Deep learning based tac membrane pre-drying intelligent temperature control system

Abstract

The application discloses a TAC film pre-drying intelligent temperature control system based on deep learning, and relates to the technical field of TAC film pre-drying temperature control, and comprises the following steps: collecting the space temperature of multiple positions above the surface of a TAC film to obtain environmental temperature data; collecting the film surface temperature and solvent evaporation rate of the TAC film to obtain film surface change data; respectively performing feature cleaning treatment to obtain space temperature evaporation data and film surface temperature data; performing correlation analysis on the space temperature evaporation data and the film surface temperature data, and constructing a film surface temperature prediction model based on a space-time graph neural network; and predicting the film surface temperature change by using the film surface temperature prediction model, and controlling the heating device; the application is used to solve the problem that the existing TAC film pre-drying temperature control technology cannot accurately predict the change of the film surface temperature and timely adjust the film surface temperature when regulating the film surface temperature of the TAC film according to the temperature change of the space where the TAC film is located and the change of the solvent evaporation rate.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of TAC film pre-drying temperature control, in particular to a TAC film pre-drying intelligent temperature control system based on deep learning. BACKGROUND

[0002] The TAC film pre-drying temperature control technology refers to a special technology for precisely regulating the drying space environment temperature, film surface temperature and related heat transfer process in the pre-drying process of TAC film production through the cooperation of temperature control equipment, real-time monitoring, intelligent feedback algorithm and process adaptation. The core goal is to realize uniform solvent evaporation and consistent film surface temperature under the premise of guaranteeing the thermal stability of TAC film, and to provide a physical performance standard film material for subsequent stretching, curing and other processes.

[0003] When the existing TAC film pre-drying temperature control technology regulates the film surface temperature of the TAC film, it usually collects the film surface temperature through a temperature sensor, and then automatically adjusts the heating power through a PID controller according to the deviation between the set temperature and the actual temperature, so as to realize temperature control. However, there is a physical lag in temperature change in TAC film pre-drying. For example, after adjusting the heating power, it takes a certain time for the heat to be transferred to the film surface, and it also takes a certain time for the film surface temperature monitoring to be fed back to the controller. Temperature monitoring and PID control are passive response logic that adjusts only after the temperature deviation occurs. Before the deviation occurs, no matter how other process parameters change, the PID will not actively adjust. This leads to fluctuations in the film surface temperature. Temperature fluctuations will directly change the solvent evaporation rate, causing local solvent residue to exceed the standard or excessive drying. In addition, sensor instantaneous failure and local process abnormalities may occur in TAC film pre-drying. Temperature monitoring and PID control lack the ability to identify and filter abnormal data, and are prone to misadjustment. Therefore, when the existing TAC film pre-drying temperature control technology regulates the film surface temperature of the TAC film, it cannot accurately predict the change in the film surface temperature according to the change in the temperature of the space where the TAC film is located and the change in the solvent evaporation rate, and adjust it in time. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the prior art, by collecting the space temperature of multiple positions above the TAC film surface to obtain environmental temperature data; and collecting the film surface temperature and solvent evaporation rate of the TAC film to obtain film surface change data; and respectively performing feature cleaning processing to obtain space temperature evaporation data and film surface temperature data; performing correlation analysis on the space temperature evaporation data and the film surface temperature data, and constructing a film surface temperature prediction model based on a spatio-temporal graph neural network; predicting the film surface temperature change using the film surface temperature prediction model during the TAC film pre-drying process, and controlling the heating device; to solve the problem that the existing TAC film pre-drying temperature control technology cannot accurately predict the change of the film surface temperature and timely adjust it when regulating the film surface temperature of the TAC film according to the temperature change of the space where the TAC film is located and the change of the solvent evaporation rate.

[0005] To achieve the above-mentioned purpose, the present application provides a TAC film pre-drying intelligent temperature control system based on deep learning, comprising a data collection module, a feature cleaning module, an intelligent model module and a temperature control module;

[0006] The data collection module is used to collect the space temperature of multiple positions above the TAC film surface to obtain environmental temperature data; and collect the film surface temperature and solvent evaporation rate of the TAC film to obtain film surface change data;

[0007] The feature cleaning module is used to perform feature cleaning processing on the environmental temperature data and the film surface change data respectively to obtain space temperature evaporation data and film surface temperature data;

[0008] The intelligent model module comprises an analysis unit and a construction unit, the analysis unit is used to perform correlation analysis on the space temperature evaporation data and the film surface temperature data, and the construction unit constructs a film surface temperature prediction model based on a spatio-temporal graph neural network;

[0009] The temperature control module is used to predict the film surface temperature change using the film surface temperature prediction model during the TAC film pre-drying process, and control the heating device.

[0010] Further, the data collection module is configured with a data collection strategy, and the data collection strategy comprises:

[0011] The plane where the TAC film is located is denoted as the film plane, the plane where the heating device above the TAC film is located is denoted as the heating plane, and the space region between the film plane and the heating plane is denoted as the heating transition region;

[0012] A plurality of planes with different heights and parallel to the film plane and the heating plane are uniformly arranged in the heating transition region, denoted as temperature collection planes, and a plurality of points are uniformly selected on each temperature collection plane, denoted as temperature collection points;

[0013] A spatial rectangular coordinate system is established in the heating transition region, denoted as the transition region coordinate system, and a temperature sensor is arranged at each temperature collection point to obtain the actual position coordinates of each temperature sensor in the transition region coordinate system, denoted as sensor position information;

[0014] A plurality of points are uniformly selected on the film plane, denoted as film temperature collection points, and the position coordinates of each film temperature collection point in the transition region coordinate system are obtained, denoted as collection point position information.

[0015] Further, the data collection strategy further includes:

[0016] For any one TAC film pre-drying process, denoted as the first drying process, the measured values of each temperature sensor are synchronously obtained at a first time interval from the beginning to the end of the first drying process, the collection time is recorded, and the corresponding temperature sensors are classified and sorted in time sequence, denoted as the spatial temperature sequence of the corresponding temperature sensor, wherein the first time interval is t1;

[0017] The corresponding film surface temperature of the TAC film at each film temperature collection point is synchronously obtained at the first time interval, the collection time is recorded, and the corresponding film temperature collection points are classified and sorted in time sequence, denoted as the film surface temperature sequence of the corresponding film temperature collection point;

[0018] The solvent evaporation rate of the TAC film is synchronously obtained at the first time interval, the collection time is recorded, and sorted in time sequence, denoted as the evaporation rate sequence;

[0019] All spatial temperature sequences are denoted as the environmental temperature data of the first drying process, and the film surface temperature sequence and the evaporation rate sequence are denoted as the film surface change data of the first drying process;

[0020] The environmental temperature data and the film surface change data of a plurality of TAC film pre-drying processes of the same kind are repeatedly collected.

[0021] Further, the feature cleaning module is configured with a feature cleaning strategy, and the feature cleaning strategy includes:

[0022] For the environmental temperature data and the film surface change data of the first drying process, any one spatial temperature sequence or film surface temperature sequence is denoted as a first temperature sequence; and any one temperature value in the first temperature sequence is denoted as T(i), wherein i represents the position serial number;

[0023] The temperature change rate BT(i) corresponding to T(i) is calculated, wherein BT(i)=(T(i+1)-T(i)) / t1; the temperature change rates corresponding to all temperature values in the first temperature sequence are repeatedly obtained and arranged in corresponding order, denoted as a first change rate sequence; and the average value of the first change rate sequence is calculated, denoted as BT0;

[0024] Starting from the first temperature change rate of the first change rate sequence, a single first temperature change rate is recorded as a first change segment, and the subsequent temperature change rates are retrieved in turn; if the absolute difference between the subsequent temperature change rate and the average value of the first change segment is not greater than BT0 or the number of temperature change rates in the first change segment is less than k1, the corresponding temperature change rate is merged into the first change segment, the average value of the first change segment is recalculated, and the retrieval is continued in turn;

[0025] If the absolute difference between the subsequent temperature change rate and the average value of the first change segment is greater than BT0 and the number of temperature change rates in the first change segment is not less than k1, the first change segment at this time is taken as a change rate segment, and starting from the first temperature change rate of the remaining part of the first change rate sequence, the detection of the segment is repeated; the first change rate sequence is divided into multiple change rate segments, and if the number of temperature change rates of the last change rate segment is less than k1, the last change rate segment is merged with the adjacent change rate segment, where k1 is a set number threshold;

[0026] A plurality of change rate segments are obtained, which are sequentially recorded as segment 1-segment n1 in order of position; n1 is the number of change rate segments;

[0027] The temperature change rate at the starting position of segment 2-segment n1 is obtained and recorded as a segment change rate; the middle position of the two temperature values corresponding to the segment change rate is taken as a division position; all division positions are obtained, and the first temperature sequence is divided into multiple subsequences according to the division positions, and any one subsequence is recorded as a first subsequence.

[0028] Further, the feature cleaning strategy further comprises:

[0029] Any one temperature value in the first subsequence is recorded as a first temperature value, the temperature values in the first subsequence are arranged in ascending order, and the corresponding 25th percentile AQ1 and 75th percentile AQ2 are obtained, and the corresponding quartile range QR is calculated, where QR=AQ2-AQ1;

[0030] The temperature values in the first subsequence that are not located in [AQ1-k2*QR, AQ2+k2*QR] are removed, and linear fitting is performed according to the remaining temperature values to obtain a corresponding fitting function; and the fitting temperature value corresponding to the position of the first temperature value is obtained according to the corresponding fitting function, and the first temperature value is replaced, where k2 is a set proportion coefficient;

[0031] The replacement of all temperature values in the first subsequence is repeated, and after completion, the corresponding cleaned subsequence is obtained; the cleaned subsequences of all subsequences of the first temperature sequence are repeatedly obtained, and are merged according to the corresponding positions to obtain a cleaned temperature sequence of the first temperature sequence;

[0032] Repeat to obtain the cleaning temperature sequence corresponding to all the space temperature sequences, denoted as the space cleaning sequence, and repeat to obtain the cleaning temperature sequence corresponding to the membrane surface temperature sequence, denoted as the membrane surface cleaning sequence.

[0033] Further, the feature cleaning strategy further comprises:

[0034] For any one solvent evaporation rate in the evaporation rate sequence of the first drying process, denoted as the first rate; set the window size as k3; take the first rate as the window center position, and obtain the median of the solvent evaporation rate in the window at this time, and replace the first rate;

[0035] Repeat the replacement for all solvent evaporation rates in the evaporation rate sequence, and after completion, obtain the first rate sequence; uniformly divide the first rate sequence into multiple subsequences, and denote any one subsequence as the first rate subsequence;

[0036] Set the smoothing window proportion of the LOWESS smoothing algorithm as k4, the polynomial order as k5, and the iteration number as k6, and process the first rate subsequence by using the LOWESS smoothing algorithm to obtain the rate cleaning subsequence corresponding to the first rate subsequence;

[0037] Repeat to obtain all the rate cleaning subsequences, and merge them to obtain the evaporation cleaning sequence corresponding to the evaporation rate sequence;

[0038] Denote all the space cleaning sequences and evaporation cleaning sequences of the first drying process as the space temperature evaporation data of the first drying process, and denote all the membrane surface cleaning sequences of the first drying process as the membrane surface temperature data of the first drying process;

[0039] Repeat to obtain the collected space temperature evaporation data and membrane surface temperature data of all the pre-drying processes.

[0040] Further, the analysis unit is configured with an analysis strategy, and the analysis strategy comprises:

[0041] For the space temperature evaporation data and the membrane surface temperature data of the first drying process; any one membrane surface cleaning sequence or space cleaning sequence; denoted as the first cleaning sequence, and the corresponding temperature sensor or membrane temperature collection point is denoted as the first collection point;

[0042] Calculate the Pearson correlation coefficient of the first cleaning sequence and the evaporation cleaning sequence, and take the absolute value; denoted as the correlation coefficient of the solvent evaporation rate and the first collection point in the first drying process;

[0043] Repeat to calculate the correlation coefficient of the solvent evaporation rate and the first collection point in all the collected pre-drying processes, and take the average value, denoted as the correlation coefficient of the solvent evaporation rate and the first collection point;

[0044] The correlation coefficient of the solvent evaporation rate and all temperature sensors or film temperature collection points is repeatedly obtained, and is recorded as first correlation data.

[0045] Further, the analysis strategy further includes:

[0046] All positions of the temperature sensors and all film temperature collection points are collectively recorded as temperature collection points; the Euclidean distance of each temperature collection point from the nearest temperature collection point is obtained, and is recorded as adjacent distance; and the minimum value and the standard deviation of all adjacent distances are obtained, and are recorded as AR and BR in sequence, respectively, and CR=AR-BR is recorded as distance correlation reference;

[0047] For the spatial temperature evaporation data and the film surface temperature data of the first drying process; any two film surface cleaning sequences or spatial cleaning sequences are recorded as second cleaning sequence and third cleaning sequence, respectively, and the corresponding temperature sensors or film temperature collection points are recorded as second collection point and third collection point, respectively;

[0048] The Euclidean distance of the second collection point and the third collection point is calculated, and is recorded as L0, and the first correlation coefficient G1 of the second collection point and the third collection point is calculated, wherein G1=CR / L0; the Pearson correlation coefficient of the second cleaning sequence and the third cleaning sequence is calculated, and the absolute value is taken; recorded as the second correlation coefficient G2 of the second collection point and the third collection point; the correlation coefficient G0 of the second collection point and the third collection point in the first drying process is calculated, wherein G0=q1*G1+q2*G2, q1 and q2 are set weights;

[0049] The correlation coefficient of the second collection point and the third collection point in all pre-drying processes is repeatedly calculated, and the average value is calculated, and is recorded as the correlation coefficient of the second collection point and the third collection point;

[0050] The correlation coefficient of any temperature sensor or film temperature collection point and any temperature sensor or film temperature collection point is repeatedly obtained, and second correlation data is obtained.

[0051] Further, the construction unit is configured with a construction strategy, and the construction strategy includes:

[0052] All temperature sensors, all film temperature collection points and solvent evaporation rates are sequentially recorded as node 1-node m; and an adjacency matrix of m*m size is constructed according to the first correlation data and the second correlation data, and is recorded as H=(h uv ), wherein h uv is the correlation coefficient corresponding to the node u and the node v, h uv =1 when u=v; u∈[1, m], v∈[1, m];

[0053] According to the collected space temperature volatilization data and film surface temperature data of all pre-drying processes, the data of all nodes at the same collection time are associated with the adjacency matrix to construct a space-time feature matrix at each collection time, and model training data are obtained;

[0054] The space-time graph neural network is denoted as an initial prediction model, the model output is set as the film surface temperature of all film temperature collection points, the initial prediction model is trained by using the model training data, and after the training, a film surface temperature prediction model is obtained.

[0055] Further, the temperature control module is configured with a temperature control strategy, and the temperature control strategy comprises:

[0056] When the corresponding TAC film of the same kind starts pre-drying, the corresponding environmental temperature data and film surface change data are collected at a first time interval, and feature cleaning processing is performed to obtain corresponding space temperature volatilization data and film surface temperature data; and the film surface temperature prediction model is periodically used to predict the film surface temperature change of each film surface collection point in the future;

[0057] When the film surface temperature of more than half of the film surface collection points shows a downward trend in the future, the heating power of the heating device is increased;

[0058] When the film surface temperature of more than half of the film surface collection points shows an upward trend in the future, the heating power of the heating device is reduced.

[0059] The present application has the following advantages: the present application collects the space temperature of multiple positions above the TAC film surface to obtain environmental temperature data; and collects the film surface temperature and solvent volatilization rate of the TAC film to obtain film surface change data; the environmental temperature data and the film surface change data are respectively subjected to feature cleaning processing to obtain space temperature volatilization data and film surface temperature data; the space temperature volatilization data and the film surface temperature data are subjected to correlation analysis, and a film surface temperature prediction model is constructed based on a space-time graph neural network; the film surface temperature prediction model is used to predict the film surface temperature change during the pre-drying process of the TAC film, and the heating device is controlled; when the film surface temperature of the TAC film is regulated during the pre-drying process, the change of the film surface temperature can be accurately predicted according to the change of the temperature of the space where the TAC film is located and the change of the solvent volatilization rate, and timely adjustment can be made;

[0060] The application can divide the temperature change of the pre-drying process into several physical stages by automatically segmenting the temperature sequence according to the temperature change rate, and can remove short-time burst interference and retain the temperature change trend by replacing the original temperature value with a linearly fitted temperature value, so that the model training is more stable, and the prediction error is reduced; extreme values are removed by quartile range, and the original value is replaced by local linear fitting, so as to avoid fitting deviation caused by single-point noise; by taking the minimum value and the standard deviation of the adjacent distance of all temperature sampling points as the distance reference, and then normalizing the distance influence value, the extreme weight can be avoided, and the value is more reasonable; the first correlation coefficient based on distance and the second correlation coefficient based on Pearson correlation coefficient are combined into G0 according to the weight, so that the correlation coefficient can reflect the degree of geometric proximity and the degree of statistical coupling; this can enable the model to directly learn the spatial propagation characteristics, and improve the accuracy of temperature prediction. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 It is a principle block diagram of the system of the application;

[0062] Figure 2 It is a step flow chart of the method of the application;

[0063] Figure 3 It is a cleaning temperature sequence acquisition flow chart of the application;

[0064] Figure 4 It is a structural schematic diagram of the electronic device of the application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0066] Embodiment 1, please refer to Figure 1 As shown in the figure, the application provides a TAC film pre-drying intelligent temperature control system based on deep learning, which includes a data collection module, a feature cleaning module, an intelligent model module and a temperature control module.

[0067] The data collection module is used to collect the spatial temperature of multiple positions in the space above the TAC film surface, to obtain environmental temperature data; and to collect the film surface temperature and the solvent evaporation rate of the TAC film, to obtain film surface change data.

[0068] The data collection module is configured with a data collection strategy, and the data collection strategy comprises: taking a plane where the TAC film is located as a film plane, taking a plane where the heating device above the TAC film is located as a heating plane, and taking a space region between the film plane and the heating plane as a heating transition region; in order to ensure the easy volatilization of the TAC film, the heating device is usually arranged directly above the TAC film;

[0069] A plurality of planes with different heights and parallel to the film plane and the heating plane are uniformly arranged in the heating transition region, and are denoted as temperature collection planes; the temperature collection planes are artificially set and have no physical entity; the planes with different heights can capture the temperature stratification or gradient in the vertical direction, because heat transfer is continuous and needs time, when the power of the heating device fluctuates or changes, the position closer to the heating device in the heating transition region changes first, and by obtaining the temperature change of the position and the relative position, the temperature change of the TAC film can be predicted;

[0070] A space rectangular coordinate system is established in the heating transition region, denoted as a transition region coordinate system, and a temperature sensor is arranged at each temperature collection point to obtain actual position coordinates of each temperature sensor in the transition region coordinate system, denoted as sensor position information; the Euclidean distance is calculated, because the temperature collection point and the temperature sensor at the point can have a slight deviation due to installation errors, and therefore the actual position coordinates of the temperature sensor are obtained for the accuracy of subsequent prediction;

[0071] A plurality of points are uniformly selected on the film plane, denoted as film temperature collection points, and position coordinates of each film temperature collection point in the transition region coordinate system are obtained, denoted as collection point position information.

[0072] For any TAC film pre-drying process, denoted as a first drying process; starting from the first drying process to the end, the measurement values of each temperature sensor are synchronously obtained at a first time interval, the collection time is recorded, and the corresponding temperature sensors are classified and sorted in time sequence, denoted as a spatial temperature sequence of the corresponding temperature sensors, wherein the first time interval is t1; in this embodiment, t1=0.2 seconds, which can be adjusted according to actual application scenarios;

[0073] The film surface temperature of the TAC film at each film temperature collection point is synchronously obtained at the first time interval, the collection time is recorded, and the corresponding film temperature collection points are classified and sorted in time sequence, denoted as a film surface temperature sequence of the corresponding film temperature collection points; the film surface temperature of each point is a key variable that needs to be predicted and controlled finally;

[0074] The solvent volatilization rate of the TAC film is synchronously obtained at the first time interval, the collection time is recorded, and the volatilization rate sequence is sorted in time sequence.

[0075] After the heating power is changed, it will directly affect the temperature of the surrounding air at the first time. When the power increases, the temperature of the surrounding space can rise within a few seconds. The hot gas flow is the carrier of heat transfer to the TAC film, and needs to complete its own temperature change first, and then transfer the heat to the film surface. The core driving factor of the solvent evaporation rate is the temperature difference between the environment and the film surface. After the heating power is changed, even if the film surface temperature has not changed significantly, the temperature of the hot gas flow will rise or fall first, which will directly change the environmental temperature gradient and the evaporation rate of the solvent vapor near the film surface. Therefore, the change of the solvent evaporation rate can also be used to predict the change of the film surface temperature in advance.

[0076] All space temperature sequences are recorded as environmental temperature data of the first drying process, and the film surface temperature sequence and the evaporation rate sequence are recorded as film surface change data of the first drying process.

[0077] Repeat the collection of environmental temperature data and film surface change data of the pre-drying process of multiple TAC films of the same kind.

[0078] In the specific implementation process, different kinds of TAC films will have different film surface temperature changes under the same environment due to different physical sizes and different solution ratios. Subsequently, a prediction model of the same kind of TAC film needs to be established. The data characteristics of different kinds of TAC films are different, which will interfere with the subsequent model training, so they cannot be used for subsequent processing. Only the data of the same kind of TAC film under the same pre-drying environment can be used for subsequent processing.

[0079] The feature cleaning module is configured to clean the environmental temperature data and the film surface change data respectively, and obtain the space temperature evaporation data and the film surface temperature data.

[0080] The feature cleaning module is configured with a feature cleaning strategy. The feature cleaning strategy includes: referring to Figure 3 As shown in the figure, for the environmental temperature data and the film surface change data of the first drying process, any one space temperature sequence or film surface temperature sequence is recorded as a first temperature sequence. Any one temperature value in the first temperature sequence is recorded as T(i), where i represents the position sequence number.

[0081] Calculate the temperature change rate BT(i) corresponding to T(i), where BT(i)=(T(i+1)-T(i)) / t1. Repeat the calculation of the temperature change rate corresponding to all temperature values in the first temperature sequence, and arrange them in corresponding order, recorded as the first change rate sequence. Calculate the average value of the first change rate sequence, recorded as BT0. BT0 is a judgment standard provided by dynamic segmentation, which avoids misjudging noise as trend inflection point or misjudging trend inflection point as noise.

[0082] Starting from the first temperature change rate of the first change rate sequence, a single first temperature change rate is recorded as a first change segment, and the subsequent temperature change rates are retrieved in turn; if the absolute difference between the subsequent temperature change rate and the average value of the first change segment is not greater than BT0 or the number of temperature change rates in the first change segment is less than k1, the corresponding temperature change rate is merged into the first change segment, the average value of the first change segment is recalculated, and the retrieval is continued in turn.

[0083] If the absolute difference between the subsequent temperature change rate and the average value of the first change segment is greater than BT0 and the number of temperature change rates in the first change segment is not less than k1, the first change segment at this time is taken as a change rate segment, and the detection of the segment is repeated starting from the first temperature change rate of the remaining part of the first change rate sequence; the first change rate sequence is divided into multiple change rate segments, and if the number of temperature change rates of the last change rate segment is less than k1, the last change rate segment is merged with the adjacent change rate segment, wherein k1 is a set number threshold; in this embodiment, k1=10, which ensures that each segment has sufficient data to support linear fitting;

[0084] The obtained multiple change rate segments are sequentially recorded as segment 1-segment n1 in order of position; n1 is the number of change rate segments;

[0085] The temperature change rate at the starting position of segment 2-segment n1 is obtained and recorded as a segment change rate; the middle position of the two temperature values corresponding to the segment change rate is taken as a division position; all division positions are obtained, and the first temperature sequence is divided into multiple subsequences according to the division positions, and any one subsequence is recorded as a first subsequence; the pre-drying process exhibits different temperature change rates at different stages; the temperature sequence is split into several subsequences according to the trend change through the temperature change rate, which not only retains the slow change trend of the temperature, but also enables the subsequent linear fitting of each subsequence to capture the characteristics of each segment, thereby improving the accuracy of the subsequent linear fitting and avoiding the problems of non-compliance with the trend or inaccurate fitting of the traditional fixed window linear fitting.

[0086] Any one temperature value in the first subsequence is recorded as a first temperature value, the temperature values in the first subsequence are arranged in ascending order, and the corresponding 25th percentile AQ1 and 75th percentile AQ2 are obtained, and the corresponding quartile range QR is calculated, wherein QR=AQ2-AQ1;

[0087] The temperature values in the first sub-sequence that are not located in [AQ1-k2*QR, AQ2+k2*QR] are removed, and a linear fitting is performed on the remaining temperature values to obtain a corresponding fitting function; and a fitting temperature value corresponding to the position of the first temperature value is obtained according to the corresponding fitting function, and the first temperature value is replaced, where k2 is a set proportion coefficient; in this embodiment, k2 = 1.5, which can be flexibly adjusted; the replacement is performed regardless of whether the first temperature value is removed or not; removing the temperature values that are not located in [AQ1-k2*QR, AQ2+k2*QR] can further improve the accuracy of linear fitting; replacing the original value by the fitting function can retain the slow trend in the segment and filter out high-frequency random noise.

[0088] The replacement of all temperature values in the first sub-sequence is repeated, and a corresponding cleaning sub-sequence is obtained after completion; the cleaning sub-sequences of all sub-sequences of the first temperature sequence are repeatedly obtained, and are combined according to the corresponding positions to obtain a cleaning temperature sequence of the first temperature sequence;

[0089] The cleaning temperature sequences corresponding to all spatial temperature sequences are repeatedly obtained and are recorded as spatial cleaning sequences, and the cleaning temperature sequences corresponding to the membrane surface temperature sequence are repeatedly obtained and are recorded as membrane surface cleaning sequences.

[0090] For any one solvent evaporation rate in the evaporation rate sequence of the first drying process, it is recorded as a first rate; the window size is set as k3; the first rate is taken as the center position of the window, and the median of the solvent evaporation rates in the window at this time is obtained and is used to replace the first rate; in this embodiment, k3 = 5, which can be flexibly adjusted; the pulse noise with large amplitude and short duration can be accurately removed without affecting the normal data and overall trend; the median can effectively skip the abnormal value, while the mean value can be affected by the abnormal value;

[0091] The replacement of all solvent evaporation rates in the evaporation rate sequence is repeated, and a first rate sequence is obtained after completion; the first rate sequence is evenly divided into multiple sub-sequences, and any one sub-sequence is recorded as a first rate sub-sequence;

[0092] The smoothing window proportion of the LOWESS smoothing algorithm is set as k4, the polynomial order is set as k5, and the iteration number is set as k6; the first rate sub-sequence is processed by using the LOWESS smoothing algorithm to obtain a rate cleaning sub-sequence corresponding to the first rate sub-sequence; in this embodiment, k4 = 0.2, k5 = 1, and k6 = 2, which can be flexibly adjusted according to actual application scenarios; the LOWESS smoothing algorithm can fit the nonlinear decay trend of the evaporation rate that is fast at first and slow at last, filter out medium and high frequency noise, and retain the small amplitude fluctuations of the evaporation rate caused by temperature changes.

[0093] The rate cleaning sub-sequences are repeatedly obtained and are combined to obtain an evaporation cleaning sequence corresponding to the evaporation rate sequence.

[0094] all the space cleaning sequences and volatile cleaning sequences of the first drying process are recorded as the space temperature volatile data of the first drying process, and all the film surface cleaning sequences of the first drying process are recorded as the film surface temperature data of the first drying process;

[0095] all the space temperature volatile data and film surface temperature data of the collected pre-drying processes are repeatedly obtained;

[0096] In the pre-drying process, the change characteristics of the temperature and the change characteristics of the solution volatile rate are different, so they are processed respectively. The essence of the temperature sequence is a smooth trend plus random noise. The temperature change should be continuous and smooth due to the time required for heat transfer, and there should be no sharp mutation. The film surface temperature lags behind the ambient temperature. After the ambient temperature changes, the film surface temperature needs a certain time to respond. The trends of the two are consistent, but there is a time difference. Random noise mainly comes from sensor electronic noise, local airflow turbulence, and small equipment vibration. The essence of the volatile rate sequence is slow decay plus random noise. At the beginning of pre-drying, the solvent residue is high, and the volatile rate is fast. As the solvent is lost, the film material gradually gels, and the volatile rate slowly decreases and finally stabilizes in a low rate interval. The volatile rate will rise slightly when the ambient temperature rises, but the trend is still mainly decay, and the overall trend is fast first and then slow. The noise mainly comes from the sensor;

[0097] The intelligent model module includes an analysis unit and a construction unit. The analysis unit is used for correlation analysis of the space temperature volatile data and the film surface temperature data. The construction unit constructs a film surface temperature prediction model based on a spatio-temporal graph neural network;

[0098] The analysis unit is configured with an analysis strategy. The analysis strategy includes: for the space temperature volatile data and the film surface temperature data of the first drying process; any one film surface cleaning sequence or space cleaning sequence is recorded as a first cleaning sequence, and the corresponding temperature sensor or film temperature collection point is recorded as a first collection point;

[0099] The Pearson correlation coefficient of the first cleaning sequence and the volatile cleaning sequence is calculated, and the absolute value is taken. The correlation coefficient of the solvent volatile rate and the first collection point in the first drying process is recorded. The Pearson correlation coefficient quantifies the correlation strength of the two in the time sequence. Because the solvent volatile rate and the first collection point have no distance concept, the Pearson correlation is directly used to quantify the correlation strength;

[0100] The correlation coefficient of the solvent volatile rate and the first collection point in all the collected pre-drying processes is repeatedly calculated, and the average value is taken. The correlation coefficient of the solvent volatile rate and the first collection point is recorded;

[0101] The correlation coefficient of the solvent evaporation rate and all temperature sensors or film temperature collection points is repeatedly obtained, and is recorded as first correlation data.

[0102] All positions of the temperature sensors and all film temperature collection points are collectively recorded as temperature collection points; the Euclidean distance between each temperature collection point and the nearest temperature collection point is obtained, and is recorded as adjacent distance; and the minimum value and the standard deviation of all adjacent distances are obtained, and are recorded as AR and BR in sequence, respectively, and CR=AR-BR is recorded as distance correlation reference; it is helpful to keep the first correlation coefficient in a reasonable numerical interval and avoid numerical instability;

[0103] For the spatial temperature evaporation data and the film surface temperature data of the first drying process; any two film surface cleaning sequences or spatial cleaning sequences are recorded as second cleaning sequences and third cleaning sequences, respectively, and the corresponding temperature sensors or film temperature collection points are recorded as second collection points and third collection points, respectively;

[0104] The Euclidean distance between the second collection points and the third collection points is calculated, and is recorded as L0, and the first correlation coefficient G1 of the second collection points and the third collection points is calculated, wherein G1=CR / L0; the Pearson correlation coefficient of the second cleaning sequence and the third cleaning sequence is calculated, and the absolute value is taken; it is recorded as the second correlation coefficient G2 of the second collection points and the third collection points; the correlation coefficient G0 of the second collection points and the third collection points in the first drying process is calculated, wherein G0=q1*G1+q2*G2, q1 and q2 are set weights; in this embodiment, q1=q2=0.5, the heat transfer is affected by the spatial distance, the first correlation coefficient quantifies the correlation in the physical distance, the closer the distance between two points, the more similar the temperature change of the two points, and the stronger the coupling;

[0105] The correlation coefficient of the second collection points and the third collection points in all pre-drying processes is repeatedly calculated, and the average value is recorded as the correlation coefficient of the second collection points and the third collection points;

[0106] The correlation coefficient of any temperature sensor or film temperature collection point and any temperature sensor or film temperature collection point is repeatedly obtained, and the second correlation data is obtained.

[0107] The construction unit is configured with a construction strategy, and the construction strategy comprises: sequentially recording all temperature sensors, all film temperature collection points and the solvent evaporation rate as node 1-node m; and constructing an adjacency matrix of m*m size according to the first correlation data and the second correlation data, and recording it as H=(h uv ), wherein h uv is the correlation coefficient corresponding to the node u and the node v, h uv =1 when u=v; u∈[1, m], v∈[1, m]; H directly determines how the temperature change propagates in space;

[0108] According to the collected spatial temperature volatilization data and film surface temperature data of all pre-drying processes, the data of all nodes at the same collection time is combined with the adjacency matrix to construct a space-time feature matrix at each collection time, and model training data is obtained;

[0109] The space-time graph neural network is taken as an initial prediction model, the model output is set as the film surface temperature of all film temperature collection points, the initial prediction model is trained by using the model training data, and a film surface temperature prediction model is obtained after the training is completed;

[0110] In the specific implementation process, at each collection time, the nodes are stacked into a vector or a matrix, and then input into the space-time graph neural network together with the adjacency matrix H. The space-time graph neural network learns the interaction between nodes and the time dynamics by performing spatial convolution on the graph structure, thereby predicting the change of the film surface temperature.

[0111] The temperature control module is used for predicting the change of the film surface temperature by using the film surface temperature prediction model during the pre-drying process of the TAC film, and controlling the heating device;

[0112] The temperature control module is configured with a temperature control strategy, and the temperature control strategy includes: when the corresponding same type of TAC film starts pre-drying, the corresponding environmental temperature data and film surface change data are collected at a first time interval, and feature cleaning processing is performed to obtain corresponding spatial temperature volatilization data and film surface temperature data; and periodically predicting the film surface temperature change of each film surface collection point in the future by using the film surface temperature prediction model; the cycle size can be set according to the actual application scenario;

[0113] When the film surface temperature of more than half of the film surface collection points shows a downward trend in the future, the heating power of the heating device is increased; the prediction model provides an estimate of the film surface temperature in the future for a period of time, so that the controller can predict the temperature change in advance, and can take measures to mitigate or prevent before the trend is formed, thereby avoiding temperature fluctuations caused by adjustment lag;

[0114] When the film surface temperature of more than half of the film surface collection points shows an upward trend in the future, the heating power of the heating device is reduced;

[0115] In the specific implementation process, the specific adjustment rules of the heating device can be set according to the actual application scenario.

[0116] Embodiment 2, please refer to Figure 2 As shown in the figure, the application provides a deep learning-based TAC film pre-drying intelligent temperature control method, which includes the following steps:

[0117] Step S1, collect the spatial temperature of multiple positions in the space above the TAC film surface to obtain environmental temperature data; and collect the film surface temperature and solvent volatilization rate of the TAC film to obtain film surface change data; step S1 includes the following substeps:

[0118] Step S101, the plane where the TAC film is located is recorded as the film plane, the plane where the heating device above the TAC film is located is recorded as the heating plane, and the space region between the film plane and the heating plane is recorded as the heating transition region;

[0119] Step S102, a plurality of planes with different heights and parallel to the film plane and the heating plane are uniformly arranged in the heating transition region, recorded as temperature collection planes, and a plurality of points are uniformly selected on each temperature collection plane, recorded as temperature collection points;

[0120] Step S103, a space rectangular coordinate system is established in the heating transition region, recorded as the transition region coordinate system, and a temperature sensor is arranged at each temperature collection point to obtain the actual position coordinates of each temperature sensor in the transition region coordinate system, recorded as sensor position information;

[0121] Step S104, a plurality of points are uniformly selected on the film plane, recorded as film temperature collection points, and the position coordinates of each film temperature collection point in the transition region coordinate system are obtained, recorded as collection point position information.

[0122] Step S105, for any TAC film pre-drying process, recorded as the first drying process; starting from the first drying process to the end, the measured values of each temperature sensor are synchronously obtained at a first time interval, the collection time is recorded, and the corresponding temperature sensors are classified and sorted in time sequence, recorded as the spatial temperature sequence of the corresponding temperature sensor, wherein the first time interval is t1;

[0123] Step S106, the corresponding film surface temperature of the TAC film at each film temperature collection point is synchronously obtained at the first time interval, the collection time is recorded, and the corresponding film temperature collection points are classified and sorted in time sequence, recorded as the film surface temperature sequence of the corresponding film temperature collection point;

[0124] Step S107, the solvent evaporation rate of the TAC film is synchronously obtained at the first time interval, the collection time is recorded, and sorted in time sequence, recorded as the evaporation rate sequence;

[0125] Step S108, all spatial temperature sequences are recorded as the environmental temperature data of the first drying process, and the film surface temperature sequence and the evaporation rate sequence are recorded as the film surface change data of the first drying process;

[0126] Step S109, the environmental temperature data and the film surface change data of a plurality of TAC film pre-drying processes of the same kind are repeatedly collected.

[0127] Step S2, the environmental temperature data and the film surface change data are respectively subjected to feature cleaning processing to obtain spatial temperature evaporation data and film surface temperature data; Step S2 includes the following sub-steps:

[0128] Step S201, for the ambient temperature data and the film surface variation data of the first drying process, record any one of the spatial temperature sequence or the film surface temperature sequence as the first temperature sequence; and record any one of the temperature values in the first temperature sequence as T(i), wherein i represents the position sequence number;

[0129] Step S202, calculate the temperature variation rate BT(i) corresponding to T(i), wherein BT(i)=(T(i+1)-T(i)) / t1; repeat to obtain the temperature variation rates corresponding to all the temperature values in the first temperature sequence, and arrange them in the corresponding order to record as the first variation rate sequence; calculate the average value of the first variation rate sequence, and record as BT0;

[0130] Step S203, from the first temperature variation rate of the first variation rate sequence, record the single first temperature variation rate as the first variation segment, and sequentially retrieve the temperature variation rates thereafter; if the absolute difference between the temperature variation rate thereafter and the average value of the first variation segment is not greater than BT0 or the number of the temperature variation rates in the first variation segment is less than k1, then combine the corresponding temperature variation rate to the first variation segment, and re-calculate the average value of the first variation segment, and continue to sequentially retrieve;

[0131] Step S204, if the absolute difference between the temperature variation rate thereafter and the average value of the first variation segment is greater than BT0 and the number of the temperature variation rates in the first variation segment is not less than k1, then take the first variation segment at this time as the variation rate segment, and from the first temperature variation rate of the remaining part of the first variation rate sequence, repeat to detect the segment; divide the first variation rate sequence into multiple variation rate segments, and if the number of the temperature variation rates of the last variation rate segment is less than k1, then combine it with the adjacent variation rate segment, wherein k1 is a set number threshold;

[0132] Step S205, obtain multiple variation rate segments, and sequentially record them as segment 1-segment n1 in the order of position; n1 is the number of the variation rate segments;

[0133] Step S206, obtain the temperature variation rate at the starting position of segment 2-segment n1, and record it as the segment variation rate; take the middle position of the two temperature values corresponding to the segment variation rate as the division position; obtain all the division positions, and divide the first temperature sequence into multiple subsequences according to the division positions, and record any one of the subsequences as the first subsequence.

[0134] Step S207, record any one of the temperature values in the first subsequence as the first temperature value, arrange the temperature values in the first subsequence in the order from small to large, and obtain the corresponding 25th percentile AQ1 and 75th percentile AQ2, and calculate the corresponding quartile range QR, wherein QR=AQ2-AQ1;

[0135] Step S208, the temperature values in the first sub-sequence not located in [AQ1-k2*QR, AQ2+k2*QR] are removed, and a linear fitting is performed according to the remaining temperature values to obtain a corresponding fitting function; and a fitting temperature value corresponding to the position of the first temperature value is obtained according to the corresponding fitting function, and the first temperature value is replaced by the fitting temperature value, wherein k2 is a set proportion coefficient;

[0136] Step S209, the replacement of all temperature values in the first sub-sequence is repeated, and after completion, a corresponding cleaning sub-sequence is obtained; the cleaning sub-sequences of all sub-sequences of the first temperature sequence are repeatedly obtained, and are combined according to the corresponding positions to obtain a cleaning temperature sequence of the first temperature sequence;

[0137] Step S210, the cleaning temperature sequences corresponding to all spatial temperature sequences are repeatedly obtained and recorded as spatial cleaning sequences, and the cleaning temperature sequences corresponding to the membrane surface temperature sequences are repeatedly obtained and recorded as membrane surface cleaning sequences.

[0138] Step S211, for any one solvent evaporation rate in the evaporation rate sequence of the first drying process, recorded as a first rate; the window size is set to k3; the first rate is taken as the window center position, and the median of the solvent evaporation rate in the window at this time is obtained and replaces the first rate;

[0139] Step S212, the replacement of all solvent evaporation rates in the evaporation rate sequence is repeated, and after completion, a first rate sequence is obtained; the first rate sequence is evenly divided into multiple sub-sequences, and any one sub-sequence is recorded as a first rate sub-sequence;

[0140] Step S213, the smoothing window proportion of the LOWESS smoothing algorithm is set to k4, the polynomial order is k5, and the iteration number is k6, and the first rate sub-sequence is processed by using the LOWESS smoothing algorithm to obtain a rate cleaning sub-sequence corresponding to the first rate sub-sequence;

[0141] Step S214, all rate cleaning sub-sequences are repeatedly obtained and combined to obtain an evaporation cleaning sequence corresponding to the evaporation rate sequence;

[0142] Step S215, all spatial cleaning sequences and evaporation cleaning sequences of the first drying process are recorded as spatial temperature evaporation data of the first drying process, and all membrane surface cleaning sequences of the first drying process are recorded as membrane surface temperature data of the first drying process;

[0143] Step S216, the spatial temperature evaporation data and the membrane surface temperature data of all pre-drying processes collected are repeatedly obtained.

[0144] Step S3, correlation analysis is performed on the spatial temperature volatilization data and the film surface temperature data, and a film surface temperature prediction model is constructed based on a space-time graph neural network; step S3 includes the following sub-steps:

[0145] Step S301, for the spatial temperature volatilization data and the film surface temperature data of the first drying process; any one film surface cleaning sequence or spatial cleaning sequence is recorded as a first cleaning sequence, and the corresponding temperature sensor or film temperature collection point is recorded as a first collection point;

[0146] Step S302, the Pearson correlation coefficient of the first cleaning sequence and the volatilization cleaning sequence is calculated, and the absolute value is taken; recorded as the correlation coefficient of the solvent volatilization rate and the first collection point in the first drying process;

[0147] Step S303, the correlation coefficient of the solvent volatilization rate and the first collection point in all collected pre-drying processes is repeatedly calculated, and the average value is taken, recorded as the correlation coefficient of the solvent volatilization rate and the first collection point;

[0148] Step S304, the correlation coefficient of the solvent volatilization rate and all temperature sensors or film temperature collection points is repeatedly obtained, recorded as first correlation data.

[0149] Step S305, all temperature sensors and all film temperature collection points are uniformly recorded as temperature collection points; for each temperature collection point, the Euclidean distance from the nearest temperature collection point is obtained, recorded as the adjacent distance; and the minimum value and the standard deviation of all adjacent distances are obtained, in order recorded as AR and BR, respectively, and CR=AR-BR, recorded as distance correlation reference;

[0150] Step S306, for the spatial temperature volatilization data and the film surface temperature data of the first drying process; any two film surface cleaning sequences or spatial cleaning sequences are recorded as a second cleaning sequence and a third cleaning sequence, respectively, and the corresponding temperature sensor or film temperature collection point is recorded as a second collection point and a third collection point, respectively;

[0151] Step S307, the Euclidean distance of the second collection point and the third collection point is calculated, recorded as L0, and the first correlation coefficient G1 of the second collection point and the third collection point is calculated, wherein G1=CR / L0; the Pearson correlation coefficient of the second cleaning sequence and the third cleaning sequence is calculated, and the absolute value is taken; recorded as the second correlation coefficient G2 of the second collection point and the third collection point; the correlation coefficient G0 of the second collection point and the third collection point in the first drying process is calculated, wherein G0=q1*G1+q2*G2, q1 and q2 are set weights;

[0152] Step S308, the correlation coefficient of the second collection point and the third collection point in all pre-drying processes is repeatedly calculated, and the average value is taken, recorded as the correlation coefficient of the second collection point and the third collection point;

[0153] Step S309, the correlation coefficient of any temperature sensor or film temperature collection point and any temperature sensor or film temperature collection point is repeatedly obtained, and second correlation data is obtained.

[0154] Step S310, all temperature sensors, all film temperature collection points and solvent evaporation rates are sequentially recorded as node 1-node m; and an m*m size adjacency matrix is constructed according to the first correlation data and the second correlation data, recorded as H=(h uv ), wherein h uv is the correlation coefficient corresponding to node u and node v, h uv =1 when u=v; u∈[1, m], v∈[1, m];

[0155] Step S311, and according to the collected spatial temperature evaporation data and film surface temperature data of all pre-drying processes, the data of all nodes at the same collection time is constructed with the adjacency matrix to construct a space-time feature matrix at each collection time, and model training data is obtained;

[0156] Step S312, the space-time graph neural network is recorded as an initial prediction model, the model output is set as the film surface temperature of all film temperature collection points, the initial prediction model is trained by using the model training data, and after completion, a film surface temperature prediction model is obtained.

[0157] Step S4, the film surface temperature prediction model is used to predict the film surface temperature change in the TAC film pre-drying process, and the heating device is controlled; step S4 includes the following sub-steps:

[0158] Step S401, when the corresponding same type of TAC film starts pre-drying, the corresponding environmental temperature data and film surface change data are collected at a first time interval, and feature cleaning processing is performed, and corresponding spatial temperature evaporation data and film surface temperature data are obtained; and the film surface temperature prediction model is periodically used to predict the film surface temperature change of each film surface collection point in the future;

[0159] Step S402, when the film surface temperature of more than half of the film surface collection points shows a downward trend in the future, the heating power of the heating device is increased;

[0160] Step S403, when the film surface temperature of more than half of the film surface collection points shows an upward trend in the future, the heating power of the heating device is reduced.

[0161] Example 3, please refer to Figure 4 , Figure 4An example is provided for a structural diagram of an electronic device, which can include a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete mutual communication through the communication bus. The memory stores computer readable instructions, and the processor can invoke the instructions in the memory, and when the computer readable instructions are executed by the processor, steps in the method for intelligent temperature control of TAC film pre-drying based on deep learning are run to achieve the following functions: collecting spatial temperatures at multiple positions in a space above a TAC film surface to obtain environmental temperature data; and collecting a film surface temperature of the TAC film and a solvent evaporation rate to obtain film surface change data; performing feature cleaning processing on the environmental temperature data and the film surface change data respectively to obtain spatial temperature evaporation data and film surface temperature data; performing correlation analysis on the spatial temperature evaporation data and the film surface temperature data, and constructing a film surface temperature prediction model based on a spatio-temporal graph neural network; and predicting film surface temperature changes using the film surface temperature prediction model during a TAC film pre-drying process, and controlling a heating device.

[0162] In addition, the logical instructions in the memory described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0163] In embodiment 4, the present application also provides a computer readable storage medium, and the present application provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to run the steps in the method for intelligent temperature control of TAC film pre-drying based on deep learning to achieve the following functions: collecting spatial temperatures at multiple positions in a space above a TAC film surface to obtain environmental temperature data; and collecting a film surface temperature of the TAC film and a solvent evaporation rate to obtain film surface change data; performing feature cleaning processing on the environmental temperature data and the film surface change data respectively to obtain spatial temperature evaporation data and film surface temperature data; performing correlation analysis on the spatial temperature evaporation data and the film surface temperature data, and constructing a film surface temperature prediction model based on a spatio-temporal graph neural network; and predicting film surface temperature changes using the film surface temperature prediction model during a TAC film pre-drying process, and controlling a heating device.

[0164] Through the description of the above embodiments, the embodiments of the present application can be provided as a method, a system or a computer program product. Based on such an understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0165] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are only illustrative, for example, the division of modules or units is only a logical function division, and in actual implementation, there can be another division manner, for example, a plurality of modules or units can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some communication interface, indirect coupling or communication connection between the systems, modules and units can be electrical, mechanical or other forms.

Claims

1. A deep learning-based intelligent temperature control system for TAC membrane pre-drying, characterized in that, It includes a data collection module, a feature cleaning module, an intelligent model module, and a temperature control module; The data collection module is used to collect the spatial temperature at multiple locations above the TAC membrane to obtain ambient temperature data; and to collect the membrane surface temperature and solvent evaporation rate of the TAC membrane to obtain membrane surface change data. The feature cleaning module is used to perform feature cleaning processing on ambient temperature data and membrane surface change data respectively to obtain space temperature volatilization data and membrane surface temperature data. The intelligent model module includes an analysis unit and a construction unit. The analysis unit is used to perform correlation analysis on spatial temperature volatilization data and membrane surface temperature data. The construction unit constructs a membrane surface temperature prediction model based on a spatiotemporal graph neural network. The temperature control module is used to predict the temperature change of the membrane surface during the TAC membrane pre-drying process using a membrane surface temperature prediction model, and to control the heating device. The build unit is configured with a build strategy, which includes: All temperature sensors, all membrane temperature acquisition points, and solvent evaporation rates are sequentially denoted as node 1 to node m; and an m*m adjacency matrix is ​​constructed based on the first and second correlation data, denoted as H = (h... uv ), where h uv Let h be the correlation coefficient between node u and node v, where h is the correlation coefficient when u = v. uv =1;u∈[1,m], v∈[1,m]; Based on the collected spatial temperature volatilization data and membrane surface temperature data of all pre-drying processes, the spatiotemporal feature matrix of each acquisition time is constructed by combining the data of all nodes at the same acquisition time with the adjacency matrix, thus obtaining the model training data; The spatiotemporal graph neural network is denoted as the initial prediction model. The model output is set to the membrane surface temperature of all membrane temperature acquisition points. The initial prediction model is trained using the model training data. After completion, the membrane surface temperature prediction model is obtained.

2. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 1, characterized in that, The data collection module is configured with data collection strategies, which include: The plane where the TAC membrane is located is called the membrane plane, the plane where the heating device above the TAC membrane is located is called the heating plane, and the space between the membrane plane and the heating plane is called the heating transition region. Multiple planes of different heights, parallel to the membrane plane and the heating plane, are uniformly set in the heating transition region and are denoted as temperature acquisition planes. Multiple points are uniformly selected on each temperature acquisition plane and are denoted as temperature acquisition points. A spatial rectangular coordinate system is established in the heating transition region, denoted as the transition region coordinate system. A temperature sensor is set at each temperature acquisition point to obtain the actual position coordinates of each temperature sensor in the transition region coordinate system, which is denoted as the sensor position information. Multiple points are evenly selected on the membrane plane and recorded as membrane temperature acquisition points. The position coordinates of each membrane temperature acquisition point in the transition region coordinate system are obtained, and the position information of the acquisition point is obtained.

3. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 2, characterized in that, Data collection strategies also include: For any TAC membrane pre-drying process, it is referred to as the first drying process. From the beginning to the end of the first drying process, the measured values ​​of each temperature sensor are acquired synchronously at the first time interval, and the acquisition time is recorded. The temperature sensor is classified according to its corresponding temperature sensor and sorted in chronological order, which is recorded as the spatial temperature sequence of the corresponding temperature sensor, where the first time interval is t1. The membrane surface temperature of the TAC membrane at each membrane temperature acquisition point is acquired synchronously at the first time interval. The acquisition time is recorded, and the membrane surface temperature sequence of the corresponding membrane temperature acquisition point is recorded according to the classification of the corresponding membrane temperature acquisition point and sorted in chronological order. The solvent evaporation rate of the TAC membrane was acquired synchronously at the first time interval, and the acquisition time was recorded and sorted in chronological order to form an evaporation rate sequence. All spatial temperature sequences are recorded as ambient temperature data for the first drying process, and the film surface temperature sequence and evaporation rate sequence are recorded as film surface change data for the first drying process. Ambient temperature data and membrane surface change data were repeatedly collected for the pre-drying process of multiple TAC membranes of the same type.

4. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 3, characterized in that, The feature cleaning module is configured with feature cleaning strategies, which include: For the ambient temperature data and membrane surface change data of the first drying process, any spatial temperature sequence or membrane surface temperature sequence is denoted as the first temperature sequence; and any temperature value in the first temperature sequence is denoted as T(i), where i represents the position number; Calculate the temperature change rate BT(i) corresponding to T(i), where BT(i) = (T(i+1) - T(i)) / t1; repeatedly obtain the temperature change rate corresponding to all temperature values ​​in the first temperature sequence, and arrange them in the corresponding order, denoted as the first change rate sequence; calculate the average value of the first change rate sequence, denoted as BT0; Starting from the first temperature change rate in the first rate of change sequence, the first individual temperature change rate is recorded as the first change segment, and the subsequent temperature change rates are retrieved sequentially. If the absolute difference between the subsequent temperature change rate and the average value of the first change segment is not greater than BT0 or the number of temperature change rates in the first change segment is less than k1, the corresponding temperature change rate is merged into the first change segment, the average value of the first change segment is recalculated, and the sequential retrieval continues. If the absolute difference between the subsequent temperature change rate and the average value of the first change segment is greater than BT0 and the number of temperature change rates in the first change segment is not less than k1, then the first change segment at this time is taken as a change rate segment, and the segment detection is repeated starting from the first temperature change rate of the remaining part of the first change rate sequence; the first change rate sequence is divided into multiple change rate segments, and if the number of temperature change rates in the last change rate segment is less than k1, then it is merged with the adjacent change rate segment, where k1 is the set number threshold; Multiple rate of change segments are obtained, and they are denoted as segment 1 to segment n1 in order of their positions; n1 is the number of rate of change segments. Obtain the temperature change rate at the starting position of segment 2 to segment n1, and denote it as the segment change rate; take the midpoint between the two temperature values ​​corresponding to the segment change rate as the division position; obtain all division positions, and divide the first temperature sequence into multiple subsequences according to the division positions, and denote any one of the subsequences as the first subsequence.

5. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 4, characterized in that, Feature cleaning strategies also include: Record any temperature value in the first subsequence as the first temperature value. Arrange the temperature values ​​in the first subsequence in ascending order and obtain the corresponding 25th percentile AQ1 and 75th percentile AQ2. Calculate the corresponding interquartile range QR, where QR = AQ2 - AQ1. Temperature values ​​in the first subsequence that are not located in [AQ1-k2*QR, AQ2+k2*QR] are removed, and linear fitting is performed on the remaining temperature values ​​to obtain the corresponding fitting function; the fitted temperature value corresponding to the position of the first temperature value is obtained according to the corresponding fitting function, and the first temperature value is replaced, where k2 is a set scaling factor; Repeatedly replace all temperature values ​​in the first subsequence to obtain the corresponding cleaned subsequence; repeatedly obtain the cleaned subsequences of all subsequences of the first temperature sequence and merge them according to their corresponding positions to obtain the cleaned temperature sequence of the first temperature sequence. Repeatedly obtain the cleaning temperature sequence corresponding to all space temperature sequences, and record it as the space cleaning sequence. Also, repeatedly obtain the cleaning temperature sequence corresponding to the membrane surface temperature sequence, and record it as the membrane surface cleaning sequence.

6. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 5, characterized in that, Feature cleaning strategies also include: For any solvent evaporation rate in the evaporation rate sequence of the first drying process, denote it as the first rate; set the window size to k3; take the first rate as the center position of the window, obtain the median of the solvent evaporation rate in the window at this time, and replace the first rate; Repeatedly replace all solvent evaporation rates in the evaporation rate sequence to obtain the first rate sequence; divide the first rate sequence evenly into multiple subsequences, and denote any one of the subsequences as the first rate subsequence; The smoothing window ratio of the LOWESS smoothing algorithm is set to k4, the polynomial order to k5, and the number of iterations to k6. The LOWESS smoothing algorithm is used to process the first rate subsequence to obtain the rate cleansing subsequence corresponding to the first rate subsequence. Repeatedly obtain all rate cleaning subsequences and merge them to obtain the evaporation cleaning sequence corresponding to the evaporation rate sequence; All spatial cleaning sequences and evaporation cleaning sequences of the first drying process are recorded as spatial temperature evaporation data of the first drying process, and all membrane surface cleaning sequences of the first drying process are recorded as membrane surface temperature data of the first drying process. Repeatedly acquire all collected space temperature evaporation data and membrane surface temperature data from the pre-drying process.

7. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 6, characterized in that, The analysis unit is configured with analysis strategies, which include: For the space temperature evaporation data and membrane surface temperature data of the first drying process; any membrane surface cleaning sequence or space cleaning sequence; is denoted as the first cleaning sequence, and the corresponding temperature sensor or membrane temperature acquisition point is denoted as the first acquisition point; Calculate the Pearson correlation coefficient between the first cleaning sequence and the evaporation cleaning sequence, and take the absolute value; denoted as the correlation coefficient between the solvent evaporation rate and the first sampling point in the first drying process; Repeatedly calculate the correlation coefficient between the solvent evaporation rate and the first sampling point during all pre-drying processes, and calculate the average value, which is recorded as the correlation coefficient between the solvent evaporation rate and the first sampling point; Repeatedly obtain the correlation coefficients between the solvent evaporation rate and all temperature sensors or membrane temperature acquisition points, and record them as the first correlation data.

8. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 7, characterized in that, The analysis strategy also includes: All temperature sensor locations and all membrane temperature acquisition points are collectively referred to as temperature acquisition points; for each temperature acquisition point, the Euclidean distance to the nearest temperature acquisition point is obtained and recorded as the proximity distance; and the minimum value and standard deviation of all proximity distances are obtained and recorded as AR and BR in order, respectively. Let CR = AR - BR, and record it as the distance-related benchmark. For the space temperature evaporation data and membrane surface temperature data of the first drying process; any two membrane surface cleaning sequences or space cleaning sequences; are respectively recorded as the second cleaning sequence and the third cleaning sequence, and the corresponding temperature sensors or membrane temperature acquisition points are respectively recorded as the second acquisition point and the third acquisition point; Calculate the Euclidean distance between the second and third sampling points, denoted as L0, and calculate the first correlation coefficient G1 between the second and third sampling points, where G1 = CR / L0; calculate the Pearson correlation coefficient between the second and third cleaning sequences, and take the absolute value; denoted as the second correlation coefficient G2 between the second and third sampling points; calculate the correlation coefficient G0 between the second and third sampling points in the first drying process, where G0 = q1*G1 + q2*G2, where q1 and q2 are the set weights; Repeatedly calculate the correlation coefficient between the second and third sampling points in all pre-drying processes, and calculate the average value, which is recorded as the correlation coefficient between the second and third sampling points; Repeatedly obtain the correlation coefficient between any temperature sensor or membrane temperature acquisition point and any other temperature sensor or membrane temperature acquisition point to obtain the second correlation data.

9. The intelligent temperature control system for TAC membrane pre-drying based on deep learning according to claim 8, characterized in that, The temperature control module is configured with a temperature control strategy, which includes: When the corresponding TAC membrane of the same type begins to pre-dry, the corresponding ambient temperature data and membrane surface change data are continuously collected at the first time interval, and feature cleaning processing is performed to obtain the corresponding space temperature volatilization data and membrane surface temperature data; and the membrane surface temperature prediction model is periodically used to predict the membrane surface temperature changes at each membrane surface collection point in the future. If the membrane surface temperature at more than half of the membrane surface collection points shows a decreasing trend in the future, the heating power of the heating device should be increased. If the membrane surface temperature at more than half of the membrane surface sampling points shows an upward trend in the future, the heating power of the heating device should be reduced.

Images