A method and system for real-time management of a quick-frozen food production workshop based on the Internet of Things

Abstract

This invention relates to the field of Internet of Things (IoT) control technology, specifically to a real-time management method and system for a quick-frozen food production workshop based on IoT. The method includes: collecting quick-frozen food production parameters via IoT and preprocessing these parameters; extracting a core temperature state vector reflecting the core temperature change and freezing progress characteristics of the quick-frozen food based on the preprocessed production parameters and an adaptive core temperature compensation algorithm for fat content; and predicting the core temperature of the quick-frozen food at the end of a specified freezing time using the core temperature state vector and a long short-term memory network model incorporating spray state masking weights. This invention achieves high-precision perception, intelligent prediction, and closed-loop control of the core temperature state throughout the quick-freezing process through IoT control, effectively overcoming the problem of unstable freezing quality caused by differences in fat content, spray interference, and fluctuations in operating conditions in traditional quick-freezing processes.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) control technology, specifically to a method and system for real-time management of a frozen food production workshop based on IoT. Background Technology

[0002] In the production of frozen foods, the freezing quality is highly dependent on the precise control of the core temperature. Traditional management methods rely mainly on manual experience or fixed process parameters, making it difficult to dynamically adapt to the influence of different food components (especially differences in fat content), equipment operating condition fluctuations, and environmental interference. High-fat foods, due to their complex phase transition characteristics and low latent heat release, exhibit nonlinear, wide-temperature-range thermal response behavior during freezing. This means that traditional monitoring methods based on single temperature sensors cannot accurately reflect the true freezing state, easily leading to underfreezing or overfreezing, affecting product quality and safety. Furthermore, existing quick-freezing equipment (such as liquid nitrogen spraying) mostly uses open-loop control, lacking linkage with real-time production data, and cannot achieve dynamic optimization of operating parameters. Although IoT technology has been gradually applied in the industrial field, existing systems have not yet effectively integrated food thermophysical mechanisms, time-series sensing, and intelligent prediction models, making it difficult to support refined closed-loop control. Summary of the Invention

[0003] The purpose of this invention is to address the problems existing in the background technology by proposing a real-time management method and system for quick-frozen food production workshops based on the Internet of Things.

[0004] The technical solution of this invention: A real-time management method for a frozen food production workshop based on the Internet of Things, comprising: S1. Collect frozen food production parameters through the Internet of Things and preprocess the frozen food production parameters. S2. Based on the pre-processed production parameters of quick-frozen foods, a core temperature state vector reflecting the core temperature change and freezing progress characteristics of quick-frozen foods is extracted using a fat content adaptive core temperature compensation algorithm. S3. Using the core temperature state vector, based on a long short-term memory network model with introduced spray state masking weights, the core temperature of quick-frozen food at the end of the specified freezing time is predicted, and the core temperature state evaluation result is output. S4. Based on the core temperature status assessment results, the operating parameters of quick-freezing equipment in the quick-freezing food production process are managed in real time through the Internet of Things.

[0005] As a further improvement to this technical solution, in S1, the production parameters for quick-frozen food include at least the core temperature of the food, the surface temperature of the food, and the operating status parameters of the conveyor belt.

[0006] As a further improvement to this technical solution, in step S2, the core temperature state vector reflecting the core temperature change and freezing progress characteristics of quick-frozen food is extracted based on the fat content adaptive core temperature compensation algorithm, including the following steps: S2.1 Determine the effective sampling range of the core temperature of the pretreated food based on the pretreated conveyor belt operating status parameters; S2.2 Within the effective sampling interval, perform time window statistical processing on the core temperature of food at multiple consecutive moments, and correct the core temperature parameters of food based on the fat content adaptive core temperature compensation algorithm, and calculate the corrected core temperature parameters of food at the current moment. S2.3 Calculate the freezing progress characteristic parameters of quick-frozen food at the current moment based on the corrected core temperature parameters of food; S2.4. Based on the corrected food core temperature parameters within the effective sampling interval, calculate the trend parameters of the corrected food core temperature parameters over time. S2.5 Combine the corrected food core temperature parameters, trend parameters, food surface temperature and freezing progress characteristic parameters to construct a core temperature state vector that reflects the core temperature change and freezing progress characteristics of quick-frozen food.

[0007] As a further improvement to this technical solution, in step S2.2, the food core temperature parameters are corrected based on the fat content adaptive core temperature compensation algorithm, and the corrected food core temperature parameters at the current moment are calculated, including the following steps: S2.21. Within the effective sampling interval, for continuous... The core food temperature data at each time point are statistically processed using a time window to calculate the original core food temperature statistical value corresponding to the current time point. ; S2.22. Obtain the fat content characteristic parameters of quick-frozen foods by using a dynamic characteristic parameter extraction method based on temperature-sensitive fat response; S2.23. Based on the characteristic parameters of fat content, construct an adaptive compensation coefficient for fat content. ; Among them, the adaptive compensation coefficient of fat content Including static basic compensation items With dynamic process compensation items ; S2.24, Adaptive compensation coefficient based on fat content Statistical values ​​of the core temperature of the original food Make corrections and incorporate the normalized change intensity factor. Generates the corrected core temperature parameters of the food at the current moment.

[0008] As a further improvement to this technical solution, in step S2.22, the fat content characteristic parameters corresponding to quick-frozen foods are obtained through a dynamic characteristic parameter extraction method based on temperature-sensitive fat response, including the following steps: Based on food core temperature statistics The rate of change of the core temperature statistics of food over time was calculated, and the characteristic parameter of the maximum rate of temperature change used to characterize the influence of fat content was extracted. The characteristic parameter of the maximum temperature change rate A mapping relationship is established with the original fat content characteristics to generate a dynamic correction factor. The original fat content characteristics were combined with dynamic correction factors. Weighted fusion is performed to generate the final fat content characteristic parameters.

[0009] As a further improvement to this technical solution, in step S3, based on a long short-term memory network model that incorporates spray state masking weights, the core temperature of the quick-frozen food at the end of the specified freezing time is predicted, including the following steps: S3.1 Apply spray state temporal alignment and temporal masking weighting to the food surface temperature component in the core temperature state vector output in step S2 to generate an enhanced core temperature state vector. Construct the enhanced core temperature state vector into the input sequence of the long short-term memory network model in chronological order, and normalize the input sequence. S3.2 Construct a long short-term memory network model architecture based on the normalized input sequence; S3.3. Collect the preprocessed core temperature state vector from historical production data as the training input sequence, and use the actual food core temperature at the end of freezing as the label to train the long short-term memory network model. S3.4 Input the core temperature state vector generated during real-time production into the trained long short-term memory network model to predict the core temperature of the quick-frozen food at the end of the specified freezing time. S3.5 Calculate the deviation between the predicted core temperature of the food and the preset target core temperature standard; S3.6 Output the core temperature status assessment results based on the deviation calculation results.

[0010] As a further improvement to this technical solution, in step S3.1, the food surface temperature component in the core temperature state vector output in step S2 is subjected to spray state timing alignment and temporal masking weighting processing to generate an enhanced core temperature state vector, including the following steps: S3.11. The spraying status of the liquid nitrogen spraying control system is collected in real time through the Internet of Things, and the spraying status at each time point is encoded to generate a spraying status sequence. S3.12, For each spray state transition time Constructing based on this transition moment A transition period of 5 seconds before and after the center; S3.13 Constructing spray state masking weights based on spray state sequence and transition time interval. ; S3.14. Apply spray state masking weights to the food surface temperature component in the original core temperature state vector. The weighted food surface temperature is generated, and the enhanced core temperature state vector is generated.

[0011] As a further improvement to this technical solution, in step S3.3, the long short-term memory network model architecture is constructed based on the normalized input sequence, including the following steps: The normalized input sequence is used as the input to the long short-term memory network model. Each input normalized input sequence has a dimension of 4, and a multi-layer long short-term memory network is constructed. The Long Short-Term Memory (LSTM) network layer outputs the hidden state sequence at each time step. An attention layer is added after the LSM layer to calculate attention weights based on the hidden state sequence, and spray state masking weights are introduced during the attention weight calculation process. An inhibition factor is applied to the hidden state of the time step during the transition period of the spray state switching; The hidden state sequence is weighted and summed according to the attention weights through the attention layer to generate a context vector; the context vector is then mapped to a single numerical value through a fully connected output layer, which is the output of the predicted core temperature of the frozen food at the end of the specified freezing time.

[0012] As a further improvement to this technical solution, in step S4, the operating parameters of the quick-freezing equipment in the quick-freezing food production process are managed in real time through the Internet of Things based on the core temperature status assessment results, including the following steps: The adjustment range of the quick-freezing equipment's operating parameters is calculated based on the core temperature status assessment results. The adjustment range is then sent to the quick-freezing equipment via the Internet of Things control interface for real-time management of the quick-freezing equipment's operating parameters. The operating parameters of the quick-freezing equipment include at least the liquid nitrogen injection flow rate, fan speed, and conveyor belt speed.

[0013] On the other hand, the present invention provides a real-time management system for a frozen food production workshop based on the Internet of Things, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the above-mentioned real-time management method for a frozen food production workshop based on the Internet of Things.

[0014] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: high-precision perception, intelligent prediction and closed-loop control of the core temperature state of the entire quick-freezing process are realized through Internet of Things control, which effectively overcomes the problem of unstable freezing quality caused by differences in fat content, spray interference and fluctuations in operating conditions in traditional quick-freezing processes; in particular, by using Internet of Things control to dynamically adjust the operating parameters of quick-freezing equipment (such as liquid nitrogen injection flow rate, fan speed and conveyor belt speed) at the millisecond level, freezing efficiency and product consistency are significantly improved, while reducing energy consumption and the risk of over-freezing, and ensuring the safety and quality stability of quick-frozen foods. Attached Figure Description

[0015] Figure 1 This is a flowchart of the overall 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: Please refer to Figure 1 As shown, this embodiment provides a real-time management method for a frozen food production workshop based on the Internet of Things, including the following steps: S1. Collect frozen food production parameters through the Internet of Things and preprocess the frozen food production parameters (preprocessing includes time synchronization, outlier removal, and smoothing based on sliding time windows). In this embodiment, the production parameters for quick-frozen food include at least the core temperature of the food, the surface temperature of the food, and the operating status parameters of the conveyor belt (core temperature sensors and surface temperature sensors are arranged in the quick-freezing or liquid nitrogen spray freezing zone on the quick-freezing production line to collect the core temperature and surface temperature data of the food in real time, and at the same time, the operating status parameters of the conveyor belt (including running / stopping status, running speed or cycle time information) are read from the conveyor belt control system through the industrial bus or control interface).

[0018] S2. Based on the pre-processed production parameters of quick-frozen foods, a core temperature state vector reflecting the core temperature change and freezing progress characteristics of quick-frozen foods is extracted using a fat content adaptive core temperature compensation algorithm. In this embodiment, the core temperature state vector reflecting the core temperature change and freezing progress characteristics of quick-frozen foods is extracted based on the fat content adaptive core temperature compensation algorithm, including the following steps: S2.1. Determine the effective sampling range of the core temperature of the pre-treated food based on the pre-treated conveyor belt operating status parameters. Only select the core temperature data of the food when the quick-freezing equipment is in operation and the conveyor belt speed is not lower than the preset speed threshold, so as to avoid interference from shutdown, stagnation or abnormal working conditions on the core temperature sampling results. S2.2 Within the effective sampling interval, perform time window statistical processing on the core temperature of food at multiple consecutive moments, and correct the core temperature parameters of food based on the fat content adaptive core temperature compensation algorithm. Calculate the corrected core temperature parameters of food at the current moment to characterize the true thermal state level of quick-frozen food at the current freezing stage. The core temperature compensation algorithm based on fat content comprehensively analyzes the static fat content information of food and the dynamically captured thermal response characteristics during freezing. It constructs a compensation coefficient related to fat content and freezing stage to intelligently correct the originally collected core temperature statistics. Ultimately, the algorithm outputs an effective core temperature parameter that more closely approximates the true thermodynamic state of the food, thereby overcoming temperature monitoring deviations caused by differences in material composition and providing reliable state input for accurate judgment and precise control of the subsequent freezing state. The specific heat capacity and phase transition characteristics of fats differ significantly from those of proteins and water. During the freezing process of high-fat foods, fats exhibit a wide range of phase transition behaviors, with their main phase transition temperature range typically higher than the freezing point of water, and the phase transition process is not concentrated at a single temperature point. Simultaneously, the latent heat released during fat solidification is relatively small, making it difficult for traditional temperature sampling to accurately reflect the overall freezing state of the food. An adaptive core temperature compensation algorithm based on fat content was used to achieve adaptive compensation for core temperature measurements. This algorithm not only considers the static influence of fat content but also captures the dynamic characteristics of fat phase transitions during freezing, thus significantly improving the accuracy of judging the freezing state of high-fat foods. This overcomes the control deviation caused by traditional temperature monitoring methods ignoring differences in the thermal properties of materials, providing a reliable state input for subsequent intelligent control. The process involves correcting the food's core temperature parameters based on an adaptive core temperature compensation algorithm for fat content, and calculating the corrected core temperature parameters at the current moment, including the following steps: S2.21. Within the effective sampling interval, for continuous... The core food temperature data at each moment is statistically processed using a time window. By assigning higher weights to data closer to the current moment, the original statistical value of the core food temperature corresponding to the current moment is calculated. ( For time, the statistical value of the core temperature of the original food. It is obtained by weighting and statistically analyzing the core temperature data of food from multiple consecutive sampling times, and is used to characterize the average thermal state under sampling noise and operating condition fluctuations. S2.22. Obtain the fat content characteristic parameters of quick-frozen foods using a dynamic characteristic parameter extraction method based on temperature-sensitive fat response. (Used to characterize the differences in thermal conductivity and phase change behavior of food due to differences in fat content during freezing). In this embodiment, the dynamic feature parameter extraction method based on temperature-sensitive fat response monitors the rate of change of the core temperature of food in the early stage of freezing in real time, extracts its maximum cooling rate as a key feature, and dynamically corrects and updates the fat content parameter based on the physical relationship between this feature and the known fat content (the higher the fat content, the lower the maximum cooling rate usually is). The dynamic feature parameter extraction method based on temperature-sensitive fat response addresses the specific problem of discrepancies between the actual thermal behavior of fat in frozen foods and static formulation data. Traditional methods rely on fixed formulation fat content, but during actual freezing, due to differences in raw materials, uneven mixing, or different pretreatment, the solidification temperature and phase transition rate of fat will dynamically change, causing static parameters to fail to accurately reflect the true thermal response of the current batch of food, thus affecting the compensation accuracy. This method dynamically corrects the fat content parameter by monitoring the core temperature change rate in real time, achieving an upgrade from "fixed formulation value" to "dynamic response value". It uses the temperature drop rate during the freezing process as a "sensor" of the actual thermal behavior of fat, enabling the fat content feature parameter to adaptively adjust according to the real-time performance of the material, thereby significantly improving the adaptability of the compensation algorithm to complex working conditions and batch fluctuations in the production site. Furthermore, the fat content characteristic parameters of quick-frozen foods are obtained through a dynamic feature parameter extraction method based on temperature-sensitive fat response, including the following steps: Based on food core temperature statistics Calculate the rate of change of the statistical value of food core temperature over time. The maximum temperature change rate characteristic parameter was extracted to characterize the extent to which fat content is affected. ( The maximum rate of temperature change from the start of freezing to the current moment is used as a key characteristic of the dynamic response of fat content, and the maximum rate of temperature change characteristic parameter is used. Characteristics of original fat content Establish mapping relationship (original fat content characteristics) The extraction steps are as follows: First, obtain the food's formula information or nutritional component labeling data, and extract the basic indicators related to fat, such as fat mass fraction, the proportion of main fat-containing ingredients, and the fat content range per unit mass of food. Simultaneously, collect the food's physical property parameters, including total mass, volume, geometric dimensions, and morphological category, and calculate the fat distribution density per unit volume or unit mass. Then, standardize and normalize the above fat-related indicators to eliminate the influence of different food specifications and labeling scales. Finally, generate a single scalar form of static fat content characteristic parameter, i.e., the original fat content characteristic, by weighted combination of the normalized multidimensional fat-related indicators. ), generating dynamic correction factors (In the formula, This is a temperature sensitivity adjustment coefficient, used to adjust the strength of the effect of the rate of temperature change on the dynamic correction amplitude of fat. It is obtained through experimental calibration or regression of historical samples. , As a reference rate of temperature change, used for... Normalization was performed to obtain experimentally determined values ​​based on standard fat content foods; the formula here... This is the characteristic parameter of the maximum temperature change rate. Characteristics of original fat content The mapping relationship will be used to determine the original fat content characteristics. With dynamic correction factor Weighted fusion is performed to generate the final fat content characteristic parameters. ; In the formula, The fusion weight is used to adjust the weight ratio between the original fat content parameter and the dynamically corrected fat content parameter, and is obtained through experimental statistics, cross-validation or empirical setting. S2.23, Based on fat content characteristic parameters Construct an adaptive compensation coefficient for fat content. ; Among them, the adaptive compensation coefficient of fat content Including static basic compensation items With dynamic process compensation items ; ; Static basic compensation items Dynamic process compensation term used to reflect the differences in overall thermal response during the freezing process of foods with different fat contents. It is used to reflect the dynamic impact of the fat phase transition stage on the core temperature change during the freezing process; ; ; In the formula, The fat content of the standard reference food is taken as 0 or the experimentally determined benchmark value. This is the static gain coefficient (unit: ℃ / %), with a value range of 0.05~0.20℃ / %, representing the steady-state temperature deviation caused by a one-unit change in fat content. It is obtained through experimental comparison (e.g., lean meat vs. meat with a good balance of fat and lean). This is the median of the main phase transition temperature range for fats (determined through experimental calibration based on food type; in this example, it is taken as -2℃). This is a dynamic compensation amplitude coefficient related to fat content, used to adjust the maximum influence of foods with different fat contents on the core temperature compensation amplitude during the fat phase transition stage. The value ranges from 0.1 to 1.0℃ and is determined through experimental calibration, historical sample statistics, or regression analysis. This is a parameter related to the width of the phase transition temperature sensitive range, with a value range of 1.0~5.0℃, used to adjust the decay rate of the dynamic compensation term when the core temperature deviates from the fat phase transition temperature range; S2.24, Adaptive compensation coefficient based on fat content Statistical values ​​of the core temperature of the original food Make corrections and incorporate the normalized change intensity factor. (Dimensionless) Generates the corrected core temperature parameters of the food at the current moment. (In the formula, This is a freezing rate modulation coefficient related to fat content, used to characterize the differences in the sensitivity of foods with different fat contents to the rate of change of core temperature during the non-phase transition stage. The unit is °C, and the value ranges from -0.5 to 0.5 °C. It was determined experimentally. To obtain effective state parameters after incorporating fat thermal response mechanism compensation based on this statistical estimate, which will be used for subsequent freezing state determination and equipment control. ; In the formula, The rate of change of the core temperature of food over time. The core temperature change rate is determined based on experimental data, historical statistics, or process experience of standard fat-content foods; compensation coefficient It primarily targets the dynamic compensation of the phase transition hysteresis effect of fat within a specific temperature range, correcting the illusion of "artificially high temperature" caused by the exothermic release of heat during the slow solidification of fat; while the change intensity factor Based on the deviation between the current actual cooling rate and the standard rate, the overall thermal inertia of the non-phase change stage is modulated, thereby ultimately outputting an effective temperature parameter that reflects both the material characteristics (fat content) and the current thermodynamic process, laying the foundation for subsequent accurate state determination and control decisions. S2.3 Calculate the freezing progress characteristic parameters of quick-frozen food at the current moment based on the corrected core temperature parameters of the food. (In the formula, The initial core temperature of the quick-frozen food when it enters the liquid nitrogen spray freezing zone (in this embodiment, the liquid nitrogen spray freezing zone is the processing area on the quick-frozen production line that rapidly reduces the core temperature of the food by liquid nitrogen spraying to achieve quick-freezing). The preset target core temperature standard value, (This refers to the corrected core temperature parameter for food), used to characterize the degree to which the core temperature has moved towards the target standard value; S2.4. Based on the corrected core temperature parameters of the food within the effective sampling interval, calculate the trend parameter of the corrected core temperature parameters over time. This trend parameter reflects the direction and rate of temperature change during the freezing process of quick-frozen foods. The trend parameter should include at least one of the following: a decreasing temperature trend, a stable temperature trend, or a rising temperature trend. Specifically, first, calculate the corrected core temperature parameters of the food at consecutive time intervals. The time series is smoothed to eliminate short-term noise fluctuations. Then, the temperature change rate of adjacent time steps is calculated. According to the temperature change pattern during the freezing process of quick-frozen food, the intervals where the rate exceeds the lower threshold a (in this embodiment, a is -0.03℃) are marked as a temperature decreasing trend, the intervals where the rate is below the upper threshold b (in this embodiment, b is 0.03℃) are marked as a temperature increasing trend, and the intervals between the lower threshold a and the upper threshold b are marked as a temperature stabilizing trend. Finally, the trend category of each time step is encoded or quantified into a one-dimensional trend parameter. For example, a decreasing trend is encoded as -1, a stabilizing trend as 0, and an increasing trend as 1, which is used to reflect the direction and rate of change of the core temperature over time, providing a basis for freezing status determination and equipment control. S2.5. Combine the corrected food core temperature parameters, trend parameters, food surface temperature, and freezing progress characteristic parameters to construct a core temperature state vector reflecting the core temperature change and freezing progress characteristics of quick-frozen foods: (In the formula, For the core temperature state vector, This is a component of the food surface temperature. (Parameters for changing trends).

[0019] S3. Using the core temperature state vector, based on the Long Short-Term Memory (LSTM) network model with introduced spray state masking weights, the core temperature of quick-frozen food at the end of the specified freezing time is predicted, and the core temperature state evaluation result is output. In this embodiment, a Long Short-Term Memory (LSTM) network model with introduced spray state masking weights is used to predict the core temperature of quick-frozen food at the end of a specified freezing time. The steps include: S3.1 Apply spray state temporal alignment and temporal masking weighting to the food surface temperature component in the core temperature state vector output in step S2 to generate an enhanced core temperature state vector. Construct the enhanced core temperature state vector into the input sequence of the Long Short-Term Memory (LSTM) network model in chronological order, and normalize the input sequence to eliminate dimensional differences and improve the convergence speed of the LSTM network model. In this embodiment, in the liquid nitrogen spray freezing zone of quick-frozen food production, the periodic switching control of the liquid nitrogen nozzles causes short-term periodic fluctuations in the ambient temperature (e.g., ±3-5℃) lasting 1-2 minutes. These precisely periodic short-term temperature shocks interfere with the readings of the food surface temperature sensor, thereby affecting the accuracy of the core temperature vector calculated based on the food surface temperature. By combining the spray control signal with a temporal masking mechanism through spray state timing alignment and temporal masking weighting, intelligent dynamic evaluation of the value of surface temperature data is achieved. It assigns differentiated confidence weights to surface temperature data at different time periods based on the spray state (on, off, transition period), thereby suppressing periodic spray impact noise while maximizing the preservation of effective temperature information reflecting the actual freezing process of food, significantly improving the quality of feature vectors input to the prediction model. Specifically, the food surface temperature component in the core temperature state vector output in step S2 is subjected to spray state temporal alignment and temporal masking weighting processing to generate an enhanced core temperature state vector, including the following steps: S3.11. The spraying status of the liquid nitrogen spraying control system is collected in real time through the Internet of Things (IoT). The spraying status at each time point is encoded to generate a spraying status sequence. The spraying status sequence is precisely aligned with the timestamp of the core temperature state vector to ensure that the core temperature data at each time point corresponds to the corresponding spraying status. Specifically, the liquid nitrogen spraying control system is connected through the IoT interface. The spraying switch signal is read at fixed time intervals (such as every second or shorter time steps). The spraying on state is encoded as 1 and the spraying off state is encoded as 0. The encoded values ​​at each time point are arranged in chronological order to generate a spraying status sequence that is precisely aligned with the timestamp of the core temperature state vector. This sequence is used for subsequent spraying status masking weight calculation and core temperature state vector enhancement. The spray status at each time point is coded as follows: ; S3.12, For each spray state transition time (i.e., the time it takes for the sprinkler system to switch from on to off, or from off to on), construct a system based on this transition time. A transition period of 5 seconds before and after the center. This transition period is used to capture the drastic fluctuations in food surface temperature during liquid nitrogen spraying switching; S3.13 Constructing spray state masking weights based on spray state sequence and transition time interval. , According to the spraying status And whether it is in a transition period is determined, which is used to adjust the contribution of food surface temperature to the core temperature vector at different time points; In this embodiment: ; In the formula, Indicates from the transition time Begin with a 5-second transition period; The transition option has the lowest weight (0.2), which is mainly applicable to the first 5 seconds after the spray is turned on: when the spray is turned on, liquid nitrogen is sprayed out rapidly, and the food surface temperature drops sharply and is very unstable. The weight given is the smallest (0.2), which means that the contribution of the food surface temperature to the core temperature vector is greatly weakened during this period. The second lowest weight (0.3) for the transition period is mainly applicable to the first 5 seconds after the spray is turned off: as soon as the spray is turned off, the surface temperature of the food begins to slowly stabilize. The weight is slightly higher (0.3) to reflect the gradual stabilization of the temperature, but it is still not completely reliable. The stable spray period is moderate (0.8), which is suitable for the time when the spray is continuously on and not in any transition period: the spray is continuously on, the food surface temperature is under control under liquid nitrogen cooling, the weight is moderate (0.8), the surface temperature data is relatively reliable, and it can participate in the core temperature state vector to a large extent. The highest stable shutdown period (1.0) is applicable to the time when the spray is continuously shut off and not in any transition period: the spray is shut off, the surface temperature changes steadily, there is no liquid nitrogen impact, the weight is the largest (1.0), the data is the most reliable, and it contributes the most to the core temperature vector; S3.14. Apply spray state masking weights to the food surface temperature component in the original core temperature state vector. The weighted average food surface temperature is generated. And generate an enhanced core temperature state vector (using the weighted food surface temperature). Replace the food surface temperature component in the original core temperature state vector (while keeping other temperature components unchanged, thereby generating an enhanced core temperature state vector). S3.2 Construct a Long Short-Term Memory (LSTM) network model architecture based on the normalized input sequence; Furthermore, a Long Short-Term Memory (LSTM) network model architecture is constructed based on the normalized input sequence, including the following steps: The normalized input sequence is used as the input of the Long Short-Term Memory (LSTM) network model. The dimension of each input normalized input sequence is 4-dimensional, corresponding to each component of the enhanced core temperature state vector (including food core temperature, weighted food surface temperature, freezing progress feature parameters, and change trend parameters), ensuring that the time series information and the features of different temperature components are completely preserved, and a multi-layer long short-term memory network (LSTM) is constructed. The Long Short-Term Memory (LSTM) network layer is used to capture the temporal dependencies of the input sequence and output the hidden state sequence at each time step. An attention layer is added after the LSM network layer to calculate attention weights based on the hidden state sequence, and spray state masking weights are introduced in the attention weight calculation process. (Introducing spray state masking weights) This method is used to adaptively adjust the importance of hidden states at different time steps based on the stability of the liquid nitrogen spraying state. This reduces the interference of temperature fluctuations caused by liquid nitrogen shocks during the spraying start-up or shut-down transition period on the model's judgment, suppresses the weight of hidden states at unstable time steps in feature fusion, and highlights the effective contribution of hidden states in the stable spraying stage to core temperature prediction, thereby improving the stability and accuracy of food core temperature prediction results at the end of freezing. Furthermore, a spray state masking weight is introduced into the attention weight calculation process. Specifically: Calculate the initial attention score : , ; Introducing spray state masking weights into the attention weight calculation process for: ; Normalized to attention weights : ; In the formula, These are the trainable weight vectors for the attention layer. For transpose operation, The hyperbolic tangent activation function is used. A trainable matrix used to train the hidden state sequence Mapped to attention space, This is a trainable bias vector. For time steps, The length of the input sequence is the total number of time steps in the core temperature state vector. For time steps Attention score after introducing spray state masking weights For sequence index, For time steps Attention score after introducing spray state masking weights; The hidden state sequence is weighted and summed according to the attention weights through the attention layer to generate a context vector. This context vector integrates the important information of each time step within the sliding window. The context vector is then mapped to a single value through a fully connected output layer, which outputs the predicted core temperature of the quick-frozen food at the end of the specified freezing time. This output is used for subsequent deviation calculation and freezing status determination. S3.3. Collect the preprocessed core temperature state vector from historical production data as the training input sequence, and use the actual food core temperature at the end of freezing as the label to train the Long Short-Term Memory (LSTM) network model. Optimize the network parameters by minimizing the error between the predicted and actual values. Specifically: First, construct the LSTM input sequence in chronological order using the core temperature state vector from historical production data that has undergone time synchronization, anomaly removal, and smoothing. Normalize each component to eliminate dimensional differences. Then, use the actual core temperature of the corresponding batch of food at the end of freezing as the output label. Perform forward propagation on the LSTM network batch by batch to calculate the predicted value. By calculating the error (such as mean square error) between the predicted value and the actual core temperature label, adjust the network weights and bias parameters using the backpropagation algorithm. Repeat the iteration until the loss function converges or reaches the preset number of training rounds, thereby obtaining an LSTM model that can accurately predict the core temperature of food at the end of freezing. S3.4 Input the core temperature state vector generated during real-time production into the trained Long Short-Term Memory (LSTM) network model to predict the core temperature of the frozen food at the end of the specified freezing time. S3.5 Calculate the deviation between the predicted core temperature of the food and the preset target core temperature standard: ; In the formula, The core temperature deviation at the end of freezing, in °C (degrees Celsius). The predicted core temperature of the food, the value output by the LSTM model at the end of the specified freezing time, in °C. The preset target core temperature standard, in °C; S3.6 Output the core temperature status assessment results based on the deviation calculation results, including the frozen qualified status ( Insufficient freezing () Over-freezing ), The allowable deviation threshold can be set according to the food type and process requirements (e.g., ±1℃).

[0020] S4. Based on the core temperature status assessment results, the operating parameters of quick-freezing equipment in the quick-freezing food production process are managed in real time through the Internet of Things. In this embodiment, based on the core temperature status assessment results, the operating parameters of the quick-freezing equipment in the quick-freezing food production process are managed in real time via the Internet of Things, including the following steps: The adjustment range of the quick-freezing equipment's operating parameters is calculated based on the core temperature status assessment results. The adjustment range is then sent to the quick-freezing equipment via the Internet of Things control interface for real-time management of the quick-freezing equipment's operating parameters. Among them, the operating parameters of quick-freezing equipment include at least liquid nitrogen injection flow rate (or spray valve opening / pressure), fan speed and conveyor belt speed; Real-time management of quick-freezing equipment operating parameters: The adjustment range of quick-freezing equipment operating parameters is based on the core temperature deviation. Calculations, for example: for insufficient freezing ( The increase in liquid nitrogen injection flow rate is calculated as follows: Adjustment, the increase in fan speed is based on Adjustment, the conveyor belt speed decreases by the following amount Adjust and not lower than the minimum safe speed For excessive freezing ( The reduction in liquid nitrogen injection flow rate is calculated as follows: Adjustment, the fan speed decreases by the following amount Adjust the conveyor belt speed increase rate according to... Adjust and do not exceed the maximum safe speed For those that are frozen and qualified ( ), with all parameters remaining unchanged, among which, To adjust the gain coefficient for liquid nitrogen injection flow rate, This refers to the gain coefficient for fan speed regulation. The gain coefficient for conveyor belt speed adjustment can be obtained through historical data or experimental calibration, achieving a linear mapping from core temperature deviation to equipment parameter adjustment. In this embodiment, The value is 8.0 (L / min) / °C. It is 1.2 (m / s) / °C. It is 0.15 (m / min) / °C.

[0021] Example 2: This example provides a real-time management system for a frozen food production workshop based on the Internet of Things (IoT), including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the real-time management method for a frozen food production workshop based on the IoT described in Example 1.

[0022] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for real-time management of a frozen food production workshop based on the Internet of Things, characterized in that, include: S1. Collect frozen food production parameters through the Internet of Things and preprocess the frozen food production parameters. S2. Based on the pre-processed production parameters of quick-frozen foods, a core temperature state vector reflecting the core temperature change and freezing progress characteristics of quick-frozen foods is extracted using a fat content adaptive core temperature compensation algorithm. S3. Using the core temperature state vector, based on a long short-term memory network model with introduced spray state masking weights, the core temperature of quick-frozen food at the end of the specified freezing time is predicted, and the core temperature state evaluation results are output. S4. Based on the core temperature status assessment results, the operating parameters of quick-freezing equipment in the quick-freezing food production process are managed in real time through the Internet of Things.

2. The method for real-time management of a frozen food production workshop based on the Internet of Things according to claim 1, characterized in that, In S1, the production parameters for quick-frozen food include at least the core temperature of the food, the surface temperature of the food, and the operating status parameters of the conveyor belt.

3. The method for real-time management of a frozen food production workshop based on the Internet of Things according to claim 2, characterized in that, In step S2, the core temperature state vector reflecting the core temperature change and freezing progress characteristics of quick-frozen foods is extracted based on the fat content adaptive core temperature compensation algorithm, including the following steps: S2.1 Determine the effective sampling range of the core temperature of the pretreated food based on the pretreated conveyor belt operating status parameters; S2.2 Within the effective sampling interval, perform time window statistical processing on the core temperature of food at multiple consecutive moments, and correct the core temperature parameters of food based on the fat content adaptive core temperature compensation algorithm, and calculate the corrected core temperature parameters of food at the current moment. S2.3 Calculate the freezing progress characteristic parameters of quick-frozen food at the current moment based on the corrected core temperature parameters of food; S2.

4. Based on the corrected food core temperature parameters within the effective sampling interval, calculate the trend parameters of the corrected food core temperature parameters over time. S2.5 Combine the corrected food core temperature parameters, trend parameters, food surface temperature and freezing progress characteristic parameters to construct a core temperature state vector that reflects the core temperature change and freezing progress characteristics of quick-frozen food.

4. The method for real-time management of a frozen food production workshop based on the Internet of Things according to claim 3, characterized in that, In step S2.2, the food core temperature parameters are corrected based on the fat content adaptive core temperature compensation algorithm, and the corrected food core temperature parameters at the current moment are calculated, including the following steps: S2.

21. Within the effective sampling interval, for continuous... The core food temperature data at each time point are statistically processed using a time window to calculate the original core food temperature statistical value corresponding to the current time point. ; S2.

22. Obtain the fat content characteristic parameters of quick-frozen foods by using a dynamic characteristic parameter extraction method based on temperature-sensitive fat response; S2.

23. Based on the characteristic parameters of fat content, construct an adaptive compensation coefficient for fat content. ; Among them, the fat content adaptive compensation coefficient Including static basic compensation items With dynamic process compensation items ; S2.24, Adaptive compensation coefficient based on fat content Statistical values ​​of the core temperature of the original food Make corrections and incorporate the normalized change intensity factor. Generates the corrected core temperature parameters of the food at the current moment.

5. A real-time management method for a quick-frozen food production workshop based on the Internet of Things, as described in claim 4, is characterized in that... In step S2.22, the fat content characteristic parameters of quick-frozen foods are obtained through a dynamic characteristic parameter extraction method based on temperature-sensitive fat response, including the following steps: Based on food core temperature statistics The rate of change of the core temperature statistics of food over time was calculated, and the characteristic parameter of the maximum rate of temperature change used to characterize the influence of fat content was extracted. The characteristic parameter of the maximum temperature change rate A mapping relationship is established with the original fat content characteristics to generate a dynamic correction factor. The original fat content characteristics were combined with dynamic correction factors. Weighted fusion is performed to generate the final fat content characteristic parameters.

6. The method for real-time management of a frozen food production workshop based on the Internet of Things according to claim 1, characterized in that, In step S3, based on a long short-term memory network model that incorporates spray state masking weights, the core temperature of the quick-frozen food is predicted at the end of the specified freezing time, including the following steps: S3.1 Apply spray state temporal alignment and temporal masking weighting to the food surface temperature component in the core temperature state vector output in step S2 to generate an enhanced core temperature state vector. Construct the enhanced core temperature state vector into the input sequence of the long short-term memory network model in chronological order, and normalize the input sequence. S3.2 Construct a long short-term memory network model architecture based on the normalized input sequence; S3.

3. Collect the preprocessed core temperature state vector from historical production data as the training input sequence, and use the actual food core temperature at the end of freezing as the label to train the long short-term memory network model. S3.4 Input the core temperature state vector generated during real-time production into the trained long short-term memory network model to predict the core temperature of the quick-frozen food at the end of the specified freezing time. S3.5 Calculate the deviation between the predicted core temperature of the food and the preset target core temperature standard; S3.6 Output the core temperature status assessment results based on the deviation calculation results.

7. A real-time management method for a quick-frozen food production workshop based on the Internet of Things, as described in claim 6, is characterized in that... In step S3.1, the food surface temperature component in the core temperature state vector output in step S2 is subjected to spray state temporal alignment and temporal masking weighting processing to generate an enhanced core temperature state vector, including the following steps: S3.

11. The spraying status of the liquid nitrogen spraying control system is collected in real time through the Internet of Things, and the spraying status at each time point is encoded to generate a spraying status sequence. S3.12, For each spray state transition time Constructing based on this transition moment A transition period of 5 seconds before and after the center; S3.13 Constructing spray state masking weights based on spray state sequence and transition time interval. ; S3.

14. Apply spray state masking weights to the food surface temperature component in the original core temperature state vector. The weighted food surface temperature is generated, and the enhanced core temperature state vector is generated.

8. A real-time management method for a quick-frozen food production workshop based on the Internet of Things, as described in claim 6, is characterized in that... In step S3.3, the long short-term memory network model architecture is constructed based on the normalized input sequence, including the following steps: The normalized input sequence is used as the input to the long short-term memory network model. Each input normalized input sequence has a dimension of 4, and a multi-layer long short-term memory network is constructed. The Long Short-Term Memory (LSTM) network layer outputs the hidden state sequence at each time step. An attention layer is added after the LSM layer to calculate attention weights based on the hidden state sequence, and spray state masking weights are introduced during the attention weight calculation process. An inhibition factor is applied to the hidden state of the time step during the transition period of the spray state switching; The hidden state sequence is weighted and summed according to the attention weights through the attention layer to generate a context vector; the context vector is then mapped to a single numerical value through a fully connected output layer, which is the output of the predicted core temperature of the frozen food at the end of the specified freezing time.

9. A real-time management method for a frozen food production workshop based on the Internet of Things according to claim 1, characterized in that, In step S4, based on the core temperature status assessment results, the operating parameters of the quick-freezing equipment in the quick-freezing food production process are managed in real time via the Internet of Things, including the following steps: The adjustment range of the quick-freezing equipment's operating parameters is calculated based on the core temperature status assessment results. The adjustment range is then sent to the quick-freezing equipment via the Internet of Things control interface for real-time management of the quick-freezing equipment's operating parameters. The operating parameters of the quick-freezing equipment include at least the liquid nitrogen injection flow rate, fan speed, and conveyor belt speed.

10. A real-time management system for a frozen food production workshop based on the Internet of Things, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the Internet of Things-based real-time management method for frozen food production workshops as described in any one of claims 1-9.

Images