition (such as one directory per day). Write real-time monitoring data to a time-series database such as Redis or InfluxDB. Monitor resource usage (CPU, memory, and network) of each operator in the task through the Flink web UI. Adjust operator parallelism or optimize data processing logic based on backpressure metrics (such as Input / Output Buffer Usage). Configure job restart strategies (such as fixed delay restart or exponential backoff restart). Regularly verify the recoverability of checkpoints and savepoints to ensure rapid recovery in the event of failures.

[0033] In this embodiment, various types of output data are received from the data processing module of the cloud computing platform, including real-time process parameters (such as temperature, pressure, flow, liquid level) in the chemical production process, equipment operating status data (rotating speed, vibration frequency, energy consumption), anomaly detection results, and data analysis and prediction results; the received data are cleaned to remove invalid values, missing values, and duplicate data; data in different formats are uniformly converted into a standard format that can be recognized by a visual interface, such as JSON or CSV format; they are classified according to data type and purpose, such as classifying process parameters into one category and equipment status data into another category; at the same time, associations between data are established, such as associating the operating status data of a certain equipment with the process parameters of the production link where the equipment is located. association; design the layout of the visual interface according to the chemical production process and user needs; display the key process parameters in the core area of ​​the interface in the form of dashboards, trend charts, etc.; present the equipment operation status in the form of equipment topology diagrams, status indicators, etc.; highlight abnormal information in the form of pop-up windows, highlighted warning boxes, etc.; for real-time process parameters, use line charts and bar charts to show the trend of parameter changes over time, such as updating the temperature line chart once every second; for the equipment operation status, use dynamic equipment model diagrams to intuitively display the operation, fault, standby and other states of the equipment through color, flashing and other effects, such as green for normal operation and red for fault; the abnormality detection results are displayed in the form of Gantt charts or timelines to show the time, duration and impact range of the abnormality Add interactive elements to the visual interface, and users can zoom in to view data for a specific time period, view detailed equipment parameters, and trace the history of abnormal events by clicking and dragging the mouse; combine chemical production process knowledge and equipment operation principles to conduct in-depth analysis of historical data and real-time data processed by the cloud computing platform; for example, analyze product quality data under different process parameter combinations to find out the key factors affecting product quality; study the relationship between equipment operating parameters and energy consumption, and tap the potential for energy saving; establish a production optimization mathematical model based on data analysis results; for example, establish a product quality prediction model to predict product quality by inputting process parameters; build an equipment maintenance optimization model to predict equipment failure time according to the equipment operation status and formulate a suitable For process parameter optimization, when a parameter is found to be outside the optimal range, the system calculates adjustment suggestions based on the optimization model, such as "The current reactor temperature is too high, it is recommended to lower the temperature to XX°C to improve the product qualification rate." For equipment operation optimization, if an abnormal equipment vibration frequency is detected, combined with the equipment maintenance optimization model, a decision suggestion is given: "The equipment vibration frequency exceeds the normal range, it is recommended to shut down and inspect the bearings within the next X hours." Considering the overall optimization of the production process, the connection efficiency between various production links is analyzed, and suggestions such as adjusting the production plan and optimizing the material transportation path are proposed. A dedicated area is set up in the visual interface to display production optimization decision suggestions, clearly presenting the content, basis and expected results of the suggestions in the form of text and icons.For example, suggestions are marked with a yellow background box and accompanied by relevant data charts as a basis. A user feedback interface is provided, allowing operators and managers to evaluate, adopt, or modify decision suggestions. The system records user feedback information and uses it to further optimize the decision suggestion generation model to improve the accuracy and practicality of the suggestions. The implementation of adopted decision suggestions is tracked, and changes in production data after the suggestions are implemented are monitored in real time to evaluate the optimization results. If the results do not meet expectations, the data is re-analyzed and the decision suggestions are adjusted, forming a closed-loop optimization process.

[0034] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A chemical production monitoring system based on cloud computing, characterized in that: The system includes: The field data acquisition module is used to collect chemical production data in real time through sensors at the chemical production site, and filter and denoise the collected data to obtain pre-processed data; The network transmission module is used to transmit the pre-processed data to the cloud computing platform module by combining industrial Ethernet and wireless network, encrypt the transmitted data, and use data compression technology to reduce the data transmission volume; The cloud computing platform module is used to receive data transmitted by the network transmission module, and uses distributed storage technology to disperse and store the data on multiple storage nodes. It classifies and processes the data according to the data type and application requirements, and uses streaming computing technology to perform real-time analysis to detect data anomalies. When the data exceeds the preset safety threshold, an alarm mechanism is immediately triggered; The monitoring application module is used to display various parameters of the chemical production process and the operating status of equipment in real time on a visual interface based on the processing results of the cloud computing platform module, and to provide decision-making suggestions for production optimization in combination with chemical production processes and equipment.

2. A chemical production monitoring system based on cloud computing according to claim 1, characterized in that: The field data acquisition module includes: The conversion submodule is used for sensors at chemical production sites to collect production data in real time according to the frequency, convert the collected original analog signals into digital signals and output them to complete data collection; The filtering submodule is used to perform preliminary filtering on the collected original digital signal using the mean filtering method to remove the impulse noise and salt and pepper noise in the data to obtain the filtered data; The reconstruction submodule is used to input the filtered data into the denoising diffusion probability model, input the clean signal data reconstructed by the denoising diffusion probability model into the cyclic variational autoencoder, reconstruct the processed output data, and obtain the preprocessed data.

3. A chemical production monitoring system based on cloud computing according to claim 2, characterized in that: The reconstruction submodule includes: The filtered data is divided into time periods of fixed length according to time. The data in each time period constitutes a sequence. Noise is gradually added to the data until the data reaches the set maximum noise level. Using the UNet model, starting from the final noisy state, the noise is gradually predicted and removed until the final reconstructed clean data is obtained. The clean data is arranged in chronological order to obtain time series data, and then input into the recurrent variational autoencoder. The encoder generates a hidden state sequence to capture the temporal dependency, calculates the parameters of the variational posterior distribution through the hidden state, samples the latent variables from the posterior distribution, inputs the latent variables into the decoder, initializes the decoder hidden state, generates a reconstructed sequence time step by time, and obtains the preprocessed data.

4. The chemical production monitoring system based on cloud computing according to claim 1, characterized in that: The network transmission module includes: The capture submodule is used to divide the preprocessed data into blocks according to time windows. Each block extracts temporal features through the Transformer encoder and uses the self-attention mechanism to capture long-range dependencies in the data. The compression submodule is used to input the feature vector output by the Transformer into the INN and compress the data dimension through reversible transformation; The perturbation submodule is used to generate adversarial samples using quantum neural networks, perturb the compressed data, map the perturbed data into quantum state space, and enhance encryption through quantum superposition and entanglement; The encryption submodule is used by the sender and receiver to generate session keys through the quantum key distribution protocol, and to encrypt the compressed data using a symmetric encryption algorithm. The key is derived from the dynamically generated session key and the key matrix embedded in the INN, and the session key is encrypted using an asymmetric encryption algorithm.

5. A chemical production monitoring system based on cloud computing according to claim 4, characterized in that: The compression submodule includes: The feature vector output by the Transformer encoder is used to generate a three-dimensional key matrix based on the current timestamp, device ID, and random number. The three-dimensional key matrix is ​​mapped to the same dimensional space as the feature vector through a hash function. The INN network is divided into multiple coupling layers. Each coupling layer includes a coupling layer, a transformation function, and a reversible transformation. After each coupling layer, the feature dimensions are rearranged through a channel shuffle operation, the feature selection mechanism is applied to discard redundant dimensions, and the attention mechanism is used to assign weights to different features. The feature vector after multi-layer coupling transformation is reshaped to obtain the compressed latent space representation, and the compression ratio is calculated. The check code is added to generate the final output vector.

6. A chemical production monitoring system based on cloud computing according to claim 4, characterized in that: The disturbance submodule includes: Input the compressed data into the quantum neural network and generate a perturbation vector that maximizes the loss function by adjusting the parameters until the perturbation reaches a preset intensity. Normalize the perturbed data vector, use the normalized data elements as the amplitude of the quantum state, map them to the quantum register, apply the Hadamard gate operation, convert the quantum bit from the quantum bit initialization to the superposition state, and encode the data characteristics into the phase and amplitude of the quantum state through the parameterized rotation gate; Non-local correlations are created between quantum bits to generate highly entangled states, and quantum entanglement-enhanced encryption is performed through controlled rotation gates.

7. The chemical production monitoring system based on cloud computing according to claim 1, characterized in that: The cloud computing platform module includes: The collection submodule is used to collect hardware indicators, network bandwidth and historical load data of storage nodes, and generate multi-dimensional features for each node, where the multi-dimensional features include at least processing power, storage capacity and reliability score; The allocation submodule abstracts storage nodes as vertices in a graph, with physical connections between nodes as edges. It applies a graph convolutional network to learn node embeddings, calculates similarity based on node embeddings, groups similar nodes into data shards, and allocates shards based on data type and access pattern. It monitors node load in real time and, when hotspots emerge, uses a graph convolutional network to predict load trends and dynamically adjust shard allocation. The classification submodule is used to automatically identify data types through data metadata and classify structured data, semi-structured data, and unstructured data; Set up submodules to deploy data stream processing tasks in Apache Flink, set parallelism and watermarks to process out-of-order data, and extract statistical features from time series data using sliding windows. The construction submodule is used to construct a causal graph. The TC-CNN model is used to capture the temporal dependencies and causal relationships between variables. The residual between the actual observation value and the TC-CNN model prediction value is calculated. When the residual exceeds the preset threshold and violates the causal graph relationship, it is marked as an anomaly.

8. A chemical production monitoring system based on cloud computing according to claim 7, characterized in that: The process of constructing the causal diagram includes: Filter out key variables from the data and organize them into structured multivariate time series data, where the key variables include at least temperature, pressure, flow rate and liquid level; A constraint-based causal discovery algorithm is used to perform causal relationship analysis on multivariate time series data. Taking the data within the time window as the unit, the conditional independence between variables is calculated, and irrelevant edges are gradually eliminated to construct a preliminary causal graph structure. The preliminary causal graph structure is then verified and corrected to obtain the final causal graph.

9. A chemical production monitoring system based on cloud computing according to claim 8, characterized in that: The building block includes: Construct a TC-CNN network and train it using multivariate time series data to obtain a TC-CNN model. The TC-CNN network uses causal convolutional layers at the bottom layer, adds an attention mechanism to the middle layer, and outputs the predicted value of each variable through a fully connected layer at the top layer. The trained TC-CNN model is deployed in a real-time computing environment and integrated with the data stream processing task. For the real-time incoming data, the model predicts the value of each variable at a fixed time step and outputs the predicted value at the future time point. For each time step, the difference between the actual observation value and the TC-CNN prediction value is calculated to obtain the residual sequence. Based on the constructed causal graph, the relationship between the variable with residual anomaly and its causal association variable is checked to see whether it meets the expectations. When the residual exceeds the preset threshold and violates the causal graph relationship, it is marked as an anomaly.

10. The chemical production monitoring system based on cloud computing according to claim 1, characterized in that: The processing results output by the cloud computing platform module include at least real-time process parameters in the chemical production process, equipment operation status data, anomaly detection results, and data analysis and prediction results.

Images