Raw material conveying control system for candy production

Abstract

The invention relates to the field of raw material conveying control, in particular to a raw material conveying control system for candy production, which comprises a data acquisition module, a prediction model establishment module, a prediction module, a control execution module, an image color analysis module, a feedback adjustment module and the like, and is characterized in that the prediction model establishment module is used for cleaning and normalizing acquired data, training an LSTM (Long Short Term Memory) model and constructing a conveying prediction model; the prediction module outputs optimal transmission parameters according to the real-time data; the control execution module accurately controls the conveying equipment according to the predicted value; the image color analysis module collects a raw material image and judges whether the color of the raw material is abnormal or not through multi-step processing; the feedback adjustment module processes the abnormal raw materials, records related information and stores the related information in a database; raw material conveying can be accurately controlled, the product quality is guaranteed, the functions of abnormal raw material processing and data tracing are achieved, and the problems existing in traditional raw material conveying control are effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of raw material transportation control, and specifically to a raw material transportation control system for candy production. Background Art

[0002] With the booming development of the candy market, consumers' demands for the quality, variety, and personalized customization of candies are increasing day by day. This puts higher requirements on the raw material transportation link in the candy production process. However, the current raw material transportation technology for candy production has many shortcomings and is difficult to meet the needs of the industry's development;

[0003] The color of raw materials is one of the important indicators for judging their quality. Abnormal color often implies quality problems with the raw materials, such as deterioration and contamination. In the traditional raw material transportation process, there is a lack of means for real-time monitoring and analysis of the color of raw materials, and it is impossible to detect and handle them in a timely manner when the color of the raw materials is abnormal; this may cause raw materials with quality problems to enter the subsequent production links, affecting the quality and safety of candies and damaging the market reputation of the enterprise; in the process of candy production, a large amount of historical data and real-time data related to raw material transportation have been accumulated. However, the traditional transportation control system has not fully tapped the value of these data. Due to the lack of an effective data analysis and learning mechanism, it is impossible to establish an accurate transportation prediction model, difficult to conduct intelligent control over the raw material transportation process, and unable to achieve efficient and intelligent production management.

[0004] To solve the above problems, a technical solution is proposed now. Summary of the Invention

[0005] To solve the technical problems raised in the above background art, the present invention provides a raw material transportation control system for candy production.

[0006] The purpose of the present invention can be achieved through the following technical solutions: The present invention is a raw material transportation control system for candy production, including a data acquisition module, a prediction model establishment module, a prediction module, a control execution module, an image color analysis module, a feedback adjustment module, and a database.

[0007] The data acquisition module is used to collect the viscosity and density characteristic data of raw materials in real time, as well as the environmental temperature and air pressure condition data in the production workshop, and connect to the order management program port of the enterprise to obtain the data of production orders. The specific process is as follows:

[0008] Install viscometers and densitometers in each raw material storage tank to collect the viscosity value and density value of the raw materials in the raw material storage tank in real time; install various environmental temperature sensors in different areas of the production workshop, and different areas include the raw material storage area and the conveying pipeline, to collect the environmental temperature value of the raw materials in real time; install an air pressure sensor on the ceiling of the production workshop to collect the air pressure value in the workshop in real time;

[0009] Install a conveying parameter sensor on the conveying pipeline and conveying equipment to monitor the conveying speed of raw materials in real time;

[0010] Connect the port of the data acquisition module to the port of the enterprise's order management program, set the data acquisition frequency to once per hour, and obtain the data of production orders. The data includes product name, product specification, batch number, and production quantity.

[0011] The prediction model establishment module cleans and normalizes the collected historical data and real-time data, and trains the model with the processed data. The specific steps are as follows:

[0012] The prediction model establishment model is provided with a preprocessing unit and a model training unit;

[0013] The preprocessing unit cleans the missing values and error values in the real-time data during the acquisition process. Specifically: detect the missing data in the real-time data, fill it with the median, then sequentially obtain the outliers in the real-time data. The acquisition method is Z-score, and the obtained outliers are deleted to obtain the processed raw material data, environmental data, and order data;

[0014] Convert the processed raw material data, environmental data, and order data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Specifically: arrange the processed data into a data set, calculate the mean value of the original data points in the data set to obtain the original data mean, then subtract the mean value of each original data point from the value of the original data point to obtain the original data difference, and calculate the standard deviation of the data set. Divide the original data difference by the standard deviation of the data set to obtain the standard normal distribution data, and construct the normalization process in this way;

[0015] Extract the time series features from the normalized data. The time series features include seasonal features and trend features;

[0016] Use the time series decomposition method to decompose the raw material data, environmental data, and order data into a trend term, a seasonal term, and a residual term, and use these three terms as seasonal feature points;

[0017] Use a linear regression model to fit the time series data. Specifically: let the time series be Yt, the time index be t, establish a linear regression model Yt = β0 + β1t + εt and obtain the trend coefficient β1, where β0 represents the intercept and ε represents the error term;

[0018] The model training unit selects the long short-term memory network LSTM as the deep learning model, including an input layer, an LSTM layer, and an output layer. Specifically:

[0019] First, divide the dataset into a training set and a temporary set with a ratio of 7:3. Then, divide the temporary set into a validation set and a test set with a ratio of 2:1. Send the seasonal feature points and trend coefficients as input vectors to the input layer. The LSTM selects Tanh as the activation function to map the input vectors to the interval [-1, 1]. The output layer selects the mean squared error as the loss function for training, thus constructing a conveying prediction model.

[0020] According to the real-time input raw material characteristics, production order requirements, and environmental conditions, the prediction module uses the trained conveying prediction model to output the optimal conveying parameter values. The specific steps are as follows:

[0021] The prediction module receives the viscosity value, density value, environmental temperature value, air pressure value, and production demand. It loads the viscosity value, density value, environmental temperature value, air pressure value, and production demand as input feature quantities into the conveying prediction model. The conveying value is calculated as the feature vector value of the output layer. Through continuous iterative calculations of forward propagation until the optimal conveying value is obtained at the output layer, the optimal conveying value is sent to the control execution module;

[0022] According to the optimal conveying value predicted by the model, the control execution module controls the operation of the conveying equipment to achieve precise control of the raw material conveying process. The specific steps are as follows:

[0023] Receive the optimal conveying value sent by the prediction module, extract the preset standard conveying value in the database, and take the difference between the optimal conveying value and the standard conveying value to obtain the conveying compensation value. If the conveying compensation value is positive, generate a control enhancement instruction and send it to the conveying pump control terminal, valve adjustment terminal, and fresh air device. The conveying pump control terminal uses the standard conveying value as the initial conveying capacity and increases it according to the preset rotational speed value of the motor, increasing once per second. Then, continuously collect the current conveying speed through the data acquisition module, and adjust the preset rotational speed value of the motor until the conveying compensation value reaches 0. The valve adjustment terminal uses the current pressure value as the initial pressure point and extracts the standard pressure value in the database to reduce the initial pressure point until it reaches the standard pressure value. Similarly, the fresh air device uses the current temperature value as the initial temperature, extracts the standard temperature in the data, and cools the pipeline until it reaches the standard temperature;

[0024] If the conveying compensation value is negative, it is determined that the current speed is too high. Generate a control reduction instruction and send it to the conveying pump control terminal, valve adjustment terminal, and fresh air device. Use the standard conveying value as the initial conveying capacity and reduce the conveying speed according to the preset rotational speed value, and so on until the conveying compensation value reaches 0; The valve adjustment terminal uses the current pressure value as the initial pressure point and enhances the pressure until it reaches the standard pressure value,

[0025] The fresh air device uses the current temperature value as the initial temperature and heats the pipeline until it reaches the standard temperature.

[0026] The image color analysis module collects images of the conveying process of each raw material, identifies different raw materials, obtains corresponding candidate color templates, and compares them with the color images of the real-time raw materials. The specific steps are as follows:

[0027] Install high-definition cameras in the areas of each conveying pipeline, number the areas of each conveying pipeline, and collect the initial raw material images of each numbered pipeline in real time. First, convert each initial raw material image into a grayscale image, obtain the corresponding raw material gradient amplitude and direction of each grayscale image through an edge detection algorithm, use the position of the local maximum value of the gradient amplitude as the boundary to obtain the raw material contour information of each initial raw material image, and extract each preset raw material contour image in the database for coincidence matching in turn. Mark the preset raw material contour images with a matching similarity exceeding the first similarity threshold as raw materials to be detected; obtain the images of the raw materials to be detected, compare the grayscale values of the central pixel and the neighboring pixels of the initial raw material image through the local binary pattern, binarize and encode the grayscale values of the neighboring pixels according to their size relationship with the central pixel to obtain a binary code, count the pixels in the initial raw material image and generate an initial histogram. Similarly, perform the above steps on the images of the raw materials to be detected to obtain a histogram to be detected, and calculate the sum of the ratios of the squared differences of the corresponding intervals of the initial histogram to the values of the corresponding intervals of the histogram to be detected to obtain the chi-square distance;

[0028] Extract the candidate color templates of the raw material images to be detected in the database, calculate the sum of the squares of the differences between the candidate color templates and the color values of the corresponding pixels in each local area of the initial raw material image. Assume that the size of template T is m×n, and the size of the initial raw material image L is M×N. In the image L, the local area L with the position (x,y) as the upper left corner x,y The squared difference matching value SSD(x,y) with template T, which is expressed by the formula: where i and j are index variables used to traverse each element in template T;

[0029] And so on, slide template T on the image L, calculate the value SSD(x,y) for each position (x,y), obtain the squared difference matching values of the image, and perform an average calculation to obtain the average matching value. Extract the preset matching degree threshold in the data center. If the average matching value is greater than the preset matching degree threshold, it indicates that the color is abnormal. Mark the raw material as an abnormal raw material, mark the number of the location, generate an adjustment instruction and send it to the feedback adjustment module. Otherwise, it indicates that the color of the raw material conforms to the standard color.

[0030] The feedback adjustment module processes the marked raw materials according to the received adjustment instructions. The specific process is as follows:

[0031] Upon receiving an adjustment instruction, obtain the location number of the corresponding abnormal raw material, then obtain the control valve terminal with the obtained number, generate a blocking signal and send it to the control valve. The control valve blocks the connection between the conveying pipeline where the abnormal raw material is located and the subsequent production link, guides the abnormal raw material to a dedicated isolation area or temporary storage container, records the source of the abnormal raw material, the conveying pipeline number, the discovery time, and the type of abnormality. The type of abnormality includes information such as too dark, too light, or the appearance of variegated colors, etc., and sends it to the database for storage, which is convenient for subsequent traceability and analysis.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows: The image color analysis module collects raw material images by installing high-definition cameras in the conveying pipeline area, identifies the raw materials through operations such as grayscale conversion, edge detection, and contour matching, and judges whether the color of the raw materials is abnormal by calculating with local binary pattern and sum of squared differences matching. Once an abnormality is found, it is marked in time and the adjustment instruction is sent to the feedback adjustment module. This function can effectively prevent raw materials with abnormal colors from entering the subsequent production link, ensuring the color and quality of the candy products and enhancing the market competitiveness of the products.

[0033] The control execution module compares the optimal conveying value of the prediction module with the preset standard conveying value in the database, and precisely controls the operation of the conveying equipment by calculating the conveying compensation value. When the conveying compensation value is positive, the conveying pump, valve, and fresh air device work together to increase the conveying capacity, adjust the pressure and temperature; when it is negative, it adjusts in the opposite direction. This process realizes the precise control of the conveying speed, pressure, and temperature of the raw materials, ensures the raw materials are conveyed under suitable conditions, improves the accuracy of raw material conveying, and guarantees the stable quality of candy production. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. The following drawings are not deliberately drawn to scale in actual size, and the focus is on showing the gist of the present invention.

[0035] Figure 1 It is a schematic diagram of the module connection of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts also belong to the scope of protection of the present invention.

[0037] Please refer to Figure 1As shown in the figure, the present invention is a raw material transportation control system for candy production, including a data acquisition module, a prediction model establishment module, a prediction module, a control execution module, an image color analysis module, a feedback adjustment module, and a database.

[0038] The data acquisition module is used to collect the viscosity and density characteristic data of the raw materials in real time, as well as the environmental temperature and air pressure condition data in the production workshop, and connect to the order management program port of the enterprise to obtain the data of production orders. The specific process is as follows:

[0039] Install viscometers and densitometers in each raw material storage tank to collect the viscosity values and density values of the raw materials in the raw material storage tank in real time; install environmental temperature sensors in different areas of the production workshop, and different areas include the raw material storage area and the conveying pipeline, to collect the environmental temperature values of the raw materials in real time; install a barometric pressure sensor on the ceiling of the production workshop to collect the barometric pressure value in the workshop in real time;

[0040] Install conveying parameter sensors on the conveying pipeline and conveying equipment to monitor the conveying speed of the raw materials in real time;

[0041] Connect the port of the data acquisition module to the port of the enterprise's order management program, set the data acquisition frequency to once per hour, and obtain the data of production orders. The data includes product name, product specification, batch number, and production quantity.

[0042] The prediction model establishment module cleans and normalizes the collected historical data and real-time data, and trains the model with the processed data. The specific steps are as follows:

[0043] The prediction model establishment model is provided with a preprocessing unit and a model training unit;

[0044] The preprocessing unit cleans the missing values and error values in the real-time data during the collection process. Specifically: detect the missing data in the real-time data, fill it with the median, and then sequentially obtain the outliers in the real-time data. The acquisition method is Z-score, and the obtained outliers are deleted to obtain the processed raw material data, environmental data, and order data;

[0045] Convert the processed raw material data, environmental data, and order data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Specifically: arrange the processed data into a data set, calculate the mean of the values of each original data point in the data set to obtain the original data mean, then subtract the value of each original data point from the original data mean to obtain the original data difference, and calculate the standard deviation of the data set. Divide the original data difference by the standard deviation of the data set to obtain the standard normal distribution data, and thus construct the normalization process;

[0046] Extract time series features from the normalized data. The time series features include seasonal features and trend features. It should be noted that seasonal features refer to the regular changes presented during the alternation of seasons in raw material production and transportation. For example, the low temperature in winter or high temperature in summer lead to fluctuations in the viscosity and density of raw materials. The seasonal fluctuations in order volume and product type form a periodic correlation with raw material consumption. Trend features reflect the continuous upward, downward, or stable change trend of raw material transportation over a long time span;

[0047] Use the time series decomposition method to decompose the raw material data, environmental data, and order data into trend terms, seasonal terms, and residual terms, and use these three terms as seasonal feature points;

[0048] Use a linear regression model to fit the time series data. Specifically: Let the time series be Yt, the time index be t, establish a linear regression model Yt = β0 + β1t + εt and obtain the trend coefficient β1, where β0 represents the intercept and ε represents the error term;

[0049] The model training unit selects the long short-term memory network (LSTM) as the deep learning model, including an input layer, an LSTM layer, and an output layer. Specifically:

[0050] First, divide the dataset into a training set and a temporary set with a ratio of 7:3. Then, divide the temporary set into a validation set and a test set with a ratio of 2:1. Send the seasonal feature points and trend coefficients as input vectors to the input layer. LSTM selects Tanh as the activation function to map the input vectors to the interval [-1, 1]. The output layer selects the mean squared error as the loss function for training, thus constructing a transportation prediction model.

[0051] The prediction module, based on the real-time input raw material characteristics, production order requirements, and environmental conditions, uses the trained transportation prediction model to output the optimal transportation parameter values. The specific steps are as follows:

[0052] The prediction module receives the viscosity value, density value, environmental temperature value, air pressure value, and production demand. Load the viscosity value, density value, environmental temperature value, air pressure value, and production demand as input feature quantities into the transportation prediction model. Calculate the transportation value as the feature vector value of the output layer. Through continuous iterative calculations of forward propagation until the output layer obtains the optimal transportation value, and send the optimal transportation value to the control execution module;

[0053] The control execution module, based on the optimal transportation value predicted by the model, controls the operation of the transportation equipment to achieve precise control of the raw material transportation process. The specific steps are as follows:

[0054] Receive the optimal delivery value sent by the prediction module, extract the preset standard delivery value in the database, subtract the optimal delivery value from the standard delivery value to obtain the delivery compensation value. If the delivery compensation value is positive, generate a control enhancement instruction and send it to the delivery pump control terminal, the valve adjustment terminal, and the fresh air device. The delivery pump control terminal uses the standard delivery value as the initial delivery capacity and increases it according to the preset rotational speed value of the motor, increasing once per second. Then, continuously collect the current transmission speed through the data acquisition module, and adjust the preset rotational speed value of the motor until the delivery compensation value reaches 0. The valve adjustment terminal uses the current pressure value as the initial pressure point, extracts the standard pressure value in the database, and reduces the initial pressure point until it reaches the standard pressure value. Similarly, the fresh air device uses the current temperature value as the initial temperature, extracts the standard temperature in the data, and cools the pipeline until it reaches the standard temperature;

[0055] If the delivery compensation value is negative, it is determined that the current speed is too high. Generate a control reduction instruction and send it to the delivery pump control terminal, the valve adjustment terminal, and the fresh air device. Use the standard delivery value as the initial delivery capacity and reduce the delivery speed according to the preset rotational speed value, and so on until the delivery compensation value reaches 0; The valve adjustment terminal uses the current pressure value as the initial pressure point and enhances the pressure until it reaches the standard pressure value,

[0056] The fresh air device uses the current temperature value as the initial temperature and heats the pipeline until it reaches the standard temperature.

[0057] The image color analysis module collects images of the delivery process of each raw material, identifies different raw materials, obtains the corresponding candidate color templates, and compares them with the color images of the real-time raw materials. The specific steps are as follows:

[0058] Install high-definition cameras in each conveying pipeline area, number each conveying pipeline area, and collect the initial raw material images of each numbered pipeline in real time. First, convert each initial raw material image into a grayscale image, and obtain the corresponding raw material gradient amplitude and direction of each grayscale image through an edge detection algorithm. Use the position of the local maximum value of the gradient amplitude as the boundary to obtain the raw material contour information of each initial raw material image, and extract each preset raw material contour image in the database for coincidence matching in turn. Mark the preset raw material contour images with a matching similarity exceeding the first similarity threshold as raw materials to be detected; obtain the images of the raw materials to be detected, compare the grayscale values of the central pixel and the neighborhood pixels of the initial raw material image through the local binary pattern, binarize the grayscale values of the neighborhood pixels according to their size relationship with the central pixel to obtain binary codes, count the pixels in the initial raw material image and generate an initial histogram. Similarly, perform the above steps on the image of the raw material to be detected to obtain a histogram to be detected, and calculate the sum of the ratios of the squared differences of the corresponding intervals of the initial histogram to the values of the corresponding intervals of the histogram to be detected to obtain the chi-square distance; it should be noted that the chi-square distance is used to measure the difference between two histograms. The smaller the chi-square distance, the more similar the two histograms are, indicating that the image textures are more similar; extract the second chi-square distance threshold. If the chi-square distance is less than the second chi-square distance threshold, it is determined that the raw material in the image of the raw material to be detected matches the corresponding initial raw material image;

[0059] Extract the candidate color templates of the raw material images to be detected in the database. It should be noted that the candidate color templates are the images selected from a large number of collected images with uniform raw material colors and typical features as candidate templates; calculate the sum of the squares of the differences between the candidate color templates and the color values of the corresponding pixels in each local area of the initial raw material image. Let the size of template T be m×n, and the size of the initial raw material image L be M×N. m, n, M, and N respectively represent rows and columns. Among them, in the image L, the local area L x,y The squared difference matching value SSD(x,y) with template T is expressed by the formula: where i and j are index variables used to traverse each element in template T;

[0060] And so on, slide template T on the initial raw material image L, calculate the value of SSD(x,y) for each position (x,y), obtain the squared difference matching values of the image, and perform an average calculation to obtain the average matching value. Extract the preset matching degree threshold in the data center. If the average matching value is greater than the preset matching degree threshold, it indicates that the color is abnormal. Mark the raw material as an abnormal raw material, mark the number of the position where it is located, generate an adjustment instruction and send it to the feedback adjustment module. Otherwise, it indicates that the raw material color conforms to the standard color.

[0061] The feedback adjustment module processes the marked raw materials according to the received adjustment instructions. The specific process is as follows:

[0062] Upon receiving the adjustment instruction, obtain the location number of the corresponding abnormal raw material, then obtain the control valve terminal of the location number, generate a blocking signal and send it to the control valve. The control valve blocks the connection between the conveying pipeline where the abnormal raw material is located and the subsequent production process, guides the abnormal raw material to a dedicated isolation area or a temporary storage container, records the source of the abnormal raw material, the conveying pipeline number, the discovery time and the type of abnormality. The type of abnormality includes information such as too dark, too light or variegated colors, etc., and sends it to the database for storage to facilitate subsequent traceability and analysis.

[0063] The above is the description of the present invention and should not be regarded as a limitation thereof. Although several exemplary embodiments of the present invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is the description of the present invention and should not be considered limited to the specific embodiments disclosed, and modifications to the disclosed embodiments as well as other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. A raw material conveying control system for candy production, comprising an image color analysis module, a feedback adjustment module, and a database, characterized in that: The image color analysis module collects images of the raw material conveying process, identifies different raw materials, obtains corresponding candidate color templates, and compares them with the color images of the real-time raw materials; converts each initial raw material image into a grayscale image, obtains the corresponding raw material gradient amplitude and direction of each grayscale image through an edge detection algorithm, uses the position of the local maximum of the gradient amplitude as the boundary to obtain the raw material contour information of each initial raw material image, performs coincidence matching in sequence, and marks the preset raw material contour images with a matching similarity exceeding the first similarity threshold as raw materials to be detected; obtains the images of the raw materials to be detected, compares the gray values of the central pixels and neighborhood pixels of the initial raw material images through the local binary pattern, binary-encodes the gray values of the neighborhood pixels according to their size relationship with the central pixels to obtain binary codes, counts the pixels in the initial raw material images and generates initial histograms, processes the images of the raw materials to be detected to obtain the histograms to be detected, and calculates the sum of the ratios of the squared differences of the corresponding intervals of the initial histograms to the values of the corresponding intervals of the histograms to be detected to obtain the chi-square distance; if the chi-square distance is less than the second chi-square distance threshold, it is determined that the raw materials in the images of the raw materials to be detected match the corresponding initial raw material images. Calculate the sum of the squares of the differences between the candidate color template and the color values of the corresponding pixels in each local area of the initial raw material image, and obtain the squared difference matching value between the local area with the position as the upper left corner and the template in the initial raw material image; slide the template on the initial raw material image, calculate for each position, obtain the squared difference matching values of the image, and perform mean calculation to obtain the average matching value. If the average matching value is greater than the preset matching degree threshold, mark the raw material as an abnormal raw material and generate an adjustment instruction to send to the feedback adjustment module.

2. The raw material conveying control system for candy production according to claim 1, wherein, After receiving the adjustment instruction, the feedback adjustment module processes the marked raw materials. The specific process is as follows: Obtain the location number where the corresponding abnormal raw material is located, then obtain the control valve terminal of the location number, generate a blocking signal to send to the control valve, the control valve blocks the connection between the conveying pipeline where the abnormal raw material is located and the subsequent production link, guides the abnormal raw material to a dedicated isolation area or a temporary storage container, records the source, conveying pipeline number, discovery time, and abnormal type of the abnormal raw material. The abnormal type includes information data such as too dark color, too light color, and the appearance of miscellaneous colors, and sends it to the database for storage.

3. The raw material conveying control system for candy production according to claim 1, wherein It also includes a prediction model establishment module. The prediction model establishment module cleans and normalizes the collected historical data and real-time data, and trains the model with the processed data. The specific steps are as follows: The prediction model establishment model is provided with a preprocessing unit and a model training unit; The preprocessing unit cleans the missing values and error values in the real-time data during the collection process. Specifically: detect the missing data in the real-time data, fill it with the median, then sequentially obtain the abnormal values in the real-time data. The obtaining method is Z-score, delete the obtained abnormal values to obtain the processed raw material data, environmental data, and order data. Convert the processed raw material data, environmental data, and order data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Specifically: Arrange the processed data into a data set, calculate the mean of the original data points in the data set to obtain the original data mean, then subtract the original data mean from the value of each original data point to obtain the original data difference, and calculate the standard deviation of the data set. Divide the original data difference by the standard deviation of the data set to obtain the standard normal distribution data, and thus construct the normalization process; Extract time series features from the normalized data. The time series features include seasonal features and trend features; Use the time series decomposition method to decompose the raw material data, environmental data, and order data into a trend term, a seasonal term, and a residual term, and use these three terms as seasonal feature points; Use a linear regression model to fit the time series data. Specifically: Let the time series be Yt and the time index be t. Establish a linear regression model Yt = β0 + β1t + εt and obtain the trend coefficient β1, where β0 represents the intercept and ε represents the error term.

4. A raw material conveying control system for candy production according to claim 3, characterized in that, The model training unit selects the long short-term memory network (LSTM) as the deep learning model, including an input layer, an LSTM layer, and an output layer. Specifically: First, divide the data set into a training set and a temporary set with a ratio of 7:

3. Then divide the temporary set into a validation set and a test set with a ratio of 2:

1. Send the seasonal feature points and the trend coefficient as input vectors to the input layer. The LSTM selects Tanh as the activation function to map the input vector to the interval [-1, 1]. The output layer selects the mean squared error as the loss function for training, and thus construct the conveying prediction model.

5. A raw material conveying control system for candy production according to claim 4, characterized in that, It also includes a prediction module, a control execution module, and a data acquisition module. The prediction module uses the trained conveying prediction model to output the optimal conveying parameter value according to the real-time input raw material characteristics, production order requirements, and environmental conditions. The specific steps are as follows: The prediction module receives the viscosity value, density value, environmental temperature value, air pressure value, and production demand. Load the viscosity value, density value, environmental temperature value, air pressure value, and production demand as input feature quantities into the conveying prediction model. Calculate the conveying value as the feature vector value of the output layer. Through continuous iterative calculation of forward propagation until the output layer obtains the optimal conveying value, and send the optimal conveying value to the control execution module.

6. The raw material conveying control system for candy production according to claim 5, characterized in that, The control execution module controls the operation of the conveying equipment according to the optimal conveying value predicted by the model to achieve precise control of the raw material conveying process. The specific steps are as follows: Receive the optimal delivery value sent by the prediction module, extract the preset standard delivery value in the database, subtract the optimal delivery value from the standard delivery value to obtain the delivery compensation value. If the delivery compensation value is positive, generate a control enhancement instruction and send it to the delivery pump control terminal, the valve adjustment terminal, and the fresh air device. The delivery pump control terminal uses the standard delivery value as the initial delivery capacity, increases it according to the preset rotational speed value of the motor once per second, and then continuously collects the current transmission speed through the data acquisition module. By adjusting the preset rotational speed value of the motor until the delivery compensation value reaches 0, the valve adjustment terminal uses the current pressure value as the initial pressure point, extracts the standard pressure value in the database, and reduces the initial pressure point until it reaches the standard pressure value. Similarly, the fresh air device uses the current temperature value as the initial temperature, extracts the standard temperature in the data, and cools the pipeline until it reaches the standard temperature; If the delivery compensation value is negative, it is determined that the current speed is too high. Generate a control reduction instruction and send it to the delivery pump control terminal, the valve adjustment terminal, and the fresh air device. Use the standard delivery value as the initial delivery capacity and reduce the delivery speed according to the preset rotational speed value, and so on until the delivery compensation value reaches 0; the valve adjustment terminal uses the current pressure value as the initial pressure point and enhances the pressure until it reaches the standard pressure value. The fresh air device uses the current temperature value as the initial temperature and heats the pipeline until it reaches the standard temperature.

7. The raw material conveying control system for candy production according to claim 6, characterized in that, The data acquisition module is used to collect the viscosity and density characteristic data of the raw materials in real time, as well as the environmental temperature and air pressure condition data in the production workshop, and connect to the order management program port of the enterprise to obtain the data of production orders. The specific process is as follows: Install viscometers and densitometers in each raw material storage tank to collect the viscosity value and density value of the raw materials in the raw material storage tank in real time; install environmental temperature sensors in different areas of the production workshop, where different areas include the raw material storage area and the conveying pipeline, to collect the environmental temperature value of the raw materials in real time; install air pressure sensors on the ceiling of the production workshop to collect the air pressure value in the workshop in real time; Install delivery parameter sensors on the conveying pipeline and conveying equipment to monitor the delivery speed of the raw materials in real time; Connect the port of the data acquisition module to the order management program port of the enterprise, set the data acquisition frequency to once per hour, and obtain the data of production orders. The data includes product name, product specification, batch number, and production quantity.

Images