Calcium formate production management method and system based on Internet of Things

Abstract

The invention relates to the technical field of production management, and discloses a calcium formate production management method and system based on the Internet of Things. The method comprises the steps of collecting time sequence data of a production process through Internet of Things equipment deployed on a calcium formate production line, dividing data segments according to different time windows, and training prediction models of the corresponding windows; an optimal time window is automatically selected based on model expression, and dynamic division and prediction of real-time production time sequence data are realized; detecting abnormal production by comparing the deviation between the actual data and the predicted data; when the deviation exceeds a preset threshold value, a reliable parameter space is delimited based on historical production parameters, multiple production parameter combinations are generated, optimized production parameters are output through a quality prediction model, and intelligent optimization and active regulation and control of the production process are achieved. According to the method, the anomaly detection capability and the parameter optimization efficiency in the calcium formate production process can be improved, and the quality and the production stability of the calcium formate can be improved.

Description

Technical Field

[0001] This application relates to the field of production management technology, and in particular to a method and system for calcium formate production management based on the Internet of Things. Background Technology

[0002] Calcium formate, as an important chemical product, has wide applications in feed, building materials, and pharmaceuticals. Its production process involves multiple steps, including reaction, crystallization, separation, and drying. The parameters such as temperature, pressure, pH, and concentration in each step vary significantly and have a substantial impact on the quality of the final product.

[0003] With the development of IoT technology, real-time data acquisition and intelligent analysis have become possible. How to achieve dynamic monitoring, anomaly early warning, and parameter optimization of the production process based on time-series data has become crucial for improving the production management level of calcium formate. However, existing production management relies heavily on manual experience and fixed monitoring points, making it difficult to achieve real-time dynamic monitoring and anomaly early warning across the entire process and multiple parameters. This has the following shortcomings: traditional methods typically preset fixed time windows for monitoring key parameters, making it difficult to capture abnormal patterns at different time scales (such as instantaneous fluctuations and slow drifts), leading to missed anomalies or false alarms; production anomalies are often discovered only after the fact, making troubleshooting difficult and inefficient, lacking a data-driven real-time early warning mechanism; when production deviations occur, parameter adjustments rely heavily on trial and error based on experience, lacking scientific and systematic optimization methods, making it difficult to quickly find the optimal combination of production parameters, affecting product quality and production efficiency.

[0004] Therefore, the present invention provides a method and system for calcium formate production management based on the Internet of Things. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and system for calcium formate production management based on the Internet of Things (IoT); wherein the method for calcium formate production management based on the IoT includes: Step S1: Collect time-series data of the production process by deploying IoT devices on the calcium formate production line. Divide the time-series data of each production process into multiple different data segments according to different time windows. Step S2: Based on the data segments divided for each time window, train the prediction model for the corresponding time window, select the optimal time window based on multiple prediction models, obtain the current production process, and obtain the optimal time window for the corresponding production process. Step S3: Divide the real-time collected time series data into corresponding first real-time time series segments and second real-time time series segments based on the optimal time window. Input the first real-time time series segment into the prediction model corresponding to the optimal time window to obtain the second prediction time series segment. Calculate the deviation value between the second real-time time series segment and the second prediction time series segment. Step S4: If the deviation value is greater than the preset deviation threshold, define a reliable parameter space based on the historical production parameters of the corresponding production process, generate multiple combinations of production parameters based on the reliable parameter space, and generate optimized production parameters based on the production parameter combinations and the quality prediction model.

[0006] It also proposed an IoT-based calcium formate production management system, including: The data segmentation module collects time-series data of the production process through IoT devices deployed on the calcium formate production line, and divides the time-series data of each production process into multiple different data segments according to different time windows; The window selection module trains a prediction model for each time window based on the data segments divided for each time window, selects the optimal time window based on multiple prediction models, obtains the current production process, and obtains the optimal time window for the corresponding production process. The deviation calculation module divides the real-time acquired time series data into corresponding first real-time time series segments and second real-time time series segments based on the optimal time window. The first real-time time series segment is input into the prediction model corresponding to the optimal time window to obtain the second prediction time series segment. The deviation value between the second real-time time series segment and the second prediction time series segment is calculated. The parameter optimization module, if the deviation value is greater than the preset deviation threshold, defines a reliable parameter space based on the historical production parameters of the corresponding production process, generates multiple combinations of production parameters based on the reliable parameter space, and generates optimized production parameters based on the production parameter combinations and the quality prediction model.

[0007] Compared with the prior art, the beneficial effects of the present invention are at least as follows: The technical solution provided in this application automatically identifies the time scale that best reflects production process anomalies through multi-time window division and model training, improving the accuracy and adaptability of anomaly detection and overcoming the limitations of fixed-window monitoring; based on the prediction model, it compares actual and predicted data in real time to achieve early anomaly detection and warning, changing passive handling to proactive intervention, reducing product defect rate and production risks; it constructs a reliable parameter space through historical data and combines it with a quality prediction model to quickly generate optimized parameter combinations, realizing intelligent adjustment of production parameters and improving product quality and process stability. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1This is a schematic diagram of one embodiment of a calcium formate production management method based on the Internet of Things in this application. Figure 3 This is a schematic diagram of a two-dimensional mesh cell in an embodiment of this application; Figure 2 This is a schematic diagram of one embodiment of an IoT-based calcium formate production management system in this application. Detailed Implementation

[0010] This application provides a method and system for calcium formate production management based on the Internet of Things. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0011] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the IoT-based calcium formate production management method in this application includes: Step S1: Collect time-series data of the production process by deploying IoT devices on the calcium formate production line. Divide the time-series data of each production process into multiple different data segments according to different time windows.

[0012] Specifically, in order to obtain data on the calcium formate production process and facilitate real-time and accurate quality testing and process optimization to improve the quality of calcium formate production, various IoT devices are deployed on the calcium formate production line to collect time-series data of multiple production processes. The time-series data includes the reactor temperature, reactor pressure, pH value, and stirring speed in the reaction process; the solution concentration, cooling rate, and seed crystal addition amount in the crystallization process; and the mother liquor conductivity, drying temperature, and product moisture content in the separation and drying process.

[0013] In the complex calcium formate production process, anomalies that cause the final product to "deviate from the normal pattern" can occur at any stage of the process (such as uneven mixing, temperature runaway in the middle of the reaction, abnormal crystallization rate, etc.). Traditional methods usually rely on experience to preset one or a few fixed monitoring windows (for example, only focusing on the period of highest temperature in the reactor), which is highly unreliable. If the anomaly occurs outside the preset window, the system will completely fail. Traditional inspections usually focus on certain links based on experience, but it is difficult to scientifically prove which time period parameter anomalies have the greatest impact on product quality. Post-incident investigation is often difficult and inefficient. In order to detect anomalies in the production process in advance, the time series data of each production process is first divided into multiple different data segments according to different time windows. For example, a set of time series data contains 12 sets of data obtained at different time points. The sequence can be divided into multiple data segments according to the size of the time window. For example, the step size of the time window can be 2, 3, 4, etc., so the 12 data sequences can be divided into 6 data segments, 4 data segments, 3 data segments, and 2 data segments. Assuming that the 12 time series data are D1, D2, D3, D4, D5, D6, D7, D8, D9, D10, D11, D12, and the time window sizes are 2, 3, 4, 5, and 6 respectively, then when the time window is 2, it is divided into 6 data segments: Data segment 1: {D1, D2}, Data segment 2: {D3, D4}, Data segment 3: {D5, D6}, Data segment 4: {D7, D8}, Data segment 5: {D9, D10}, Data segment 6: {D11, D12}.

[0014] Step S2: Based on the data segments divided for each time window, train the prediction model for the corresponding time window, select the optimal time window based on multiple prediction models, obtain the current production process, and obtain the optimal time window for the corresponding production process.

[0015] Specifically, abnormal patterns affecting quality may have different time scales. Some are instantaneous spikes (such as short-term sensor failures or instantaneous valve action), some are slow drifts (such as a slow decline in catalyst activity), and some are deviations throughout a specific stage (such as lower temperatures throughout the crystallization stage). A fixed-length window cannot effectively capture all types of anomalies simultaneously. To obtain the observation point that best distinguishes between normal and potential anomalies for each production process, i.e., the optimal time window, we first train a prediction model for each time window based on the data segments divided for each time window. The prediction model can predict the data for the second half of the data segment based on the first half of the data segment. Then, we select the optimal time window based on the prediction model. The specific method for obtaining the optimal time window will be explained in detail later. After obtaining the optimal time window for each production process, for the current production process to be detected, we obtain the optimal time window for the corresponding production process based on the above method, which facilitates timely and accurate detection of anomalies.

[0016] Step S3: Divide the real-time collected time series data into corresponding first real-time time series segments and second real-time time series segments based on the optimal time window. Input the first real-time time series segment into the prediction model corresponding to the optimal time window to obtain the second predicted time series segment. Calculate the deviation value between the second real-time time series segment and the second predicted time series segment.

[0017] Specifically, based on the optimal time window, the real-time collected time-series data of the corresponding production process is divided into a first real-time time-series segment and a second real-time time-series segment. The first real-time time-series segment is input into the prediction model corresponding to the time window to obtain the corresponding second predicted time-series segment. Then, the deviation value between the second real-time time-series segment and the second predicted time-series segment is calculated. The deviation value represents the difference between the actual second time-series segment and the corresponding prediction result. The larger the difference, the greater the difference between the change in the corresponding production time-series data and the prediction. Since the second predicted time-series segment is obtained based on the change trend and change characteristics of the historical first time-series segment, the larger the difference, the more the actual second real-time time-series segment deviates from the earlier first real-time time-series segment, indicating a greater possibility of anomalies and a greater possibility of substandard quality of the corresponding calcium formate. Therefore, based on the deviation value and the preset deviation threshold, it is determined whether there is an anomaly. If so, the subsequent production parameters are optimized to improve the quality of the generated calcium formate as much as possible. Through the above steps, anomalies in the production process can be predicted in advance, thereby enabling early adjustments and avoiding the scrapping or downgrading of the produced calcium formate products.

[0018] Step S4: If the deviation value is greater than the preset deviation threshold, define a reliable parameter space based on the historical production parameters of the corresponding production process, generate multiple combinations of production parameters based on the reliable parameter space, and generate optimized production parameters based on the production parameter combinations and the quality prediction model.

[0019] Specifically, to automatically and quickly generate optimized production parameters when deviation warnings occur, a reliable parameter space is first defined based on historical production parameters of the corresponding production process. This space defines reliable boundaries to narrow the search range for subsequent optimization. Then, various combinations of production parameters are generated based on this reliable parameter space, a conservative space constructed based on historical experience. Narrowing the search range using this reliable parameter space facilitates the rapid identification of optimal production parameters. Finally, optimized production parameters are generated based on the production parameter combinations and a quality prediction model. These optimized parameters are applied to subsequent production processes of calcium formate, ensuring that the quality of calcium formate remains optimal throughout subsequent production, compensating for quality defects in previous production, and maximizing the quality of calcium formate. Through the coordination of these steps and methods, production management can be shifted from a passive to a proactive mode, predicting anomalies in advance based on models and quickly and promptly generating optimized production parameters when anomalies are detected.

[0020] In one specific embodiment, based on the data segments divided for each time window, a prediction model for the corresponding time window is trained, specifically including the following steps: Each data segment is divided into a first time series segment and a second time series segment. The first time series segment is used as input data and the second time series segment is used as output data to train a prediction model, enabling the prediction model to learn to predict the second time series segment from the first time series segment.

[0021] Specifically, in order to quickly and accurately determine whether an anomaly has occurred in each data segment, it is necessary to train a corresponding prediction model. The prediction model can predict the second half of the data segment based on the first half of the data segment. First, for multiple data segments divided according to time windows, each data segment is further divided into a first time segment and a second time segment. The first time segment is the first half of the data segment, and the second time segment is the second half of the data segment. The prediction model is trained using the first time segment as input data and the second time segment as output data. This enables the prediction model to predict the second time segment under the ideal state based on the input first time segment, which facilitates subsequent judgment on whether anomalies have occurred in the time series data based on the prediction model.

[0022] In one specific embodiment, selecting the optimal time window based on multiple prediction models includes the following steps: For each time window, all corresponding first time series segments are input into the prediction model to obtain the corresponding predicted second time series segment. The difference between the actual second time series segment and the corresponding predicted second time series segment is calculated. Data segments with differences greater than a preset threshold are defined as first data segments, and data segments with differences less than or equal to the preset threshold are defined as second data segments. Evaluation index values ​​of the first data segments and the second data segments are calculated, and the optimal time window is determined based on the evaluation index values.

[0023] Specifically, in order to address the problem that in the complex and continuous calcium formate production process, it is impossible to know in advance which production process and which time scale data will best reveal potential quality defects, for each time window, firstly, all first time series segments corresponding to the time window are input into the prediction model to obtain the corresponding predicted second time series segments. Then, the difference between the actual second time series segment and the corresponding predicted second time series segment is calculated. Each difference corresponds to a data segment. By calculating the difference, the difference between the predicted data and the actual data is transformed into specific numerical values.

[0024] To accurately determine whether anomalies exist, the data segment is divided into a first data segment and a second data segment based on the data characteristics under normal and abnormal conditions. Under normal conditions, the various parameters of the production process remain within a relatively stable range, and the difference between the model's predicted values ​​and the actual values ​​is small. However, when anomalies occur, the actual values ​​may deviate from the normal range due to certain external influences, resulting in a larger difference between the predicted and actual values. Therefore, the data segment with a difference greater than a preset threshold is defined as the first data segment, and the data segment with a difference less than or equal to the preset threshold is defined as the second data segment. Evaluation index values ​​for the first and second data segments are also calculated. The specific calculation method will be explained in detail later. Finally, the optimal time window is determined based on the rating index values.

[0025] Further, the difference between the actual second time series segment and the corresponding predicted second time series segment is calculated, including: For each data parameter in the second time series segment, calculate the absolute value of the difference between the predicted second time series segment and the actual second time series segment for each data parameter, calculate the variance of the absolute value corresponding to each data parameter, and use the variance value as the difference between the actual second time series segment and the corresponding predicted second time series segment.

[0026] Specifically, for each data parameter in the second time series segment, the absolute value of the difference between the predicted result and the actual data parameter in the second time series segment is calculated. Then, the variance of the absolute values ​​of all data parameters is calculated. The variance is used as the difference between the actual second time series segment and the corresponding predicted second time series segment. The difference between the predicted result and the actual value is calculated based on the above method, which facilitates subsequent calculations based on the difference.

[0027] In one specific embodiment, determining the optimal time window based on the evaluation index value specifically includes the following steps: The differences corresponding to all first data segments are defined as the first differences. The standard deviation of all first differences is calculated as the evaluation index value of the first data segment and is called the first evaluation index value. The differences corresponding to all second data segments are defined as the second differences. The standard deviation of all second differences is calculated as the evaluation index value of the first data segment and is called the second evaluation index value. The optimal time window is determined based on the first evaluation index value and the second evaluation index value.

[0028] Specifically, the first evaluation index value refers to the standard deviation of the difference divided into the first data segment, and the second evaluation value refers to the standard deviation of the difference divided into the second data segment. These values ​​are used to assess the dispersion of the difference data. Different time windows can capture the changing characteristics of data at different time scales. By analyzing the distribution of differences in different time series data, the optimal time window can be determined. Subsequently, the data in the optimal time window can be used to determine whether there are any anomalies, which can improve the accuracy of the judgment and ensure timely detection of anomalies.

[0029] In one specific embodiment, determining the optimal time window based on the first evaluation index value and the second evaluation index value specifically includes the following steps: Obtain the first evaluation index value and the second evaluation index value corresponding to each time window. Take the maximum value among all the first evaluation index values ​​as the first reference value and the minimum value among all the second evaluation index values ​​as the second reference value. Take the combination of the first reference value and the second reference value as the reference point and the combination of the first evaluation index value and the second evaluation index value corresponding to each time window as the judgment point. Calculate the distance between each judgment point and the standard point and take the time window corresponding to the judgment point with the smallest distance as the optimal time window.

[0030] Specifically, a perfect time window should satisfy two characteristics. For the first data segment, due to its large differences, its production parameters are judged to fluctuate significantly, thus the first data segment is judged to be potentially abnormal data. Therefore, the differences corresponding to the first data segment should be as large as possible. For the second data segment, due to its small differences, its production parameters are judged to fluctuate less, thus the differences corresponding to the second data segment should be as small as possible. To better distinguish whether anomalies exist, the first evaluation index value and the second evaluation index value corresponding to each time window are first obtained. The maximum value among all the first evaluation index values ​​is taken as the first reference value, and the minimum value among all the second evaluation index values ​​is taken as the second reference value. The first reference value and the second reference value represent two ideal targets. Therefore, the combination of the first reference value and the second reference value is taken as the reference point. For any time window, the combination of the corresponding first evaluation index value and the second evaluation index value is taken as the judgment point. The distance between each judgment point and the reference point is calculated in a two-dimensional coordinate system, and the time window corresponding to the judgment point with the smallest distance is taken as the optimal time window.

[0031] By constructing data-driven reference points and quantifying the overall gap between the actual performance and the ideal target for each time window, the smaller the distance, the higher the degree of normal data aggregation and abnormal data dispersion within that time window, i.e., the better the comprehensive detection performance, and the more accurately it can determine whether there is an anomaly. The above method transforms the complex problem into a clear and calculable mathematical problem, and intelligently and objectively locks the most valuable time window for detection in the calcium formate production process.

[0032] In one specific embodiment, defining a reliable parameter space based on historical production parameters includes the following steps: The multidimensional space of historical production parameters is divided into multidimensional grid cells. The grid cells containing historical production parameter data points are identified as core data cells. The core data cells are used as reliable parameter spaces. Based on the core data cells, a complete reliable parameter space is generated by iteratively executing preset connection rules.

[0033] Specifically, to obtain corresponding optimized production parameters for the production process, the multidimensional space of historical production parameters is first acquired and divided into multidimensional grid cells. Production parameters include data parameters from each stage of the production process. Assuming the production process is a reaction stage, and to clearly explain this technical solution, let's assume the data parameters corresponding to the reaction stage include reactor temperature and reactor internal pressure. Dividing the multidimensional space of production parameters into multidimensional grid cells means obtaining the minimum and maximum values ​​of temperature and pressure. Based on these minimum and maximum values, multiple intervals are evenly divided along the temperature and pressure dimensions to generate two-dimensional grid cells, such as... Figure 2 The diagram shows the divided two-dimensional grid cells. Grid cells containing historical production parameter data points are identified as core data cells. A core data cell is defined as a grid cell where a corresponding historical data parameter combination exists. If such a combination exists, the corresponding grid cell is designated as the core data cell. For example, if a grid cell has a temperature range of 70–75°C and a pressure range of 0.1–0.15, and historical production parameter data combinations such as (71, 0.11) and (73, 0.12) fall within the temperature range of this grid cell, then this grid cell is designated as the core data cell, forming the initial reliable parameter space. The initial reliable parameter space may contain production parameters that could improve production quality, and therefore requires close attention. Besides the reliable parameter space, adjacent spaces may also contain production parameter combinations that could improve production quality, even if they did not exist in the historical production process. Therefore, based on the core data cells, a complete reliable parameter space is subsequently generated by iteratively executing preset connection rules.

[0034] The above method first obtains the initial core data unit, and then obtains the reliable parameter space based on the core data unit. This narrows the search range from the entire two-dimensional grid cell to the reliable parameter space, which allows for a faster finding of relatively reliable production parameter data combinations and a faster acquisition of the optimal production parameters.

[0035] In one specific embodiment, a complete reliable parameter space is generated by iteratively executing preset connection rules, specifically including the following steps: Choose any parameter dimension as the target parameter dimension, and obtain all core data units that are the same in the target parameter dimension but different in other parameter dimensions as the first core unit. If there is a blank grid cell between any two first core units, the blank grid cell is a cell that is not a core data unit in the target parameter dimension but exists between two core data units. Add all blank grid cells to the reliable parameter space, and also regard the blank grid cells added to the reliable parameter space as core data units. Take the next input parameter dimension as the target parameter dimension, and repeat this step until all input parameter dimensions have been selected as the target parameter dimensions, and end this step.

[0036] Specifically, we first select temperature as the target parameter dimension, and then obtain core data units that are the same in the temperature dimension but different in other parameter dimensions as the first core unit, such as... Figure 2 As shown, assuming units 1 and 4 are the first core units, in terms of the temperature parameter, units 2 and 3 between units 1 and 4 are blank grid units. A blank grid unit refers to a combination in the space of a blank grid unit that does not contain historical parameter data and exists between two core data units. Units 2 and 3 are incorporated into the reliable parameter space, and the blank grid units added to the reliable parameter space are also regarded as core data units. Assuming that in the initial state, units 1, 4, 14, and 15 are core data units, then after iteration in the temperature dimension, units 2 and 3 are incorporated into the reliable parameter space and are also regarded as core data units. Subsequently, in the pressure dimension, units 6 and 10 between units 2 and 14, and units 7 and 11 between units 3 and 15 can also be incorporated into the reliable parameter space based on the same rules. After iteration in two dimensions, the space formed by the combination of units 1, 2, 3, 4, 6, 7, 10, 11, 13, and 15 is the complete reliable parameter space, and subsequent production parameter combinations can be generated based on the reliable parameter space.

[0037] In one specific embodiment, generating multiple combinations of production parameters based on a reliable parameter space includes the following steps: The data parameter combination corresponding to the center point of each core data unit in the reliable parameter space is taken as the first data parameter combination. The historical production parameter data contained in the core data unit is input into the quality prediction model. The historical production parameter data with good quality prediction is taken as the second data parameter combination. Several sampling points are randomly selected in each core data unit, and the data parameter combination corresponding to the sampling points is taken as the third data parameter combination. The first data parameter combination, the second data parameter combination, and the third data parameter combination are taken as the production parameter combination.

[0038] Specifically, to obtain multiple different combinations of production parameters and facilitate faster and more accurate acquisition of optimal production parameters based on these combinations and the production quality prediction model, the data parameter combination corresponding to the center point of each core data unit in the reliable parameter space is firstly used as the first data parameter combination. Historical production parameter data contained within the core data unit is then input into the quality prediction model to obtain corresponding quality evaluation parameters. Production parameter combinations whose quality evaluation parameters exceed a preset first threshold are used as the second data parameter combination. The second data parameter combination indicates that the calcium formate produced in the historical production process was of good quality, thus increasing the likelihood of producing high-quality calcium formate in subsequent production processes. Therefore, the second data parameter combination is obtained. To compensate for the shortcomings of the production parameter combinations, several sampling points are randomly selected within each core data unit, and the data parameter combination corresponding to these sampling points is used as the third data parameter combination. The first, second, and third data parameter combinations are then used as the production parameter combination. Through these steps, multiple possible production parameter combinations that result in high-quality calcium formate are obtained, covering as many potential high-quality production parameter combinations as possible and avoiding the omission of potential optimal solutions.

[0039] Furthermore, optimized production parameters are generated based on the combination of production parameters and the quality prediction model, including: The quality prediction model can predict the corresponding calcium formate product quality score based on any combination of input production parameters. By inputting all combinations of production parameters into the quality prediction model, the corresponding production quality score is obtained. The production parameter combination with the highest production quality score is taken as the optimal production parameter. The production plan is adjusted based on the optimal production parameter, and the adjusted production plan is applied to the subsequent production process.

[0040] By using the above methods, abnormalities can be detected in a timely manner during the production process. Then, when abnormalities are detected, the production plan can be adjusted quickly to remedy the abnormalities as much as possible, thereby achieving the goal of improving the quality of calcium formate products as much as possible.

[0041] The above describes an IoT-based calcium formate production management method according to embodiments of this application. The following describes an IoT-based calcium formate production management system according to embodiments of this application. Please refer to [link / reference]. Figure 3 One embodiment of the IoT-based calcium formate production management system in this application includes: The data segmentation module collects time-series data of the production process through IoT devices deployed on the calcium formate production line, and divides the time-series data of each production process into multiple different data segments according to different time windows; The window selection module trains a prediction model for each time window based on the data segments divided for each time window, selects the optimal time window based on multiple prediction models, obtains the current production process, and obtains the optimal time window for the corresponding production process. The deviation calculation module divides the real-time acquired time series data into corresponding first real-time time series segments and second real-time time series segments based on the optimal time window. The first real-time time series segment is input into the prediction model corresponding to the optimal time window to obtain the second prediction time series segment. The deviation value between the second real-time time series segment and the second prediction time series segment is calculated. The parameter optimization module, if the deviation value is greater than the preset deviation threshold, defines a reliable parameter space based on the historical production parameters of the corresponding production process, generates multiple combinations of production parameters based on the reliable parameter space, and generates optimized production parameters based on the production parameter combinations and the quality prediction model.

[0042] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0043] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for managing the production of calcium formate based on the Internet of Things, characterized in that, The method includes: Step S1: Collect time-series data of the production process by deploying IoT devices on the calcium formate production line. Divide the time-series data of each production process into multiple different data segments according to different time windows. Step S2: Based on the data segments divided for each time window, train the prediction model for the corresponding time window, select the optimal time window based on multiple prediction models, obtain the current production process, and obtain the optimal time window for the corresponding production process. Step S3: Divide the real-time collected time series data into corresponding first real-time time series segments and second real-time time series segments based on the optimal time window. Input the first real-time time series segment into the prediction model corresponding to the optimal time window to obtain the second prediction time series segment. Calculate the deviation value between the second real-time time series segment and the second prediction time series segment. Step S4: If the deviation value is greater than the preset deviation threshold, define a reliable parameter space based on the historical production parameters of the corresponding production process, generate multiple combinations of production parameters based on the reliable parameter space, and generate optimized production parameters based on the production parameter combinations and the quality prediction model.

2. The method according to claim 1, characterized in that, Based on the data segments divided for each time window, a prediction model for the corresponding time window is trained, including: Each data segment is divided into a first time series segment and a second time series segment. The first time series segment is used as input data and the second time series segment is used as output data to train a prediction model, enabling the prediction model to learn to predict the second time series segment from the first time series segment.

3. The method according to claim 1, characterized in that, The optimal time window is selected based on multiple prediction models, including: For each time window, all corresponding first time series segments are input into the prediction model to obtain the corresponding predicted second time series segment. The difference between the actual second time series segment and the corresponding predicted second time series segment is calculated. Data segments with differences greater than a preset threshold are defined as first data segments, and data segments with differences less than or equal to the preset threshold are defined as second data segments. Evaluation index values ​​of the first data segments and the second data segments are calculated, and the optimal time window is determined based on the evaluation index values.

4. The method according to claim 3, characterized in that, Determining the optimal time window based on evaluation index values ​​includes: The differences corresponding to all first data segments are defined as the first differences. The standard deviation of all first differences is calculated as the evaluation index value of the first data segment and is called the first evaluation index value. The differences corresponding to all second data segments are defined as the second differences. The standard deviation of all second differences is calculated as the evaluation index value of the first data segment and is called the second evaluation index value. The optimal time window is determined based on the first evaluation index value and the second evaluation index value.

5. The method according to claim 4, characterized in that, The optimal time window is determined based on the values ​​of the first and second evaluation indicators, including: Obtain the first evaluation index value and the second evaluation index value corresponding to each time window. Take the maximum value among all the first evaluation index values ​​as the first reference value and the minimum value among all the second evaluation index values ​​as the second reference value. Take the combination of the first reference value and the second reference value as the reference point and the combination of the first evaluation index value and the second evaluation index value corresponding to each time window as the judgment point. Calculate the distance between each judgment point and the standard point and take the time window corresponding to the judgment point with the smallest distance as the optimal time window.

6. The method according to claim 1, characterized in that, Delineate a reliable parameter space based on historical production parameters, including: The multidimensional space of historical production parameters is divided into multidimensional grid cells. The grid cells containing historical production parameter data points are identified as core data cells. The core data cells are used as reliable parameter spaces. Based on the core data cells, a complete reliable parameter space is generated by iteratively executing preset connection rules.

7. The method according to claim 6, characterized in that, A complete reliable parameter space is generated by iteratively executing preset connection rules, including: Choose any parameter dimension as the target parameter dimension, and obtain all core data units that are the same in the target parameter dimension but different in other parameter dimensions as the first core unit. If there is a blank grid cell between any two first core units, the blank grid cell is a cell that is not a core data unit in the target parameter dimension but exists between two core data units. Add all blank grid cells to the reliable parameter space, and also regard the blank grid cells added to the reliable parameter space as core data units. Take the next input parameter dimension as the target parameter dimension, and repeat this step until all input parameter dimensions have been selected as the target parameter dimensions, and end this step.

8. The method according to claim 7, characterized in that, Multiple combinations of production parameters are generated based on a reliable parameter space, including: The data parameter combination corresponding to the center point of each core data unit in the reliable parameter space is taken as the first data parameter combination. The historical production parameter data contained in the core data unit is input into the quality prediction model. The historical production parameter data with good quality prediction is taken as the second data parameter combination. Several sampling points are randomly selected in each core data unit, and the data parameter combination corresponding to the sampling points is taken as the third data parameter combination. The first data parameter combination, the second data parameter combination, and the third data parameter combination are taken as the production parameter combination.

9. An IoT-based calcium formate production management system, used to implement the IoT-based calcium formate production management method as described in any one of claims 1-8, characterized in that, The system includes: The data segmentation module collects time-series data of the production process through IoT devices deployed on the calcium formate production line, and divides the time-series data of each production process into multiple different data segments according to different time windows; The window selection module trains a prediction model for each time window based on the data segments divided for each time window, selects the optimal time window based on multiple prediction models, obtains the current production process, and obtains the optimal time window for the corresponding production process. The deviation calculation module divides the real-time acquired time series data into corresponding first real-time time series segments and second real-time time series segments based on the optimal time window. The first real-time time series segment is input into the prediction model corresponding to the optimal time window to obtain the second prediction time series segment. The deviation value between the second real-time time series segment and the second prediction time series segment is calculated. The parameter optimization module, if the deviation value is greater than the preset deviation threshold, defines a reliable parameter space based on the historical production parameters of the corresponding production process, generates multiple combinations of production parameters based on the reliable parameter space, and generates optimized production parameters based on the production parameter combinations and the quality prediction model.

Images