Evaporator fouling prediction control method and system based on multi-source operation data

Abstract

The present application relates to the technical field of evaporator control, in particular to an evaporator scaling prediction control method and system based on multi-source operation data, comprising the following steps: collecting real-time state data in the operation process of the evaporator and calling historical operation record data, correlating and arranging different source operation data to form a data sequence, inputting the data sequence into a machine learning model to analyze the operation state change relationship and generate an evaporator scaling state prediction result, adjusting the prediction process according to the deviation between the actual operation state and the prediction result, and generating a control instruction according to the corrected scaling state prediction result. Through multi-source operation data correlation processing and prediction correction, the present application forms a data closed loop between the evaporator scaling state prediction and operation control, improves the accuracy of operation state analysis and control adjustment, and provides data support and control execution capability improvement for state management in the operation process of the evaporator.

Description

Technical Field

[0001] This invention relates to the field of evaporator control technology, and in particular to an evaporator scaling prediction and control method and system based on multi-source operating data. Background Technology

[0002] The field of evaporator control technology involves the status detection, data processing, and operation adjustment technologies during the operation of evaporation equipment. It typically revolves around the information interaction between the evaporator body, detection components, control equipment, and actuators to acquire, analyze, and adjust the operating status and parameters. This involves collecting and processing the status information during equipment operation and adjusting the relevant execution processes based on the analysis results.

[0003] The traditional evaporator scaling prediction and control method refers to the method of predicting and controlling the scaling state generated during the operation of the evaporator. It usually acquires operating status data by setting detection components on the evaporator and its connecting pipelines, and transmits the collected data to the control equipment for processing and analysis. The scaling state inside the evaporator is judged based on the data changes, and control commands are generated based on the judgment results. The actuator adjusts the relevant operations during the operation of the evaporator. At the same time, the operating data is stored in the local storage area or server for subsequent status analysis and control calls.

[0004] In the current evaporator operation monitoring and control process, equipment status information is mainly obtained by relying on a single operating parameter or a fixed periodic detection method. It is difficult to uniformly correlate and process operating data from different sources, which makes it difficult to analyze the scaling state change process in a timely manner. There is a discrepancy between the predicted results and the actual operating state. Control adjustments lack continuous data basis, affecting the state judgment and control execution of the evaporator operation process. At the same time, it is difficult to adapt to the scaling trend analysis needs under continuously changing operating conditions, and to provide data support for operating status analysis and control strategy adjustment. Summary of the Invention

[0005] This invention aims to provide a method and system for predicting and controlling evaporator scaling based on multi-source operating data. By collecting multi-source state data during the operation of the evaporator and combining it with historical operating records to establish a data analysis foundation, and using a machine learning model to analyze changes in operating state, the method corrects the deviation between the predicted results and the actual operating state, and then executes control adjustments based on the corrected prediction results, thereby establishing a correlation between the prediction of evaporator scaling state and the operation control process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an evaporator scaling prediction and control method based on multi-source operating data, comprising the following steps: Real-time status data of the evaporator during operation is collected by a detection device set at the evaporator's operating location. Historical operation records are retrieved from a historical operation database that receives and stores the operation data output by the detection device according to the acquisition time. The real-time status data and the historical operation records are correlated and organized according to the operation time to form a data sequence for scaling status analysis. The real-time status data acquisition process further includes setting up multiple detection devices according to the operating status corresponding to different detection positions inside the evaporator, collecting operating status data of the corresponding positions through the multiple detection devices, and sorting and associating the real-time status data according to the data acquisition time corresponding to each detection device to form a status data record with a time correspondence.

[0007] The process of calling historical operation record data also includes extracting historical operation record data within the corresponding time period from the historical operation database according to a preset time range, and marking the historical operation record data with time. The data records corresponding to different operation cycles are associated with the real-time status data to form a data sequence for input to the machine learning model.

[0008] When associating and organizing the real-time status data and the historical operation record data, the process also includes performing data integrity checks on the real-time status data and the historical operation record data, filtering out status records that do not meet preset data conditions, and combining the remaining status records according to the corresponding operation process to form a continuous data sequence.

[0009] The data sequence formed based on the real-time status data and the historical operation record data is input into the machine learning model. The machine learning model analyzes the status change characteristics in the data sequence to generate a prediction result of evaporator scaling status. The machine learning model is a multivariate regression model. The normalized change characteristics of operating temperature, pressure, flow rate, and concentration are input and weighted sequentially. The weighted features and their unidirectional, inverse, or hold-alive relationships are linearly combined according to the corresponding output regression coefficients and a bias is added to generate status prediction vectors and scaling status prediction values ​​for operating temperature, pressure, flow rate, and concentration, respectively. The status prediction vectors, scaling status prediction values, and prediction time together constitute the prediction result of evaporator scaling status.

[0010] In the process of analyzing the data using a machine learning model, based on the evaporator operating temperature change data, pressure change data, flow rate change data, and concentration change data in the data sequence, the difference between adjacent records of each field is divided by the pre-fixed positive reference change amount of that field to determine the normalized change amplitude and trend within the corresponding time period. The relationship between the changes of different types of operating data is then correlated to generate the scaling state change correlation results for machine learning model analysis.

[0011] Furthermore, the data change relationships corresponding to the operating temperature change data, the pressure change data, the flow rate change data, and the concentration change data are input into the machine learning model. Based on the combination of data changes under different operating conditions, the predicted output corresponding to the scaling state change is determined, and the predicted output is converted into the evaporator scaling state prediction result.

[0012] The evaporator operating status data for the current time period is collected by the detection device. The measured values ​​of operating temperature, pressure, flow rate and concentration for the current time period are respectively compared with the predicted values ​​of the corresponding fields of the state prediction vector for the same time period in the evaporator scaling state prediction result to form the deviation of each field. When the deviation of at least one field exceeds the preset deviation range of the field, the prediction parameters of the machine learning model or the input data processing method are adjusted according to the deviation to generate a corrected scaling state prediction result.

[0013] Specifically, when comparing the evaporator operating status data with the evaporator scaling status prediction results, the state prediction vectors in the evaporator operating status data and the evaporator scaling status prediction results are matched according to the prediction time. The measured values ​​of operating temperature, pressure, flow rate, and concentration are subtracted from the corresponding field prediction values ​​to form the state deviation of each field. The state deviation of each field is judged according to the preset deviation range of the field. When the state deviation of at least one field exceeds its corresponding preset deviation range, the input weight of the corresponding operating status data in the machine learning model is adjusted.

[0014] Based on the corrected scaling state prediction results and the evaporator operating state data, control conditions are determined. When the preset control conditions are met, a control command is generated and sent to the evaporator actuator to adjust the evaporator operation process.

[0015] Specifically, when judging control conditions based on the corrected scaling state prediction result and the evaporator operating status data, the predicted scaling state corresponding to the corrected scaling state prediction result is matched with the current operating status corresponding to the evaporator operating status data. When the predicted relative scaling degree value with a range of [0, 1] reaches the control trigger threshold preset according to the same calibration scale, a control instruction for increasing the material circulation flow of the evaporator is generated, and the control result is recorded based on the operating status data after the control instruction is executed.

[0016] The present invention also provides an evaporator scaling prediction and control system based on multi-source operating data, including a data acquisition unit, a data processing unit, a prediction and analysis unit, a feedback correction unit, and a control execution unit.

[0017] The data acquisition unit is used to collect real-time status data during the operation of the evaporator, and to retrieve historical operation record data from the historical operation database that receives and stores the operation data output by the detection device according to the acquisition time.

[0018] The data processing unit is used to organize the real-time status data and the historical operation record data to form a data sequence for scaling status analysis.

[0019] The predictive analysis unit is used to input the data sequence into a machine learning model to analyze the changes in the evaporator's operating status and generate a prediction result of the evaporator's scaling status.

[0020] The feedback correction unit is used to acquire evaporator operating status data. According to the same prediction time, it subtracts the measured values ​​of operating temperature, pressure, flow rate and concentration in the evaporator operating status data from the corresponding field prediction values ​​of the state prediction vector in the evaporator scaling state prediction result, and compares the deviation of each field with the preset deviation range of the field. The prediction process is adjusted according to the comparison results.

[0021] The control execution unit is used to generate control commands based on the corrected scaling state prediction results, and send the control commands to the evaporator actuator.

[0022] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention collects real-time status data during evaporator operation and combines it with historical operation records in the data storage area for time correlation and processing to form a data sequence for scaling status analysis. It uses a machine learning model to analyze the relationship between changes in operating status to generate prediction results, and corrects the results based on the deviation between the actual operating status and the prediction results. This allows the scaling status prediction process to be adjusted in accordance with actual operating changes. At the same time, it generates control commands based on the corrected prediction results, realizing the data correlation between prediction analysis and operation control, reducing the impact of prediction result deviations on the control process, and lowering the risk of status misjudgment in the control process. Attached Figure Description

[0023] Figure 1 This is the main flowchart of the evaporator scaling prediction and control based on multi-source operating data of the present invention; Figure 2 This is a data flow diagram that shows how real-time status data and historical operation record data form a continuous data sequence in this invention. Figure 3 This is a data flow diagram for the state feature calculation and prediction of the multivariate regression machine learning model of this invention; Figure 4 This is the feedback correction diagram for adjusting input weights based on field state deviation in this invention; Figure 5 This is a diagram showing the collaborative and closed-loop data interaction of the evaporator scaling prediction and control system units in this invention. Detailed Implementation

[0024] The technical solution described in this invention is used to predict the scaling state based on multi-source operating data generated during evaporator operation, and to perform feedback correction on the prediction process using the current operating state. When the corrected prediction result meets the control trigger condition, a control command is generated to adjust the evaporator operation process. The following embodiments uniformly use real-time status data, historical operating record data, evaporator operating status data, data sequences, status characteristic data, scaling state change correlation results, status prediction vectors, scaling state prediction values, evaporator scaling state prediction results, status deviations, corrected scaling state prediction results, and control commands as data objects.

[0025] The real-time status data consists of data continuously acquired by a detection device located at the evaporator's operating position during evaporator operation. Historical operation record data is existing operation data obtained from a historical operation database that receives and stores the operation data output by the detection device according to the acquisition time. Evaporator operating status data is data acquired by the detection device within the current time period when the feedback comparison phase begins. Real-time status data, historical operation record data, and evaporator operating status data all contain at least data records corresponding to operating temperature, pressure, flow rate, and concentration, and have acquisition times that can determine the order of acquisition.

[0026] The data sequence is a continuous state record formed by organizing real-time status data and historical operation records according to operating time. State characteristic data is the model input formed by operating temperature change data, pressure change data, flow rate change data, and concentration change data from the data sequence. The scaling state change correlation results represent the magnitude and direction of change of the above different types of operating data within the same corresponding time period, as well as the synchronous change relationship between different types of operating data. The machine learning model adopts a multivariate regression model capable of receiving multivariate continuous state characteristics and outputting predicted states. Its online operation part includes at least three continuous processing stages: input weights, prediction parameters, and prediction output.

[0027] Throughout a complete prediction cycle, the machine learning model not only generates predicted scaling conditions representing the degree of scaling change, but also simultaneously retains a state prediction vector corresponding to the current comparison cycle. The field order in the state prediction vector is consistent with the field order of the current evaporator operating status data, corresponding at least sequentially to operating temperature, pressure, flow rate, and concentration. Therefore, during the feedback correction phase, the actual evaporator operating status data can be compared with the state prediction vector in the model output item by item according to the same fields, rather than directly comparing data objects with different physical meanings. The predicted scaling conditions and the corresponding state prediction vector together constitute the evaporator scaling condition prediction result.

[0028] Before running the machine learning model, a defined order of input fields, corresponding input weights for each input field, prediction parameters for forming the state prediction vector, and prediction parameters for forming the scale state prediction value are maintained. During online execution, the model receives state feature data in the chronological order of the data sequence. When the feedback correction condition is not met, the current input weights and prediction parameters are maintained. When the feedback correction condition is met, the corresponding input weights or prediction parameters are modified according to the correction rules in the following embodiment, and the state prediction vector and scale state prediction value are re-formed.

[0029] Example 1 Please see Figures 1 to 4 This embodiment provides a method for predictive control of evaporator scaling based on multi-source operating data. The method employs a multivariate regression machine learning model as the prediction model. First, real-time status data obtained from multiple detection locations and historical operating records are combined to form a continuous data sequence according to operating time. Then, the changes in operating temperature, pressure, flow rate, and concentration between adjacent time records are calculated, and the resulting state feature data is input into the machine learning model. The machine learning model forms a state prediction vector and a scaling state prediction value according to the input weights and prediction parameters. The evaporator operating status data detected within the current time period and the state prediction vector are used to form a state deviation item by item. When the state deviation exceeds a preset deviation range, the input weights of the corresponding state feature data are adjusted, the prediction calculation is re-executed, and a corrected scaling state prediction result is formed. Finally, control conditions are determined based on this corrected result.

[0030] S1. Real-time status data of the evaporator during operation is collected by a detection device set at the operating position of the evaporator, and historical operation record data is retrieved from the historical operation database that receives and stores the operation data output by the detection device according to the collection time. The real-time status data and historical operation record data are correlated and organized according to the operation time to form a data sequence for scaling status analysis.

[0031] S111, multiple detection devices are set up according to the operating status corresponding to different detection locations inside the evaporator. Each detection device acquires the operating status data of its corresponding location. Each operating status record retains at least the acquisition time and the corresponding data content of operating temperature, pressure, flow rate, and concentration for that detection location. The data output by each detection device is first sorted in chronological order according to the acquisition time. Data from the same detection location maintains its own continuous time relationship, while data from different detection locations establish a time correspondence based on their respective acquisition times.

[0032] During data processing, the acquisition time is used as the correlation benchmark between real-time status data. Status records from different detection sources at the same corresponding time position are grouped into one time status record; records whose acquisition time cannot be determined are not included in the current data sequence; records whose acquisition time can be determined but cannot form a corresponding relationship with the current data processing interval are retained in their original running records and are not included in the data sequence of the current prediction period. In this way, each status record in the data sequence has a definite time position.

[0033] S112, extract historical operation record data from the historical operation database according to the preset time range, and retain corresponding time stamps for the historical operation record data. For historical operation records formed in different operation cycles, organize them in chronological order within their respective operation cycles, and then establish a time relationship between the data records of the same or corresponding operation stages and the real-time status data.

[0034] When historical operation records are added to the current data sequence, the technical meaning of operating temperature, pressure, flow rate, and concentration is not changed; they are simply rearranged according to the field order required by the model. The same field uses the same name and data identity in both real-time status data and historical operation records to ensure that subsequent calculations of changes in operating temperature, pressure, flow rate, and concentration can be performed continuously along the same field.

[0035] S113, Perform data integrity checks on real-time status data and historical operation record data. The data integrity check sequentially checks whether the acquisition time can be identified, whether the status fields can be read according to the predetermined field types, and whether the chronological order of preceding and following status records can be determined. Status records lacking acquisition time and whose chronological order cannot be determined are excluded; if a status record lacks a status field required for the current prediction calculation, that status record will not participate in the calculation of adjacent changes dependent on that field; if duplicate status records with completely identical content appear at the same acquisition time, only one corresponding record is retained; if conflicting records appear at the same acquisition time and their unique status values ​​cannot be determined, the conflicting record will not be used as input for the current prediction period.

[0036] After completing the data integrity check, the retained real-time status data and historical operation record data are combined according to the operation time to form a continuous data sequence. Once the record order in the data sequence is fixed, the time order will not be changed again in subsequent model processing.

[0037] S114: From the continuous data sequence, retrieve two adjacent state records one by one, using the state record that appears later in time as the current record and the state record that appears immediately before it as the preceding record. Calculate the state changes for operating temperature, pressure, flow rate, and concentration respectively. The magnitude of the change in a certain type of operating data is determined by the difference between the corresponding value of the current record and the corresponding value of the preceding record; the direction of change is determined by the relationship between the current record and the preceding record: when the current record is greater than the preceding record, the record shows an upward trend; when the current record is less than the preceding record, the record shows a downward trend; when the two are equal, the record shows a stable trend. Before performing correlation calculations or input weighting on different types of operating data, divide the difference between each field by a pre-fixed positive reference change for that field; subsequent change magnitudes refer to the absolute value of the obtained normalized difference, and the direction of change remains unchanged.

[0038] The operating temperature change data, pressure change data, flow rate change data, and concentration change data are generated sequentially in the manner described above. Each change data corresponds to a specific preceding acquisition time and the current acquisition time, thus ensuring that the state changes are not isolated values ​​detached from time sequence, but rather can be located to the state characteristics of a specific operating time period.

[0039] S115: Recombine the existing operating temperature change data, pressure change data, flow rate change data, and concentration change data according to the same time period. For the same time period, record the magnitude and direction of change of each of the four types of operating data, and compare the directional relationships between the four types of operating data. When two operating data both show an increase or both show a decrease in the same time period, they are recorded as having a same-direction change relationship; when one shows an increase and the other shows a decrease, they are recorded as having an opposite-direction change relationship; if any operating data remains unchanged in the time period, its trend is retained and combined with the direction of change of other operating data as the state change combination for that time period.

[0040] The state feature data consists of four types of change magnitude, four types of change direction, and the change relationships between these types of operational data corresponding to a time period. The state feature data for each time period are still arranged in the chronological order of the original data sequence and are used as the input to the S2 model.

[0041] S2 inputs the data sequence formed by real-time status data and historical operation record data, as well as the status feature data formed by the data sequence, into the machine learning model. The machine learning model analyzes the status change characteristics and generates the evaporator scaling status prediction result.

[0042] S211: The state characteristics of each time period are arranged in the order of operating temperature, pressure, flow rate, and concentration. The signed difference of adjacent records in the same field is divided by the pre-fixed positive reference change for that field to obtain a normalized change value. The absolute value of this value represents the magnitude of the change, and 1, -1, and 0 are used to encode the rising, falling, and holding directions, respectively. The product of the four magnitude and four direction codes, along with the six direction codes arranged in the order of temperature-pressure, temperature-flow rate, temperature-concentration, pressure-flow rate, pressure-concentration, and flow rate-concentration, forms a fourteen-dimensional feature vector. The feature vectors are then arranged row by row in chronological order and input into a multivariate regression model. The positive reference change is determined and saved by the historical operating change magnitude of each field before model training and remains unchanged during online prediction and feedback correction. Time information is used for recording sorting and prediction time matching and does not participate in the numerical summation of different physical quantities.

[0043] The machine learning model employs a multivariate regression prediction structure. During the online prediction phase, the model assigns corresponding input weights to each type of state feature and saves prediction parameters used to form the state prediction vector and the scale state prediction value. The input weights determine the contribution of a particular type of state change feature to the prediction calculation, while the prediction parameters determine the weighted effect of different state features on each predicted output field. By applying the input weights w to each term of the fourteen-dimensional feature vector x, the following is obtained: Each output is set according to Calculate and sum the fourteen input features using the index k. The first four outputs form the state prediction vector, corresponding to the regression coefficients. and bias The data is fitted based on the historical measured values ​​of each field; the fifth output is the scaling state regression output, truncated to the [0, 1] interval as the predicted scaling state value. The regression coefficients and biases are fitted and saved using historical operating samples and corresponding output labels. The scaling label is set as 0 for no scaling after cleaning and 1 for reaching the preset cleaning judgment state, and is linearly normalized between the two calibration states according to the calibrated scaling degree; this value represents the relative scaling degree and does not represent temperature, pressure, flow rate or concentration.

[0044] S212, when calculating for a time period to be predicted, the machine learning model first reads the continuous state feature records formed before that time period. The amplitudes of temperature change, pressure change, flow rate change, and concentration change are weighted according to their current input weights. The larger the input weight, the greater the influence of the corresponding state change amplitude in subsequent prediction combinations; the smaller the input weight, the smaller the influence of the corresponding state change amplitude in subsequent prediction combinations.

[0045] The direction of change is not mixed with the magnitude of change in the same data. When the machine learning model reads each state feature record, it simultaneously reads the upward trend, downward trend, or static trend, and determines the direction of the change's effect on the current prediction output based on the directional interaction relationships already established in the prediction parameters. Therefore, two state records with the same magnitude of temperature change but opposite directions will not be processed as having the same state feature.

[0046] S213 combines the unidirectional, inverse, and constant relationships among operating temperature, pressure, flow rate, and concentration within the same time period, and combines this combination with the change amplitude of the four types of operating data after input weighting to form the scaling state change correlation result.

[0047] The scaling state change correlation results are entered into the model prediction calculation according to a fixed field order. For the operating temperature prediction field in the state prediction vector, the machine learning model reads the operating temperature change characteristics and the associated pressure, flow rate, and concentration changes. Based on the prediction parameters corresponding to this output field, it combines the contributions of each input to form the operating temperature prediction value. The pressure, flow rate, and concentration prediction fields use the same data flow method, with their corresponding prediction parameters combining the relevant state features to sequentially form the pressure, flow rate, and concentration prediction values.

[0048] The four predicted values ​​are arranged in a fixed order of operating temperature, pressure, flow rate, and concentration to form a state prediction vector. The state prediction vector retains the corresponding prediction time period, enabling the feedback correction stage to compare it field-by-field with the evaporator operating status data actually detected during the same time period.

[0049] S214, while forming the state prediction vector, the machine learning model reads the weighted operating temperature change features, pressure change features, flow rate change features, concentration change features, and scaling state change correlation results, and determines the contribution of each input feature to the scaling state prediction in sequence according to the prediction parameters used for scaling prediction output.

[0050] In the specific calculation, the weighted variation amplitude of various types of operating data is first processed according to the corresponding prediction parameters. Then, the direction of change of these types of operating data is combined to determine whether they have an increasing, decreasing, or maintaining effect on the predicted value of scaling status. Subsequently, the various effects within the same time period are combined according to the fixed input order of the model, and the correlation effects corresponding to the same-direction change relationship, opposite-direction change relationship, and maintaining relationship are added to the prediction combination for that time period to form the predicted value of scaling status for that time period. The combination is calculated according to the fourteen-dimensional input order determined in S211 and the above regression formula. The predicted value of scaling status is the relative scaling degree truncated to [0, 1].

[0051] For the state characteristics of multiple consecutive time periods, the above processing is performed sequentially in chronological order, so that the prediction calculations for later times can be based on the established continuous change relationship, rather than randomly rearranging different operating time periods. The final scale state prediction value, the state prediction vector, and the corresponding prediction time together constitute the evaporator scale state prediction result.

[0052] S215, the evaporator scaling state prediction result is transmitted to S3. S3 receives the state prediction vector and the corresponding prediction time, and obtains the evaporator operating state data formed by actual detection during the same time period; the scaling state prediction value is retained and used to form the corrected scaling state prediction result after the feedback correction is completed.

[0053] S3. Collect evaporator operating status data within the current time period through detection equipment. Subtract the measured values ​​of operating temperature, pressure, flow rate and concentration within the current time period from the predicted values ​​of the corresponding fields of the state prediction vector in the same time period in the evaporator scaling state prediction result to form the deviation of each field. When there is a state deviation between the two that meets the correction condition, adjust the input weight of the corresponding operating status data in the machine learning model according to the state deviation, and regenerate the corrected scaling state prediction result.

[0054] S311: Based on the predicted time in the evaporator scaling state prediction result, obtain evaporator operating state data for the same time period from the detection equipment, and form an actual operating state record according to the field order of operating temperature, pressure, flow rate, and concentration. After the state prediction vector and the actual operating state record have the same field order, calculate the deviations of the operating temperature field, pressure field, flow rate field, and concentration field respectively. Each field deviation is compared with its corresponding preset deviation range.

[0055] Each field's deviation is formed by the difference between the actual detected value and the corresponding predicted value in the state prediction vector. A positive deviation relationship is maintained when the actual detected value is higher than the predicted value, and a negative deviation relationship is maintained when the actual detected value is lower than the predicted value. When both are the same, the corresponding deviation for that field is considered indifferent. The magnitude of the difference for each field is also obtained to determine the degree of prediction error for that field.

[0056] S312 compares the deviations of the operating temperature field, pressure field, flow rate field, and concentration field with their respective preset deviation ranges. The preset deviation ranges have a lower boundary and an upper boundary. When a field deviation is within the corresponding range or at the boundary, the field is determined to be within the deviation range; when a field deviation is greater than the upper boundary or less than the lower boundary, the field is determined to be outside the deviation range.

[0057] If none of the four fields exceed their respective preset deviation ranges, the input weights of the current machine learning model are maintained, and the current evaporator scaling state prediction result is used as the corrected result for this cycle and entered into S4. If at least one field exceeds its corresponding preset deviation range, the out-of-range field is identified, and the state change feature corresponding to that field is identified as the object requiring input weight correction.

[0058] S313, determine the adjustment direction for the input weights that need correction. First, read the direction of the deviation between the actual detected value and the predicted value corresponding to the out-of-range field, then read the direction of change of this field in the current state features and the direction of action of the corresponding prediction parameter.

[0059] When the actual detected value is higher than the predicted value, and the state feature of that field has a positive effect on the corresponding predicted output after processing by the current prediction parameters, the input weight corresponding to that field is increased. When the actual detected value is higher than the predicted value, but the state feature of that field has a negative effect on the corresponding predicted output, the input weight corresponding to that field is decreased. When the actual detected value is lower than the predicted value, the corresponding input weight is adjusted in the opposite direction. Through the above relationships, the direction of input weight adjustment is always directed towards reducing the difference between the actual detected state and the state prediction vector.

[0060] The adjustment amount of the input weight is determined by the over-limit ratio. Let the deviation of this field be e, and the preset deviation range be [L, U], where L < 0 < U; when e > U, take... When e < L, take Within the range, r = 0. The model pre-stores the threshold. >0 and two defined positive correction values , ,and < When 0 < r ≤ Time to take correction amount When r > Time to take correction amount No adjustment is made when r = 0. The corresponding correction amount is added or subtracted from the original input weight according to the direction determined by S313. If the adjustment result is lower than zero, it is taken as zero; the same deviation corresponds to only one correction amount.

[0061] When multiple fields simultaneously exceed their respective preset deviation ranges, each field determines its input weight adjustment direction according to its own deviation direction. Weight adjustments for different fields are not interchangeable; a re-prediction is performed only after all adjustments are complete. If data conflicts occur simultaneously in the same field or a unique actual detection value cannot be determined, that field is excluded from the current weight adjustment to prevent conflicting data from directly altering the model's input contribution.

[0062] S314, re-execute the prediction calculations corresponding to S212 to S214 using the adjusted input weights. During the recalculation, the current data sequence, field order, and prediction time remain unchanged, and only the updated corresponding input weights are used to reconstruct the weighted state features, state prediction vector, and scaling state prediction values.

[0063] The newly formed state prediction vector is then mapped back to the evaporator operating status data for the current time period. The newly formed scaling state prediction value and the state prediction vector together constitute the corrected scaling state prediction result, which is then passed to S4. This forms a feedback loop from model prediction, actual detection, deviation calculation, weight correction to re-prediction.

[0064] S4. Based on the corrected scaling state prediction results and evaporator operating status data, the control conditions are judged. When the preset control conditions are met, a control command is generated and sent to the evaporator actuator to adjust the evaporator operation process.

[0065] S411: Obtain the predicted scaling state value q from the corrected scaling state prediction result, and simultaneously read the evaporator operating status data for the current time period. The value range of q is [0, 1], where 0 corresponds to the no-scaling calibration state after cleaning, and 1 corresponds to the preset cleaning judgment state. Control trigger threshold. Before operation, set and save the parameters according to the same calibration scale, and 0 < ≤1; when q≥ When the control trigger condition is met, when q < The control triggering conditions are not met at the time.

[0066] S412: When the preset control trigger condition is not met, no control command is generated to increase the material circulation flow rate of the evaporator. The current operating status continues to be collected by the detection equipment and enters the subsequent prediction cycle. When the preset control trigger condition is met, a control command to increase the material circulation flow rate of the evaporator is generated based on the current predicted scaling status.

[0067] The objective of the control command is to change the corresponding operating process of the evaporator actuator. During the control execution phase, a one-to-one transmission relationship is maintained between the control command and the evaporator actuator, preventing the machine learning model from bypassing control condition judgments and directly driving the evaporator actuator. The control command is sent to the variable frequency drive of the material circulation pump. Upon triggering, the current circulation flow rate F is read, and the flow rate setpoint is determined using the formula F target = min(F + ΔF, F upper limit), where ΔF is a pre-set positive flow rate increment, and F upper limit is the maximum allowable circulation flow rate of the equipment. The variable frequency drive adjusts the circulation pump speed according to the flow rate setpoint, maintaining the setpoint when the current flow rate has reached the upper limit.

[0068] S413 sends a control command to the evaporator actuator. Upon receiving the control command, the evaporator actuator performs the corresponding operational adjustments. After execution, the detection equipment retrieves the evaporator operating status data again and writes the adjusted operating status data into the real-time status data record for subsequent time periods.

[0069] S414 records the operational status data after the control command is executed, along with the corresponding execution time. At the start of subsequent prediction cycles, this operational status data enters a new data sequence according to the time processing rules in S1, and then undergoes state change calculation, machine learning model prediction, and feedback correction in sequence, ensuring that the result of one control execution can become the input source for the next prediction cycle.

[0070] Through the above processing, the model calculation chain specifically forms a continuous closed loop of "adjacent state records form change amplitude and change direction - different operating data form correlation change relationship - input weight adjusts the contribution of each state feature - prediction parameters form state prediction vector and scale state prediction value - actual detection state forms field deviation - out-of-range fields change corresponding input weight - re-prediction - control trigger judgment - execution state re-enters the data sequence".

[0071] Example 2 Please see Figure 1 and Figure 3 This embodiment provides another method for predictive control of evaporator scaling based on multi-source operating data. The difference between this embodiment and Embodiment 1 is that the feedback correction occurs in the prediction parameters of the machine learning model, rather than in the input weights of the state feature data. Embodiment 1 completes the correction by changing the contribution of a certain state feature when it enters the model; this embodiment keeps the input weights unchanged and corrects the prediction parameters that generate the corresponding prediction output based on the state deviation between the actual operating state and the state prediction vector, so that the prediction parameters change the mapping degree between the state feature and the prediction output in the next calculation.

[0072] S1 continues to collect and organize real-time status data and historical operation record data according to the running time, and forms operating temperature change data, pressure change data, flow rate change data, and concentration change data through adjacent status records. The data integrity detection, time sorting, change amplitude, change direction, and the formation process of same-direction and opposite-direction change relationships are kept the same as in Example 1, and finally form status feature data arranged continuously in time.

[0073] S2 employs the same multivariate regression machine learning model structure as Example 1. The features of each state type are first weighted according to the current input weights, and then the prediction parameters are used to form state prediction vectors and scaling state prediction values, respectively. The state prediction vectors are still arranged in the order of operating temperature, pressure, flow rate, and concentration. The evaporator scaling state prediction result still consists of the state prediction vector, the scaling state prediction value, and the corresponding prediction time.

[0074] S3 obtains the evaporator operating status data for the current time period, calculates the deviations of the operating temperature field, pressure field, flow field, and concentration field according to the same field order, and determines whether the deviation of each field exceeds the corresponding preset deviation range.

[0075] When all field deviations are within their respective preset deviation ranges, the current prediction parameters are maintained. When at least one field deviation exceeds its corresponding preset deviation range, the out-of-range field is identified, and the prediction parameters used to generate the predicted value for that field are determined. For the operating temperature field deviation, the corresponding prediction parameters used to generate the operating temperature predicted value are adjusted; for the pressure, flow rate, and concentration field deviations, the corresponding prediction parameters used to generate the predicted values ​​for each field are adjusted respectively.

[0076] When adjusting the prediction parameters, first determine the direction of the deviation between the actual detected value and the predicted value. When the actual detected value is higher than the predicted value, the corresponding predicted value for the next input state with the same or similar conditions needs to be corrected in the direction of increasing; when the actual detected value is lower than the predicted value, the corresponding predicted value for the next input state needs to be corrected in the direction of decreasing.

[0077] For a state feature involved in the calculation of the prediction field, the feature value of the state feature after processing with the input weights and the original direction of action of the prediction parameter are read. If it is necessary to increase the prediction value, the prediction parameters that have a positive effect on the prediction output are strengthened, and the prediction parameters that have a negative effect are weakened; if it is necessary to decrease the prediction value, the positively acting parameters are weakened, and the negatively acting parameters are strengthened. The parameter correction amount is determined based on the degree to which the deviation of the corresponding field exceeds the preset deviation range and the preset parameter correction step size.

[0078] Parameter correction is performed sequentially for fields exceeding the deviation range. When multiple fields simultaneously exceed the deviation range, the prediction parameter set corresponding to each field is first determined, and then updates are completed based on the deviation of each field. When a prediction parameter participates in the calculation of multiple prediction fields and forms conflicting adjustment directions, the excess ratio is calculated for each conflicting field: when the deviation e is greater than the upper boundary U, it is taken as... When the deviation e is less than the lower boundary L, take Where L < 0 < U. The adjustment direction and correction amount determined by the field with the largest r are selected to update the parameter; when r are equal, the first field is selected in the fixed field order of operating temperature, pressure, flow rate, and concentration. The remaining non-conflicting prediction parameters are adjusted independently according to their respective field deviations.

[0079] After updating the prediction parameters, keeping the current data sequence and input weights unchanged, the formation of the scaling state change correlation results, the calculation of the state prediction vector, and the calculation of the scaling state prediction value are re-executed to generate new evaporator scaling state prediction results. The new prediction results are then used as the corrected scaling state prediction results in S4.

[0080] S4 processes the corrected scaling state prediction result according to the same control condition judgment relationship as in Example 1. When the scaling state prediction value reaches the preset control trigger condition, a control command is generated and sent to the evaporator actuator. The operating status after execution is collected again by the detection device and enters the next prediction cycle.

[0081] The complete technical chain formed in this embodiment is as follows: continuously running data forms state feature data, the input weights remain fixed, prediction parameters are subjected to multivariate regression processing on the weighted state features to form a state prediction vector and a scale state prediction value; the actual running state and the state prediction vector form a field deviation; fields exceeding the preset deviation range determine the corresponding prediction parameters and parameter correction directions; the updated prediction parameters are reapplied to the current state features to obtain the corrected scale state prediction result; after the correction result reaches the control trigger condition, control execution is driven. Compared with Embodiment 1, this embodiment changes the mapping parameters between the internal state features of the model and the prediction output, rather than changing the contribution weights of each state feature at the model input.

[0082] Example 3 Please see Figure 3 , Figure 4 and Figure 5 This embodiment provides an evaporator scaling prediction and control system based on multi-source operating data. The difference between this embodiment and the aforementioned method embodiments lies in that it employs a data acquisition unit, a data processing unit, a predictive analysis unit, a feedback correction unit, and a control execution unit to carry out the aforementioned data processing, machine learning prediction, deviation calculation, feedback correction, and control command generation processes. The input-output connections between these units are established using a unified data object.

[0083] The data acquisition unit receives real-time status data obtained by the detection equipment during evaporator operation, and simultaneously retrieves historical operation records from a historical operation database that receives and stores the operation data output by the detection equipment according to the acquisition time. The data acquisition unit outputs real-time status data and historical operation records to the data processing unit. Upon entering the current feedback cycle, the evaporator operation status data obtained by the detection equipment also serves as the actual status input to the feedback correction unit.

[0084] The data processing unit receives real-time status data and historical operation record data, sorts and correlates them according to the acquisition time, and performs data integrity checks. The data processing unit continues to read adjacent status records, generating data on operating temperature changes, pressure changes, flow rate changes, and concentration changes, determining the magnitude and direction of change for each type of data, and establishing unidirectional, inverse, or constant relationships among different types of operating data within the same time period. The processed continuous status characteristic data is sent as part of a data sequence to the predictive analysis unit.

[0085] A machine learning model is set up within the predictive analysis unit. The machine learning model adopts the aforementioned multivariate regression prediction structure, and its model data includes at least the input weights corresponding to each state feature, the prediction parameters corresponding to the state prediction vector, and the prediction parameters corresponding to the scale state prediction value. After receiving the state feature data arranged in a fixed field order, the predictive analysis unit first uses the input weights to form a weighted input for each state feature, and then determines the direction of action of each feature in the prediction calculation based on the direction of state change and the correlation between different types of operational data.

[0086] Subsequently, the predictive analysis unit generates predicted values ​​for operating temperature, pressure, flow rate, and concentration, respectively, and assembles them into a state prediction vector according to a fixed field order. Simultaneously, based on the weighted characteristics of operating temperature, pressure, flow rate, and concentration changes, as well as the correlation results of scaling state changes, a scaling state prediction value is generated. The state prediction vector, the scaling state prediction value, and the corresponding prediction time are collectively sent to the feedback correction unit as the evaporator scaling state prediction result.

[0087] The feedback correction unit receives the predicted scaling status of the evaporator and simultaneously obtains the evaporator operating status data output by the detection equipment for the current time period. The feedback correction unit first matches the actual operating status records according to the predicted time, and then compares the actual detected values ​​with the corresponding predicted values ​​in the status prediction vector item by item according to the fixed field order of operating temperature, pressure, flow rate, and concentration, forming the field deviation.

[0088] The feedback correction unit has a deviation comparison processing relationship, whereby the deviation of each field is compared with its corresponding preset deviation range. When all fields are within their respective deviation ranges, the current prediction result is output as the corrected scaling state prediction result; when any field exceeds the deviation range, the system proceeds to either input weight correction or prediction parameter correction according to the correction path adopted by the system.

[0089] When using the correction path corresponding to Example 1, the feedback correction unit determines the input weight adjustment direction based on the out-of-range field, the field deviation direction, the state change direction, and the direction of action of the corresponding prediction parameter, and updates the input weight by calling the weight correction amount corresponding to the field deviation interval. The updated input weight is returned to the prediction analysis unit, which recalculates the state prediction vector and the scale state prediction value.

[0090] When using the correction path corresponding to Example 2, the feedback correction unit keeps the input weights unchanged, determines the prediction parameters for the predicted field based on the out-of-range field, determines the correction direction of the prediction parameters based on the relationship between the actual detected value and the predicted value, and corrects the prediction parameters based on the degree to which the deviation of the corresponding field exceeds the preset range and the parameter correction step size. The corrected prediction parameters are returned to the prediction analysis unit to re-execute the prediction calculation.

[0091] One of the determined feedback correction paths is used for the same actual operating cycle. After the feedback correction unit completes the correction, it forms the corrected scale state prediction result by combining the final state prediction vector, the scale state prediction value, and the prediction time, and sends it to the control execution unit.

[0092] The control execution unit receives the corrected scaling state prediction result and simultaneously obtains the current evaporator operating status data. The control execution unit reads the scaling state prediction value and compares it with preset control trigger conditions. When the control trigger conditions are met, a control command to increase the evaporator material circulation flow rate is generated and sent to the evaporator actuator; if the control trigger conditions are not met, the current operation is maintained and the unit waits for the next prediction cycle.

[0093] After the evaporator actuator responds to the control command, the detection equipment continues to collect operational status data after the control execution. This operational status data is returned to the data acquisition unit to form new real-time status data, which is then correlated with existing historical operational records in the next processing cycle. The system thus forms a closed loop: "Data acquisition unit outputs operational data—data processing unit forms status characteristics—predictive analysis unit completes multivariate regression prediction—feedback correction unit calculates field deviations and updates the model—control execution unit generates control commands—operational status data re-enters the data acquisition unit after execution."

[0094] When there are records in the input data whose time order cannot be determined, the data processing unit will not send the record to the predictive analysis unit; when conflicting state records occur at the same acquisition time, the conflicting record will not be used for the current predictive calculation; when the current evaporator operating status data cannot establish a correspondence with the prediction time of the evaporator scaling state prediction result, the feedback correction unit will not use the record to change the input weights or prediction parameters, but will wait to form an operating status record that corresponds to the prediction time; when no new operating status data has been obtained after control execution, the unreturned operating status will not be used to form the next feedback deviation.

[0095] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting and controlling evaporator scaling based on multi-source operating data, characterized in that, Includes the following steps: Real-time status data of the evaporator during operation is collected by a detection device set at the evaporator's operating location. Historical operation records are retrieved from a historical operation database that receives and stores the operation data output by the detection device according to the acquisition time. The real-time status data and the historical operation records are correlated and organized according to the operation time to form a data sequence for scaling status analysis. The data sequence is input into a multivariate regression machine learning model that uses feature weighting, linear combination, and bias calculation to generate evaporator scaling state prediction results, including state prediction vectors and scaling state prediction values ​​for operating temperature, pressure, flow rate, and concentration. The evaporator operating status data within the current time period is collected by the detection device, and the corresponding fields of the status prediction vector in the same time period in the evaporator scaling status prediction result are subtracted one by one. When the deviation of at least one field exceeds its preset deviation range, the prediction parameters of the machine learning model or the input data processing method are adjusted according to the deviation to generate a corrected scaling status prediction result. Based on the corrected scaling state prediction results and the evaporator operating state data, control conditions are determined. When the preset control conditions are met, a control command is generated and sent to the evaporator actuator to adjust the evaporator operation process.

2. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of collecting real-time status data of the evaporator during operation by setting up detection devices at the operating position of the evaporator also includes: setting up multiple detection devices according to the operating status corresponding to different detection positions inside the evaporator, collecting operating status data of the corresponding positions through the multiple detection devices, and sorting and associating the real-time status data according to the data collection time corresponding to each detection device to form a status data record with time correspondence.

3. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of retrieving historical operation record data from the historical operation database that receives and stores the operation data output by the detection device according to the acquisition time also includes: extracting historical operation record data within the corresponding time period from the historical operation database according to a preset time range, and time-marking the historical operation record data, and time-correlating the data records corresponding to different operation cycles with the real-time status data to form a data sequence for input to the machine learning model.

4. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of associating and organizing real-time status data and historical operation record data to form a data sequence for scaling status analysis also includes: performing data integrity checks on the real-time status data and the historical operation record data, filtering out status records that do not meet preset data conditions, and combining the remaining status records according to the corresponding operation process to form a continuous data sequence.

5. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of inputting the data sequence formed by real-time status data and historical operation record data into the machine learning model also includes: arranging the status features of each time period in the order of operating temperature, pressure, flow rate, and concentration; dividing the signed difference of adjacent records in the same field by the pre-fixed positive reference change amount of that field to obtain the normalized change value, using its absolute value to represent the change amplitude, and using 1, -1, and 0 to encode the rising, falling, and holding directions respectively; multiplying the four change amplitudes, four directions, and six directions of the codes arranged in the order of temperature-pressure, temperature-flow rate, temperature-concentration, pressure-flow rate, pressure-concentration, and flow rate-concentration to form a fourteen-dimensional feature vector; and then inputting each feature vector into a multivariate regression model after arranging them row by row in chronological order.

6. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of analyzing the state change characteristics in the data sequence using a machine learning model also includes: based on the evaporator operating temperature change data, pressure change data, flow rate change data, and concentration change data in the data sequence, dividing the difference between adjacent records of each field by the pre-fixed positive reference change amount of that field, determining the normalized change amplitude and change trend within the corresponding time period, and performing correlation calculations on the change relationships between different types of operating data to generate scaling state change correlation results for machine learning model analysis.

7. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of generating evaporator scaling state prediction results by analyzing the state change characteristics in the data sequence through a machine learning model also includes: inputting the data change relationships corresponding to the operating temperature change data, pressure change data, flow rate change data, and concentration change data into the machine learning model, determining the prediction output corresponding to the scaling state change based on the combination of data changes under different operating conditions, and converting the prediction output into evaporator scaling state prediction results.

8. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of comparing the evaporator operating status data with the evaporator scaling status prediction results also includes: matching the state prediction vectors in the evaporator operating status data and the evaporator scaling status prediction results according to the prediction time; subtracting the corresponding field prediction values ​​from the measured values ​​of operating temperature, pressure, flow rate, and concentration to form the state deviation of each field; and judging based on the state deviation of each field and the preset deviation range of that field. When the state deviation of at least one field exceeds its corresponding preset deviation range, the input weights of the corresponding operating status data in the machine learning model are adjusted.

9. The evaporator scaling prediction and control method based on multi-source operating data according to claim 1, characterized in that, The process of determining control conditions based on the corrected scaling state prediction results and evaporator operating status data also includes: matching the predicted scaling state corresponding to the corrected scaling state prediction results with the current operating status corresponding to the evaporator operating status data; when the predicted relative scaling degree value with a range of [0, 1] reaches the control trigger threshold preset according to the same calibration scale, a control command for increasing the material circulation flow rate of the evaporator is generated, and the control result is recorded based on the operating status data after the control command is executed.

10. An evaporator scaling prediction and control system based on multi-source operating data, characterized in that, The system includes a data acquisition unit, a data processing unit, a predictive analysis unit, a feedback correction unit, and a control execution unit; The data acquisition unit is used to collect real-time status data during the operation of the evaporator, and to retrieve historical operation record data from the historical operation database that receives and stores the operation data output by the detection device according to the acquisition time. The data processing unit is used to organize the real-time status data and the historical operation record data to form a data sequence for scaling status analysis; The predictive analysis unit is used to input the data sequence into a machine learning model to analyze the changes in the operating status of the evaporator and generate a prediction result of the evaporator scaling status. The feedback correction unit is used to acquire evaporator operating status data. According to the same prediction time, it subtracts the measured values ​​of operating temperature, pressure, flow rate and concentration in the evaporator operating status data from the corresponding field prediction values ​​of the state prediction vector in the evaporator scaling state prediction result item by item, compares the deviation of each field with the preset deviation range of the field, and adjusts the prediction process according to the comparison results. The control execution unit is used to generate control commands based on the corrected scaling state prediction results, and send the control commands to the evaporator actuator.

Images