Feed production line intelligent regulation and control system and method based on Internet of Things

Abstract

The invention relates to the technical field of feed production intelligent control, and particularly discloses a feed production line intelligent regulation and control system and method based on the Internet of Things. Comprising a sensing acquisition module, a data processing module, a model construction module, an intelligent regulation and control module and a collaborative optimization module, according to the method, anomaly recognition is realized through DBSCAN clustering, and a production parameter anomaly recognition result is obtained; constructing a correlation mapping matrix between the production parameters and feed quality, production efficiency and energy consumption indexes according to an abnormal recognition result, and obtaining an optimal target production parameter set through multi-target optimization; generating a production parameter adjustment amount based on a deep reinforcement learning model; and finally, multi-device and multi-process cooperative regulation and control are completed through edge cloud collaboration. Real-time sensing and precise regulation and control of the whole feed production process are achieved, the technical problems that a traditional production line is poor in parameter collaboration, large in quality fluctuation, high in energy consumption and the like are solved, and the intelligent level of feed production is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for feed production, specifically to an intelligent control system and method for a feed production line based on the Internet of Things. Background Technology

[0002] Feed production is a crucial link in the development of animal husbandry, and its quality and efficiency directly affect the economic benefits and product safety of the industry. Traditional feed production lines rely on manual control or single-equipment automatic control, which presents the following technical challenges: First, incomplete production parameter collection; each stage of equipment operates independently, lacking comprehensive data collection and integration, making coordinated parameter control difficult. Second, delayed anomaly identification; fluctuations in production parameters cannot be detected and traced in a timely manner, easily leading to fluctuations in feed quality. Third, control methods rely heavily on experience for adjustments, making it difficult to balance feed quality, production efficiency, and energy consumption.

[0003] In the feed production sector, although some companies have adopted sensors and automated equipment, most of these efforts remain at the data acquisition level, lacking in-depth data analysis and intelligent control. Therefore, this paper proposes developing an intelligent control method and system for feed production lines based on the Internet of Things (IoT). Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent control system and method for feed production lines based on the Internet of Things, so as to solve the technical problems of poor parameter coordination, large quality fluctuations, high energy consumption, and lagging control in traditional feed production lines.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] The first aspect of this invention provides an intelligent control system for a feed production line based on the Internet of Things, comprising: The sensing and acquisition module is used to deploy IoT sensing terminals in key sections of the feed production line, build a full-link sensing network, and synchronously collect multi-dimensional production parameter data to obtain feed production line parameter data. The data processing module is used to encrypt, preprocess, and intelligently identify anomalies in the feed production line parameter data, and obtain the anomaly identification results of the production parameters. The model building module is used to construct the correlation mapping matrix between production parameters and feed quality, production efficiency, and energy consumption indicators based on the anomaly identification results of production parameters, and to solve for the optimal set of target production parameters through a multi-objective optimization algorithm. The intelligent control module is used to input the target production parameter set into the deep reinforcement learning model for intelligent control calculation to obtain the production parameter adjustment amount; The collaborative optimization module is used to achieve collaborative optimization and control of multiple devices and processes based on the edge-cloud collaborative architecture and the adjustment amount of production parameters.

[0007] A second aspect of this invention provides an intelligent control method for a feed production line based on the Internet of Things, comprising the following steps: S1. Sensing and Data Acquisition: Deploy IoT sensing terminals in key sections of the feed production line to build a full-link sensing network and synchronously collect multi-dimensional production parameter data to obtain feed production line parameter data. S2. Data Processing: Encrypt, preprocess, and intelligently identify anomalies in the feed production line parameter data to obtain the anomaly identification results for production parameters. S3. Model Construction: Based on the results of anomaly identification of production parameters, construct the correlation mapping matrix between production parameters and feed quality, production efficiency, and energy consumption indicators, and solve the optimal set of target production parameters through a multi-objective optimization algorithm; S4. Intelligent Control: Input the target production parameter set into the deep reinforcement learning model to perform intelligent control calculations and obtain the production parameter adjustment amount; S5. Collaborative Optimization: Based on an edge-cloud collaborative architecture, it achieves collaborative optimization and control of multiple devices and processes by adjusting production parameters.

[0008] The beneficial effects of this invention are: This invention utilizes IoT sensing terminals at various stages of the feed production line to collect data on raw material receiving, crushing, mixing, pelleting, and cooling / drying processes. Data fusion analysis based on edge and cloud-based processing facilitates subsequent control. A multi-level anomaly identification method based on DBSCAN clustering, Gaussian influence function, and partial least squares regression can not only classify states into normal, critically abnormal, and severely abnormal categories, but also analyze the impact of various parameter anomalies on production, thereby determining the abnormal work section, anomaly type, and deviation magnitude.

[0009] This invention establishes a mapping relationship between production parameters and abnormal states through multi-dimensional feature evaluation and abnormal impact analysis, enabling early detection and location of production anomalies, and significantly reducing batch product defects caused by excessive moisture in raw materials, equipment failures, etc.

[0010] Based on anomaly identification results and key control parameters, this invention establishes a quantitative correlation mapping matrix of three-dimensional evaluation indicators of production parameters through orthogonal experiments and Gaussian process regression. Then, combined with the weights of quality, efficiency, and energy consumption in hierarchical analysis, the optimal target production parameter set is finally solved by particle swarm optimization algorithm. This solves the problem of blindness in traditional experience-based trial-and-error parameter adjustment, ensuring that the optimal target production parameter set can not only meet the hard constraints of feed quality, but also maximize efficiency and minimize energy consumption, thereby achieving a multi-objective synergistic balance.

[0011] This invention is based on a DQN deep reinforcement learning-based intelligent control model and an edge-cloud collaborative architecture. Compared with traditional PID control, this invention overcomes its lack of adaptability to dynamic working conditions in feed production. Through a state-action-reward framework and a hierarchical collaborative strategy, it achieves synchronous control of multiple devices in a single link and high-priority parameter coordination across links, solving the problem of difficulty in coordinating parameters with different response characteristics, thereby improving the stability and control accuracy of the production line.

[0012] In summary, this invention transforms the control mode of feed production from the traditional experience-based trial-and-error approach to a data-driven approach, effectively reducing energy consumption and quality loss, improving production capacity and product stability, and providing a technical solution for intelligent feed production. Attached Figure Description

[0013] The invention will now be further described with reference to the accompanying drawings.

[0014] Figure 1 This is a system block diagram of the present invention.

[0015] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 As shown, this invention is an intelligent control system for a feed production line based on the Internet of Things (IoT), comprising: a sensing and acquisition module, a data processing module, a model building module, an intelligent control module, and a collaborative optimization module. The sensing and acquisition module is connected to the data processing module, the data processing module is connected to the model building module, the model building module is connected to the intelligent control module, and the intelligent control module is connected to the collaborative optimization module. The specific application scenario of this invention is a corn-soybean meal pig feed production line.

[0018] The sensing and acquisition module includes setting up IoT sensing terminals at key stages of the feed production line. A full-link sensing network is constructed through these IoT terminals to synchronously collect production parameter data, thereby obtaining feed production line parameter data. The specific process is as follows: Temperature and humidity sensors and moisture sensors are installed in the raw material storage section to monitor the temperature and humidity of the storage environment and the initial moisture content of the raw materials in real time, thereby obtaining process data for the raw material storage section. Vibration sensors, motor current sensors, particle size sensors, and speed sensors are installed in the crushing section to monitor the operating vibration amplitude of the crusher, the motor operating current, the particle size distribution of the crushed material, and the main shaft speed of the crusher, thereby obtaining the process data of the crushing section. Weighing sensors, stirring speed sensors, liquid addition flow sensors, and mixing uniformity sensors are installed in the mixing section to monitor the added weight of each component raw material, the stirring shaft speed of the mixer, the added flow rate of liquid auxiliary materials, and the uniformity of the mixed material, thereby obtaining the process data of the mixing section. A ring die temperature sensor, a ring die pressure sensor, a conditioning temperature sensor, and a steam pressure sensor are installed in the pelleting section to monitor the working temperature of the pelleting ring die, the die hole pressure, the material conditioning and preheating temperature, and the steam supply pressure, thereby obtaining process data for the pelleting section. Wind speed sensors, discharge temperature sensors, and finished product moisture sensors are installed in the cooling and drying section to monitor the cooling fan speed, finished product discharge temperature, and final moisture content of the finished product, thereby obtaining process data for the cooling and drying section. The IoT gateway converts, encrypts, and timestamps the process data collected from each section, including raw material storage, crushing, mixing, pelleting, and cooling and drying. The data is then standardized and structured by edge computing nodes to obtain feed production line parameter data.

[0019] In one specific embodiment, the feed production line is equipped with IoT-based data collection to achieve comprehensive acquisition of parameters throughout the entire production process. In the raw material storage section, high-precision SHT30 temperature and humidity sensors are installed in the middle and top of the storage silo to avoid monitoring deviations caused by blind spots. These sensors are resistant to dust interference and can capture real-time fluctuations in storage environment temperature and humidity to prevent raw material mold growth. High-frequency capacitive moisture sensors are inserted into the raw material pile to detect the initial moisture content of raw materials such as corn and soybean meal in real time, providing basic data for subsequent conditioning process parameter adjustments.

[0020] In the crushing section, piezoelectric vibration sensors are attached to key parts of the crusher casing to monitor the vibration amplitude and frequency during crusher operation, and to determine whether there are faults such as bearing wear or rotor imbalance. The operating current of the crusher motor is collected by a current transformer to indirectly reflect changes in the crushing load. A laser particle size sensor is installed at the crusher outlet to detect the particle size distribution of the crushed material in real time, ensuring that the particle size meets the requirements for subsequent mixing and granulation. The crusher spindle speed is monitored by an incremental encoder to achieve speed feedback.

[0021] In the mixing section, weighing sensors are installed at the raw material addition stage to collect the weight of the raw materials, so as to achieve accurate proportioning of each component and control the error within ±0.5%. A magnetoelectric speed sensor is installed at the mixing shaft end of the mixer to monitor the mixing speed in real time and ensure the uniformity of mixing. For liquid excipients such as oils and vitamins, electromagnetic flow sensors are used to monitor the addition flow rate to ensure that the amount of excipients added is controllable. A near-infrared spectral sensor is installed at the discharge port of the mixer to quickly detect the compositional uniformity of the mixed material through spectral analysis and avoid local component segregation.

[0022] In the pelleting section, ring die temperature and pressure are parameters affecting pellet forming quality. A sheathed thermocouple temperature sensor embedded inside the ring die monitors the working temperature in real time to prevent overheating and damage to feed nutrients. A piezoresistive pressure sensor installed at the ring die outlet monitors the die pressure during material extrusion and helps determine if the ring die is clogged or worn. A temperature sensor installed at the conditioner outlet monitors the temperature of the material after steam conditioning to ensure proper maturation. A pressure transmitter installed on the steam pipeline provides real-time feedback on the steam supply pressure, ensuring stable conditioning results.

[0023] In the cooling and drying section, wind speed sensors are installed at the air inlet and outlet of the cooler to monitor the cooling air circulation efficiency; an infrared temperature sensor is installed at the outlet of the cooler to detect the finished product discharge temperature in a non-contact manner, avoiding excessive temperature that could cause the finished product to absorb moisture; and a high-frequency resistance moisture sensor is used to detect the final moisture content of the finished product, ensuring that the moisture content is controlled within the acceptable range of 12%-14%.

[0024] All production parameter data collected by IoT sensing terminals are locally aggregated through a LoRa / Wi-Fi dual-mode IoT gateway, which converts the Modbus and MQTT protocols. The data is then encrypted using the AES encryption algorithm before being uploaded to the edge computing node. The edge computing node timestamps and synchronizes the data from each work section to eliminate sampling time differences between different sensors. Min-max normalization is used to eliminate dimensional differences and abnormal pulse data is removed. Finally, the data is integrated into structured parameter data, providing high-quality data support for subsequent data processing and control decisions.

[0025] The data processing module is used to encrypt, preprocess, and intelligently identify anomalies in the feed production line parameter data, resulting in anomaly identification results. The specific process is as follows: The parameter data uploaded by the IoT gateway is encrypted using the TLS protocol to ensure data transmission security. Wavelet threshold denoising is performed on the encrypted parameter data to obtain the denoised parameter data; The raw material storage section process data in the denoised parameter data is synchronized in the time domain, curve fitted and feature extracted to obtain the raw material storage section feature data. The process data of the crushing section, the process data of the mixing section, and the process data of the granulation section in the denoised parameter data are processed separately to obtain the characteristic data of the crushing section, the characteristic data of the mixing section, and the characteristic data of the granulation section, respectively. Dynamic curve fitting and feature extraction are performed on the cooling and drying process data in the denoised parameter data to obtain the feature data of the cooling and drying section. The characteristic data of the raw material storage section, the crushing section, the mixing section, the granulation section, and the cooling and drying section are combined to construct a production target characteristic dataset; Based on the production target feature dataset, DBSCAN clustering and anomaly threshold determination are performed to obtain the results of production parameter anomaly identification.

[0026] In one specific embodiment, the parameter data of the feed production line faces risks of dust interference, electromagnetic interference, and data leakage during transmission. Therefore, a secure transmission channel is established using the TLS protocol during data transmission, and the data is encrypted end-to-end to prevent tampering or theft. To address issues such as sensor noise and electromagnetic interference in the parameter data, a wavelet threshold denoising method is used for data purification. The parameter data is decomposed into low-frequency useful signals and high-frequency noise signals using the db6 wavelet basis function. A soft thresholding function is then applied to process the high-frequency coefficients, eliminating sharp noise interference while preserving the data's trend characteristics to the greatest extent possible, resulting in denoised parameter data.

[0027] The process data of the raw material storage section were synchronized in the time domain and curve fitted to obtain the temperature change curve, humidity change curve, and initial moisture content change curve of the raw material storage environment. Feature extraction was performed on the change curves to obtain the raw material storage temperature gradient, temperature uniformity index, humidity gradient, humidity uniformity index, initial moisture content gradient, and moisture content uniformity index, which are used as feature data of the raw material storage section. It should be noted that the slope of the curve is used as the gradient, and the ratio of the mean to the standard deviation of the data in the curve is used as the uniformity index.

[0028] The specific processing method for the crushing process data in the denoised parameter data is as follows: Time-frequency analysis is performed on the vibration amplitude of the crusher to obtain characteristic data such as the proportion of low-frequency vibration energy, the amplitude of the dominant vibration frequency, and the vibration amplitude fluctuation coefficient. It should be noted that the time-series signal of the vibration amplitude is divided into multiple data segments using a Hanning sliding window with a length of 265 data points and a 50% overlap. Fourier transform is performed on each data segment, and the segments are spliced ​​to obtain a two-dimensional time-frequency spectrum. The proportion of low-frequency vibration energy is extracted from the two-dimensional time-frequency spectrum to reflect the stability of the crusher's operation; the proportion of energy in the low-frequency range (0-50Hz) in the entire frequency domain is used. The amplitude of the dominant vibration frequency reflects the intensity of the core vibration source, and is the mean amplitude corresponding to the highest energy frequency. The vibration amplitude fluctuation coefficient reflects the risk of vibration abrupt changes, and is the ratio of the standard deviation to the mean of the dominant frequency amplitude.

[0029] Principal component analysis was performed on the particle size distribution data of the pulverized material to obtain characteristic data of particle size distribution. It should be noted that covariance matrix decomposition was performed on the particle size distribution data to extract the principal components with the strongest variance explanation capability. The principal components compress the data dimensionality while aggregating core particle size distribution information, and extracting principal component features to form particle size distribution characteristic data, which reflects the global particle size distribution. Curve fitting was performed on the motor operating current and the pulverizer spindle speed to obtain the motor operating current variation curve and the spindle speed variation curve. The current stabilization duration, current stabilization value (average current during stable operation), and current fluctuation coefficient (ratio of standard value to mean) were extracted as feature data. The spindle stabilization duration, current stabilization value, and speed fluctuation coefficient were also extracted as feature data.

[0030] The specific processing method for the mixing process data in the denoised parameter data is as follows: The weight of each component raw material is synchronized in the time domain and curve-fitted to obtain the weight addition curves for each component. Piecewise linear fitting is performed on the curves, and the entire curve is divided into three continuous segments using a piecewise least squares fitting method: the initial addition segment, the stable addition segment, and the completion addition segment. The slope of each segment is extracted as the weight addition gradient, reflecting the addition rate. The deviation coefficient of each segment is calculated to evaluate the uniformity of the raw material ratio change, serving as the raw material weight addition characteristic data, namely the raw material weight addition gradient and weight addition uniformity index. Similarly, the mixing shaft speed and liquid additive addition flow rate are analyzed to obtain the mixing shaft speed gradient and speed uniformity index, and the addition flow rate gradient and flow rate uniformity index. Principal component analysis is performed on the uniformity of the mixed material to extract principal component features and form material uniformity characteristic data.

[0031] The specific processing method for the granulation process data in the denoised parameter data is as follows: The working temperature of the granulation ring die, the material conditioning and preheating temperature, and the steam supply pressure are dynamically fitted to obtain the ring die temperature change curve, the conditioning and preheating temperature change curve, and the steam supply pressure change curve. The ring die temperature gradient is extracted to reflect the ring die temperature uniformity index, conditioning temperature gradient, conditioning temperature uniformity index, steam pressure gradient, and pressure stabilization duration. Principal component analysis is performed on the die orifice pressure data to extract principal component features and form die orifice pressure feature data.

[0032] The dynamic change curves of cooling fan speed, finished product discharge temperature, and final moisture content of the finished product in the cooling and drying section data are fitted, and five features are extracted: wind speed rise rate, stable cooling time, temperature drop gradient, moisture decay coefficient, and final moisture deviation. These features form cooling and drying efficiency characteristic data, which comprehensively reflects the cooling and drying effect.

[0033] The extracted feature parameters are integrated to construct a production target feature dataset. The DBSCAN clustering algorithm is then used to perform cluster analysis on this dataset. This DBSCAN algorithm does not require pre-setting the number of clusters and can adaptively identify density clusters for different production states. The clustering is determined using historical normal production data. The clustering results are divided into three categories: normal production cluster, critical abnormal cluster, and severe abnormal cluster, based on the domain radius and minimum number of points. Based on the feed production process standards, the threshold range of each feature parameter is set, and the clustering results are screened a second time to identify feature samples that exceed the normal threshold range. The types of abnormal parameters are identified, such as the abnormal production sections where raw material moisture exceeds the standard, abnormal ring die pressure, and unqualified finished product moisture. Finally, the abnormal production parameter identification results are obtained, providing a targeted basis for subsequent parameter optimization.

[0034] Furthermore, based on the production target feature dataset, DBSCAN clustering and anomaly threshold determination are performed to obtain the production parameter anomaly identification results. The specific process is as follows: Statistical analysis is performed on the normal sample data of the production target feature dataset to calculate the mean, standard deviation and range of each feature parameter, and to construct a standard feature space model of production parameters. DBSCAN clustering is set based on the standard feature space model. Using the neighborhood radius and minimum number of points, DBSCAN clustering is performed on the production target feature dataset to obtain clustering results for normal production clusters, critical abnormal clusters, and severe abnormal clusters; A Gaussian influence function is constructed for each feature parameter, centered on the mean of the standard feature space model. The degree of influence of each feature parameter deviating from the center is calculated using the Gaussian influence function. Partial least squares regression analysis was performed on the production target feature dataset and the influence value to obtain the feature weight matrix; Based on the feature weight matrix and the influence degree value, the abnormal influence of each feature parameter is weighted and summed to obtain the comprehensive abnormality index value; Based on the comprehensive anomaly index value and clustering results, parameter combinations and anomaly levels that exceed the normal range are identified, resulting in anomaly identification results for production parameters.

[0035] In one specific embodiment, production target feature samples corresponding to historical normal production data of the feed production line are screened. Through statistical analysis, the mean (μ), standard deviation (σ), and range (R) of each feature parameter are calculated dimension by dimension. The mean reflects the normal central level of the feature parameter, the standard deviation characterizes the normal fluctuation range, and the range defines the extreme threshold boundary. Based on the mean, standard deviation, and range of each feature parameter, a standard feature space model of the production parameters is constructed, and the normal value range of each feature parameter is set. Based on this standard feature space model of the production parameters, and based on the distribution density of the feature parameters, the core parameter of DBSCAN clustering is set: 0.6 times the standard deviation of each feature dimension is used as the threshold. The neighborhood radius is used to ensure that normal samples form dense clusters; the minimum number of points is set to three times the number of feature dimensions to avoid isolated noise points forming false clusters. The DBSCAN clustering algorithm is used to cluster the current production target feature dataset, dividing the feature samples into three categories: normal production clusters, critically abnormal clusters, and severely abnormal clusters. Normal production clusters are those with sample density meeting requirements and all feature parameters within the standard range; critically abnormal clusters are those with low sample density and some feature parameters approaching the standard threshold boundary; severely abnormal clusters are those with extremely low sample density and multiple feature parameters significantly exceeding the standard range. It should be noted that: low sample density means the sample density is within a preset range below the standard range; some feature parameters approaching the standard threshold boundary means some feature parameters are within the preset range corresponding to the standard threshold boundary; extremely low sample density means the sample density is within the preset extremely low range; multiple feature parameters significantly exceeding the standard range means multiple feature parameters exceed the standard range by a certain margin.

[0036] Calculate the impact of anomalies in each characteristic parameter on production quality, construct a Gaussian influence function for each characteristic parameter, and use the mean of the standard feature space model. Centered on, with standard deviation Let be the diffusion coefficient, and the expression for the Gaussian influence function is: When the feature parameter x equals the mean At that time, the degree of influence value This indicates that there is no negative impact on production quality; when x deviates... The larger, The closer a value is to 0, the greater the negative impact on production quality. The influence value of each characteristic parameter is calculated using a Gaussian influence function. To highlight the dominant role of key characteristic parameters, partial least squares regression analysis is performed on the production target characteristic dataset and its corresponding influence values ​​to effectively handle the multicollinearity problem among features. By extracting principal components, a regression model of features and anomaly influence is established, resulting in a feature weight matrix. The matrix elements of the feature weight matrix reflect the contribution of each characteristic parameter to the overall anomaly.

[0037] Based on the weight coefficients in the feature weight matrix ( ) and the degree of influence of each feature parameter ( ), through formula Calculate the comprehensive anomaly index value, where The range of values ​​is , The closer the value is to 1, the more severe the abnormality. Within the normal range This is a critical anomaly. This is considered a severe anomaly. Based on the DBSCAN clustering results and the comprehensive anomaly index value, a final determination is made regarding the current production parameter status: if the sample belongs to the normal production cluster and If the sample belongs to the critical anomaly cluster, it is determined to be in normal production; if the sample belongs to the critical anomaly cluster and If a sample is identified as a critical anomaly, its feature parameters that are close to the threshold are labeled; if the sample belongs to a severely anomalous cluster and The result is determined to be a serious anomaly, and the feature parameter with the largest deviation and the corresponding work section are traced back. The final output of the production parameter anomaly identification result includes the anomaly level, the abnormal work section, the type of abnormal parameter, and the degree of deviation, providing a precise target direction for subsequent parameter optimization.

[0038] The model building module is used to construct a correlation mapping matrix between production parameters and feed quality, production efficiency, and energy consumption indicators based on the anomaly identification results of production parameters, and to solve for the optimal set of target production parameters using a multi-objective optimization algorithm; the specific process is as follows: Based on the results of anomaly identification of production parameters, key indicators affecting feed quality, production efficiency, and energy consumption are determined, and a three-dimensional evaluation index system is constructed. Among them, feed quality indicators include nutritional uniformity, pellet hardness, and moisture content; production efficiency indicators include hourly output and equipment utilization rate; and energy consumption indicators include electricity consumption per unit product and heat consumption per unit product. A statistical analysis was conducted on the correspondence between key indicators and three-dimensional evaluation indicators to obtain a preliminary impact relationship table. Based on the preliminary influence relationship table, an orthogonal experimental scheme was designed, and multiple batches of production tests were carried out to obtain measured data of evaluation indicators under different parameter combinations. The measured data are modeled using the Gaussian process regression algorithm to establish the mathematical mapping relationship between each key parameter and each evaluation index, thus obtaining the correlation mapping matrix between production parameters and evaluation indicators. Based on the correlation mapping matrix, a multi-objective optimization model is constructed with feed quality compliance, production efficiency maximization, and energy consumption minimization as optimization objectives. The weight coefficients of each optimization objective are calculated using the analytic hierarchy process to obtain the weighted and fused objective function. The objective function after weighted fusion is solved iteratively using the particle swarm optimization algorithm to finally obtain the optimal set of objective production parameters.

[0039] In one specific embodiment, based on the results of anomaly identification of production parameters, the impact of various parameters on the feed production process is analyzed in depth to identify key parameters affecting feed quality, production efficiency, and energy consumption. These key parameters are the core regulatory factors in the feed production process, directly or indirectly affecting the quality of the final product and the efficiency of the production process. Simultaneously, a three-dimensional evaluation index system is constructed, which measures the feed production process from three dimensions: feed quality, production efficiency, and energy consumption. Nutritional uniformity reflects the balanced distribution of various nutrients in the feed, a key factor in ensuring animals receive comprehensive nutrition; pellet hardness affects the palatability and storage stability of the feed; excessively hard or soft pellets may adversely affect animal feeding and digestion; moisture content relates to the shelf life and quality stability of the feed; appropriate moisture content effectively prevents feed from molding and spoiling. Production efficiency indicators include hourly output and equipment utilization rate. Hourly output directly reflects the production capacity of the production line, while equipment utilization rate reflects the effective working time of the equipment during the production process, reflecting the efficiency of equipment use. Energy consumption indicators include electricity consumption per unit of product and heat consumption per unit of product, which directly reflect the energy consumption in the production process and are of great significance for reducing production costs and improving energy efficiency.

[0040] A detailed statistical analysis was conducted on the correspondence between the identified key parameters and the three-dimensional evaluation indicators. All known key parameter deviation samples were extracted from the production parameter anomaly identification results, and the mapping relationship between these parameter deviations and feed quality test results, production efficiency data, and energy consumption data was investigated in depth. Various statistical methods, such as cluster analysis, correlation analysis, and regression trend comparison, were used to analyze the degree and direction of influence of each key parameter on each evaluation indicator. For example, the impact of changes in raw material ratios on feed nutritional uniformity, pellet hardness, and moisture content was analyzed; the effects of adjusting the grinder speed on hourly output, equipment utilization rate, and unit product power and heat consumption were also analyzed. Finally, a preliminary influence relationship table was formed, which clearly presents the preliminary correlation between each key parameter and different evaluation indicators, providing an important basis for experimental design and model construction.

[0041] An orthogonal experimental design was created based on a preliminary influence relationship table. Orthogonal experimental design is a scientific and efficient experimental method that uses a minimal set of parameter levels to cover all main effects and some interaction effects, ensuring rich information acquisition while effectively controlling the experimental scale. In the feed production experiment, representative combinations of key parameters were selected for multiple batch production tests. For example, different level combinations were set for key parameters such as raw material ratio, grinder speed, mixing time, pelleting temperature, and conditioning pressure, and production operations were carried out according to the requirements of orthogonal experiments. During the experimental execution, each set of parameters corresponded to a complete set of evaluation index test results. These test results covered feed quality indicators, production efficiency indicators, and energy consumption indicators. All experimental samples together constitute a complete measured dataset of evaluation indicators covering multiple operating conditions. This dataset records in detail the specific performance of each evaluation indicator under different combinations of key parameters, providing data support for subsequent modeling and analysis. The measured dataset of evaluation indicators was modeled and analyzed using the Gaussian process regression algorithm. Gaussian process regression is a non-parametric Bayesian modeling method that does not require prior knowledge of the specific form of the function. It captures complex nonlinear relationships with its powerful learning ability and provides the mean and confidence interval of the predicted values. This Gaussian process regression algorithm uses the covariance function as its core component. By learning the correlation patterns between key parameters, it uncovers the underlying patterns in the data and exhibits good generalization ability, accurately predicting evaluation indicators under new parameter combinations. After training with the Gaussian process model, a correlation mapping matrix between production parameters and three-dimensional evaluation indicators is constructed. Each element in the correlation mapping matrix corresponds to the influence function of a certain key parameter on a certain evaluation indicator, realizing the function mapping from the parameter space to the quality response space, production efficiency response space, and energy consumption response space. The mapping relationship not only has good interpretability, clearly showing the influence mechanism of each parameter on the evaluation indicator, but also has strong predictive ability, providing a theoretical basis for multi-objective optimization. Based on the correlation mapping matrix, a multi-objective optimization model is constructed with feed quality compliance, maximizing production efficiency, and minimizing energy consumption as optimization objectives. Based on the varying degrees of importance of different optimization objectives in actual feed production, a weighted approach (AHP) is used to assign weights to each objective. For example, industry experts are invited to conduct pairwise comparisons of the three objectives, creating a pairwise comparison matrix. It is assumed that experts consider achieving feed quality standards slightly more important than maximizing production efficiency, and maximizing production efficiency more important than minimizing energy consumption. Data is collected based on these comparison results, and the relative weight coefficient of each objective is calculated through consistency checks. The weighted and fused objective function is then used. , Let represent the objective functions corresponding to achieving feed quality standards, maximizing production efficiency, and minimizing energy consumption, respectively. These represent the relative weight coefficients corresponding to achieving feed quality standards, maximizing production efficiency, and minimizing energy consumption, respectively. This standard function integrates factors such as feed quality, production efficiency, and energy consumption, providing a unified evaluation standard for optimization solutions.

[0042] A multi-objective optimization model is iteratively solved using the Particle Swarm Optimization (PSO) algorithm. PSO is a biomimetic intelligent algorithm that simulates the foraging behavior of bird flocks, where multiple particles collaboratively search within a parameter space. Each particle represents a key parameter combination. In each iteration, the particle adjusts its flight direction and speed based on its current optimal position and the global optimal position. The individual optimal position reflects the optimal solution found by the particle in previous iterations, while the global optimal position is the optimal solution found by the entire particle swarm in the current iteration. By continuously learning the optimal information of individuals and the global swarm, particles gradually converge towards the optimal solution region. During the solution process, the PSO algorithm continuously evaluates the fitness of each particle using a comprehensive objective function, updating the particle's speed and position based on the evaluation results until the algorithm converges. As the number of iterations increases, the particle swarm gradually finds the parameter combination that satisfies the optimal solution of the comprehensive objective function. The final output particle positions represent the target production parameter set, which includes the optimal values ​​of key parameters such as raw material ratio, crusher speed, mixing time, granulation temperature, and conditioning pressure. These optimal parameter combinations not only ensure that feed quality meets preset standards, but also maximize production efficiency and minimize energy consumption. They also have good stability and anti-disturbance capabilities, providing a basis for the optimized control of feed production.

[0043] The intelligent control module is used to input the target production parameter set into a deep reinforcement learning model for intelligent control calculations to obtain the production parameter adjustment amount; the specific process is as follows: The difference between the target production parameter set and the current measured production parameter values ​​is calculated to obtain parameter deviation data; A deep reinforcement learning model is constructed, using the DQN structure, with parameter deviation data as state input, production parameter adjustment amount as action output, and the sum improvement rate of feed quality, production efficiency, and energy consumption as reward function. The deep reinforcement learning model is trained based on historical production data, and the model parameters are optimized through an experience replay mechanism to obtain the trained deep reinforcement learning model. By inputting real-time parameter deviation data into the trained deep reinforcement learning model, the adjustment direction and adjustment range of process parameters in each stage can be obtained. Based on the operating constraints of the equipment, the adjustment range is limited to obtain the final production parameter adjustment amount.

[0044] In a specific embodiment, the difference between the target production parameter set and the current measured production parameter values ​​is calculated. The difference is obtained item by item along the corresponding dimension of the two sets of parameters to acquire the current parameter deviation data. The difference directly reflects the specific degree of deviation between the current production conditions and the theoretical optimal setting, serving as an input variable for intelligent control. Each parameter deviation corresponds to a potential risk point in production quality, efficiency, or energy consumption. For example, if the target speed of the crusher is 1500 r / min, and the current measured value is 1400 r / min, the corresponding speed deviation is 100 r / min. This deviation may lead to coarser raw material particle size, thus affecting the nutritional uniformity in the mixing process. Similarly, if the target pelleting temperature is 85℃, and the measured value is 78℃, a 7℃ temperature deviation will result in insufficient pellet hardness, reducing the feed product's resistance to breakage.

[0045] A deep reinforcement learning model is constructed, using a DQN (Deep Q-Network) structure to build an end-to-end control decision model. The core components of the control decision model must strictly match the control requirements of the feed production line, based on a state-action-reward ternary decision framework: Parameter deviation data is used as the state input, with the number of input layer nodes corresponding to the dimension of key process parameters. If the target production parameter set includes N types of core parameters such as raw material ratio, crusher speed, mixing time, pelleting temperature, and conditioning pressure, then the input layer has N nodes, transforming the parameter deviation data into a structured state vector to ensure the control decision model can fully capture the deviation characteristics of the current operating condition. The production parameter adjustment amount is used as the action output, with the number of output layer nodes consistent with the dimension of the input layer parameters. Each node corresponds to a control command for a process parameter, and the output result includes not only the adjustment direction (positive increase, negative decrease) but also other adjustment parameters. The reduction also includes specific adjustment ranges, such as adjusting the rotation speed by +50 r / min and the mixing time by +1 min, forming a motion space that can directly guide equipment operation. The comprehensive improvement rate of feed quality, production efficiency, and energy consumption is used as the reward function to quantify the quality of the model's control actions. First, the improvement rate of quality indicators such as nutrient uniformity, pellet hardness, and moisture content after control is calculated separately, as well as the improvement rate of efficiency indicators such as hourly output and equipment utilization rate, and the reduction rate of unit product power consumption and unit product heat consumption. Then, the improvement rates of each dimension are weighted and summed according to the weights determined by the analytic hierarchy process to obtain the comprehensive improvement rate. If the comprehensive improvement rate is positive, a positive reward is given, and the higher the improvement rate, the greater the reward value. If the comprehensive improvement rate is negative or does not reach the preset threshold, a negative reward is given, guiding the model to learn a coordinated control strategy that achieves quality standards, optimal efficiency, and lowest energy consumption.

[0046] The deep reinforcement learning model is trained based on historical production data. The training process utilizes an experience replay mechanism to optimize model parameters, addressing the issues of strong correlation and unstable training data in reinforcement learning. Specifically, an experience pool is first constructed to store the "state-action-reward-next state" quadruple experience data generated during training. Parameter deviations (states) under different working conditions in historical production, corresponding control actions, the overall improvement rate after action execution (reward), and parameter deviations under the new working conditions after action execution (next state) are stored one by one in the experience pool. The pool capacity is set to 10,000-50,000 records to ensure coverage of various typical working conditions in feed production (such as fluctuations in raw material moisture content and changes in equipment load). During training, a batch of experience data (e.g., 32 or 64 records) is randomly sampled from the experience pool each time to avoid interference from temporal correlations in the model training. The sampled data is input into the DQN model, and the gradient descent algorithm is used to minimize the error between the model's predicted Q-value and the target Q-value (calculated based on the optimal reward of the next state). The network weights of the model are iteratively updated, including the connection weights between the input layer and hidden layers, and the weights within the hidden layers. During training, the learning rate is set to 0.0001-0.001, and the number of training iterations is 5000-10000. A dual-network structure of the target network and the current network is introduced. Every fixed number of iterations (e.g., 100 steps), the parameters of the current network are copied to the target network to further improve training stability, until the model's loss function converges to a preset threshold (e.g., 10^6). -4 ), complete model training.

[0047] Real-time collected parameter deviation data is input into the trained deep reinforcement learning model. The deep reinforcement learning model first standardizes the input deviation data to eliminate the differences in the units of different parameters, such as the influence of the unit of rotation speed deviation (r / min) versus the unit of time deviation (min) on the model's inference. Then, the standardized deviation data is input into the input layer of the DQN. Through nonlinear transformation of the hidden layer, the ReLU activation function is used to extract high-order coupling features from the deviation data, such as the synergistic effect of rotation speed deviation and granulation temperature deviation, mapping the original deviation features into high-dimensional abstract features. Based on these abstract features, the deep reinforcement learning model traverses the action space, evaluates the comprehensive benefits corresponding to different control actions through Q-value calculation, selects the action with the largest Q-value as the optimal control decision, and outputs the adjustment direction and adjustment range of each process parameter. For example, when the real-time input deviation data is "grinder rotation speed deviation -80r / min, mixing time deviation -1min, granulation temperature deviation -5℃", the model outputs the control command "grinder rotation speed +75r / min, mixing time +0.8min, granulation temperature +4.5℃" through inference, clarifying the adjustment direction and specific range of each parameter.

[0048] Based on the operational constraints of the equipment, the adjustment range of the model output is limited, ultimately yielding the adjustment amounts for production parameters. The operational constraints of the feed production line equipment, based on the rated performance and technological requirements of the equipment, are key boundary conditions for ensuring production safety and preventing equipment damage. For example, the rated operating range of the grinder speed is 1200-1800 r / min, and the adjustment range should not exceed 200 r / min to avoid motor overload; the process threshold for pelleting temperature is 70-95℃, and the adjustment rate should not exceed 5℃ / min to prevent excessive gelatinization or insufficient forming of the raw materials; the reasonable range for mixing time is 4-8 min, and the adjustment range should not exceed 2 min to avoid excessively long adjustments leading to decreased production efficiency or excessively short adjustments leading to uneven mixing. For each adjustment range output by the model, verify whether it meets the above constraints: if the adjusted parameter value is within the allowable range of the equipment / process, retain the adjustment range directly; if the adjusted parameter value exceeds the constraint range, truncate the adjustment range to the maximum allowable value. For example, if the model outputs a crusher speed adjustment of +250 r / min, exceeding the maximum adjustment range of 200 r / min, the final adjustment range will be corrected to +200 r / min. The adjustment direction and range after constraint verification are the final production parameter adjustment amounts.

[0049] The collaborative optimization module, based on an edge-cloud collaborative architecture, achieves collaborative optimization and control of multiple devices and processes according to production parameter adjustments; the specific process is as follows: Construct an edge-cloud collaborative control architecture, where the edge layer is responsible for real-time device control and the cloud layer is responsible for global optimization and decision-making; Based on the adjustment amount of production parameters, the control tasks are divided into raw material receiving control sub-tasks, crushing control sub-tasks, mixing control sub-tasks, granulation control sub-tasks and cooling and drying control sub-tasks. Establish a coupling degree matrix between subtasks, calculate the mutual influence coefficients between each subtask, and obtain a task coordination requirement table. Based on the task collaboration requirement table, synchronous control of multiple devices in the same stage is realized at the edge layer, and collaborative control of different stages is realized at the cloud layer. Set safety boundaries for parameter adjustments, including upper and lower limits for equipment operating parameters and limits on adjustment rates; through an event triggering mechanism, when production parameters exceed safety boundaries or the anomaly identification result is a serious anomaly, trigger the emergency control process to ensure the stable operation of the production line.

[0050] In one specific embodiment, an edge-cloud collaborative control architecture is constructed. The edge layer is integrated into the local control cabinet of each link in the feed production line, and is equipped with an industrial-grade edge gateway as an edge computing node to directly establish communication connections with the execution units of equipment such as crushers, mixers, and pellet mills, such as frequency converters, heating controllers, and motor drivers. The cloud layer is deployed on a server cluster of a remote industrial internet platform and issues globally optimized control commands to the edge layer. The edge layer and the cloud layer interact with each other through 5G industrial internet.

[0051] Based on the production stage attribution corresponding to the adjustment of production parameters, the control tasks are divided into raw material receiving control sub-tasks, crushing control sub-tasks, mixing control sub-tasks, granulation control sub-tasks, and cooling and drying control sub-tasks. For example, adjusting parameters such as crusher speed and particle size is classified as a crushing control sub-task, with corresponding equipment including the crusher, dust removal equipment in the crushing zone, and conveyor belt; adjusting parameters such as mixing time and mixer torque is classified as a mixing control sub-task, with corresponding equipment including the mixer and material flow sensor; adjusting parameters such as granulation temperature, conditioning pressure, and particle hardness is classified as a granulation control sub-task, with corresponding equipment including the granulator, steam heating device, and ring die pressure controller. Each sub-task includes a control objective, responsible equipment, and execution sequence, resulting in a sub-task list for convenient subsequent coordinated control. A control objective might be adjusting the crushing speed to 1500 r / min; the execution sequence might be synchronized with the mixing sub-task.

[0052] Because there are strong coupling relationships between various stages of feed production, such as the grinding effect directly affecting the mixing uniformity and the mixing quality affecting the pelleting rate, if the control subtasks are executed independently, it is easy to cause control conflicts. For example, if the grinding speed is increased but the mixing time is not extended synchronously, it will still lead to uneven nutrition. Therefore, the mutual influence between subtasks is quantified by the coupling degree matrix, as follows: A 5×5 coupling degree matrix is ​​constructed with 5 control subtasks as row vectors and column vectors. The matrix elements are the mutual influence coefficients between subtask i and subtask j, with values ​​ranging from 0 to 1, where 0 represents no influence and 1 represents strong coupling. The larger the value, the stronger the mutual influence. The mutual influence coefficients are determined based on the correlation analysis of historical production data. For example, after the speed is adjusted in the grinding control subtask, the probability and adjustment range of the mixing time in the mixing control subtask need to be adjusted synchronously are statistically analyzed. The correlation coefficient between the two is calculated and normalized as the corresponding coupling degree value.

[0053] Example coupling matrix: Subtask Raw material receiving Crushed mix Granulation Cooling and drying Raw material receiving 1 0.8 0.6 0.5 0.2 Crushed 0.8 1 0.9 0.8 0.3 mix 0.6 0.9 1 0.9 0.4 Granulation 0.5 0.8 0.9 1 0.5 Cooling and drying 0.2 0.3 0.4 0.5 1

[0054] Based on the coupling degree matrix results, collaborative priorities are divided. Tasks with a mutual influence coefficient ≥ 0.8 are considered high priority and require real-time synchronous control; those with a mutual influence coefficient between 0.5 and 0.8 are considered medium priority and require synchronous control as needed; and those with a mutual influence coefficient < 0.5 are considered low priority and can be executed independently. Sub-task combinations corresponding to each priority are extracted, and collaborative control requirements are clarified to form a task collaboration requirement table. For example, "crushing-mixing" and "mixing-granulation" are high-priority collaborative combinations that require synchronous execution of control commands; "raw material receiving-cooling and drying" is a low-priority combination that can be controlled independently.

[0055] Based on the task collaboration requirement table, a layered collaborative approach is used to ensure synchronous response of multiple devices in the same stage and collaborative optimization of sub-tasks across stages. The edge layer is responsible for the synchronous control of multiple devices in the same stage. For the linkage requirements of multiple devices in the same stage, local edge nodes uniformly issue synchronous control commands, thereby avoiding control imbalances caused by response delays between devices. The cloud layer is responsible for the collaborative control of high-priority sub-tasks across stages. The cloud platform coordinates the control timing and magnitude of edge nodes in each stage, achieving global optimization of parameters throughout the entire process. The edge layer uploads the adjusted device operation data to the cloud layer in real time. The cloud layer compares the control effect with the expected target. If there is a deviation, the adjustment amount of each relevant stage is dynamically corrected and reissued to the corresponding edge nodes for execution, ensuring the accuracy of collaborative control.

[0056] A safety boundary for parameter adjustment is established to construct a production safety protection system. An event-triggered mechanism enables emergency control, ensuring stable production line operation. The safety boundary is set based on equipment rated performance parameters, feed processing technology requirements, and historical fault data. Specifically, it comprises two types of constraints: first, absolute upper and lower limits for equipment operating parameters to prevent equipment overload or damage; and second, limits on parameter adjustment rates to prevent sudden parameter changes that could cause equipment shocks or process fluctuations. The event-triggered mechanism monitors in real-time whether parameters touch the safety boundary or exhibit anomalies. Once triggered, an emergency control process is immediately initiated. First, the edge layer collects equipment operating parameters in real-time and compares them with the safety boundary, while simultaneously receiving anomaly identification results from the data processing module. When parameters exceed the safety boundary or the anomaly identification result is deemed severe, the edge layer immediately triggers a local emergency response, suspending the corresponding equipment operation, cutting off power to hazardous processes, and sending an emergency alarm to the cloud platform. Upon receiving the alarm, the cloud platform quickly invokes historical fault handling schemes and generates emergency control instructions. After emergency control is completed, the cloud platform tracks the equipment operating status, confirms that parameters have returned to the safe range, gradually restores production line operation, and records the cause of the anomaly and the control process for subsequent model optimization and process improvement.

[0057] Please see Figure 2 As shown, this invention is an intelligent control method for a feed production line based on the Internet of Things, comprising the following steps: S1. Sensing and Data Acquisition: Deploy IoT sensing terminals in key sections of the feed production line to build a full-link sensing network and synchronously collect multi-dimensional production parameter data to obtain feed production line parameter data. S2. Data Processing: Encrypt, preprocess, and intelligently identify anomalies in the feed production line parameter data to obtain the anomaly identification results for production parameters. S3. Model Construction: Based on the results of anomaly identification of production parameters, construct the correlation mapping matrix between production parameters and feed quality, production efficiency, and energy consumption indicators, and solve the optimal set of target production parameters through a multi-objective optimization algorithm; S4. Intelligent Control: Input the target production parameter set into the deep reinforcement learning model to perform intelligent control calculations and obtain the production parameter adjustment amount; S5. Collaborative Optimization: Based on an edge-cloud collaborative architecture, it achieves collaborative optimization and control of multiple devices and processes by adjusting production parameters.

[0058] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. An intelligent control system for a feed production line based on the Internet of Things, characterized in that, include: The sensing and acquisition module installs IoT sensing terminals in each corresponding section of the feed production line. The IoT sensing terminals synchronously collect production data of the feed production line to obtain feed production line parameter data. The data processing module is used to encrypt, preprocess, and intelligently identify anomalies in the feed production line parameter data, and obtain the anomaly identification results of the production parameters. The model building module is used to construct the correlation mapping matrix between production parameters and feed quality, production efficiency, and energy consumption indicators based on the anomaly identification results of production parameters, and to solve for the optimal set of target production parameters through a multi-objective optimization algorithm. The intelligent control module is used to input the target production parameter set into the deep reinforcement learning model for intelligent control calculation to obtain the production parameter adjustment amount; The collaborative optimization module is used to perform collaborative optimization and control of multiple devices and processes based on production parameters, using an edge-cloud collaborative architecture.

2. The intelligent control system for a feed production line based on the Internet of Things according to claim 1, characterized in that, Production data from the feed production line is collected synchronously through IoT sensing terminals, specifically including: Temperature and humidity sensors and moisture sensors are installed in the raw material storage section to monitor the temperature and humidity of the storage environment and the initial moisture content of the raw materials in real time, thereby obtaining process data for the raw material storage section. Vibration sensors, motor current sensors, particle size sensors, and speed sensors are installed in the crushing section to monitor the operating vibration amplitude of the crusher, the motor operating current, the particle size distribution of the crushed material, and the main shaft speed of the crusher, thereby obtaining the process data of the crushing section. Weighing sensors, stirring speed sensors, liquid addition flow sensors, and mixing uniformity sensors are installed in the mixing section to monitor the added weight of each component raw material, the stirring shaft speed of the mixer, the added flow rate of liquid auxiliary materials, and the uniformity of the mixed material, thereby obtaining the process data of the mixing section. A ring die temperature sensor, a ring die pressure sensor, a conditioning temperature sensor, and a steam pressure sensor are installed in the pelleting section to monitor the working temperature of the pelleting ring die, the die hole pressure, the material conditioning and preheating temperature, and the steam supply pressure, thereby obtaining process data for the pelleting section. Wind speed sensors, discharge temperature sensors, and finished product moisture sensors are installed in the cooling and drying section to monitor the cooling fan speed, finished product discharge temperature, and final moisture content of the finished product, thereby obtaining process data for the cooling and drying section. The IoT gateway converts, encrypts, and timestamps the process data collected from each section, including raw material storage, crushing, mixing, pelleting, and cooling and drying. The data is then standardized and structured by edge computing nodes to obtain feed production line parameter data.

3. The intelligent control system for a feed production line based on the Internet of Things according to claim 1, characterized in that, The data processing module specifically includes: The feed production line parameter data uploaded by the IoT gateway is encrypted using the TLS protocol, and wavelet threshold denoising is performed on the encrypted parameter data to obtain the denoised parameter data. Time-domain synchronization, curve fitting, and feature extraction were performed on the process data of the raw material storage section to obtain the characteristic data of the raw material storage section; Process data from the crushing section, mixing section, and granulation section are processed to obtain characteristic data for the crushing section, mixing section, and granulation section. Dynamic curve fitting and feature extraction were performed on the process data of the cooling and drying section to obtain the characteristic data of the cooling and drying section; The characteristic data of the raw material storage section, the crushing section, the mixing section, the granulation section, and the cooling and drying section are combined to form the production target characteristic dataset; For the production target feature dataset, DBSCAN clustering and anomaly threshold determination are performed to obtain the production parameter anomaly identification results.

4. The intelligent control system for a feed production line based on the Internet of Things according to claim 3, characterized in that, Based on the aforementioned production target feature dataset, DBSCAN clustering and anomaly threshold determination are performed to obtain production parameter anomaly identification results, including: Statistical analysis is performed on the normal sample data of the production target feature dataset to calculate the mean, standard deviation and range of each feature parameter, and to construct a standard feature space model of production parameters. DBSCAN clustering is set based on the standard feature space model. Using the neighborhood radius and minimum number of points, DBSCAN clustering is performed on the production target feature dataset to obtain clustering results for normal production clusters, critical abnormal clusters, and severe abnormal clusters; A Gaussian influence function is constructed for each feature parameter, centered on the mean of the standard feature space model. The degree of influence of each feature parameter deviating from the center is calculated using the Gaussian influence function. Partial least squares regression analysis was performed on the production target feature dataset and the influence value to obtain the feature weight matrix; Based on the feature weight matrix and the influence degree value, the abnormal influence of each feature parameter is weighted and summed to obtain the comprehensive abnormality index value; Based on the comprehensive anomaly index value and clustering results, parameter combinations and anomaly levels that exceed the normal range are identified, resulting in anomaly identification results for production parameters.

5. The intelligent control system for a feed production line based on the Internet of Things according to claim 1, characterized in that, The model building module includes: Based on the results of anomaly identification of production parameters, key parameters affecting feed quality, production efficiency, and energy consumption are determined, and a three-dimensional evaluation index system is constructed. A statistical analysis was conducted on the correspondence between key parameters and three-dimensional evaluation indicators to obtain a preliminary influence relationship table. Based on the preliminary influence relationship table, an orthogonal experimental scheme was established, and multiple batches of production tests were carried out to obtain measured data of evaluation indicators under different parameter combinations. The Gaussian process regression algorithm is used to model the measured data, establish the mathematical mapping relationship between each key parameter and each evaluation index, and obtain the correlation mapping matrix between production parameters and evaluation indicators; Based on the correlation mapping matrix, a multi-objective optimization model is constructed with feed quality compliance, production efficiency maximization, and energy consumption minimization as optimization objectives. The weight coefficients of each optimization objective are calculated using the analytic hierarchy process to obtain the weighted and fused objective function. The objective function after weighted fusion is solved iteratively using the particle swarm optimization algorithm to finally obtain the optimal set of objective production parameters.

6. The intelligent control system for a feed production line based on the Internet of Things according to claim 1, characterized in that, The intelligent control module includes: The difference between the target production parameter set and the current measured production parameter values ​​is calculated to obtain parameter deviation data; A deep reinforcement learning model is constructed, using the DQN structure, with parameter deviation data as state input, production parameter adjustment amount as action output, and the sum improvement rate of feed quality, production efficiency, and energy consumption as reward function. The deep reinforcement learning model is trained based on historical production data, and the model parameters are optimized through an experience replay mechanism to obtain the trained deep reinforcement learning model. By inputting real-time parameter deviation data into the trained deep reinforcement learning model, the adjustment direction and adjustment range of process parameters in each stage can be obtained. Based on the operating constraints of the equipment, the adjustment range is limited to obtain the final production parameter adjustment amount.

7. The intelligent control system for a feed production line based on the Internet of Things according to claim 6, characterized in that, The collaborative optimization module includes: Construct an edge-cloud collaborative control architecture, including an edge layer and a cloud layer; Based on the adjustment amount of production parameters, the control tasks are divided into raw material receiving control sub-tasks, crushing control sub-tasks, mixing control sub-tasks, granulation control sub-tasks and cooling and drying control sub-tasks. Establish a coupling degree matrix between subtasks, calculate the mutual influence coefficients between each subtask, and obtain a task coordination requirement table. Based on the task collaboration requirement table, synchronous control of multiple devices in the same stage is achieved at the edge layer, and collaborative control of different stages is achieved at the cloud layer. Set safety boundaries for parameter adjustments, including upper and lower limits for equipment operating parameters and limits on adjustment rates; trigger emergency control procedures through an event-triggered mechanism when production parameters exceed safety boundaries or when anomaly identification results in a serious anomaly.

8. A method for intelligent control of a feed production line based on the Internet of Things (IoT), implementing the intelligent control system for a feed production line based on the IoT as described in claims 1-7, characterized in that, Includes the following steps: S1. Sensing and Data Acquisition: Deploy IoT sensing terminals in key sections of the feed production line to build a full-link sensing network and synchronously collect multi-dimensional production parameter data to obtain feed production line parameter data. S2. Data Processing: Encrypt, preprocess, and intelligently identify anomalies in the feed production line parameter data to obtain the anomaly identification results for production parameters. S3. Model Construction: Based on the results of anomaly identification of production parameters, construct the correlation mapping matrix between production parameters and feed quality, production efficiency, and energy consumption indicators, and solve the optimal set of target production parameters through a multi-objective optimization algorithm; S4. Intelligent Control: Input the target production parameter set into the deep reinforcement learning model to perform intelligent control calculations and obtain the production parameter adjustment amount; S5. Collaborative Optimization: Based on an edge-cloud collaborative architecture, it achieves collaborative optimization and control of multiple devices and processes by adjusting production parameters.

Images