Noodle production line equipment operation and maintenance management system based on Internet of Things

Abstract

The invention provides a noodle production line equipment operation and maintenance management system based on the Internet of Things. The system comprises a raw material management module, an environment monitoring module, a parameter optimization module and an operation and maintenance management module. The raw material management module is used for collecting raw material data for producing noodles, inputting the raw material data into a preset double-branch LSTM model to predict raw material characteristics in a future time period, and generating a characteristic prediction result; the environment monitoring module is used for collecting environment data; the parameter optimization module is used for constructing a multi-target problem set for optimizing the noodle quality according to the characteristic prediction result, the environmental data and a preset process database, and solving the multi-target problem set through a preset second-generation non-dominated sorting genetic algorithm to obtain production process parameters; and the operation and maintenance management module is used for generating an operation and maintenance management instruction according to the production process parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment operation and maintenance management, and in particular to a noodle production line equipment operation and maintenance management system based on the Internet of Things. BACKGROUND

[0002] With the wide application of the Internet of Things technology in the operation and maintenance management of modern noodle production lines, various sensors can collect a large amount of equipment operation and environmental data in real time. However, current production line operation and maintenance management still highly depends on the manual experience of technical personnel to interpret and make decisions on the data, and fails to effectively utilize the Internet of Things data to realize intelligent closed loop. This management mode relying on manual experience has problems of response lag and strong subjectivity of decision-making, resulting in that the production process parameters cannot be accurately adjusted in real time according to the fluctuation of raw material characteristics and environmental changes, which not only restricts the improvement of production management efficiency, but also makes it difficult to guarantee the quality stability of products. SUMMARY

[0003] The present application provides a noodle production line equipment operation and maintenance management system based on the Internet of Things, for improving production management efficiency and improving the quality stability of noodles.

[0004] In a first aspect, the present application provides a noodle production line equipment operation and maintenance management system based on the Internet of Things, which comprises a raw material management module, an environment monitoring module, a parameter optimization module and an operation and maintenance management module. The raw material management module is configured to collect raw material data for producing noodles, input the raw material data into a preset double-branch LSTM model to predict the raw material characteristics in a future period, and generate a characteristic prediction result. The environment monitoring module is configured to collect environmental data. The parameter optimization module is configured to construct a multi-objective problem set for optimizing the quality of noodles according to the characteristic prediction result, the environmental data and a preset process database, solve the multi-objective problem set by a preset second-generation non-dominated sorting genetic algorithm, and obtain production process parameters. The operation and maintenance management module is configured to generate operation and maintenance management instructions according to the production process parameters.

[0005] In some embodiments, the raw material data includes moisture content, ash content, pH value, dosage of additives, storage temperature, storage humidity, farinograph parameters and tensile parameters.

[0006] In some embodiments, the preset double-branch LSTM model comprises a feature extraction layer, a main branch, a secondary branch and a plurality of parallelly arranged fully connected layers. When the raw material management module is used to input the raw material data into the preset double-branch LSTM model to predict the raw material characteristics in a future period and generate a characteristic prediction result, it is specifically configured to perform: Extract local features from the raw material data and perform dimensionality reduction to generate a raw material feature vector; The raw material feature vector is input into the main branch of the dual-branch LSTM model, and bidirectional temporal dependency features are extracted through the forward LSTM layer and the backward LSTM layer to generate the first hidden state vector set. The raw material feature vector is input into the main branch of the dual-branch LSTM model. Bidirectional temporal modeling is performed through forward LSTM and backward LSTM layers to extract long-term dependency features in both directions and generate the first hidden vector. Dynamic attention weights are generated based on the storage temperature and the storage humidity. The dynamic attention weights are then used to enhance the features of the first hidden vector set to generate a second hidden vector. Based on the fully connected layer, the first characteristic prediction set corresponding to the second hidden vector set is mapped respectively. The first characteristic prediction set includes: water absorption rate prediction value, formation time prediction value, stability prediction value and tensile resistance prediction value. The first feature prediction set is input into the secondary branch of the dual-branch LSTM model to perform adversarial gradient calculation and backpropagation, thereby generating the second feature prediction set. The first characteristic prediction set and the second characteristic prediction set are weighted and fused, and constrained and calibrated based on the statistical distribution of historical raw material characteristic data to generate the characteristic prediction result.

[0007] In some embodiments, the characteristic prediction results include: predicted water absorption rate, predicted formation time, predicted stability, and predicted tensile resistance. The parameter optimization module, when executing the construction of a multi-objective problem set for optimizing noodle quality based on the characteristic prediction results, the environmental data, and a preset process database, specifically executes: Based on the predicted water absorption rate, the predicted formation time, the predicted stability, the predicted tensile resistance, and the environmental data, a range of process parameters is obtained from the preset process database. The range of process parameters includes: dough water content, maturation time, rolling ratio, and drying rate. The process parameter range is defined as the decision variable search space for a multi-objective optimization problem, wherein the amount of water used for mixing, the maturation time, and the drying rate are continuous variables, and the rolling ratio sequence is an ordered discrete variable set. Within the search space of decision variables, a multi-objective optimization problem set is constructed using the sensory score of noodles, the cooking loss rate, and the comprehensive index of textural properties as optimization objectives.

[0008] In some embodiments, when the parameter optimization module is used to solve the multi-objective problem set using a preset second-generation non-dominated sorting genetic algorithm to obtain production process parameters, it is specifically used to perform the following: The set of multi-objective optimization problems is used to generate a parent population through a hybrid coding strategy, wherein the amount of water used for mixing, the maturation time, and the drying rate are encoded using floating-point numbers, and the rolling ratio sequence is encoded using integer permutation. Calculate the first comprehensive cost of each individual in the parent population based on the preset total comprehensive cost function; Preset process constraints are imposed on the population, and the degree of violation of the constraints is quantified as a penalty cost, which is added to the first comprehensive cost to obtain the second comprehensive cost; Based on the second comprehensive cost, fast non-dominated sorting and crowding calculation are performed on the parent population, and binary tournament selection is performed according to the non-dominated level and crowding to generate a mating pool. Simulated binary crossover and polynomial mutation operations are sequentially performed on the mating pool to generate offspring populations. The offspring populations are then merged with the parent populations for a new round of iteration until the maximum number of generations is reached, and the Pareto optimal solution set is output. Select the second comprehensive cost optimal solution set from the Pareto optimal solution set, and decode the second comprehensive cost optimal solution set into the production process parameters.

[0009] In some embodiments, the preset total comprehensive cost function is: ; Where F is the first comprehensive cost, T is the total number of time periods divided into the entire noodle production process; C1 is the unit electricity price for time period t; P1 is the total electricity consumption in time period t during the dough mixing, cooking, rolling, and drying stages; C2 is the cost coefficient per unit mass of raw materials; L1 is the raw material loss mass caused by moisture evaporation, dough scraps, etc., in time period t; C3 is the unit penalty cost coefficient for product quality deviation from the standard; D is the absolute value of the deviation of the sensory score, cooking loss rate, and comprehensive score of the texture characteristics of the noodles produced in time period t from the target value; K is the number of adjustable key process equipment in the production process; and C4 is the cost per unit adjustment action of the kth piece of equipment. This represents the absolute value of the process parameter adjustment for the k-th device during time period t.

[0010] In some embodiments, the compression ratio sequence is a set of multiple preset stages of compression ratios that decrease sequentially from large to small, wherein the number of preset stages is 3-5, the compression ratio range of the first stage is 0.4-0.5, and the compression ratio range of the last stage is 0.1-0.15.

[0011] This application provides an IoT-based operation and maintenance management system for noodle production line equipment. The system includes: a raw material management module, an environmental monitoring module, a parameter optimization module, and an operation and maintenance management module. The raw material management module collects raw material data for noodle production, inputs the raw material data into a pre-set dual-branch LSTM model to predict raw material characteristics for future periods, and generates characteristic prediction results. The environmental monitoring module collects environmental data. The parameter optimization module constructs a multi-objective problem set for optimizing noodle quality based on the characteristic prediction results, environmental data, and a pre-set process database. It then solves the multi-objective problem set using a pre-set second-generation non-dominated sorting genetic algorithm to obtain production process parameters. The operation and maintenance management module generates operation and maintenance management instructions based on the production process parameters. In the above method, data is automatically collected through the raw material management module and the environmental monitoring module, and the characteristics of raw materials are intelligently predicted through a dual-branch LSTM model. Based on the prediction results, a multi-objective problem is constructed through the parameter optimization module and solved automatically using a genetic algorithm. This effectively transforms the experience-based decision-making of technicians into data-driven optimization algorithm decision-making, reducing the degree of manual intervention. Finally, a management closed loop is formed through the operation and maintenance management module, realizing the equipment operation and maintenance process from data perception to intelligent decision-making and then to automatic execution. This improves production management efficiency while maintaining product quality consistency. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A schematic block diagram of an IoT-based noodle production line equipment operation and maintenance management system provided for embodiments of this application; Figure 2 A schematic flowchart illustrating a prediction result generation method provided in an embodiment of this application; Figure 3 A schematic flowchart illustrating a method for constructing a multi-objective problem set provided in an embodiment of this application; Figure 4 This is a schematic flowchart illustrating a method for solving production process parameters provided in an embodiment of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described below with reference to the accompanying drawings.

[0015] The terms "first" and "second," etc., used in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0016] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0017] It should be understood that in this application, "at least one (item)" means one or more, "more than one" means two or more, "at least two (items)" means two or three or more, and "and / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0018] Please see Figure 1 , Figure 1 This is a schematic block diagram of an IoT-based operation and maintenance management system for noodle production line equipment, provided in an embodiment of this application. Figure 1 As shown, the IoT-based noodle production line equipment operation and maintenance management system 100 includes: a raw material management module 101, an environmental monitoring module 102, a parameter optimization module 103, and an operation and maintenance management module 104.

[0019] The raw material management module 101 is used to collect raw material data for noodle production, input the raw material data into a preset dual-branch LSTM model to predict the raw material characteristics in future periods, and generate characteristic prediction results.

[0020] For example, IoT sensors are used to continuously monitor materials such as flour and additives in a raw material warehouse, acquiring multi-dimensional indicators reflecting the status of the raw materials. After data cleaning and processing, the collected raw material data is input into a pre-built dual-branch LSTM model for analysis and calculation.

[0021] The pre-built dual-branch LSTM model is a long short-term memory neural network with a dual-channel parallel processing architecture. This model can simultaneously extract and learn the time-series features of raw material data from multiple perspectives. The feature extraction layer within the model performs preliminary processing on the input raw material data, identifying key patterns in the data through convolution and pooling operations, and mapping the high-dimensional raw data to a low-dimensional feature space, forming a vector representation containing the essential characteristics of the raw materials. The extracted data is then fed into two independent processing branches. The main branch employs a bidirectional long short-term memory network structure, capable of capturing the evolution of raw material characteristics over time, considering both the impact of historical data on the current state and the effect of future trends on the prediction results. The secondary branch introduces an adversarial training mechanism to generate alternative solutions that differ from the prediction results of the main branch. Through competition and cooperation between the two branches, the overall accuracy and stability of the prediction are improved. The model output layer contains multiple fully connected neuron layers, each responsible for predicting different types of characteristic indicators to output characteristic prediction results. These characteristic prediction results include key parameters such as flour's water absorption capacity, dough formation characteristics, stability performance, and tensile strength.

[0022] The environmental monitoring module 102 is used to collect environmental data.

[0023] For example, various types of environmental parameter monitoring devices are deployed at key locations in the production site to sense environmental factors affecting production quality, such as temperature, humidity, and cleanliness. The collected environmental data is aggregated to a data processing center via wireless or wired communication. After format conversion and unit standardization, standardized environmental status descriptions are generated. Changes in workshop temperature directly affect the fermentation speed of dough and the drying effect of noodles. Excessive temperature may lead to over-fermentation, affecting taste, while insufficient temperature prolongs processing time and reduces production efficiency. Workshop humidity determines the evaporation rate of moisture on the noodle surface. Excessive humidity prolongs drying time and may even cause mold growth, while insufficient humidity can easily cause surface cracking, affecting appearance. Air cleanliness is related to product hygiene and safety; substandard cleanliness may introduce microbial contamination or foreign matter, endangering consumer health. Airflow speed affects the uniform distribution of temperature and humidity within the workshop; proper airflow organization ensures consistent environmental conditions across all production areas. The environmental monitoring module 102 not only records instantaneous environmental parameter values, but also performs statistical analysis on environmental data, calculates characteristic quantities such as the average value, fluctuation range, and rate of change of environmental parameters, identifies abnormal fluctuations and potential risks in environmental status, and provides decision support for environmental control for production management personnel.

[0024] The parameter optimization module 103 is used to construct a multi-objective problem set for optimizing noodle quality based on the characteristic prediction results, environmental data and the preset process database. The multi-objective problem set is solved by the preset second-generation non-dominated sorting genetic algorithm to obtain the production process parameters.

[0025] For example, the parameter optimization module 103 constructs a mathematical model for optimizing noodle quality based on raw material characteristic prediction information and environmental monitoring data, combined with production experience knowledge, and performs intelligent solution.

[0026] Historical process solutions matching the current production conditions are retrieved from a pre-defined process database. This database stores production records and process parameter configurations for different raw material batches and environmental conditions. A similarity matching algorithm filters out historical cases with high reference value. Based on the characteristic prediction results and the process parameter search range determined by environmental data, a problem set containing multiple optimization objectives is constructed. These objectives often have interdependent relationships; improving product quality may increase production costs, while reducing energy consumption may extend the production cycle. A reasonable balance needs to be found among these objectives. The pre-defined second-generation non-dominated sorting genetic algorithm is a heuristic optimization algorithm simulating biological evolution. This algorithm initializes a batch of random combinations of process parameters as an initial population, with each individual representing a possible production solution. The algorithm evaluates the fitness of individuals in the population, calculates the performance of each solution on each optimization objective, and ranks individuals according to Pareto dominance. Non-dominated individuals are considered excellent solutions in the current population. New individuals are generated through genetic operations such as selection, crossover, and mutation. These new individuals inherit some of the superior characteristics of their parents and introduce random variations, allowing the search process to explore a broader solution space. Through multiple rounds of iterative evolution, the overall quality of the population has been continuously improved, gradually converging to the optimal solution region of the problem, resulting in a set of production process parameter configuration schemes that perform well in multiple objectives.

[0027] The operation and maintenance management module 104 is used to generate operation and maintenance management instructions based on production process parameters.

[0028] For example, the operation and maintenance management module 104 is used to calculate the required operating states and action sequences of each piece of equipment according to the requirements of the production process parameters, and generate an instruction set containing detailed control information. The operation and maintenance management instructions are sent to the controllers of each piece of equipment through the industrial communication network. The equipment controllers parse the received instruction content and drive the actuators to complete the corresponding actions.

[0029] The dough mixing equipment adjusts the amount of water and mixing parameters according to the operation and maintenance management instructions to ensure that the flour and water are fully mixed to form a uniform dough. The softness and elasticity of the dough directly affect the subsequent processing and the taste of the final product.

[0030] The maturation equipment allows the dough to rest in a suitable environment for the time set by the instructions. During this process, the gluten network inside the dough is fully formed and stabilized, improving the dough's processability.

[0031] The rolling equipment adjusts the distance between the rollers according to the instructions, and rolls the dough into strips that meet the thickness requirements step by step. The control of rolling force and rolling speed affects the uniformity of the strips and the extensibility of the gluten.

[0032] The drying equipment controls the distribution of temperature and humidity fields according to instructions, so that the moisture in the noodles migrates outward and evaporates at an appropriate rate, avoiding the phenomenon of surface hardening and internal moisture retention caused by drying too quickly.

[0033] The operation and maintenance management module 104 continuously collects equipment operation feedback information during the control process, compares the actual operating parameters with the target parameters, and makes timely adjustments when deviations are found to achieve a closed-loop management system, ensuring that the production process is stable and controllable and that the quality of noodle products meets the expected standards.

[0034] This application provides an IoT-based operation and maintenance management system for noodle production line equipment. The system includes: a raw material management module, an environmental monitoring module, a parameter optimization module, and an operation and maintenance management module. The raw material management module collects raw material data for noodle production, inputs the raw material data into a pre-set dual-branch LSTM model to predict raw material characteristics for future periods, and generates characteristic prediction results. The environmental monitoring module collects environmental data. The parameter optimization module constructs a multi-objective problem set for optimizing noodle quality based on the characteristic prediction results, environmental data, and a pre-set process database. It then solves the multi-objective problem set using a pre-set second-generation non-dominated sorting genetic algorithm to obtain production process parameters. The operation and maintenance management module generates operation and maintenance management instructions based on the production process parameters. In the above method, data is automatically collected through the raw material management module and the environmental monitoring module, and the characteristics of raw materials are intelligently predicted through a dual-branch LSTM model. Based on the prediction results, a multi-objective problem is constructed through the parameter optimization module and solved automatically using a genetic algorithm. This effectively transforms the experience-based decision-making of technicians into data-driven optimization algorithm decision-making, reducing the degree of manual intervention. Finally, a management closed loop is formed through the operation and maintenance management module, realizing the equipment operation and maintenance process from data perception to intelligent decision-making and then to automatic execution. This improves production management efficiency while maintaining product quality consistency.

[0035] To more clearly illustrate the technical solution of this application, the technical solution of this application will be described below through specific embodiments. It should be noted that the specific embodiments are used to expand the description of the technical solution of this application, and are not intended to limit this application.

[0036] In some embodiments, the raw material data includes: moisture content, ash content, pH value, amount of additives, storage temperature, storage humidity, powder parameters, and stretching parameters.

[0037] In some embodiments, such as Figure 2As shown, the preset dual-branch LSTM model includes: a feature extraction layer, a main branch, a secondary branch, and multiple parallel fully connected layers. When the raw material management module 101 is used to execute the preset dual-branch LSTM model to predict the characteristics of raw materials in future periods and generate characteristic prediction results, it specifically executes: S21-S27.

[0038] S21. Extract local features from the raw material data and perform dimensionality reduction to generate raw material feature vectors.

[0039] For example, after processing the raw material data through a feature extraction layer, a convolutional neural network is used to perform local pattern recognition on multi-dimensional information such as moisture content, ash content, pH value, additive dosage, storage temperature, storage humidity, powder parameters, and stretching parameters. The convolutional kernel slides across the time window of the raw material data, capturing the changing patterns and interrelationships of raw material indicators between adjacent time points. The local features extracted by the convolution operation contain the fluctuation characteristics and combination patterns of the raw material data within a short time scale. These features have high dimensionality and need to be dimensionality reduced to reduce computational complexity and avoid overfitting. Dimensionality reduction uses principal component analysis or an autoencoder structure to map high-dimensional local features to a low-dimensional space while retaining the most discriminative information components, resulting in a raw material feature vector. The dimensionality of the raw material feature vector is typically controlled between thirty and fifty dimensions.

[0040] S22. Input the raw material feature vector into the main branch of the dual-branch LSTM model, extract bidirectional temporal dependency features through the forward LSTM layer and the backward LSTM layer, and generate the first hidden state vector set.

[0041] For example, the raw material feature vector is fed into the main branch of a dual-branch LSTM model for temporal feature learning. The main branch includes a forward LSTM layer that processes the raw material feature vector sequence in forward chronological order. At each time step, the hidden state of the forward LSTM layer considers both the current input features and the previous hidden state, selectively retaining or forgetting historical information through the synergistic effect of forget gates, input gates, and output gates. The backward LSTM layer processes the raw material feature vector sequence in reverse chronological order, transmitting information from future to past times and capturing the feedback effect of the subsequent evolution of raw material characteristics on the current state. The forward and backward LSTM layers each output their respective hidden state vectors. The forward hidden state vector contains temporal dependency information from the past to the present, while the backward hidden state vector contains temporal dependency information from the future to the present. The bidirectional temporal dependency feature is formed by concatenating the forward and backward hidden state vectors along the feature dimension. The dimension of the concatenated vector is twice that of the unidirectional hidden state vector, and the generated first hidden state vector set contains the complete dependency relationship of raw material characteristics along the time axis.

[0042] S23. Input the raw material feature vector into the main branch of the dual-branch LSTM model, perform bidirectional time series modeling through forward LSTM layer and backward LSTM layer, extract long-term dependency features in the forward and backward directions, and generate the first hidden vector.

[0043] For example, the raw material feature vector is modeled in the main branch using a deep bidirectional temporal modeling process employing both forward and backward LSTM layers. This modeling process not only focuses on short-term correlations between adjacent time points but also emphasizes uncovering long-term dependency features spanning extended time intervals. The forward LSTM layer, through its internal memory unit mechanism, retains important early information during information transmission across multiple time steps. Even after dozens or hundreds of time steps, key raw material characteristic changes are effectively preserved and utilized. The backward LSTM layer also possesses long-term memory capabilities, propagating backward from the end of the time series to pass forward information on future raw material characteristic trends, allowing the model to reference future evolution directions when predicting current raw material characteristics. Positive long-term dependency features reflect the cumulative effect and gradual change trend of raw material characteristics from history to the present, while negative long-term dependency features embody the reverse constraints and trend traction of raw material characteristics from the future to the present. The long-term dependency features extracted from both directions are fused to generate the first hidden vector, which integrates deep semantic information across the two temporal dimensions, providing a rich feature foundation for accurate prediction of raw material characteristics.

[0044] S24. Generate dynamic attention weights based on storage temperature and storage humidity, and use the dynamic attention weights to enhance the features of the first hidden vector set to generate the second hidden vector.

[0045] For example, storage temperature and humidity, as important environmental factors affecting the evolution of raw material characteristics, are used to calculate dynamic attention weights to enhance the model's focus on features at critical moments. The generation of dynamic attention weights is implemented through a small neural network module. This module takes the numerical inputs of storage temperature and humidity, processes them through a fully connected layer and a nonlinear activation function, and outputs an attention score vector with the same dimension as the time series. The attention score vector is then converted into a probability distribution form of dynamic attention weights through softmax normalization. The magnitude of the weight indicates the importance of the raw material characteristic information at the corresponding moment in the overall prediction. When the storage temperature or humidity increases, the dynamic attention weight at the corresponding moment increases accordingly, indicating that changes in raw material characteristics under unfavorable storage conditions are more noteworthy. When using dynamic attention weights to enhance the features of the first hidden state vector set, the hidden state vector at each moment is multiplied element-wise with its corresponding attention weight. Moments with larger weights have their feature information amplified, while moments with smaller weights have their feature information suppressed. After attention weighting, the weighted hidden state vectors at all moments are summed or average pooled to generate the second hidden vector.

[0046] S25. Based on the fully connected layer, the first characteristic prediction set corresponding to the second hidden vector set is mapped respectively. The first characteristic prediction set includes: water absorption rate prediction value, formation time prediction value, stability prediction value and tensile resistance prediction value.

[0047] For example, the second hidden vector, as a feature representation containing rich information about the raw material properties, is input into multiple parallel fully connected layers for property index prediction. Each fully connected layer is specifically responsible for predicting a particular raw material property index. The fully connected layer maps the second hidden vector to the corresponding prediction value space through a linear transformation of the weight matrix and bias vector. The mapping process may involve one or more layers of neural structures, and nonlinear transformation capabilities are introduced between layers through activation functions such as ReLU or tanh. The fully connected layer predicting water absorption rate outputs a predicted water absorption rate value, which reflects the percentage of water required by the flour mass to absorb water and reach the optimal kneading state, typically ranging from 50% to 70%. The fully connected layer predicting formation time outputs a predicted formation time value, which represents the length of time required for the dough to form a stable network structure from the start of mixing, generally in the range of three to fifteen minutes. The fully connected layer predicting stability outputs a predicted stability value, which measures the duration for which the dough maintains stable performance under continuous mixing or mechanical action; a higher value indicates that the dough is less prone to damage. The fully connected layer, which predicts tensile resistance, outputs a predicted value for tensile resistance. This value reflects the mechanical strength of the dough against tensile deformation, expressed in Newtons or centineuts. These four predicted values ​​together constitute the first characteristic prediction set.

[0048] S26. Input the first feature prediction set into the secondary branch of the dual-branch LSTM model, perform adversarial gradient calculation and backpropagation, and generate the second feature prediction set.

[0049] For example, the first feature prediction set is input into the secondary branch of a dual-branch LSTM model for adversarial training. The secondary branch is designed to improve the robustness and generalization ability of the main branch's prediction results by introducing an adversarial mechanism. The secondary branch contains neural network layers with similar structures to the main branch but independent parameters. After receiving the first feature prediction set as input, the secondary branch attempts to add carefully designed perturbation signals to the first feature prediction set. The direction and amplitude of these perturbation signals are determined by an adversarial gradient calculation method. Adversarial gradient calculation finds the perturbation direction that increases the loss function the most by calculating the gradient of the prediction loss function relative to the input feature prediction values. A small perturbation is applied in this direction to generate adversarial examples. The secondary branch reprocesses and corrects the feature prediction values ​​with added adversarial perturbations, and uses a backpropagation algorithm to transmit the adversarial loss signal back to the network parameters for updating, enabling the model to learn to resist the influence of such adversarial perturbations. After adversarial training, the secondary branch outputs a corrected and enhanced second feature prediction set. This prediction set may differ numerically from the first feature prediction set, but it has stronger anti-interference capabilities and adaptability to different data distributions.

[0050] S27. The first characteristic prediction set and the second characteristic prediction set are weighted and fused, and constrained and calibrated based on the statistical distribution of historical raw material characteristic data to generate characteristic prediction results.

[0051] For example, the first and second characteristic prediction sets represent the prediction results of the main and secondary branches, respectively. These two sets are integrated through a weighted fusion method to fully utilize their respective advantages. The weighted fusion operation assigns a fusion weight coefficient to each prediction set. The weight coefficient is determined based on the historical prediction accuracy of the two branches on the validation dataset; branches with smaller prediction errors are assigned larger weights, and those with larger prediction errors are assigned smaller weights. The fusion formula multiplies each indicator value in the first characteristic prediction set by its corresponding weight coefficient, and multiplies each indicator value in the second characteristic prediction set by its corresponding weight coefficient. The sum of the two weighted results yields the preliminary fused characteristic prediction value. The preliminary fused characteristic prediction value may exceed the physically reasonable range or deviate significantly from historical statistical patterns, requiring constraint calibration based on the statistical distribution of historical raw material characteristic data. The statistical distribution of historical raw material characteristic data is obtained through statistical analysis of a large amount of historical data, including the mean, standard deviation, quantiles, and other statistical measures of each characteristic indicator, as well as the correlation constraints between the indicators. The constraint calibration process checks whether the fused predicted values ​​fall within a reasonable range of the statistical distribution. For predicted values ​​that exceed the normal range, truncation or soft constraints are used to bring them back to the reliable range. For predicted combinations that violate physical or logical constraints, adjustments and corrections are made to obtain characteristic prediction results, which are then used as the output of the raw material management module 101101.

[0052] In some embodiments, such as Figure 3 As shown, the characteristic prediction results include: water absorption rate prediction value, formation time prediction value, stability prediction value and tensile resistance prediction value. The parameter optimization module 103 is used to execute the following steps when constructing a multi-objective problem set for optimizing noodle quality based on the characteristic prediction results, environmental data and preset process database: S31-S33.

[0053] S31. Based on the predicted values ​​of water absorption rate, formation time, stability, tensile resistance, and environmental data, the process parameter range is obtained by matching from the preset process database. The process parameter range includes: dough water content, maturation time, rolling ratio, and drying rate.

[0054] For example, the pre-defined process database uses a multi-dimensional index structure to store thousands of historical production records. Each record is associated with a specific combination of raw material characteristics, environmental conditions, corresponding process parameter configurations, and product quality evaluation results. The matching process calculates the Euclidean distance or cosine similarity between the currently input characteristic prediction value and the historical records in the database, selecting the top ten historical cases with the highest similarity as reference samples. Workshop temperature and humidity in the environmental data serve as important correction factors in the similarity calculation; when environmental conditions differ significantly, the matching weight of the historical case is reduced.

[0055] Based on the selected reference samples, statistical analysis was performed on the values ​​of the process parameters. The minimum and maximum values ​​of each parameter were calculated to determine the search boundary. The range of process parameters included: the range of water content for kneading, reflecting the mass range of water required to add per 100 kg of flour; the range of maturation time, defining the range of resting and relaxing time for the dough; the range of rolling ratio, specifying the range of thickness compression ratio of the dough in each rolling process; and the range of drying rate, limiting the range of the rate range of moisture evaporation in the noodles.

[0056] S32. Define the process parameter range as the search space of decision variables for a multi-objective optimization problem, where the amount of water used for mixing, the maturation time, and the drying rate are continuous variables, and the rolling ratio sequence is an ordered set of discrete variables.

[0057] For example, the decision variable search space is mathematically represented as a multidimensional hypercube or hybrid space, where the continuous variable dimension corresponds to the real number interval, and the discrete variable dimension corresponds to a finite set. The amount of water used for kneading, as a continuous variable, can take any real value within a preset range (30kg / 100kg to 40kg / 100kg), reflecting the need for continuous and refined control of water addition. The maturation time, as a continuous variable, varies within the range of 10 to 60 minutes, with a time granularity accurate to the second to meet the differentiated requirements of maturation degree for different raw material characteristics. The drying rate, as a continuous variable, is adjustable between 0.5kg / h and 35kg / h; the continuous adjustability of the rate allows the drying process to dynamically respond according to the moisture content of the noodles. The rolling ratio sequence, as an ordered discrete variable set, consists of the compression ratios of 3 to 5 rolling stages. The rolling ratio of each stage is selected from a predefined set of discrete values, and the rolling ratios of each stage must be arranged in descending order to conform to the process logic of progressively thinning the dough.

[0058] S33. Within the search space of decision variables, construct a multi-objective optimization problem set using the sensory score of noodles, the cooking loss rate, and the comprehensive index of textural properties as optimization objectives.

[0059] For example, the sensory score of noodles serves as an indicator for comprehensively evaluating the sensory quality of a product, including its appearance, aroma, texture, and taste. Professional reviewers score the noodles according to a standard scoring sheet, with a maximum score typically set at 100 points. Higher scores indicate better sensory quality. The cooking loss rate reflects the percentage of mass lost due to the dissolution of soluble substances and noodle breakage during boiling. A lower rate indicates better resistance to overcooking and more complete retention of nutrients. The cooking loss rate for high-quality noodles should be controlled below 8%. The comprehensive textural properties index integrates multiple physical properties, including noodle hardness (measuring the noodle's resistance to biting pressure), elasticity (characterizing the noodle's ability to recover its original shape after deformation under stress), chewiness (describing the effort required to chew the noodle to a suitable swallowing state), and cohesiveness (reflecting the density of the noodle's internal structure). These parameters are objectively measured using a texture analyzer and a comprehensive score is calculated according to a preset weighting formula. The multi-objective optimization problem set takes into account the above three optimization objectives simultaneously. It seeks a combination of process parameters that maximizes the sensory score of noodles, minimizes the cooking loss rate, and optimizes the comprehensive index of texture characteristics within the decision variable search space. There are mutual constraints among the three objectives, and a balance solution needs to be found through Pareto optimization.

[0060] In some embodiments, when the parameter optimization module 103 is used to solve the multi-objective problem set by a preset second-generation non-dominated sorting genetic algorithm to obtain production process parameters, it is specifically used to execute: S41-S46.

[0061] S41. Generate a parent population from the set of multi-objective optimization problems using a hybrid coding strategy, wherein the amount of water used for mixing, the maturation time, and the drying rate are encoded using floating-point numbers, and the rolling ratio sequence is encoded using integer permutation.

[0062] For example, the three continuous variables—mixing water quantity, maturation time, and drying rate—are encoded using floating-point numbers. Double-precision floating-point numbers directly represent the actual values ​​of these variables within their ranges. Floating-point encoding maintains the precision advantage of continuous variables and avoids the precision loss caused by discretization required by binary encoding. The rolling ratio sequence, as an ordered discrete variable, is encoded using integer permutations. Optional rolling schemes are numbered with integer indices; for example, 3-stage rolling schemes are numbered 1 to 20, 4-stage rolling schemes are numbered 21 to 50, and 5-stage rolling schemes are numbered 51 to 80. Each integer corresponds to a specific rolling ratio combination sequence. The parent population is generated through random sampling within the decision variable search space. For continuous variables, floating-point values ​​are uniformly and randomly generated within their ranges. For discrete variables, index values ​​are randomly selected from their set of optional integers. The population size is typically set to 100 to 200 individuals. The genotype of each individual is formed by concatenating the floating-point encoded segment of the continuous variable and the integer encoded segment of the discrete variable, representing a complete set of process parameter configurations.

[0063] S42. Calculate the first comprehensive cost of each individual in the parent population according to the preset total comprehensive cost function.

[0064] For example, the process parameter scheme represented by each individual in the parent population needs to be quantitatively evaluated through a cost function. The preset total comprehensive cost function comprehensively considers the economic and quality indicators of the entire noodle production process. The total comprehensive cost function divides the production cycle into several time periods. The number of time periods is determined according to the production cycle and control precision requirements, usually 10 to 20 time periods covering the complete process from dough mixing to packaging. The cost of each time period is calculated by accumulating four components, including: electricity cost, calculated by multiplying the unit electricity price by the total power consumption of all equipment such as dough mixing equipment, rolling equipment, and drying equipment in that time period; raw material loss cost, calculated by multiplying the unit mass raw material cost coefficient by the mass of raw material loss caused by moisture evaporation, dough scraps, and trimming of scraps in that time period; quality deviation penalty cost, calculated by multiplying the product quality deviation penalty cost coefficient by the absolute value of the deviation of the sensory score, cooking loss rate, and texture characteristic comprehensive score of the noodles produced in that time period from the target value; and equipment adjustment cost, calculated by summing the product of the unit adjustment cost of all adjustable equipment and the absolute value of the process parameter adjustment of that equipment in that time period. The costs of each time period are summed up over the entire production cycle, and the first comprehensive cost is used as a comprehensive indicator to measure the economic efficiency and feasibility of the process scheme.

[0065] S43. Apply preset process constraints to the population and quantify the degree of constraint violation into penalty costs, which are then added to the first comprehensive cost to obtain the second comprehensive cost.

[0066] For example, some individuals in the parent population may generate process parameter combinations that do not conform to actual production constraints due to random generation. Pre-defined process constraints are used to identify and penalize these infeasible solutions. Process constraints include physical constraints such as the amount of water used in kneading dough must not exceed the maximum water absorption capacity of the flour to prevent the dough from becoming too thin to form; temporal constraints such as the maturation time must reach the minimum duration required for the full formation of the gluten network; equipment constraints such as the drying rate must not exceed the maximum processing capacity of the drying equipment; and product constraints such as the moisture content of the final noodles must be controlled within the range that meets food safety standards. The degree of constraint violation is quantified by calculating the deviation between the actual parameter values ​​and the constraint boundaries. For example, when the amount of water used in kneading dough exceeds the allowable upper limit, the excess value is multiplied by the corresponding penalty coefficient to form the penalty cost of that constraint. The penalty costs of multiple constraints are linearly superimposed to form the total penalty cost. The penalty coefficient needs to be set large enough to ensure that infeasible solutions are at a disadvantage in the optimization process, but it cannot be too large to completely exclude potential high-quality solutions close to the feasible region boundary. The total penalty cost is added to the first comprehensive cost, and the second comprehensive cost reflects both the economy and feasibility of the solution.

[0067] S44. Based on the second comprehensive cost, perform fast non-dominated sorting and crowding calculation on the parent population, and perform binary tournament selection based on the non-dominated level and crowding to generate a mating pool.

[0068] For example, the second comprehensive cost, as a metric for individual fitness, provides the foundational data for performing fast non-dominated sorting. The fast non-dominated sorting algorithm traverses all individuals in the parent population, calculating for each individual how many other individuals dominate it and how many other individuals it dominates. Individual A is said to dominate individual B when it is not inferior to individual B on all optimization objectives and is superior to individual B on at least one objective. Individuals not dominated by any other individual constitute the first non-dominated level. After removing individuals from the first level, the search continues among the remaining individuals to find non-dominated individuals to form the second non-dominated level, and so on, dividing the entire population into multiple levels. Crowding is calculated for individuals within each non-dominated level. Crowding distance measures the distance between an individual and its neighbors in the objective space. The calculation method involves sorting individuals within the level according to each objective function value. Individuals at the two ends of the sorted hierarchy are assigned infinite crowding, while the crowding of an individual in the middle is equal to the normalized sum of the differences in objective values ​​between its immediate and adjacent individuals. The binary tournament selection process randomly selects two individuals from the parent population each time to compete. The non-dominant level of the two individuals is compared; the individual with the lower level wins. If the two individuals have the same level, their crowding distance is compared; the individual with the higher crowding distance wins. The winning individual is selected for the mating pool. The tournament selection is repeated until the mating pool reaches a preset size, containing individuals with excellent overall performance and even distribution within the population.

[0069] S45. Perform simulated binary crossover and polynomial mutation operations sequentially on the mating pool to generate offspring populations. Then merge the offspring populations with the parent populations and perform a new round of iterations until the maximum number of generations is reached, and output the Pareto optimal solution set.

[0070] For example, individuals in the mating pool generate new offspring through crossover and mutation operations. The simulated binary crossover operation is specifically designed for continuous variables encoded with floating-point numbers. This operation randomly selects two parent individuals from the mating pool as crossover targets. For the continuous variable values ​​at corresponding positions of the two parent individuals, the values ​​of the two offspring at those positions are calculated according to the simulated binary crossover formula. This crossover formula ensures that offspring values ​​are generated near the parent values ​​and is influenced by the diffusion degree controlled by the distribution exponent parameter. The multinomial mutation operation randomly perturbs the genes of the crossovered individuals. For continuous variables, a random offset following a multinomial distribution is added to the current value. The magnitude of the offset is controlled by the mutation distribution exponent and mutation probability parameters. For discrete variables, they are randomly replaced with other optional integer index values ​​with a certain probability. The mutation operation increases the diversity of the population and prevents the algorithm from prematurely converging to a local optimum. Crossover and mutation operations produce offspring populations of the same size as the parent populations. The two populations are merged to form a mixed population that doubles in size. Fast non-dominated sorting and crowding calculations are then performed on the mixed population again. Following the principle of prioritizing non-dominated levels and then crowding, the top half of the individuals in the mixed population are selected as the parents for the next generation, completing one evolutionary iteration. This iterative process continues until the preset maximum number of generations is reached or the optimal solution for multiple consecutive generations has not significantly improved, and the Pareto optimal solution set contains all individuals at the first non-dominated level in the current population.

[0071] S46. Select the second comprehensive cost optimal solution set from the Pareto optimal solution set, and decode the second comprehensive cost optimal solution set into production process parameters.

[0072] For example, the Pareto optimal solution set contains a series of process parameter schemes that achieve different trade-offs among multiple optimization objectives. These schemes are not absolutely superior or inferior; each has its applicable scenarios. The second comprehensive cost, as a single evaluation index that comprehensively considers economy, quality, and feasibility, can be used to select the scheme that best meets current production needs from the Pareto optimal solution set. The selection process traverses all individuals in the Pareto optimal solution set, comparing the second comprehensive cost values ​​of each individual. The individual with the lowest cost value indicates that it has the best comprehensive economic benefits and quality assurance capability under the premise of meeting multi-objective optimization requirements. The genotype of the individual corresponding to the solution set with the optimal second comprehensive cost needs to be decoded into actual process parameter values ​​to guide production. For floating-point encoded parameters such as dough water quantity, maturation time, and drying rate, the floating-point values ​​in their encoded segments are directly read as production process parameters, with units of kilograms per 100 kilograms of flour, minutes, and kilograms per hour, respectively. For integer-encoded rolling ratio sequences, a predefined rolling scheme mapping table is consulted based on the encoded value. The integer index is then converted into a specific sequence of rolling ratio values ​​for each stage. For example, index value 35 corresponds to rolling ratios of 0.45, 0.3, 0.2, and 0.12 for a 4-stage rolling scheme. The production process parameters are then passed to the operation and maintenance management module 104 as the output of the optimization module.

[0073] In some embodiments, the preset total comprehensive cost function is: ; Where F is the first comprehensive cost, T is the total number of time periods divided into the entire noodle production process, C1 is the unit electricity price for time period t, P1 is the total electricity consumption in time period t during the dough mixing, cooking, rolling, and drying stages, C2 is the cost coefficient per unit mass of raw materials, L1 is the raw material loss caused by moisture evaporation, dough scraps, etc. in time period t, C3 is the unit penalty cost coefficient for product quality deviation from the standard, D is the absolute value of the deviation of the sensory score, cooking loss rate, and textural characteristics comprehensive score of the noodles produced in time period t from the target value, K is the number of adjustable key process equipment in the production process, and C4 is the cost per unit adjustment action of the kth equipment. This represents the absolute value of the process parameter adjustment for the k-th device during time period t.

[0074] For example, by constructing a total comprehensive cost function comprising four components—electricity cost, raw material loss cost, quality deviation penalty cost, and equipment adjustment cost—a quantitative assessment of the economic indicators of the entire noodle production process is achieved. This cost function divides the production cycle into T time periods for segmented cumulative calculation. The product of C1 and P1 quantifies the energy consumption expenditure at each stage of dough preparation, cooking, rolling, and drying. The product of C2 and L1 quantifies the costs resulting from raw material losses such as moisture evaporation and dough scraps. The product of C3 and D transforms deviations in the noodle sensory score, cooking loss rate, and overall textural characteristic score into comparable economic penalties. The adjustment cost C4 for K key process equipment is calculated using the absolute value of parameter adjustments. The product summation quantifies the wear and control costs caused by frequent equipment adjustments. This cost function serves as the fitness evaluation standard for the genetic algorithm, enabling the parameter optimization module 103 to find the optimal balance between product quality, energy consumption, raw material utilization, and equipment stability. This avoids the local optimization problems caused by traditional methods that only focus on a single objective. Through multi-dimensional comprehensive evaluation, it achieves the global optimal configuration of production process parameters, reducing total production costs while ensuring product quality stability, and improving the intelligent decision-making level and economic benefits of noodle production line equipment operation and maintenance management.

[0075] In some embodiments, the compression ratio sequence is a set of compression ratios of multiple preset stages that decrease sequentially from large to small, wherein the number of preset stages is 3-5, the compression ratio of the first stage is 0.4-0.5, and the compression ratio of the last stage is 0.1-0.15.

[0076] For example, by setting the rolling ratio sequence as a set of multiple preset stages of compression ratios that decrease sequentially from large to small, a progressive pressing process for dough thickness is achieved. The number of preset stages is controlled within the range of 3 to 5, avoiding the problems of uneven damage to the internal structure of the dough due to too few rolling passes and increased equipment investment and reduced production efficiency due to too many rolling passes. The compression ratio range of the first stage is set to 0.4 to 0.5 to ensure that the dough obtains a large thickness compression amount during the initial rolling to quickly form the initial dough structure, while avoiding excessive compression in a single step that could cause the gluten network to break. The compression ratio range of the last stage is set to 0.1 to 0.15 to achieve fine adjustment of the dough thickness to meet product specifications. The progressively decreasing rolling ratio sequence allows the gluten network inside the dough to be fully extended and evenly distributed, improving the tensile strength and toughness of the noodles, and improving the texture and taste quality of the product.

[0077] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An IoT-based operation and maintenance management system for noodle production line equipment, characterized in that, The method includes: The raw material management module is used to collect raw material data for noodle production, input the raw material data into a preset dual-branch LSTM model to predict the raw material characteristics in future periods, and generate characteristic prediction results. The environmental monitoring module is used to collect environmental data; The parameter optimization module is used to construct a multi-objective problem set for optimizing noodle quality based on the characteristic prediction results, the environmental data, and the preset process database, and to solve the multi-objective problem set using a preset second-generation non-dominated sorting genetic algorithm to obtain production process parameters. The operation and maintenance management module is used to generate operation and maintenance management instructions based on the production process parameters.

2. The IoT-based noodle production line equipment operation and maintenance management system as described in claim 1, characterized in that, The raw material data includes: moisture content, ash content, pH value, amount of additives, storage temperature, storage humidity, powder parameters, and stretching parameters.

3. The IoT-based noodle production line equipment operation and maintenance management system as described in claim 2, characterized in that, The pre-set dual-branch LSTM model includes: a feature extraction layer, a main branch, a secondary branch, and multiple parallel fully connected layers. When the raw material management module executes the process of inputting the raw material data into the pre-set dual-branch LSTM model to predict the raw material characteristics for future periods and generate characteristic prediction results, it specifically performs the following: Extract local features from the raw material data and perform dimensionality reduction to generate a raw material feature vector; The raw material feature vector is input into the main branch of the dual-branch LSTM model, and bidirectional temporal dependency features are extracted through the forward LSTM layer and the backward LSTM layer to generate the first hidden state vector set. The raw material feature vector is input into the main branch of the dual-branch LSTM model. Bidirectional temporal modeling is performed through forward LSTM and backward LSTM layers to extract long-term dependency features in both directions and generate the first hidden vector. Dynamic attention weights are generated based on the storage temperature and the storage humidity. The dynamic attention weights are then used to enhance the features of the first hidden vector set to generate a second hidden vector. Based on the fully connected layer, the first characteristic prediction set corresponding to the second hidden vector set is mapped respectively. The first characteristic prediction set includes: water absorption rate prediction value, formation time prediction value, stability prediction value and tensile resistance prediction value. The first feature prediction set is input into the secondary branch of the dual-branch LSTM model to perform adversarial gradient calculation and backpropagation, thereby generating the second feature prediction set. The first characteristic prediction set and the second characteristic prediction set are weighted and fused, and constrained and calibrated based on the statistical distribution of historical raw material characteristic data to generate the characteristic prediction result.

4. The IoT-based noodle production line equipment operation and maintenance management system as described in claim 1, characterized in that, The characteristic prediction results include: predicted water absorption rate, predicted formation time, predicted stability, and predicted tensile resistance. The parameter optimization module is used to execute the following when constructing a multi-objective problem set for optimizing noodle quality based on the characteristic prediction results, the environmental data, and a preset process database: Based on the predicted water absorption rate, the predicted formation time, the predicted stability, the predicted tensile resistance, and the environmental data, a range of process parameters is obtained from the preset process database. The range of process parameters includes: dough water content, maturation time, rolling ratio, and drying rate. The process parameter range is defined as the decision variable search space for a multi-objective optimization problem, wherein the amount of water used for mixing, the maturation time, and the drying rate are continuous variables, and the rolling ratio sequence is an ordered discrete variable set. Within the search space of decision variables, a multi-objective optimization problem set is constructed using the sensory score of noodles, the cooking loss rate, and the comprehensive index of textural properties as optimization objectives.

5. The IoT-based noodle production line equipment operation and maintenance management system as described in claim 4, characterized in that, The parameter optimization module is used to execute the following steps when solving the multi-objective problem set using a preset second-generation non-dominated sorting genetic algorithm to obtain production process parameters: The set of multi-objective optimization problems is used to generate a parent population through a hybrid coding strategy, wherein the amount of water used for mixing, the maturation time, and the drying rate are encoded using floating-point numbers, and the rolling ratio sequence is encoded using integer permutation. Calculate the first comprehensive cost of each individual in the parent population based on the preset total comprehensive cost function; Preset process constraints are imposed on the population, and the degree of violation of the constraints is quantified as a penalty cost, which is added to the first comprehensive cost to obtain the second comprehensive cost; Based on the second comprehensive cost, fast non-dominated sorting and crowding calculation are performed on the parent population, and binary tournament selection is performed according to the non-dominated level and crowding to generate a mating pool. Simulated binary crossover and polynomial mutation operations are sequentially performed on the mating pool to generate offspring populations. The offspring populations are then merged with the parent populations for a new round of iteration until the maximum number of generations is reached, and the Pareto optimal solution set is output. Select the second comprehensive cost optimal solution set from the Pareto optimal solution set, and decode the second comprehensive cost optimal solution set into the production process parameters.

6. The IoT-based noodle production line equipment operation and maintenance management system as described in claim 5, characterized in that, The preset total comprehensive cost function is: ； Where F is the first comprehensive cost, T is the total number of time periods divided into the entire noodle production process; C1 is the unit electricity price for time period t; P1 is the total electricity consumption in time period t during the dough mixing, cooking, rolling, and drying stages; C2 is the cost coefficient per unit mass of raw materials; L1 is the raw material loss mass caused by moisture evaporation, dough scraps, etc., in time period t; C3 is the unit penalty cost coefficient for product quality deviation from the standard; D is the absolute value of the deviation of the sensory score, cooking loss rate, and comprehensive score of the texture characteristics of the noodles produced in time period t from the target value; K is the number of adjustable key process equipment in the production process; and C4 is the cost per unit adjustment action of the kth piece of equipment. This represents the absolute value of the process parameter adjustment for the k-th device during time period t.

7. The IoT-based noodle production line equipment operation and maintenance management system as described in claim 4, characterized in that, The compression ratio sequence is a set of compression ratios in multiple preset stages that decrease sequentially from large to small. The number of preset stages is 3-5, the compression ratio range of the first stage is 0.4-0.5, and the compression ratio range of the last stage is 0.1-0.15.

Images