Liquid caustic soda concentration monitoring and automatic adjusting system based on Internet of Things and data fusion

Abstract

The invention relates to the technical field of liquid caustic soda concentration regulation and control, and discloses a liquid caustic soda concentration monitoring and automatic regulating system based on Internet of Things and data fusion. A liquid caustic soda concentration data acquisition module of the system is connected with a plurality of sensors through the Internet of Things to acquire spatio-temporal information such as liquid caustic soda concentration, temperature, flow velocity and the like; the edge node operation and maintenance analysis module is used for monitoring operation and maintenance information such as memory, bandwidth and CPU utilization rate of the edge computing node; the data fusion preprocessing module performs standardization processing on the sensor data to generate a data matrix; the liquid caustic soda concentration anomaly detection module is used for extracting fusion features based on the data matrix and identifying abnormal nodes; the concentration adjustment optimization module is used for generating an optimization adjustment parameter set according to the abnormal nodes; the automatic adjusting control module is used for adjusting the valve opening and the pump speed according to the parameter set; and the distributed storage management module is used for storing sensor data and operation and maintenance information. The system realizes comprehensive monitoring and accurate adjustment of the concentration of the liquid caustic soda, and improves the operation stability and reliability of the system.

Description

Technical Field

[0001] This invention relates to the field of liquid alkali concentration control technology, specifically to a liquid alkali concentration monitoring and automatic adjustment system based on the Internet of Things and data fusion. Background Technology

[0002] In many industrial fields such as chemical engineering, metallurgy, and environmental protection, liquid caustic soda is an important chemical reagent, and the stability of its concentration directly affects the continuity of production processes, the reliability of product quality, and the safety of production. Traditional methods for monitoring the concentration of liquid caustic soda mostly rely on manual periodic sampling and testing. This method not only suffers from long testing intervals and poor real-time performance, making it difficult to capture dynamic changes in the concentration of liquid caustic soda in a timely manner, but it is also prone to inaccurate test results due to human operational errors, thereby affecting the precise control of the production process. With the improvement of industrial automation, some enterprises have begun to introduce single sensors for monitoring the concentration of liquid caustic soda. However, this type of monitoring method can often only obtain liquid caustic soda concentration data for a local area, and cannot achieve comprehensive coverage of the entire liquid caustic soda storage or transportation system. At the same time, the data collected by a single sensor lacks linkage analysis with other key environmental parameters such as temperature and flow rate, making it difficult to accurately determine the true cause of changes in liquid caustic soda concentration, resulting in a lack of targeted concentration adjustment measures. In terms of data processing, traditional systems mostly transmit the collected data directly to a remote cloud for processing. However, liquid alkali concentration data is typically characterized by high real-time requirements and large data volumes. Remote cloud processing is not only easily affected by network bandwidth limitations, leading to data transmission delays and affecting the timeliness of concentration adjustments, but may also result in data loss due to network fluctuations, reducing the stability of system operation. In addition, traditional systems have significant shortcomings in equipment operation and maintenance management. They cannot effectively monitor the status of edge computing nodes used for data acquisition and processing. When edge nodes experience problems such as excessive memory utilization or excessive CPU load, they cannot detect and address them in a timely manner, thereby affecting the efficiency of data acquisition and processing, and even causing the entire monitoring system to crash. Traditional methods of regulating liquid caustic soda concentration often employ fixed-parameter adjustment, adjusting valve opening and pump speed according to preset concentration thresholds. This approach ignores the dynamic and complex nature of changes in liquid caustic soda concentration, making it difficult to adapt to the concentration regulation needs under different production conditions. It frequently results in over- or under-regulation, wasting resources and potentially adversely affecting the production process. Therefore, the current industrial sector urgently needs an intelligent system for monitoring and regulating liquid caustic soda concentration that can achieve comprehensive sensing, accurate analysis, timely response, and stable reliability. Summary of the Invention

[0003] The purpose of this invention is to provide a liquid alkali concentration monitoring and automatic adjustment system based on the Internet of Things and data fusion, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a liquid alkali concentration monitoring and automatic adjustment system based on the Internet of Things and data fusion, the system comprising: The liquid alkali concentration data acquisition module is used to acquire data from multiple liquid alkali sensors connected to the Internet of Things and to statistically analyze the spatiotemporal information of each liquid alkali sensor, including liquid alkali concentration value, temperature value, and flow rate value. The edge node operation and maintenance analysis module is used to obtain operation and maintenance information of multiple edge computing nodes, including memory utilization, bandwidth utilization and CPU utilization. The data fusion preprocessing module standardizes the liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module to generate a standardized liquid alkali data matrix. The liquid alkali concentration anomaly detection module performs feature extraction and fusion based on the standardized liquid alkali data matrix output by the data fusion preprocessing module, and identifies nodes with abnormal liquid alkali concentration. The concentration adjustment optimization module generates an optimized adjustment parameter set based on the abnormal liquid alkali concentration nodes identified by the abnormal liquid alkali concentration detection module. An automatic adjustment and control module adjusts the liquid alkali concentration control parameters based on the optimized adjustment parameter set generated by the concentration adjustment and optimization module. The liquid alkali concentration control parameters include valve opening and pump speed. The distributed storage management module is used to store the liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module and the operation and maintenance information acquired by the edge node operation and maintenance analysis module.

[0005] Preferably, the spatiotemporal information of each liquid alkali sensor acquired by the liquid alkali concentration data acquisition module also includes the data acquisition frequency and data volume; the operation and maintenance information of each edge computing node acquired by the edge node operation and maintenance analysis module also includes the data packet processing rate.

[0006] Preferably, the standardization processing of the data fusion preprocessing module includes: The liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module is formatted to form a time-space matched dataset; For missing liquid alkali sensor data, interpolation was used to fill in the missing values; Normalize the sensor data after filling with liquid alkali. Extract the temporal and spatial features of the normalized liquid alkali sensor data; The normalized data and extracted features are combined to generate the standardized liquid alkali data matrix.

[0007] Preferably, the feature extraction and fusion of the abnormal liquid alkali concentration detection module includes: Based on the standardized liquid alkali data matrix output by the data fusion preprocessing module, a time series feature extraction unit is constructed to extract the time series features of the standardized liquid alkali data; Construct a spatial feature extraction unit to extract spatial features from standardized liquid alkali data; The extracted temporal and spatial features are weighted and fused to generate a fused feature dataset; Construct an anomaly feature correlation matrix based on the fused feature dataset; The statistical threshold detection method was used to analyze the correlation matrix of abnormal features and identify nodes with abnormal alkali concentration.

[0008] Preferably, the concentration adjustment optimization module generates an optimized adjustment parameter set including: Define an optimization objective function to quantify the degree of deviation at abnormal nodes in liquid alkali concentration; The parameter set is initialized using the simulated annealing optimization algorithm, and then iteratively optimized. During the simulated annealing evolution process, the system temperature is gradually reduced to search for the optimal parameter combination; When the temperature drops to a set threshold, the optimized parameter set is measured and the optimized adjustment parameter set is generated.

[0009] Preferably, the automatic adjustment control module adjusts the liquid alkali concentration control parameters including: Based on the optimized adjustment parameter set generated by the concentration adjustment optimization module, input the liquid alkali concentration prediction model to estimate the liquid alkali concentration adjustment amount. Adjust the valve opening according to the concentration of the liquid alkali; Adjust the pump speed according to the concentration of the alkali solution.

[0010] Preferably, the liquid alkali concentration prediction model is trained and generated based on historical liquid alkali sensor data; The training process includes: collecting historical liquid alkali sensor data and corresponding adjustment results, extracting historical feature parameters, and constructing a decision tree prediction model; The decision tree prediction model is used to estimate the adjustment amount of liquid alkali concentration.

[0011] Preferably, the distributed storage management module stores data including: Synchronously retrieve operational data from each storage port and analyze storage performance benchmarks; Based on the liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module and the operation and maintenance information acquired by the edge node operation and maintenance analysis module, the cluster data evaluation value is calculated. Match storage performance benchmarks with cluster data evaluation values, and allocate target storage ports accordingly; Distribute and store liquid alkali sensor data and operation and maintenance information to the target storage port.

[0012] Preferably, the system further includes: a load balancing configuration module, which performs load evaluation and balancing configuration based on the data stored in the distributed storage management module, including: Based on the data stored in the distributed storage management module, obtain the load parameters of each data storage node; Process the load parameters to generate a load evaluation value; Re - allocate the data storage tasks according to the load evaluation value.

[0013] Preferably, the edge node operation and maintenance analysis module generates an edge node energy efficiency characterization value based on the processing of operation and maintenance information; The liquid alkali concentration data acquisition module generates a liquid alkali sensor characteristic value based on the processing of spatio - temporal information; Match the edge node energy efficiency characterization value with the liquid alkali sensor characteristic value, and allocate the target receiving liquid alkali sensor cluster; The data fusion pre - processing module performs standardization processing on the data of the target receiving liquid alkali sensor cluster.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This liquid alkali concentration monitoring and automatic regulation system based on the Internet of Things and data fusion connects multiple liquid alkali sensors through the liquid alkali concentration data acquisition module with the help of Internet of Things technology, which can achieve comprehensive coverage of different areas of the liquid alkali storage or transportation system, and simultaneously obtain multi - dimensional spatio - temporal information such as liquid alkali concentration values, temperature values, and flow velocity values, breaking the limitation of traditional single - sensor local monitoring, enabling staff to comprehensively master the state changes of liquid alkali, and avoiding judgment errors caused by one - sided information. The data of multiple sensors confirm and complement each other, and combined with the联动 analysis of parameters such as temperature and flow velocity, can more accurately identify the internal reasons for the change of liquid alkali concentration, providing a comprehensive and reliable basis for subsequent concentration regulation.

[0015] The edge node operation and maintenance analysis module monitors in real - time the operation and maintenance information such as memory utilization rate, bandwidth utilization rate, and CPU utilization rate of multiple edge computing nodes, enabling staff to timely master the operation status of edge nodes. When problems such as abnormal resource occupation occur in edge nodes, targeted maintenance measures can be taken in advance to avoid data acquisition interruption or processing delay caused by edge node failures, ensuring the stable operation of the data acquisition and processing links of the entire system, and laying a foundation for the continuous and reliable operation of the system. It should be noted that there is a typo in the original text. "联动分析" should be "联动分析(joint analysis)", and it has been translated as "联动分析" in the above translation for the sake of consistency with the original text. If this is a specific technical term that needs a more accurate translation, it should be adjusted according to the actual situation.The data fusion preprocessing module standardizes the collected liquid alkali sensor data, generating a standardized liquid alkali data matrix. This effectively eliminates problems such as inconsistent data formats and units caused by differences in sensor models and accuracy, enabling subsequent data feature extraction and fusion analysis to proceed smoothly. The standardized dataset exhibits higher consistency and comparability, contributing to improved accuracy of data processing results and providing high-quality data support for the detection of abnormal liquid alkali concentrations. The liquid alkali concentration anomaly detection module extracts and fuses features based on a standardized liquid alkali data matrix. Through in-depth analysis of multi-dimensional data, it can accurately identify abnormal nodes in liquid alkali concentration. Compared with the traditional anomaly judgment method that only relies on a single concentration threshold, this module can more sensitively capture subtle abnormal changes in liquid alkali concentration and can clearly identify the specific location of the anomaly. This makes it easier for staff to quickly locate the problem area, reduce investigation time, and reduce losses caused by the spread of anomalies. The concentration regulation optimization module generates an optimized set of regulation parameters based on anomaly detection results. Instead of using the traditional fixed-parameter regulation mode, it develops a personalized regulation plan by considering the specific circumstances of the liquid caustic soda concentration anomaly, environmental parameters such as temperature and flow rate, and the actual needs of the production conditions. This optimized set of regulation parameters ensures that subsequent concentration regulation measures are more targeted and reasonable, avoiding the problems of over- or under-regulation in traditional methods, and achieving precise control of the liquid caustic soda concentration. The automatic adjustment control module automatically adjusts the control parameters for liquid caustic soda concentration, such as valve opening and pump speed, based on an optimized set of adjustment parameters. This completes the concentration adjustment process without manual intervention, significantly reducing human error and improving the efficiency and accuracy of concentration adjustment. Simultaneously, automatic adjustment can quickly respond to abnormal changes in liquid caustic soda concentration, shortening the adjustment cycle and ensuring the stability of liquid caustic soda concentration during production, thereby maintaining the continuity of the production process and the reliability of product quality. The distributed storage management module provides unified storage for liquid alkali sensor data and edge node operation and maintenance information. This not only enables centralized data management and efficient retrieval but also avoids the risk of data loss due to the failure of a single storage device, thus improving the security and reliability of data storage. The stored historical data can also be used for subsequent production process optimization analysis and system operation status assessment, providing strong support for enterprise production management decisions. Attached Figure Description

[0016] Figure 1 This is a timing diagram of the liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion described in this invention; Figure 2 A flowchart for a standardized data fusion preprocessing workflow; Figure 3This is a flowchart of the feature processing and identification for abnormal concentration detection of liquid alkali. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 This invention provides a liquid alkali concentration monitoring and automatic adjustment system based on the Internet of Things (IoT) and data fusion. The system includes a liquid alkali concentration data acquisition module, an edge node operation and maintenance analysis module, a data fusion preprocessing module, a liquid alkali concentration anomaly detection module, a concentration adjustment optimization module, an automatic adjustment control module, and a distributed storage management module. The liquid alkali concentration data acquisition module connects to multiple liquid alkali sensors via the IoT to acquire spatiotemporal information such as liquid alkali concentration, temperature, and flow rate, and statistically analyzes this data to form a raw dataset. The edge node operation and maintenance analysis module collects operation and maintenance information from multiple edge computing nodes, including memory utilization, bandwidth utilization, and CPU utilization, for system status monitoring. The data fusion preprocessing module standardizes the liquid alkali sensor data to generate a standardized liquid alkali data matrix, providing a unified format for subsequent analysis. The liquid alkali concentration anomaly detection module extracts and fuses features based on the standardized liquid alkali data matrix to identify abnormal liquid alkali concentration nodes. The concentration adjustment optimization module generates an optimized adjustment parameter set based on the abnormal nodes, and the automatic adjustment control module uses these parameters to adjust control parameters such as valve opening and pump speed. The distributed storage management module is responsible for storing liquid caustic soda sensor data and operation and maintenance information, ensuring data persistence and accessibility. Through modular design, this system enables real-time monitoring and automatic adjustment of liquid caustic soda concentration, improving the stability and efficiency of industrial processes.

[0019] Example 1: See Figure 2 During the operation of the liquid alkali concentration data acquisition module, each liquid alkali sensor continuously transmits multi-dimensional spatiotemporal information via the Internet of Things (IoT) protocol. This information includes not only basic parameters such as liquid alkali concentration, temperature, and flow rate, but also in-depth indicators such as data acquisition frequency and data volume. The data acquisition frequency refers to the number of times the sensor acquires data per unit time, which is periodically triggered by the timer function of the embedded system. The data volume refers to the number of bytes contained in each transmission, which is statistically summarized by the counter built into the sensor. This additional information, together with the basic data, is packaged into a transmission frame and sent to the central processing unit via the wireless communication protocol, providing a more comprehensive data dimension for subsequent analysis.

[0020] When collecting operational information from each computing node, the edge node operation and maintenance analysis module specifically acquires the key indicator of packet processing rate, in addition to memory utilization, bandwidth utilization, and CPU utilization. The packet processing rate is measured in real-time by monitoring the transmit and receive queue depth and packet count register of the network interface card. This rate reflects the actual ability of the edge node to process network packets per unit time, and together with other operational indicators, constitutes a node performance profile, providing a basis for system resource allocation. When the data fusion preprocessing module initiates the standardized processing flow, it first formats the raw liquid alkali sensor data. The formatting process aligns the data based on a unified timestamp sequence and sensor spatial coordinates, eliminating temporal offsets and positional differences between different sensors, forming a dataset structure with strict temporal and spatial matching. This dataset is stored in a cache in the form of a two-dimensional table for subsequent processing. When missing values ​​are detected in the dataset, the system automatically triggers the interpolation imputation procedure. The interpolation algorithm selects either linear interpolation or spline interpolation based on the distribution characteristics of the missing data. Linear interpolation is suitable for uniformly distributed missing points, and the imputation value is calculated by the arithmetic mean of adjacent data points. Spline interpolation is for non-uniform missing cases, and a polynomial fitting curve is used to generate a smooth imputation value. All imputation operations are logged for auditing purposes.

[0021] After handling missing values, the data enters the normalization stage. Normalization employs a min-max scaling algorithm to map the raw sensor data values ​​to a standardized range of 0 to 1. This algorithm first scans the dataset to find the minimum and maximum values ​​for each feature dimension, then calculates the normalized value for each data point using a linear transformation formula, eliminating the bias caused by different units of measurement. Subsequently, the system extracts temporal and spatial features from the normalized data. Temporal feature extraction uses a sliding window technique to calculate the moving average and first-order difference values, capturing the trend of data changes over time. Spatial feature extraction reveals the spatial correlation characteristics of the data by calculating the Euclidean distance and correlation coefficient between sensor groups. Finally, the normalized data and the extracted feature vectors are combined and concatenated to generate a standardized liquid alkali data matrix. The rows of this matrix correspond to the time series index, and the columns contain the original data values ​​and derived feature values. After generation, the matrix is ​​output to a shared memory area for use by the anomaly detection module.

[0022] Example 2: See Figure 3The system calculates residual sequences and trend components to capture the dynamic patterns of data changes over time. The time-series feature extraction unit employs a sliding window mechanism, with the window size dynamically adjusted based on the data sampling frequency to ensure effective extraction of short-term fluctuations and long-term trend features. Simultaneously, the spatial feature extraction unit performs spatial feature mining on the standardized liquid alkali data. This unit uses a density-based clustering algorithm to group sensor nodes and quantifies spatial correlation by calculating the variance and covariance matrices of data within each group. The spatial feature extraction process considers the physical layout and topology of the sensor nodes, introducing an Euclidean distance weighting factor to enhance the expressive power of spatial features. After feature extraction, the system performs a weighted fusion operation on the time-series and spatial features. The weight allocation is based on the feature contribution index obtained from principal component analysis. During the fusion process, feature splicing and dimension alignment techniques are used to generate a fused feature dataset containing multi-dimensional indicators. This dataset not only retains the statistical characteristics of the original data but also enhances the complementarity between features. When constructing anomaly feature correlation matrices based on the fused feature dataset, an improved Pearson correlation coefficient calculation method is used. A significance test mechanism is introduced to filter out spurious correlations, resulting in a symmetrical matrix structure that fully characterizes the distribution pattern of correlation strength between features. When analyzing the correlation matrix of abnormal features using the statistical threshold detection method, the system dynamically calculates the Z-score distribution of feature values ​​and sets an adaptive threshold by combining the interquartile range rule. It identifies abnormal patterns that exceed the threshold by traversing all feature correlation pairs in the matrix. The detection process adopts a multi-round iterative verification mechanism to gradually narrow down the scope of abnormal nodes.

[0023] In the process of generating the optimized adjustment parameter set by the concentration regulation optimization module, an optimization objective function based on the degree of liquid alkali concentration deviation is first defined. This function comprehensively considers multiple factors such as absolute error, cumulative error, and adjustment cost, forming a mathematical expression of a multi-objective optimization problem. When initializing the parameter set using the simulated annealing optimization algorithm, the system randomly generates an initial solution set including adjustment parameters such as valve opening and pump speed, and sets control variables such as initial temperature parameters and cooling rate. During the iterative optimization process, the algorithm accepts inferior solutions using the Metropolis criterion to avoid getting trapped in local optima. During the simulated annealing evolution, the system gradually reduces the temperature parameter according to the exponential decay law, while performing multiple neighborhood search operations at each temperature level. New solutions are generated by perturbing the current solution, and the changes in the objective function value are evaluated. The search strategy adopts an adaptive step size adjustment mechanism, dynamically adjusting the parameter change amplitude according to the search progress. When the system temperature drops to the preset threshold, the algorithm terminates the search process and outputs the current optimal parameter combination. This parameter set, after normalization and range verification, forms the final optimized adjustment parameter set. The complete optimization path and convergence curve are recorded during the parameter set generation process, providing a decision-making basis for subsequent adjustment operations. The entire feature extraction and optimization process is processed in parallel through a distributed computing framework. Data is synchronized between computing nodes using a message passing mechanism to ensure processing efficiency and result consistency.

[0024] Taking the monitoring scenario of a liquid caustic soda production line in a chemical plant as an example, the liquid caustic soda concentration anomaly detection module begins processing real-time monitoring data from the storage tank area. Twelve liquid caustic soda sensors deployed in this area transmit concentration, temperature, and flow rate readings twice per second. The data fusion preprocessing module has converted this raw data into a standardized liquid caustic soda data matrix, with the matrix dimensions being time series × sensor nodes × monitoring parameters. The time series feature extraction unit initiates sliding window analysis, with a window width set to 60 consecutive sampling points, calculating the moving average and first-order difference sequence for each sensor node. The moving average smooths short-term fluctuations and displays trend changes, while the first-order difference sequence captures instantaneous change rates, paying particular attention to sensor nodes whose difference values ​​consistently exceed the normal range. Simultaneously, the spatial feature extraction unit operates, calculating spatial correlations based on the three-dimensional distribution of sensors within the storage tank. The 12 nodes are divided into three spatial groups (bottom, body, and top), calculating the variance coefficient of sensor readings within each group and the correlation coefficient between groups. The bottom group shows higher concentration variance, while the top group exhibits a strong correlation in temperature readings. These spatial features are quantified into feature vectors.

[0025] The feature weighting fusion process assigns a weight of 0.6 to temporal features and a weight of 0.4 to spatial features. The weight ratio is determined based on the contribution of the two types of features to anomaly detection in historical data. The fused feature dataset includes 12 derived indicators such as concentration change rate and spatial consistency index. The anomaly feature correlation matrix is ​​constructed using an improved correlation coefficient calculation method, focusing on analyzing the correlation strength between concentration readings and temperature readings. Matrix analysis revealed that the concentration-temperature correlation coefficient of sensor No. 5 (middle of the tank) was significantly lower than that of other sensors, forming an anomaly signal. Statistical threshold detection sets dynamic threshold boundaries, calculating the mean and standard deviation of each feature indicator based on normal data from the past 24 hours, and comparing the current feature value with the threshold range. The concentration change rate feature value of sensor No. 5 exceeds the threshold range by 2.3 standard deviations, and its spatial consistency index is lower than the lower limit of the normal value. The system marks it as an abnormal node in liquid alkali concentration.

[0026] After receiving information about an abnormal node, the concentration regulation optimization module defines an optimization objective function to minimize the concentration deviation in the region of sensor 5, while constraining the adjustment range to avoid drastic changes. The function considers the absolute deviation between the current concentration value and the target value, the duration of the deviation, and the consistency requirements of adjacent sensor readings. The simulated annealing algorithm initializes a parameter set containing eight adjustable parameters, including the feed valve opening, the circulating pump speed, and the temperature compensation value. Initial parameter combinations are randomly generated, and the objective function value is calculated. During the algorithm iteration, an adaptive neighborhood search strategy is employed. In each iteration, a parameter value is randomly perturbed, and the quality of the new solution is evaluated. At high temperatures, more inferior solutions are accepted to expand the search range; as the temperature coefficient gradually decreases, the search gradually focuses on the region of high-quality solutions. When the system temperature drops to a set threshold, the algorithm outputs the optimal parameter combination: reducing the feed valve opening by 15%, increasing the circulating pump speed by 8%, and simultaneously activating the temperature compensation device to raise the temperature of the region by 2°C. After passing a safety check, the optimized adjustment parameter set is transmitted to the automatic regulation control module for corresponding operations. The entire process, from anomaly detection to parameter optimization, took 4.2 seconds. During this time, the system continuously monitored the data changes of sensor 5. After the adjustment was implemented, the concentration reading of the sensor gradually returned to the normal range, and the concentration-temperature correlation coefficient recovered to the population average level. The system recorded a complete processing log for this anomaly event, including the original data matrix, extracted feature values, fusion weight settings, anomaly judgment criteria, the evolution trajectory of optimized parameters, and the adjustment effect.

[0027] Example 3: When the automatic liquid alkali concentration adjustment control module performs control parameter adjustments, it calculates the liquid alkali concentration adjustment amount based on the optimized adjustment parameter set transmitted by the concentration adjustment optimization module. This parameter set includes multi-dimensional input variables such as valve opening adjustment coefficient, pump speed correction factor, and environmental compensation parameters. The system inputs these parameters into the liquid alkali concentration prediction model for inference calculation. This model generates an estimated value of the concentration adjustment amount through multi-level nonlinear transformation. Its calculation process can be expressed as follows:

[0028] in: This indicates the amount of liquid alkali concentration to be adjusted (unit: mol / L). Represents the weight coefficients of the i-th hidden layer. It is the Sigmoid activation function. This represents the scaling factor for the i-th input parameter. The corresponding value of the i-th component in the set of optimization adjustment parameters. It is the bias term of the i-th neuron. This is the temperature regulation coefficient. This indicates the number of hidden layer nodes. Based on the calculated liquid alkali concentration adjustment, the valve opening adjustment subsystem initiates a PID control algorithm to convert the concentration adjustment into a percentage change in valve opening. This conversion process employs a piecewise linear mapping strategy, with different conversion coefficients corresponding to different concentration adjustment ranges, ensuring the accuracy and stability of valve control. When adjusting the valve opening, the control system sends control commands to the actuator via the Modbus protocol. These commands include parameters such as the target opening value, rate of change, and safety check code. Upon receiving the commands, the actuator drives the motor to rotate to the designated position, while simultaneously providing real-time feedback of the actual opening value to form a closed-loop control. The pump speed adjustment subsystem synchronously responds to the concentration adjustment, using frequency conversion control technology to convert the concentration adjustment into a motor frequency signal. The conversion process considers the pump's characteristic curve and pipeline resistance factors, establishing a three-dimensional mapping relationship between frequency, flow rate, and concentration, and outputting the optimal frequency setpoint. After receiving the frequency control signal, the frequency converter adjusts the output voltage and frequency to change the pump motor speed, thereby precisely controlling the liquid alkali delivery flow rate. During speed adjustment, current and voltage parameters are monitored to achieve overload protection and energy efficiency optimization.

[0029] The training of the liquid caustic soda concentration prediction model is based on historical operational data. The training data collection covers liquid caustic soda sensor readings and corresponding adjustment results under multiple operating conditions. The data collection period covers the entire production batch, including operational data under normal and abnormal conditions, ensuring the comprehensiveness and representativeness of the training data. The historical feature parameter extraction process employs a sliding time window technique to extract multi-dimensional features from the raw data, including statistical features, time-series features, and spectral features. Statistical features include mean, variance, and extreme values; time-series features include first-order differences and autocorrelation coefficients; and spectral features are obtained by using Fast Fourier Transform to acquire the main frequency components. The decision tree prediction model is constructed using the CART algorithm framework, establishing a mapping relationship between features and adjustment values ​​through recursive partitioning. Each split node selects the feature with the smallest Gini coefficient for data partitioning until the number of samples contained in the leaf node reaches a preset threshold or the purity meets the requirements. During model training, ten-fold cross-validation is used to adjust hyperparameters, including parameters such as the maximum tree depth, minimum number of split samples, and pruning intensity, and the optimal parameter combination is determined through grid search. The trained decision tree prediction model has the ability to handle nonlinear relationships and can output an estimated value of the liquid alkali concentration adjustment based on the input parameter set. The model adopts an integrated prediction strategy during inference, and improves prediction accuracy and robustness through the voting mechanism of multiple decision trees.

[0030] Taking a liquid caustic soda production line in a chemical plant as an example, the automatic adjustment and control module receives an optimized adjustment parameter set from the concentration adjustment optimization module. This parameter set includes key values ​​such as a valve opening adjustment coefficient of 0.85, a pump speed correction factor of 1.2, and an ambient temperature compensation parameter of -0.3. The system immediately inputs these parameters into the liquid caustic soda concentration prediction model for real-time calculation. This model adopts a three-layer neural network architecture, with the hidden layer using the ReLU activation function and the output layer using a linear activation function. After forward propagation, the concentration adjustment amount is calculated to be -0.15 mol / L. Based on the calculated negative adjustment amount, the valve opening adjustment subsystem starts a fuzzy PID control algorithm to convert the adjustment amount of -0.15 mol / L into a specific adjustment command for the valve opening. The system queries the valve characteristic curve database to find the correspondence between the concentration change and the valve opening under the current operating conditions, and generates a control command to reduce the opening of feed valve No. 3 from 65% to 52%. When adjusting the valve opening, the control system sends a digital control signal to the Siemens intelligent valve positioner via the PROFINET industrial Ethernet protocol. The signal includes the target opening value of 52%, a change rate of 5% per minute, and a safety check code 0xAE43. Upon receiving the command, the valve positioner drives a stepper motor to rotate, while simultaneously detecting the valve stem position in real time using a high-precision potentiometer, forming a closed-loop position control. The entire adjustment process is completed within 8 seconds. The pump speed adjustment subsystem synchronously responds to the concentration adjustment, employing a model predictive control-based approach to convert the -0.15 mol / L adjustment into a centrifugal pump speed adjustment command. The system queries the pump characteristic curve library, combining pipeline resistance parameters and fluid characteristic data to calculate the required reduction in pump speed from 2850 rpm to 2630 rpm. After receiving the 4-20mA analog control signal, the frequency converter adjusts the output frequency of the IGBT inverter circuit to change the motor speed. During speed adjustment, the motor current and winding temperature are monitored in real time, and the soft-start protection function is automatically activated when abnormal current fluctuations are detected.

[0031] The liquid alkali concentration prediction model was trained based on the chemical plant's historical operating data over the past six months, covering operating conditions under different seasons and production loads. The training dataset contained 120,000 data samples, each with 30 feature parameters, including concentration readings, temperature values, flow rate indices, valve openings, pump speeds, and environmental conditions. Historical feature parameter extraction employed a multi-scale sliding window technique to extract statistical, frequency domain, and temporal features from the original time-series data. Statistical features included the mean, variance, and skewness within a 10-minute window; frequency domain features extracted the main frequency components using wavelet transform; and temporal features included autocorrelation functions and cross-correlation indices. The decision tree prediction model was constructed using a gradient boosting decision tree algorithm, with a maximum tree depth of 7 and a learning rate of 0.1. Early stopping was used to prevent overfitting. During training, 5-fold cross-validation was used to adjust hyperparameters, with each iteration using 80% of the randomly sampled data for training and the remaining 20% ​​for validation. The trained decision tree model contains 150 subtrees, each with an average depth of 5 layers. Feature importance analysis shows that historical concentration values, temperature gradients, and valve opening change rates are the three most important predictive features. The model is deployed using an embedded deployment scheme, burning the trained model parameters into the flash memory of edge computing nodes. During inference, the model parameters are directly called for prediction calculations. An online model update mechanism triggers incremental training every two weeks, using the latest collected production data to fine-tune the model and maintain its adaptability to changes in operating conditions.

[0032] Example 4: When the distributed storage management module starts the data storage process, it first synchronously retrieves real-time operating data from all available storage ports in the system. This operating data includes performance indicators such as port read / write speed, access latency, and error rate. Performance data acquisition is achieved through a built-in monitoring agent program. The agent program periodically queries the status register of the storage controller to obtain the latest performance readings and temporarily caches them in the shared memory area. The process of analyzing the storage performance benchmark value adopts a dynamic threshold calculation method. The system uses the historical performance data of the most recent 24 hours as the baseline and calculates the moving average of each indicator for each port plus twice the standard deviation as the benchmark reference value. The benchmark value calculation is automatically updated every half hour to adapt to the periodic changes in system load. Based on the real-time sensor data obtained by the liquid alkali concentration data acquisition module and the operation and maintenance information collected by the edge node operation and maintenance analysis module, the system calculates the cluster data evaluation value. The evaluation value calculation adopts a multi-dimensional weighted algorithm, considering factors such as data volume, expected access frequency, data retention period, and security level. Among them, real-time sensor data is given a larger weight due to its high update frequency, while operation and maintenance information is given a smaller weight due to its slower changes.

[0033] When matching storage performance benchmarks with cluster data evaluation values, the system employs an improved optimal allocation algorithm. This algorithm first establishes a correspondence table between storage port performance characteristics and data requirements, and then uses the Hungarian algorithm to solve for the optimal allocation scheme. The matching process prioritizes storing high-frequency data on low-latency ports and allocating large-volume data to high-throughput ports, achieving the best match between storage resources and data characteristics. After the decision on allocating the target storage port is completed, the system generates a detailed data distribution mapping table. This table records information such as the target port address, storage format, and number of replicas for each data block. The mapping table is synchronized to all storage nodes through a distributed consensus protocol to ensure data consistency. When distributing liquid alkali sensor data and operation and maintenance information to the target storage port, the system uses a sharding storage mechanism. The original data is divided into fixed-size data blocks, each with an attached checksum and timestamp. The data writing process uses a two-phase commit protocol: first, data is pre-written to a temporary area and verified; only after confirmation is the data formally committed to the target port, ensuring data integrity and reliability.

[0034] The load balancing configuration module initiates a load assessment process based on the storage data status, periodically acquiring real-time load parameters for each data storage node. These load parameters include core metrics such as storage space utilization, input / output operation frequency, and network bandwidth utilization, which are collected in real-time through the performance counters of the storage nodes. When processing load parameters to generate load assessment values, the system employs a multi-factor weighted scoring model, assigning dynamic weights to different load parameters. Storage space utilization has a higher weight, input / output operation frequency has a medium weight, and network bandwidth utilization has a lower weight; the weight allocation is dynamically adjusted based on the storage node type and application scenario. When reallocating data storage tasks based on the load assessment values, the system first identifies storage nodes whose load assessment values ​​exceed a threshold, then initiates a data migration process. The migration process uses hot migration technology, transferring some data blocks from high-load nodes to low-load nodes while ensuring uninterrupted service. After the migration is complete, the data distribution mapping table is updated.

[0035] The following illustrates the process of matching storage port performance benchmarks with data evaluation values: Table 1: Matching of Storage Port Performance with Data Evaluation

[0036] The entire storage management process adopts an event-driven architecture design. Any change in storage node performance or data characteristics triggers a reassessment process. The system maintains a storage resource pool, monitoring the available capacity and performance status of each node in real time. When performance deviations are detected, data reallocation is automatically triggered. Storage operation logs record the storage path, access time, and migration history of each data block in detail. This log information is used to optimize storage strategies and troubleshoot problems. The log data itself is stored in a dedicated log area using compression technology and is periodically archived to long-term storage devices. For data security, encrypted storage and access control mechanisms are employed. All sensitive data is encrypted before storage, and a distributed key management system is used, assigning different access permissions to different users and data categories. The storage system performs integrity checks periodically. By verifying and comparing data, it automatically triggers a data recovery process to repair damaged data from replicas, ensuring data persistence and availability. For performance optimization, caching acceleration technology is used. High-frequency access data is retained in a high-speed cache area, while low-frequency access data is migrated to capacity-type storage devices, achieving a balance between storage performance and cost.

[0037] Example 5: When processing the generated energy efficiency characterization values, the edge node operation and maintenance analysis module quantifies the node's energy efficiency level by comprehensively calculating the weighted average of memory utilization, bandwidth utilization, and CPU utilization. Weight allocation is dynamically adjusted based on node type and application scenario; computationally intensive nodes are assigned a higher weight to CPU utilization, while data transmission intensive nodes are emphasized based on bandwidth utilization. The energy efficiency characterization value calculation employs a sliding window mechanism, using performance data from the most recent 30 minutes as a basis to calculate the periodic performance efficiency index and eliminate the impact of instantaneous fluctuations. The calculation results retain two decimal places to ensure data accuracy. The sensor feature values ​​generated by the liquid alkali concentration data acquisition module include multi-dimensional indicators such as concentration variance, temperature gradient, and flow rate change rate. Concentration variance reflects the dispersion of concentration at the monitoring point, temperature gradient reflects thermal conductivity, and flow rate change rate characterizes fluid stability. These feature values ​​are extracted from the raw data using a time series analysis algorithm, with the sampling window size synchronized with the energy efficiency characterization value calculation period.

[0038] When matching the energy efficiency values ​​of edge nodes with the characteristic values ​​of liquid alkali sensors, the system constructs a two-dimensional matching matrix. The row dimension of the matrix corresponds to the energy efficiency values ​​of the edge nodes, and the column dimension represents the sensor characteristic values. The matching algorithm adopts an improved Hungarian algorithm, and load balancing constraints are introduced when solving the optimal allocation scheme to avoid assigning too many high-load sensor clusters to a single node. The process of allocating the target receiving liquid alkali sensor cluster adopts a multi-objective optimization strategy to minimize the overall communication latency while ensuring a relatively balanced load on each edge node. The allocation result generates a sensor-node mapping table, recording the target edge node identifier corresponding to each sensor cluster. When the data fusion preprocessing module standardizes the data of the target receiving liquid alkali sensor cluster, it first verifies the consistency of the data format. Sensor data from different manufacturers are unified into a standard data format through a format conversion middleware. The data processing adopts a pipeline architecture. The raw data undergoes format conversion, missing value imputation, numerical normalization, and feature extraction, and finally generates a standardized data matrix.

[0039] The following example illustrates the matching process between three sensor clusters and edge nodes: Cluster A exhibits high-frequency data generation characteristics (20 samples per second) and is matched to the high-performance computing node N-05; Cluster B has large data volume characteristics (2MB of data transmitted per transmission) and is assigned to the high-bandwidth node N-08; Cluster C exhibits medium data characteristics and is matched to the comprehensive performance node N-12. The entire matching process undergoes dynamic adjustments every 5 minutes, updating the matching relationships based on the system's operating status. A smooth migration mechanism is employed during adjustments to ensure no data packets are lost when sensor data streams switch. The system maintains matching operation logs, detailing the timestamp, matching parameters, allocation results, and performance metrics for each match. These logs are used for subsequent optimization of the matching algorithm parameters. The log data employs a circular storage strategy, retaining the operation records for the most recent 30 days for analysis.

[0040] The performance monitoring module tracks the system's operational status in real time after matching, collecting metrics such as data transmission latency, processing success rate, and node load. When the matching effect is found to be unsatisfactory, the system automatically triggers a re-matching process, adjusting matching parameters or changing the matching algorithm. A fault-tolerance mechanism ensures automatic reallocation of the sensor cluster in the event of a single node failure. Sensor data from the failed node is smoothly migrated to a backup node, ensuring data processing continuity during the migration process. After fault recovery, the system gradually reallocates the sensor cluster back to its original nodes, avoiding system fluctuations caused by frequent switching. The entire system adopts a distributed coordination architecture, with various functional modules communicating through message queues. Matching decision results are synchronized to all nodes via a consensus algorithm, ensuring system consistency.

[0041] The data quality assessment stage verifies the effectiveness of the matched data processing. By comparing the differences between the original sensor data and the standardized processing results, data fidelity and information integrity are evaluated. The assessment results are fed back to the matching algorithm module to optimize subsequent matching strategies. The system supports a manual intervention mode, allowing operators to adjust the automatic matching results according to actual conditions. Manually adjusted records are included in the learning sample library to improve the adaptability of the automatic matching algorithm. Long-term operational data is used to train the matching strategy optimization model. By analyzing historical matching records and system performance data, a correlation model between matching parameters and system performance is established to continuously improve the accuracy and reliability of the matching results.

[0042] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion, characterized in that, include: The liquid alkali concentration data acquisition module is used to acquire data from multiple liquid alkali sensors connected to the Internet of Things and to statistically analyze the spatiotemporal information of each liquid alkali sensor, including liquid alkali concentration value, temperature value, and flow rate value. The edge node operation and maintenance analysis module is used to obtain operation and maintenance information of multiple edge computing nodes, including memory utilization, bandwidth utilization and CPU utilization. The data fusion preprocessing module standardizes the liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module to generate a standardized liquid alkali data matrix. The liquid alkali concentration anomaly detection module performs feature extraction and fusion based on the standardized liquid alkali data matrix output by the data fusion preprocessing module, and identifies nodes with abnormal liquid alkali concentration. The concentration adjustment optimization module generates an optimized adjustment parameter set based on the abnormal liquid alkali concentration nodes identified by the abnormal liquid alkali concentration detection module. An automatic adjustment and control module adjusts the liquid alkali concentration control parameters based on the optimized adjustment parameter set generated by the concentration adjustment and optimization module. The liquid alkali concentration control parameters include valve opening and pump speed. The distributed storage management module is used to store the liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module and the operation and maintenance information acquired by the edge node operation and maintenance analysis module.

2. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion as described in claim 1, characterized in that, The spatiotemporal information of each liquid alkali sensor acquired by the liquid alkali concentration data acquisition module also includes data acquisition frequency and data volume; the operation and maintenance information of each edge computing node acquired by the edge node operation and maintenance analysis module also includes data packet processing rate.

3. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion as described in claim 2, characterized in that, The standardized processing of the data fusion preprocessing module includes: The liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module is formatted to form a time-space matched dataset; For missing liquid alkali sensor data, interpolation was used to fill in the missing values; Normalize the sensor data after filling with liquid alkali. Extract the temporal and spatial features of the normalized liquid alkali sensor data; The normalized data and extracted features are combined to generate the standardized liquid alkali data matrix.

4. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion as described in claim 3, characterized in that, The feature extraction and fusion of the abnormal concentration detection module for liquid alkali includes: Based on the standardized liquid alkali data matrix output by the data fusion preprocessing module, a time series feature extraction unit is constructed to extract the time series features of the standardized liquid alkali data; Construct a spatial feature extraction unit to extract spatial features from standardized liquid alkali data; The extracted temporal and spatial features are weighted and fused to generate a fused feature dataset; Construct an anomaly feature correlation matrix based on the fused feature dataset; The statistical threshold detection method was used to analyze the correlation matrix of abnormal features and identify nodes with abnormal alkali concentration.

5. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion according to claim 4, characterized in that, The concentration adjustment optimization module generates an optimized adjustment parameter set including: Define an optimization objective function to quantify the degree of deviation at abnormal nodes in liquid alkali concentration; The parameter set is initialized using the simulated annealing optimization algorithm, and then iteratively optimized. During the simulated annealing evolution process, the system temperature is gradually reduced to search for the optimal parameter combination; When the temperature drops to a set threshold, the optimized parameter set is measured and the optimized adjustment parameter set is generated.

6. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion according to claim 5, characterized in that, The automatic adjustment control module adjusts the liquid alkali concentration control parameters including: Based on the optimized adjustment parameter set generated by the concentration adjustment optimization module, input the liquid alkali concentration prediction model to estimate the liquid alkali concentration adjustment amount. Adjust the valve opening according to the concentration of the liquid alkali; Adjust the pump speed according to the concentration of the alkali solution.

7. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion according to claim 6, characterized in that, The liquid alkali concentration prediction model is generated based on historical liquid alkali sensor data. The training process includes: collecting historical liquid alkali sensor data and corresponding adjustment results, extracting historical feature parameters, and constructing a decision tree prediction model; The decision tree prediction model is used to estimate the adjustment amount of liquid alkali concentration.

8. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion as described in claim 1, characterized in that, The distributed storage management module stores data including: Synchronously retrieve operational data from each storage port and analyze storage performance benchmarks; Based on the liquid alkali sensor data acquired by the liquid alkali concentration data acquisition module and the operation and maintenance information acquired by the edge node operation and maintenance analysis module, the cluster data evaluation value is calculated. Match storage performance benchmarks with cluster data evaluation values, and allocate target storage ports accordingly; Distribute and store liquid alkali sensor data and operation and maintenance information to the target storage port.

9. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion as described in claim 8, characterized in that, The system also includes a load balancing configuration module, which performs load assessment and balancing configuration based on the data stored in the distributed storage management module, including: Based on the data stored in the distributed storage management module, the load parameters of each data storage node are obtained; Process load parameters to generate load assessment values; Data storage tasks are reassigned based on load assessment values.

10. The liquid alkali concentration monitoring and automatic adjustment system based on Internet of Things and data fusion according to claim 1, characterized in that, The edge node operation and maintenance analysis module generates edge node energy efficiency characterization values ​​based on operation and maintenance information processing. The liquid alkali concentration data acquisition module generates liquid alkali sensor feature values ​​based on spatiotemporal information processing. Match the energy efficiency characterization values ​​of edge nodes with the characteristic values ​​of liquid alkali sensors, and allocate target receiving liquid alkali sensor clusters; The data fusion preprocessing module performs standardized processing on the data from the target receiving liquid alkali sensor cluster.

Images