Intelligent control method and system for rice fine processing equipment

Abstract

The invention discloses an intelligent control method and system for rice finish machining equipment. The method comprises the steps that related data of rice finish machining are collected through a high-precision sensor network; constructing a rice processing quality prediction model based on a convolutional neural network and a recurrent neural network; constructing an equipment operation parameter optimization model in a mode of combining a multi-objective optimization algorithm and deep reinforcement learning; using the rice processing quality prediction model to evaluate the quality of the current batch of rice so as to obtain a rice quality prediction result after the current batch of rice is processed; and according to the rice quality prediction result after the current batch of rice is processed, using an equipment operation parameter optimization model to adjust the operation parameters of the rice fine processing equipment. Through the method provided by the invention, the rice processing quality can be predicted in real time, and the equipment parameters are adjusted in time, so that the rice processing quality is ensured, the energy consumption of the equipment is reduced, and the economic benefits of enterprises are improved.

Description

Technical Field

[0001] The present invention generally relates to the field of artificial intelligence technology. More specifically, the present invention relates to an intelligent control method and system for rice processing equipment. Background Art

[0002] In rice processing, equipment control significantly impacts processing quality. Traditional methods rely heavily on manual experience to set equipment parameters, lacking precision and adaptability.

[0003] Although some equipment currently has automated functions, such as adjusting rice mill speed and polishing machine time according to preset programs, it is difficult to respond to differences in rice quality and environmental changes in real time. Rice from different origins and varieties has variable characteristics such as moisture, hardness, and particle size, and fixed programs cannot guarantee stable processing quality. Furthermore, during processing, equipment is susceptible to interference from ambient temperature, humidity, and dust, which can affect rice quality indicators such as the polished rice rate, chalkiness, and nutrient retention. Furthermore, fault diagnosis of existing equipment is delayed, and is often discovered only after the fault is obvious, resulting in downtime for maintenance, which affects production efficiency and increases costs. The low level of intelligence makes it impossible to flexibly adjust processing parameters according to market demand to produce personalized rice products, making it difficult to meet the diverse demands of modern consumers for rice quality and characteristics.

[0004] In view of this, there is an urgent need to provide an intelligent control method for rice processing equipment to improve the quality of rice processing. Summary of the Invention

[0005] In order to at least solve one or more of the technical problems mentioned above, the present invention proposes an intelligent control method for rice processing equipment in multiple aspects.

[0006] In a first aspect, the present invention provides an intelligent control method for rice finishing equipment, comprising: collecting rice finishing-related data using a high-precision sensor network; the high-precision sensor network includes: installing equipment status sensors and image sensors at multiple preset links of the rice finishing equipment; arranging environmental sensors around the rice finishing equipment to obtain environmental change indicators; and arranging a near-infrared spectrometer at the finished rice product discharge port to perform rapid spectral scanning of the processed rice.

[0007] Based on convolutional neural networks and recurrent neural networks, a rice processing quality prediction model is constructed using historical processing data. A multi-objective optimization algorithm is combined with deep reinforcement learning to construct an equipment operation parameter optimization model. The equipment operation parameter optimization model uses the equipment's energy consumption, processing efficiency, whole rice polishing rate, and rice nutrient retention rate as optimization targets to obtain the optimal parameter adjustment strategy for the rice processing equipment.

[0008] According to the rice processing related data collected by the high-precision sensor network, the rice processing quality prediction model is used to evaluate the quality of the current batch of rice to obtain a prediction result of the rice quality after the current batch of rice is processed; according to the rice quality prediction result after the current batch of rice is processed, the equipment operation parameter optimization model is used to adjust the operating parameters of the rice processing equipment to obtain the optimal parameters of the rice processing equipment.

[0009] In some embodiments, the high-precision sensor network also includes a rice quality monitoring sensor; wherein the rice quality monitoring sensor includes: a laser particle size analyzer is set at the front end of the rice feeding conveyor belt to measure the particle size distribution of the rice; and an odor sensor is installed to detect whether the rice has an odor.

[0010] In some embodiments, the equipment status sensor includes: a micro strain gauge sensor installed on the surface of the roller inside the rice milling machine to monitor the force distribution of the roller during the rice milling process in real time; and an airflow sensor is set in the polishing chamber of the polishing machine to detect airflow data during the polishing process.

[0011] In some embodiments, the image sensor includes: installing a multispectral camera above the rice feeding conveyor belt to capture images of the appearance of the rice, and using image recognition technology to analyze the images and convert the image data into feature data.

[0012] In some embodiments, the environmental sensors include seismic sensors to monitor seismic activity in real time.

[0013] In some embodiments, the rice processing quality prediction model is constructed based on convolutional neural networks and recurrent neural networks using historical processing data, wherein the convolutional neural network includes: using an improved residual network structure as the basic architecture of the convolutional neural network, including: in the initial convolution layer, using a 7x7 convolution kernel, a step size of 2, and 64 channels to perform preliminary feature extraction on the rice image; in the middle part of the residual network, an attention mechanism module is set to reweight the feature map by calculating the spatial attention and channel attention of the feature map; at the end of the residual network, a global average pooling layer is used to convert the feature map into a feature vector, and then a fully connected layer is connected, the fully connected layer has 256 neurons, and the activation function is ReLU, which is used to further integrate features.

[0014] The recurrent neural network utilizes multi-source time series data from the processing process, including temperature, pressure, equipment operating time, and rice images, and analyzes them through a subject-gated recurrent unit network to obtain time series data on rice quality changes. Feature vectors output by the convolutional neural network and feature vectors output by the recurrent neural network are fused using a dynamic weight allocation fusion method to obtain a fused feature vector.

[0015] The fused feature vector is input into a multilayer perceptron with at least 2 hidden layers for final prediction.

[0016] In some embodiments, the method of combining a multi-objective optimization algorithm with deep reinforcement learning to construct an equipment operating parameter optimization model, wherein the equipment operating parameter optimization model takes the energy consumption, processing efficiency, whole rice rate, and rice nutrient retention rate of the equipment as optimization targets, and obtains the optimal parameter adjustment strategy for the rice processing equipment, the steps comprising:

[0017] Multiple key parameters of the equipment are encoded as chromosomes for the multi-objective optimization algorithm.

[0018] In the genetic operation of the multi-objective optimization algorithm, the selection operation adopts the tournament selection method based on crowding degree; the crossover operation adopts simulated binary crossover; and the mutation operation adopts polynomial mutation.

[0019] A multi-objective optimization algorithm is used to perform non-inferiority sorting, including performing non-inferiority sorting on individuals in a population according to the optimization objective.

[0020] In the deep reinforcement learning model, a dual-Q network and a reward function R are used, wherein the state space S in the dual-Q network includes the current rice quality characteristics evaluated by the rice processing quality prediction model, the current operating parameters of the equipment, the operating status of the equipment, and environmental information; the action space A is the parameter adjustment strategy in the multi-objective optimization algorithm.

[0021] A parameter adjustment strategy is selected based on the current state S, and the new state S' and reward R are fed back through the environmental information.

[0022] The deep reinforcement learning model repeatedly learns the state-action-reward relationship and selects the optimal parameter adjustment strategy among the non-inferior solutions generated by the multi-objective optimization algorithm.

[0023] In some embodiments, the method further comprises: dynamically adjusting equipment parameters during the processing process using an adaptive control algorithm based on deviations between real-time monitored data and expected target values. The adaptive control algorithm comprises a model reference adaptive control algorithm that, by inputting rice characteristics and initial equipment parameters, outputs desired rice quality indicators and equipment operating efficiency, and combines fuzzy logic to optimize and adjust equipment parameters.

[0024] In a second aspect, the present invention provides an intelligent control system for rice processing equipment, comprising:

[0025] A multi-source data fusion acquisition module is used to collect rice processing-related data using a high-precision sensor network. The high-precision sensor network includes: equipment status sensors and image sensors installed in multiple preset links of the rice processing equipment; environmental sensors are set around the rice processing equipment to obtain environmental change indicators; and a near-infrared spectrometer is set at the finished rice product discharge port to perform rapid spectral scanning of the processed rice.

[0026] Quality prediction model building module: used to build a rice processing quality prediction model based on convolutional neural networks and recurrent neural networks, using historical processing data.

[0027] Equipment operation parameter optimization model construction module: A multi-objective optimization algorithm is combined with deep reinforcement learning to construct an equipment operation parameter optimization model. The equipment operation parameter optimization model takes the equipment's energy consumption, processing efficiency, whole rice polishing rate, and rice nutrient retention rate as optimization targets to obtain the optimal parameter adjustment strategy for the rice processing equipment.

[0028] Quality prediction module: Based on the rice processing related data collected by the high-precision sensor network, the rice processing quality prediction model is used to evaluate the quality of the current batch of rice to obtain the rice quality prediction result after the current batch of rice is processed.

[0029] Parameter adjustment module: According to the rice quality prediction result after the current batch of rice is processed, the equipment operating parameter optimization model is used to adjust the operating parameters of the rice processing equipment to obtain the optimal parameters of the rice processing equipment.

[0030] In a third aspect, the present invention provides an intelligent control device for rice processing equipment, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to implement the intelligent control method for rice processing equipment as described in any one of the embodiments of the first aspect when executing the computer program.

[0031] Through the intelligent control method for rice processing equipment provided above, the embodiment of the present invention collects multi-source data through the use of a high-precision sensor network, and by constructing a quality prediction model and an equipment operation parameter optimization model, it is possible to more accurately evaluate the quality of rice and formulate a personalized processing plan. Furthermore, in some embodiments, by combining a multi-objective optimization algorithm with deep reinforcement learning, the equipment can achieve an optimal balance between multiple goals during operation. Therefore, the present invention can predict the quality of rice processing in real time through accurate data collection and advanced artificial intelligence algorithm models, and adjust the equipment parameters in time according to the prediction results, thereby ensuring the quality of rice processing by controlling and adjusting the equipment parameters, reducing the energy consumption of the equipment, and improving the economic benefits and market competitiveness of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an illustrative and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0033] Figure 1 An exemplary flow chart of an intelligent control method for rice processing equipment according to some embodiments of the present invention is shown;

[0034] Figure 2 An exemplary structural block diagram of an intelligent control system for rice processing equipment according to some embodiments of the present invention is shown;

[0035] Figure 3 An exemplary structural block diagram of an intelligent control device for rice processing equipment according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0037] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0038] Figure 1 An exemplary flow chart of an intelligent control method 100 for rice finishing equipment according to some embodiments of the present invention is shown.

[0039] like Figure 1As shown, in solution 100, an intelligent control method for rice processing equipment is provided, and the method steps may include:

[0040] In step S101 : using a high-precision sensor network to collect rice processing related data.

[0041] In one embodiment, the high-precision sensor network includes: installing equipment status sensors and image sensors at multiple preset links of the rice processing equipment; setting environmental sensors around the rice processing equipment to obtain environmental change indicators; and setting a near-infrared spectrometer at the rice finished product discharge port to perform rapid spectral scanning on the processed rice.

[0042] In one exemplary embodiment, the high-precision sensor network also includes a rice quality monitoring sensor. This sensor includes a laser particle size analyzer located at the front end of the rice feed conveyor to measure the rice's particle size distribution. Rice of varying particle sizes may require different processing parameters during subsequent processing. For example, larger rice particles may require more powerful grinding forces, but are also more prone to broken rice. Particle size data can provide a reference for setting processing parameters in advance.

[0043] Odor sensors are installed to detect unusual odors in rice. These odors may indicate contamination or improper storage, which can affect the quality and safety of the rice. Once an unusual odor is detected, the system can automatically flag the batch of rice and adjust the processing, such as adding cleaning and disinfection steps or subjecting the processed rice to more stringent quality testing.

[0044] In one embodiment, the equipment status sensor can be implemented by installing basic sensors at key locations (such as weight and humidity sensors at the rice feed port, pressure and temperature sensors in the rice milling process, optical sensors in the polishing process, and environmental sensors). Furthermore, the system can also include: micro strain gauge sensors installed on the surface of the rollers inside the rice milling machine to monitor the force distribution of the rollers in real time during the rice milling process. The force distribution can be used to understand the uniformity of the rice milling process. If excessive force is detected in a certain area, it may indicate uneven roller wear or uneven rice feeding. The system can then adjust the rice feeding device or prompt for roller maintenance accordingly.

[0045] An airflow sensor is installed in the polishing chamber of the polishing machine to monitor airflow data during the polishing process. Proper airflow is crucial for the polishing effect and removal of surface bran from rice. By monitoring airflow data, the polisher's fan speed and ventilation duct valve opening can be optimized to ensure consistent polishing results.

[0046] At the same time, a near-infrared spectrometer is installed at the finished rice outlet to perform rapid spectral scanning on the processed rice to obtain information on the internal components of the rice (such as protein and starch content). These data can be transmitted together with the sensor data to the central data processing unit of the intelligent control system.

[0047] In one embodiment, the image sensor includes a multispectral camera mounted above the rice feed conveyor belt. The camera captures images of the rice's appearance (e.g., grain plumpness, color, and the presence of pests and diseases). The camera then uses image recognition technology to analyze the images and convert them into feature data. This feature data can then be used by the equipment control system for further processing.

[0048] Multispectral cameras can capture images of rice grains in different spectral bands. In addition to visible light, these also include near-infrared and ultraviolet (UV) images. Near-infrared images reveal the distribution of moisture and starch content within the rice grains, while UV images help detect microorganisms and pesticide residues on the rice grain surface. Analysis of multispectral images allows for a more comprehensive assessment of rice quality and safety.

[0049] Using a deep learning image segmentation algorithm, the system accurately segments and identifies full grains, shrunken grains, and impurities in rice images. Based on the proportion of different grain types, the system automatically adjusts the parameters of pre-processing equipment, such as winnowing and screening, to improve the efficiency and accuracy of rice pre-processing. For example, if the proportion of shrunken grains is high, the wind speed of the winnowing equipment can be appropriately increased to more effectively remove the shrunken grains.

[0050] In one embodiment, the environmental sensors may also include seismic sensors to monitor seismic activity in real time. Upon detecting a seismic wave, the system can immediately halt equipment operation to prevent damage and safety incidents, while simultaneously initiating emergency response plans, such as securing equipment and evacuating personnel.

[0051] UV sensors can also be added to monitor UV intensity within the processing room. While rice processing rooms typically don't have strong UV sources, abnormal UV intensity can affect certain nutrients in rice, such as loss of B vitamins. When UV intensity exceeds the normal range, the system can prompt measures such as shading or adjusting the processing speed to reduce the rice's exposure time during processing.

[0052] During the multi-source data collection process described above, the data collection frequency can be automatically adjusted based on the equipment's operating status and processing phase. For example, during equipment startup and shutdown, as well as when adjusting key parameters during processing, the data collection frequency can be increased to more fully document the equipment's dynamic changes. For example, when adjusting the speed of a rice mill, the sensor data collection frequency can be increased from 1 to 10 times per second, and continuous collection can be performed for 10 seconds to obtain sufficient data to analyze the impact of the speed adjustment on the processing. During stable equipment operation, the collection frequency can be appropriately reduced to reduce data transmission and processing pressure. For example, the collection frequency of some non-critical parameters can be reduced from 1 per second to 1 every 5 seconds.

[0053] During data transmission, a multi-channel data transmission method is used. In addition to wired Ethernet transmission, wireless Wi-Fi and 4G / 5G modules are also equipped as backup transmission channels. If the wired network fails, the system automatically switches to the wireless transmission channel to ensure uninterrupted data transmission. Furthermore, data verification and error correction mechanisms are implemented during the data transmission process. For example, a cyclic redundancy check (CRC) algorithm is used to verify the transmitted data. If any data errors are detected, retransmission or error correction is immediately performed to ensure data accuracy and integrity. This ensures the stability, real-time nature, and security of data transmission between each acquisition device and the central control system. Data transmission is encrypted to prevent data leakage and tampering.

[0054] Through the embodiments of the present invention described above, multiple sensors can be used to monitor rice grains in multiple dimensions, including particle size, odor, external appearance, and internal spectral information, enabling comprehensive and accurate assessment of rice quality. Whether it's the physical properties, chemical composition, or potential quality defects of rice, they can all be accurately detected, providing a reliable basis for developing customized processing plans. For example, based on the rice's particle size and internal component distribution, the optimal combination of milling and polishing parameters can be precisely determined, improving the rice's head rice yield and nutrient retention.

[0055] By real-time monitoring of various process parameters, such as roller force and polishing airflow, and by timely adjusting processing strategies based on environmental changes (such as UV intensity and earthquakes), the process can be ensured to be consistently optimal. This effectively prevents degradation of processing quality due to equipment failures and environmental factors, improving the stability and consistency of rice processing. For example, monitoring and adjusting roller force can make the rice milling process more uniform and reduce broken rice; adjusting the processing speed based on ambient UV intensity ensures that the rice's nutritional content is not compromised.

[0056] Adaptive data acquisition frequency adjustment and reliable multi-channel data transmission ensure sufficient data for analysis and decision-making at critical moments while reducing data redundancy and transmission pressure. The acquisition and processing of multispectral imagery and fused spectral data provide a rich data resource for building more accurate rice processing models. This data can be further mined and analyzed to optimize equipment maintenance plans, predict equipment failures, and develop new rice processing technologies, thereby improving the overall management level and innovation capabilities of rice processing companies. For example, by analyzing large amounts of historical data, it is possible to predict the replacement time of vulnerable equipment parts in advance, schedule preventive maintenance, and reduce equipment downtime and repair costs.

[0057] Further, the process comes to step S102: based on the convolutional neural network and the recurrent neural network, using the historical processing data, a rice processing quality prediction model is constructed.

[0058] In one embodiment, the convolutional neural network may include the following implementations:

[0059] In the structural design of the convolutional neural network (CNN), the improved residual network structure (ResNet) is first used as the basic architecture of the convolutional neural network, and the network structure is deeply optimized through the deformable convolution kernel. The sampling position of the convolution kernel of the deformable convolution is no longer fixed, but is adaptively adjusted according to the input feature map. The specific formula may include:

[0060]

[0061] In the above formula, X l is the output feature map of the lth layer, W l is the convolution kernel weight matrix, X l-1 is the feature map of the previous layer, (p) represents the pixel position of the output feature map, (q) is the convolution kernel sampling point position; Ω(p) is the sampling point set, W l (q) is the convolution kernel weight, Δp l (q) is the offset of the (l)th layer at the sampling point (q), which is learned through additional convolutional layers and can effectively capture irregular features in rice images, such as the irregular shape of rice caused by pests and diseases or growth differences.

[0062] Next comes a series of residual blocks, each consisting of two 3x3 convolutional layers, separated by an adaptively weighted batch normalization layer and a ReLU activation function. Adaptive weighted batch normalization is used to further improve the model's adaptability and feature extraction capabilities for rice image data. The formula for the adaptively weighted batch normalization layer is:

[0063]

[0064] Where: x ij Represents the pixel value of the i-th channel and j-th position in the input feature map, and its dimension is [C,H,W], where C is the number of channels, H is the feature map height, and W is the feature map width.

[0065] μ j is the mean of the j-th position over all samples and all channels, calculated as Here N is the number of samples, and the superscript n represents the nth sample.

[0066] is the variance of the j-th position over all samples and all channels, calculated as:

[0067]

[0068] ∈ is a very small constant (such as 1e-5) used to prevent the denominator from being zero and ensure numerical stability.

[0069] w ij is the adaptive weight, through a small fully connected network (parameter is θ w ) is learned based on the feature map information of the current position, that is, w ij =f(x ij ,θ w ), its initial value can be set to 1, the fully connected network can be a layer, the number of neurons is set to 16 based on experience, and the activation function uses the sigmoid function to ensure that the weight value is within a reasonable range (between 0 and 1), so that the normalization strength can be dynamically adjusted according to the feature importance of different positions and channels.

[0070] β ij It is a learnable bias term with an initial value of 0. It is also continuously optimized as the model is trained. It is used to translate the normalized data to further adapt to the model requirements.

[0071] Compared with traditional batch normalization, this adaptive weighted batch normalization method can more accurately perform normalization based on the distribution differences of rice image features, making the model more effective in extracting complex features such as the appearance and texture of rice, thereby improving the subsequent rice quality assessment and the overall prediction accuracy after fusion with RNN.

[0072] Secondly, in the middle part of the residual network, an attention mechanism module is set to reweight the feature map by calculating the spatial attention and channel attention of the feature map.

[0073] The spatial attention calculation formula is as follows:

[0074] M s(X)=σ(Conv(Concat(A(X),B(X))))

[0075] Among them, A(X)=MaxPool(X), B(X)=AveragePool(X), MaxPool and AveragePool are maximum pooling and average pooling operations respectively, Conv is the convolution operation, and σ is the sigmoid function.

[0076] The channel attention calculation formula is:

[0077] M c (X)=σ(MLP(Concat(MaxPool(X),AveragePool(X))))

[0078] Among them, MLP is a multi-layer perceptron. The final feature map after attention weighting is X att =M s (X)e M c (X)e X, where e represents element-wise multiplication.

[0079] Furthermore, at the end of the residual network, a global average pooling layer is used to convert the feature map into a feature vector, and then a fully connected layer is connected. The fully connected layer has 256 neurons and the activation function is ReLU for further integrating features.

[0080] Furthermore, in one embodiment, the recurrent neural network includes: utilizing multi-source time series data received during the processing process, including temperature, pressure, equipment operating time, and rice images, and analyzing through a subject gated recurrent unit network to obtain time series data on changes in rice quality.

[0081] In the above-mentioned recurrent neural network (RNN) model structure design, an improved structure based on dynamic forgetting gate (DFG) and adaptive input modulation (AIM) can be introduced to more accurately process the complex time series data in the rice processing process.

[0082] Specifically in the dynamic forget gate, the calculation formula can be:

[0083]

[0084] in:

[0085] W f Is the basic weight matrix of the forget gate, with dimension [n h ,n h +n x ], where nh is the hidden state dimension, n x is the input dimension, which is responsible for learning the input [h t-1 ,x t ] and the general connection between the output of the forget gate. b f is the bias vector of the forget gate, dimension is n h .

[0086] σ is still the sigmoid function, which is used to compress the output to between 0 and 1, indicating the proportion of cell state information retained at the previous moment.

[0087] λ t is the dynamic weight coefficient of the time step (t), which is learned by a small one-dimensional convolutional neural network (1D-CNN) that takes the hidden state sequence h0,h1,L,h before the current time step as t-1 As input, the convolution kernel size is set to 3, the step size is 1, the number of output channels is 1, and after activation by the sigmoid function, λ is obtained. t , whose value is between 0 and 1, is used to dynamically adjust the influence of historical hidden states on the forget gate decision.

[0088] α ti is the attention coefficient, which is calculated by a multi-head attention mechanism. The multi-head attention mechanism takes the current input x t and historical hidden state sequence h0,h1,L,h t-1 As input, let the number of heads be 3 and the dimension of each head be n h / 3, after a series of calculations (the specific calculation process follows the standard multi-head attention mechanism, such as linear transformation, scaled dot product attention, etc.) we get α ti , represents the historical hidden state h i-1 The relative importance of the forget gate decision at time step (t), and satisfying

[0089] Through the dynamic forget gate, the model can more flexibly decide how much of the cell state at the previous moment to retain based on the real-time situation during the processing and the importance of historical information. For example, when a sudden abnormality occurs during the processing (such as a sudden temperature rise), the forget gate can be adjusted in time to allow the cell state to quickly adapt to the new situation.

[0090] Furthermore, in order to make the input data (x_t) better adapt to the state at the current time point, an adaptive input modulation mechanism can be introduced. The calculation formula is as follows:

[0091]

[0092] Where: W xis the basis weight matrix of the input transformation, with dimension [n h ,n x ], used to perform a preliminary linear transformation on the original input (x_t). b x is the corresponding bias vector, dimension n h .

[0093] Tanh is a hyperbolic tangent function that compresses the output to between (-1) and (1), providing suitable nonlinear characteristics for subsequent calculations.

[0094] γ t is the adaptive weight coefficient at time step (t), which is learned by a simple linear layer that takes the current input x t and the previous hidden state h t-1 is the input, the output dimension is (1), and after activation by the sigmoid function, γ is obtained t , whose value is between (0) and (1), is used to adjust the influence of the previous hidden state on the input modulation.

[0095] β tj is the attention coefficient, which is calculated by a position-based attention mechanism. t and the previous hidden state h t-1 As input, according to the input x t The position in the sequence and h t-1 The dimensional features of β are calculated tj , which represents the hidden state at the previous moment (j represents the jth dimension) the relative importance of the input modulation at time step (t), and satisfies

[0096] Input after adaptive input modulation It can better fit the current operating status, so that when the model processes continuously changing time series data such as temperature and pressure during the processing process, it can better capture the dynamic characteristics in the data and improve the accuracy of predicting the changing trend of the processing process.

[0097] Then, based on the above, the cell state is updated and output as follows:

[0098]

[0099] The output gate calculation formula is:

[0100]

[0101] h t =ot e tanh(c t )

[0102] Among them, W c 、W o They are the basic weight matrices for cell state update and output gate, with dimensions of [n h ,n h +n x ] and [n h ,n h +n x ], b c 、b o are the corresponding bias vectors, with dimension n h ,The role of these parameters is to control the update of cell state and the final output by learning the relationship between input and hidden state.

[0103] By introducing a dynamic forget gate and adaptive input modulation, the improved RNN model can more intelligently process complex time series data in the rice processing process, enhance the perception of changes in the processing process and the prediction accuracy, and provide a more reliable basis for subsequent fusion with the CNN model and decision-making of the entire hybrid artificial intelligence algorithm model.

[0104] The RNN model receives multi-source time series data from the processing process, including temperature, pressure, equipment operating time, and time series data on rice quality changes (such as the rate of change in grain plumpness) obtained through rice image analysis. The time step can be set to 15, and multiple GRU layers (for example, 3 layers) are used to learn the complex temporal relationships in the data, with 64, 32, and 16 GRU units per layer, respectively.

[0105] For the feature vector F output by the convolutional neural network CNN (dimension is 256) and the feature vector F output by the recurrent neural network RNN (dimension is 16), and the fusion method based on dynamic weight allocation is used to obtain the fused feature vector F after fusion; then: F = α(t)F CNN +(1-α(t))F RNN , where α(t) is the weight coefficient that changes with time t, and the calculation formula can be:

[0106] Here, λ is a parameter that controls the rate of change of the weight (e.g., λ = 0.5), and t0 is a reference time point (determined based on the key stages of rice processing, such as t0 = 0 at the beginning of rice milling). In the early stages of processing, α(t) is large, focusing more on the initial quality features of the rice extracted by the CNN. As processing progresses, α(t) gradually decreases, and the weight of the dynamic features learned by the RNN during the processing increases.

[0107] Finally, the fused feature vector F obtained above is input into a multi-layer perceptron with at least 2 hidden layers for final prediction.

[0108] In one embodiment, the fused feature vector F is input into a multilayer perceptron (MLP) with three hidden layers for final prediction. The first hidden layer has 128 neurons, the second hidden layer has 64 neurons, and the third hidden layer has 32 neurons. The activation function for each layer is the Reinforced Luminance (ReLU) function. The output layer of the MLP has a specific number of neurons based on the prediction task. For example, when predicting the nutrient loss rate of rice, the output layer has one neuron.

[0109] The rice processing quality prediction model constructed through the aforementioned CNN-RNN hybrid model, through complex structural design, such as the combination of residual networks and attention mechanisms, and a dynamic weight fusion strategy, can more accurately extract multiple aspects of rice grain characteristics and effectively integrate the initial quality characteristics of rice with the dynamic changes during processing. For example, for rice of different varieties and origins, it can accurately predict the changes in the head rice rate and nutrient loss during processing, reducing the prediction error by 20%-30% compared to traditional models, providing a reliable basis for the precise adjustment of processing parameters.

[0110] Further, the process comes to step S103: a multi-objective optimization algorithm is combined with deep reinforcement learning to construct an equipment operation parameter optimization model, and the equipment operation parameter optimization model takes the equipment's energy consumption, processing efficiency, whole polished rice rate and rice nutrient retention rate as optimization targets to obtain the optimal parameter adjustment strategy for the rice processing equipment.

[0111] In one embodiment, the multi-objective optimization algorithm is combined with deep reinforcement learning to construct an equipment operating parameter optimization model. The equipment operating parameter optimization model uses the equipment's energy consumption, processing efficiency, whole rice polishing rate, and rice nutrient retention rate as optimization targets to obtain the optimal parameter adjustment strategy for the rice processing equipment. The steps may include:

[0112] In an embodiment, multiple optimization objectives can be used: rice polishing rate (Q), nutrient retention rate (N), equipment energy consumption (E), and processing time (T). The higher the polishing rate and nutrient retention rate, the better, while the lower the equipment energy consumption and processing time, the better.

[0113] First, multiple key parameters of the equipment are encoded as chromosomes of the multi-objective optimization algorithm. For example, the roller speed r of the rice mill (range [r min ,rmax ], encoding length k r =8), the gap g of the rice mill (value range [g min ,g max ], encoding length k g =6), polishing pressure p of polishing machine (value range [p min ,p max ], encoding length k p =7) Temperature of drying equipment T h (Value range [T hmin ,T hmax ], encoding length ) and other parameters are encoded with real numbers. A chromosome X can be represented as X = [r, g, p, T h ,L].

[0114] Secondly, the NSGA-II algorithm is used to perform genetic operations and non-inferiority sorting.

[0115] In the genetic operation of the multi-objective optimization algorithm, the selection operation adopts the tournament selection method based on crowding degree; the crossover operation adopts the improved simulated binary crossover; and the mutation operation adopts polynomial mutation.

[0116] Specifically, the selection operation uses a tournament selection method based on crowding. During tournament selection, not only the non-inferior rank of an individual is considered, but also the crowding degree of the individual in the target space. The crowding degree reflects the density of individuals around an individual; individuals with greater crowding have a higher probability of being retained during selection.

[0117] During research, the inventors discovered that using the traditional simulated binary crossover formula to generate offspring individuals resulted in a relatively simple combination of parental gene fragments, making it difficult to fully tap the potential of the parental gene fragments. This was particularly true when dealing with coded data such as rice processing equipment parameters, which exhibit complex correlations and sensitivity differences. This method may not converge quickly to a near-global optimal solution. Therefore, improvements to the simulated binary crossover were developed.

[0118] The new simulated binary crossover formula is designed as follows:

[0119] For two parent individuals X1=[x 11 ,x 12 ,L,x 1n ] and X2=[x 21 ,x 22 ,L,x 2n ], the generated offspring individual X 1′ =[x 11′ ,x 12′ ,L,x 1n′ ] and X 2′ =[x21′ ,x 22′ ,L,x 2n′ ],, the calculation formula is as follows:

[0120] x 1i′ =0.5×[(1+μ i )×x 1i +(1-μ i )×x 2i +ρ i ×|x 1i -x 2i |]

[0121] x 2i′ =0.5×[(1-μ i )×x 1i +(1+μ i )×x 2i -ρ i ×|x 1i -x 2i |]

[0122] in:

[0123] i represents the gene bit index, ranging from 1 to n, where n is the length of the individual code, corresponding to the number of different parameters of the rice processing equipment.

[0124] μ i is a crossover probability p c (For example, c =0.9) and is based on the adaptive adjustment of the gene position sensitivity parameter, and its calculation formula is:

[0125]

[0126] Here, rand() is a random function uniformly distributed in the interval [0,1], which is used to introduce randomness to ensure the exploration ability of the algorithm; η i is the sensitivity index of gene position i, which is obtained based on the historical optimization data of equipment parameters, reflecting the sensitivity of the parameters corresponding to the gene position to the objective function (such as rice quality, energy consumption, processing efficiency, etc.) during the optimization process. i larger), μ i The range of values of is relatively small, so that the offspring individuals tend to be closer to the better value of the parent at this gene position, so as to avoid destroying the existing good gene combination; the gene position with low sensitivity (i.e. η i smaller), μ i ) has a relatively large value range, which encourages offspring individuals to conduct more extensive exploration at this gene position and broaden the search space.

[0127] ρ iIt is a parameter that is adaptively adjusted based on gene position correlation, and is used to control the degree of deviation of offspring individuals relative to their parents. Its calculation formula is:

[0128] ρ i =ω i ×(rand()-0.5)×|x 1i -x 2i |

[0129] Among them, ω i is the correlation weight of gene position i, which is also obtained based on the historical optimization data mining of equipment parameters. It measures the degree of correlation between the parameter corresponding to this gene position and other parameters. i larger), ρ i The absolute value of is relatively large, which means that the changes of offspring individuals at this gene position will be more affected by the differences of the parents to adapt to the synergistic effect between parameters; the gene position with weak correlation (i.e. ω i smaller), ρ i The absolute value of is relatively small, and the offspring individuals have relatively independent variation space at this gene position, preventing them from being overly constrained by other parameters.

[0130] Through the above-mentioned improved simulated binary crossover formula, the generation strategy of offspring individuals can be adaptively adjusted according to the characteristics of the rice processing equipment parameters. While ensuring population diversity, it accelerates the convergence to a better parameter combination, thereby improving the intelligent optimization effect of rice processing equipment and laying a solid foundation for the production of higher quality and lower energy consumption rice products.

[0131] The mutation operation uses polynomial mutation. The mutation operation is a key means to avoid the algorithm from falling into local optimality when introducing new genes. The inventors found in their research that the traditional polynomial mutation formula has the problem of blind mutation direction and difficulty in accurately adapting parameter characteristics when dealing with complex parameter optimization such as rice processing equipment. As a result, the new individuals generated after mutation may not be able to effectively push the population closer to the global optimal. Therefore, optimization and improvement are carried out based on the traditional polynomial mutation formula. For each gene x in the chromosome i , the mutated gene x i′ The calculation formula is:

[0132]

[0133] in:

[0134] and It's gene x iThe value range boundaries correspond to the maximum and minimum values allowed for a parameter of rice processing equipment, such as the upper and lower limits of the rice mill speed. These are fixed values set based on the physical properties of the equipment and actual processing requirements to ensure that the mutated parameters remain within a reasonable range and do not produce invalid or equipment-damaging parameter values.

[0135] τ i It's gene x i The current mutation iteration number gradually increases as the algorithm runs, and the value is increased by 1 for each mutation operation; It's gene x i The maximum number of mutation iterations is pre-set, which is determined based on the historical experience of equipment parameter optimization and the expected convergence speed. For example, for some parameters with high sensitivity and great influence on rice processing quality, such as rice drying temperature, the maximum number of mutation iterations is pre-set, which is based on the historical experience of equipment parameter optimization and the expected convergence speed. It may be relatively small, which means that it is expected to find a better value within a smaller number of mutations to avoid excessive mutations that destroy the existing good parameter combinations.

[0136] λ i Is related to gene x i The variation intensity adjustment coefficient closely related to the characteristics is obtained by analyzing the historical optimization data of equipment parameters, reflecting the sensitivity and change trend characteristics of the parameters corresponding to the gene during the optimization process. For highly sensitive parameters, such as the roller gap of the rice mill, λ i The smaller the value, the relatively limited variation range, to prevent small parameter deviations from causing large fluctuations in rice processing quality; for parameters with low sensitivity, such as the operating time of some auxiliary equipment, λ i A larger value allows a wider range of variation to broaden the search space and explore potential better parameter combinations.

[0137] δ i is the variable asynchronous length, and its calculation formula is:

[0138]

[0139] In the above formula, $\text{rand}()$ is a random function uniformly distributed in the interval [0,1], which is used to introduce the necessary randomness to ensure that the algorithm does not fall into a fixed pattern; $\eta_{i}$ is the variation distribution index of gene (i), which is also estimated based on the historical optimization data of device parameters. It determines the variation asynchronous length δ i The distribution of , thus affecting the radical degree of mutation. i When it is larger, the asynchronous length δ i The values of are concentrated in a small range, and the variation is relatively mild, which is suitable for fine-tuning parameters in the late stage of the algorithm and when approaching the optimal solution; when η i When it is small, the asynchronous length δ iThe value range of is wider and the mutation is more radical, which is conducive to quickly exploring the parameter space in the early stage of the algorithm.

[0140] Through the above-mentioned improved polynomial mutation formula, the mutation direction, intensity and timing can be accurately controlled according to the unique properties of each parameter of rice processing equipment, so that the algorithm can maintain sufficient innovation ability to break out of the local optimum in the process of exploring the optimal solution of parameters, and can steadily move forward in the direction that is conducive to improving rice processing quality, reducing energy consumption and other goals, thereby effectively improving the level of intelligent management and control of rice processing equipment.

[0141] Furthermore, in the non-inferiority sorting, a multi-objective optimization algorithm is used to perform non-inferiority sorting, including performing non-inferiority sorting on individuals in the population according to the optimization objective.

[0142] For example, according to multiple (such as 4) optimization objectives, the individuals in the population are sorted in a non-inferior manner. a and X b , if X a No worse than X in all four objectives b , and is better than X in at least one objective b , then X a DominateX b The population is divided into different non-inferior levels through non-inferiority sorting, and the lower the level, the better the individual.

[0143] Furthermore, in the deep reinforcement learning model, a dual-Q network and a reward function R are used, wherein the state space S in the dual-Q network includes the current rice quality characteristics evaluated by the rice processing quality prediction model, the current operating parameters of the equipment, the operating status of the equipment, and environmental information; the action space A is the parameter adjustment strategy in the multi-objective optimization algorithm.

[0144] Specifically, the deep reinforcement learning model uses a Dueling Q-network (Dueling DQN). The Dueling DQN's state space S includes the current rice quality characteristics (assessed by the rice processing quality prediction model), the equipment's current operating parameters (such as the current roller speed and polishing pressure), the equipment's operating status (such as motor temperature and equipment vibration amplitude), and environmental information (such as the workshop's temperature, humidity, and light intensity). The action space A is the parameter adjustment strategy in the NSGA-II algorithm, such as adjusting the roller speed and changing the polishing pressure.

[0145] The reward function R can be designed as: R = α1Q + α2N - α3E - α4T + α5C + α6M,

[0146] In the above formula, α1, α2, α3, α4, α5, and α6 are weight coefficients, representing the contribution of the head rice yield, nutrient retention, energy consumption, processing time, satisfaction of equipment constraints, and satisfaction of equipment maintenance requirements to the reward, respectively. C represents the equipment parameter constraints, such as roller speed not exceeding the rated speed or polishing pressure not being excessive. If the constraints are met, C = 1; otherwise, C = 0. M represents the equipment maintenance requirement indicator, such as the maintenance requirement calculated based on factors such as equipment operating time and vibration amplitude, and ranges from [0, 1].

[0147] Furthermore, a parameter adjustment strategy is selected according to the current state S, and the new state S' and reward R are fed back through the environmental information.

[0148] Dueling DQN selects an action a (i.e., parameter adjustment strategy) based on the current state s. After executing the action, the environment feeds back the new state s' and reward R.

[0149] Furthermore, the deep reinforcement learning model selects the optimal parameter adjustment strategy among the non-inferior solutions generated by the multi-objective optimization algorithm by repeatedly learning the state-action-reward relationship.

[0150] The NSGA-II model combined with deep reinforcement learning, provided by the embodiments of the present invention, achieves a better balance between the polished rice yield, nutrient retention rate, equipment energy consumption, and processing time of rice through comprehensive consideration of multiple objectives and interaction of complex genetic operations with deep reinforcement learning. For example, during actual processing, the polished rice yield can be increased by 10%-15%, the nutrient retention rate by 8%-12%, while equipment energy consumption is reduced by 15%-20%, and the processing time is shortened by 10%-15%, significantly improving the quality and efficiency of rice processing and reducing production costs.

[0151] The hybrid model described above can also dynamically adjust equipment parameters based on real-time changes in rice quality, fluctuations in equipment operating status, and changes in environmental factors. Whether it's a sudden change in the rice raw material or a minor equipment failure or performance degradation during operation, the system can respond promptly and select the optimal parameter adjustment strategy, ensuring the stability and reliability of the rice processing process, reducing the risk of processing quality degradation and equipment damage caused by external factors, and improving the intelligent management level and market competitiveness of rice processing companies.

[0152] Further, the process comes to step S104: based on the rice processing related data collected by the high-precision sensor network, the rice processing quality prediction model is used to evaluate the quality of the current batch of rice to obtain the rice quality prediction result after the current batch of rice is processed.

[0153] At step S105: according to the rice quality prediction result after the current batch of rice is processed, the equipment operating parameter optimization model is used to adjust the operating parameters of the rice processing equipment to obtain the optimal parameters of the rice processing equipment.

[0154] Furthermore, in one embodiment, the intelligent control method for rice processing equipment also includes: during the processing process, using an adaptive control algorithm to dynamically adjust equipment parameters based on the deviation between real-time monitored data and expected target values; wherein the adaptive control algorithm includes a model reference adaptive control algorithm, which outputs expected rice quality indicators and equipment operating efficiency by inputting rice characteristics and equipment initial parameters, and optimizes and adjusts equipment parameters in combination with fuzzy logic.

[0155] Specifically, the adaptive control algorithm design employed a model reference adaptive control (MRAC) algorithm, combined with fuzzy logic for optimization. First, an ideal reference model of the rice processing process was established. This model used desired rice quality indicators (such as head rice percentage and chalkiness) and equipment operating efficiency as outputs, and rice characteristics (such as variety, moisture content, and grain size) and initial equipment parameters as inputs.

[0156] The mathematical expression of the reference model is as follows:

[0157] In the above formula, y m (k) is the output of the reference model at time k, a i and b j are the coefficients of the reference model, u(k) is the control input (i.e., the adjustment amount of the equipment operating parameters), and n and m are the model orders (e.g., n=3, m=2).

[0158] The output y of the actual processing process p (k) and the reference model output y m (k) and compare them to get the error e(k)=y m (k)-y p (k).

[0159] The fuzzy logic controller adjusts device parameters based on the error e(k) and its rate of change, Δe(k) = e(k) - e(k-1). Both the input and output of the fuzzy logic controller are fuzzified, dividing the error e(k) into five fuzzy sets: negative large (NB), negative medium (NM), zero (ZO), positive medium (PM), and positive large (PB). The error rate of change, Δe(k), is also divided into the same five fuzzy sets. The output device parameter adjustment, Δu(k), is divided into five fuzzy sets: negative large (NB), negative small (NS), zero (ZO), positive small (PS), and positive large (PB).

[0160] Fuzzy reasoning uses the Mamdani method, and defuzzification uses the center of gravity method to obtain accurate equipment parameter adjustment Δu(k). For example, for the adjustment of the rice mill speed, according to the output Δu(k) of the fuzzy logic controller, the actual speed adjustment formula is: r(k)=r(k-1)+Δu(k)×k r , where r(k) is the rice mill speed at time k, k r is the speed adjustment coefficient (set according to the characteristics of the rice mill, for example, k r =10).

[0161] In the rice processing scenario, the rice processing quality prediction model is used to evaluate the current rice being processed and provide an optimization strategy for equipment operating parameters. These parameters are passed as initial recommended values to the adaptive control and decision-making unit in the intelligent control system.

[0162] Based on these initial recommended values, the adaptive control algorithm (e.g., model-referenced adaptive control combined with fuzzy logic) in the adaptive control and decision-making unit monitors in real time the difference between the actual processing output (e.g., rice quality indicators and equipment operating efficiency) and the reference model output. A fuzzy logic controller adjusts equipment parameters based on the error and its rate of change, continuously aligning the processing process with the optimal state determined by the hybrid artificial intelligence algorithm model.

[0163] Simultaneously, real-time data from the adaptive control process (such as equipment operating status and rice quality indicators) is fed back to the rice processing quality prediction model. Based on this new data, the model updates rice quality assessment and equipment parameter optimization strategies, forming a closed-loop feedback mechanism. For example, if the adaptive control process detects poor performance from equipment parameter adjustments or new changes in rice quality, the rice processing quality prediction model can recalculate a more appropriate parameter adjustment strategy and pass it back to the adaptive control component for execution, thereby continuously optimizing the rice processing process.

[0164] In summary, the intelligent control method for rice processing equipment provided by the above-mentioned embodiments of the present invention can utilize precise data collection and advanced artificial intelligence algorithm models to predict rice processing quality in real time and promptly adjust equipment parameters based on the prediction results. For example, for rice varieties with higher hardness, the rice milling pressure and time can be automatically optimized to avoid excessive milling that increases the broken rice rate, thereby effectively improving the head rice rate and reducing undesirable indicators such as chalkiness, producing higher-quality rice products and meeting market demand for high-quality rice.

[0165] At the same time, the equipment operating parameter optimization model uses energy consumption as one of its optimization objectives. By rationally adjusting equipment operating parameters, for example, automatically adjusting the drying temperature and time based on rice and ambient humidity while ensuring rice processing quality, avoiding energy waste caused by over-drying. Furthermore, the operating parameters of the rice mill and polisher are optimized to reduce overall equipment energy consumption. In the long term, this can significantly reduce production costs for rice processing companies, improving their economic benefits and market competitiveness.

[0166] Figure 2 An exemplary structural block diagram of an intelligent control system 200 for rice processing equipment according to some embodiments of the present invention is shown;

[0167] like Figure 2 As shown, in the system 200, it includes:

[0168] Multi-source data fusion acquisition module 201 is used to collect rice processing-related data using a high-precision sensor network; the high-precision sensor network includes: equipment status sensors and image sensors installed at multiple preset links of the rice processing equipment; environmental sensors are installed around the rice processing equipment to obtain environmental change indicators; and a near-infrared spectrometer is installed at the finished rice product discharge port to perform rapid spectral scanning on the processed rice;

[0169] Quality prediction model building module 202: used to build a rice processing quality prediction model based on convolutional neural network and recurrent neural network and using historical processing data;

[0170] Equipment operating parameter optimization model construction module 203: uses a multi-objective optimization algorithm combined with deep reinforcement learning to construct an equipment operating parameter optimization model, wherein the equipment operating parameter optimization model uses the equipment's energy consumption, processing efficiency, whole rice polishing rate, and rice nutrient retention rate as optimization targets to obtain the optimal parameter adjustment strategy for the rice processing equipment;

[0171] Quality prediction module 204: evaluating the quality of a current batch of rice using the rice processing quality prediction model based on the rice processing related data collected by the high-precision sensor network to obtain a prediction result of the quality of the rice after processing the current batch of rice;

[0172] Parameter adjustment module 205: Based on the rice quality prediction result after the current batch of rice is processed, the equipment operating parameter optimization model is used to adjust the operating parameters of the rice processing equipment to obtain the optimal parameters of the rice processing equipment.

[0173] It can be seen that the solution provided by system 200 can be regarded as a specific implementation of method 100. Therefore, the description in each embodiment of method 100 is also applicable here.

[0174] It should be noted that, for the purpose of simplicity, the present invention describes some methods and embodiments thereof as a series of actions and combinations thereof, but those skilled in the art will understand that the scheme of the present invention is not limited by the order of the described actions. Therefore, based on the disclosure or teachings of the present invention, those skilled in the art will understand that some of the steps therein can be performed in other orders or simultaneously. Further, those skilled in the art will understand that the embodiments described in the present invention can be regarded as optional embodiments, that is, the actions or modules involved therein are not necessarily necessary for the implementation of one or more schemes of the present invention. In addition, depending on the different schemes, the present invention also has different emphases on the description of some embodiments. In view of this, those skilled in the art will understand that the parts that are not described in detail in a certain embodiment of the present invention may also refer to the relevant descriptions of other embodiments.

[0175] Figure 3 FIG. 1 shows an exemplary structural block diagram of an intelligent control device 300 for rice processing equipment according to an embodiment of the present invention.

[0176] like Figure 3 As shown, the system 300 includes: a memory 301 and a processor 302, wherein the memory is used to store a computer program, and the processor is used to implement the intelligent control method for rice processing equipment as described in any embodiment of the method 100 when executing the computer program.

[0177] Although a number of embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art may devise numerous modifications, variations, and alternatives without departing from the concept and spirit of the present invention. It should be understood that in practicing the present invention, various alternatives to the embodiments of the present invention described herein may be employed. The appended claims are intended to define the scope of the present invention and therefore cover equivalents or alternatives within the scope of these claims.

Claims

1. An intelligent control method for rice processing equipment, characterized in that, include: Use a high-precision sensor network to collect data related to rice processing; The high-precision sensor network includes: installing equipment status sensors and image sensors at multiple preset links of the rice processing equipment; setting environmental sensors around the rice processing equipment to obtain environmental change indicators; and setting a near-infrared spectrometer at the finished rice outlet to perform rapid spectral scanning on the processed rice; Based on convolutional neural networks and recurrent neural networks, a rice processing quality prediction model was constructed using historical processing data; A multi-objective optimization algorithm is combined with deep reinforcement learning to construct an equipment operating parameter optimization model. The equipment operating parameter optimization model uses the equipment's energy consumption, processing efficiency, whole rice polishing rate, and rice nutrient retention rate as optimization targets to obtain the optimal parameter adjustment strategy for the rice processing equipment. According to the rice processing related data collected by the high-precision sensor network, the quality of the current batch of rice is evaluated using the rice processing quality prediction model to obtain a prediction result of the rice quality after the current batch of rice is processed; According to the rice quality prediction result after the current batch of rice is processed, the equipment operating parameter optimization model is used to adjust the operating parameters of the rice processing equipment to obtain the optimal parameters of the rice processing equipment.

2. The method according to claim 1, characterized in that The high-precision sensor network further includes a rice quality monitoring sensor, wherein the rice quality monitoring sensor includes: A laser particle size analyzer is set at the front end of the rice feeding conveyor belt to measure the particle size distribution of the rice; Install an odor sensor to detect whether the rice has any foreign smell.

3. The method according to claim 1, characterized in that The device status sensor includes: Install micro strain gauge sensors on the surface of the roller inside the rice mill to monitor the force distribution of the roller in real time during the rice milling process; An airflow sensor is set in the polishing chamber of the polishing machine to detect airflow data during the polishing process.

4. The method according to claim 1, wherein The image sensor comprises: A multispectral camera is installed above the rice feeding conveyor belt to capture images of the rice's appearance, and the images are analyzed using image recognition technology to convert the image data into feature data.

5. The method according to claim 1, wherein The environmental sensors include seismic sensors to monitor seismic activity in real time, and ultraviolet sensors to monitor ultraviolet intensity within the processing workshop.

6. The method according to claim 1, characterized in that The rice processing quality prediction model is constructed based on the convolutional neural network and the recurrent neural network using historical processing data, wherein the convolutional neural network includes: The improved residual network structure is used as the basic architecture of the convolutional neural network, including: deep optimization of the network structure through deformable convolution kernels to perform preliminary feature extraction on rice images; In the middle part of the residual network, an attention mechanism module is set to reweight the feature map by calculating the spatial attention and channel attention of the feature map; At the end of the residual network, a global average pooling layer is used to convert the feature map into a feature vector, and then a fully connected layer is connected. The fully connected layer has 256 neurons and the activation function is ReLU for further feature integration. The recurrent neural network comprises: utilizing multi-source time series data received during the processing process, including temperature, pressure, equipment operating time, and rice images, and obtaining time series data on rice quality changes through a structure modified based on a dynamic forget gate and adaptive input modulation; The feature vector output by the convolutional neural network and the feature vector output by the recurrent neural network are fused based on a fusion method of dynamic weight allocation to obtain a fused feature vector. The fused feature vector is input into a multilayer perceptron with at least 2 hidden layers for final prediction.

7. The method according to claim 1, characterized in that The method of combining a multi-objective optimization algorithm with deep reinforcement learning to construct an equipment operation parameter optimization model, wherein the equipment operation parameter optimization model takes the energy consumption, processing efficiency, whole rice polishing rate, and rice nutrient retention rate of the equipment as optimization targets, and obtains the optimal parameter adjustment strategy for the rice processing equipment, comprising the following steps: Multiple key parameters of the equipment are encoded as chromosomes of the multi-objective optimization algorithm; In the genetic operation of the multi-objective optimization algorithm, the selection operation adopts the tournament selection method based on crowding degree; the crossover operation adopts the improved simulated binary crossover; the mutation operation adopts the improved polynomial mutation; Performing non-inferiority sorting using a multi-objective optimization algorithm, including performing non-inferiority sorting on individuals in a population using the optimization objective; In the deep reinforcement learning model, a dual-Q network and a reward function R are used, wherein the state space S in the dual-Q network includes the current rice quality characteristics evaluated by the rice processing quality prediction model, the current operating parameters of the equipment, the operating status of the equipment, and environmental information; the action space A is the parameter adjustment strategy in the multi-objective optimization algorithm; Select a parameter adjustment strategy based on the current state S, and feed back the new state S' and reward R through the environmental information; The deep reinforcement learning model repeatedly learns the state-action-reward relationship and selects the optimal parameter adjustment strategy among the non-inferior solutions generated by the multi-objective optimization algorithm.

8. The method according to claim 1, characterized in that The method further comprises: during the processing, using an adaptive control algorithm to dynamically adjust equipment parameters according to the deviation between the real-time monitored data and the expected target value; wherein, The adaptive control algorithm includes a model reference adaptive control algorithm, which outputs expected rice quality indicators and equipment operating efficiency by inputting rice characteristics and equipment initial parameters, and optimizes and adjusts equipment parameters in combination with fuzzy logic.

9. An intelligent control system for rice processing equipment, characterized in that: include: Multi-source data fusion acquisition module: used to collect rice processing related data using a high-precision sensor network; The high-precision sensor network includes: installing equipment status sensors and image sensors at multiple preset links of the rice processing equipment; setting environmental sensors around the rice processing equipment to obtain environmental change indicators; and setting a near-infrared spectrometer at the finished rice outlet to perform rapid spectral scanning on the processed rice; Quality prediction model building module: used to build a rice processing quality prediction model based on convolutional neural networks and recurrent neural networks using historical processing data; Equipment operating parameter optimization model construction module: This module uses a multi-objective optimization algorithm combined with deep reinforcement learning to construct an equipment operating parameter optimization model. The equipment operating parameter optimization model uses the equipment's energy consumption, processing efficiency, whole rice polishing rate, and rice nutrient retention rate as optimization targets to obtain the optimal parameter adjustment strategy for the rice processing equipment. A quality prediction module is configured to evaluate the quality of a current batch of rice using the rice processing quality prediction model based on the rice processing related data collected by the high-precision sensor network to obtain a prediction result of the rice quality after the current batch of rice is processed; Parameter adjustment module: According to the rice quality prediction result after the current batch of rice is processed, the equipment operating parameter optimization model is used to adjust the operating parameters of the rice processing equipment to obtain the optimal parameters of the rice processing equipment.

10. An intelligent control device for rice processing equipment, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to implement the intelligent control method for rice processing equipment according to any one of claims 1 to 8 when executing the computer program.

Images