Wheat germ processing technology optimization suggestion generation method and system based on co-occurrence analysis

Abstract

The invention provides a wheat germ processing technology optimization suggestion generation method and system based on co-occurrence analysis, and the method comprises the steps: firstly obtaining a raw material data set of a target batch of wheat germs, covering a multi-batch historical processing parameter set and a quality evaluation index, carrying out the co-occurrence feature extraction of the raw material data set, and obtaining a co-occurrence feature set of the target batch of wheat germs; a processing co-occurrence feature set containing raw material attribute association, process parameter coupling and quality fluctuation mode features is obtained, then a pre-trained co-occurrence analysis model is called to dynamically optimize and analyze the processing co-occurrence feature set, and a process optimization suggestion set comprising a raw material preprocessing optimization item, a processing parameter adjustment item and a quality monitoring enhancement item is generated; and finally, the process optimization suggestion set is fed back to a wheat germ processing control system, processing parameter adjustment operation is triggered, and optimization of the wheat germ processing process is achieved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method and system for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis. Background Art

[0002] In the field of wheat germ processing, traditional process optimization methods have long faced many bottlenecks that are difficult to break through. Currently, the commonly used process optimization means in the industry mainly rely on empirical judgment and local test adjustment. This method is often limited to the isolated analysis of a single processing parameter or a few raw material attributes, lacking a global understanding of the internal relationships among complex factors in the processing process. Specifically, when dealing with wheat germ processing data, most of the existing technologies only perform simple statistics or comparisons on processing parameters, and fail to deeply explore the potential co-occurrence relationships between historical processing parameter groups and corresponding quality evaluation indicators, resulting in difficulty in accurately identifying the key factor combinations affecting product quality and their dynamic change laws.

[0003] In addition, the existing process optimization methods are unable to cope when faced with complex situations such as fluctuations in raw material attributes, coupling effects of process parameters, and quality fluctuation patterns. Due to the lack of systematic extraction and analysis of the associated characteristics of raw material attributes, the coupling characteristics of process parameters, and the characteristics of quality fluctuation patterns, the existing technologies cannot provide comprehensive and targeted optimization suggestions for the processing process, making it difficult to fundamentally solve the problems in the processing process, and restricting the improvement of product quality stability and production efficiency. Summary of the Invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of this application, embodiments of this application provide a method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis. The method includes: Obtain a raw material data set of wheat germ for the target batch, where the raw material data set includes historical processing parameter groups and corresponding quality evaluation indicators for multiple processing batches; Perform co-occurrence feature extraction processing on the raw material data set to obtain a processing co-occurrence feature set for each processing batch. The processing co-occurrence feature set includes associated characteristics of raw material attributes, coupling characteristics of process parameters, and characteristics of quality fluctuation patterns; Call a pre-trained co-occurrence analysis model to perform dynamic optimization analysis processing on the processing co-occurrence feature set, and generate a process optimization suggestion set for the target batch. The process optimization suggestion set includes raw material pretreatment optimization items, processing parameter adjustment items, and quality monitoring strengthening items; Feed back the process optimization suggestion set to the wheat germ processing control system to trigger processing parameter adjustment operations.

[0005] In another aspect, an embodiment of the present application further provides a processing service system, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to run the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0006] Based on the above aspects, the embodiment of the present application integrates historical processing parameters and quality evaluation indicators of multiple batches. On this basis, through co-occurrence feature extraction processing, the potential associations among raw material attributes, process parameters and quality fluctuations are made explicit in the form of a feature set. The pre-trained co-occurrence analysis model can intelligently identify the key factors affecting the processing effect through dynamic optimization analysis processing, and generate a targeted set of process optimization suggestions. This set of process optimization suggestions covers the entire process from raw material pretreatment to processing parameter adjustment and then to quality monitoring enhancement, ensuring the systematicness and coordination of optimization measures. Finally, by seamlessly feeding back the set of optimization suggestions to the wheat germ processing control system, real-time adjustment of processing parameters and continuous improvement of the process are achieved, significantly improving the processing efficiency and product quality of wheat germ. Description of the Drawings

[0007] Figure 1 It is a schematic execution flowchart of a method for generating wheat germ processing process optimization suggestions based on co-occurrence analysis provided by an embodiment of the present application.

[0008] Figure 2 It is a schematic hardware architecture diagram of a processing service system provided by an embodiment of the present application. Detailed Embodiments

[0009] The present application will be specifically described below in conjunction with the accompanying drawings of the specification. Figure 1 It is a schematic flowchart of a method for generating wheat germ processing process optimization suggestions based on co-occurrence analysis provided by an embodiment of the present application. The method for generating wheat germ processing process optimization suggestions based on co-occurrence analysis will be introduced in detail below.

[0010] Step S110, obtain a raw material data set of wheat germ of a target batch. The raw material data set includes historical processing parameter groups of multiple processing batches and corresponding quality evaluation indicators.

[0011] In this embodiment, in a certain wheat germ processing enterprise, wheat germ has been processed for a long time. For the target batch of wheat germ, the enterprise's database stores relevant data of multiple processing batches, thus forming a raw material data set. For example, in the past 10 processing batches, each batch has a detailed historical processing parameter set. Taking one of the processing batches as an example, the initial moisture content of wheat germ in this batch is 12%, the protein content is 25%, and the impurity ratio is 3%. During the processing, the cleaning duration is set to 30 minutes, the drying temperature is set to 50 degrees Celsius, the tabletting pressure is set to 100 kgf, and the storage environment humidity is 40%. At the same time, for the wheat germ after the processing of this batch, the enterprise obtained corresponding quality evaluation indicators through a series of quality inspection means. For example, the retention rate of nutritional components reached 80%, the score of the appearance color and luster was 8 points (full score 10 points), and the microbial indicators were qualified. For the other 9 processing batches, there are also similar detailed historical processing parameter sets and corresponding quality evaluation index records. These data cover all aspects from raw material input to quality inspection after processing, thus forming a rich raw material data set.

[0012] Step S120: Perform co-occurrence feature extraction processing on the raw material data set to obtain a processing co-occurrence feature set for each processing batch. The processing co-occurrence feature set includes raw material attribute association features, process parameter coupling features, and quality fluctuation mode features.

[0013] Continuing with the above-mentioned wheat germ processing enterprise as an example, when performing co-occurrence feature extraction processing on the raw material data set, first extract the raw material attribute subset and process parameter subset of each processing batch from the raw material data set. For the raw material attribute subset, for the wheat germ in a certain processing batch mentioned before, its moisture content, protein content, and impurity ratio are extracted. For the process parameter subset, the cleaning duration, drying temperature, tabletting pressure, and storage environment humidity in this batch are also extracted.

[0014] Then, synchronous sliding window sampling is performed on the subsets of raw material attributes and process parameters. Assuming that according to the characteristics of the enterprise's processing equipment, the length of the sliding window is set to 5 processing time units, and the step size is 1 processing time unit. In the first processing stage, taking the initial 5 processing time units as a sliding window, calculate the mean value of raw material attributes within this sliding window. For example, the mean moisture content may be 11.5%, the mean protein content is 24.8%, and the mean impurity ratio is 2.8%. At the same time, obtain the maximum values of process parameters, such as the maximum cleaning duration is 30 minutes, the maximum drying temperature is 50 degrees Celsius, the maximum tabletting pressure is 100 kgf, and the maximum storage environment humidity is 40%. In this way, a sequence of process parameters for one processing stage is obtained. In this manner, as the processing progresses, the sliding window is continuously shifted to obtain sequences of process parameters for multiple processing stages.

[0015] Construct a first association graph based on the co-occurrence frequency of raw material attributes and process parameters in the process parameter sequence. For example, through statistics, it is found that in multiple processing stages, when the moisture content is relatively high, the situation where the drying temperature is also relatively high appears with a relatively high frequency. This indicates a strong association between the moisture content and the drying temperature. By calculating the frequency of this co-occurrence, the co-occurrence intensity weight between them can be determined. Then, construct a first association graph with raw material attributes as nodes, process parameters as edges, and co-occurrence intensity weight as the edge weight, thereby extracting the associated characteristics of raw material attributes.

[0016] Next, construct a second association graph based on the synchronous change trend between different process parameters in the process parameter sequence. For example, during the processing, it is found that when the drying temperature increases, the tabletting pressure also shows an increasing trend. By calculating the similarity of this change trend (which can be obtained by matching using the dynamic time warping algorithm) to determine the coupling strength between process parameters, construct a second association graph with process parameters as nodes and coupling strength as the edge weight, and then extract the coupling characteristics of process parameters.

[0017] Finally, perform pattern matching on the variation range of quality evaluation indicators in multiple processing stages to extract quality fluctuation pattern characteristics. For example, during the processing, it is found that as the processing progresses, the retention rate of nutritional components shows a large decline in some stages, and the appearance color score also fluctuates. By analyzing the change patterns of these quality evaluation indicators in different processing stages, such as patterns of rising first and then falling or continuous falling, etc., the quality fluctuation pattern characteristics are extracted.

[0018] Step S130, call the pre-trained co-occurrence analysis model to perform dynamic optimization and parsing processing on the processing co-occurrence feature set, and generate a set of process optimization suggestions for the target batch. The set of process optimization suggestions includes raw material pretreatment optimization items, processing parameter adjustment items, and quality monitoring strengthening items.

[0019] For example, for the above wheat germ processing enterprise, after obtaining the processing co-occurrence feature set, a pre-trained co-occurrence analysis model is called for processing. The raw material attribute correlation features and process parameter coupling features are input into the parameter optimization module of the pre-trained co-occurrence analysis model. For example, data such as the correlation relationship between the moisture content and the drying temperature in the raw material attribute correlation features, and the coupling relationship between the drying temperature and the tableting pressure in the process parameter coupling features are input. The model analyzes and obtains the deviation weight between the current processing parameter group and the target quality index. Suppose the target quality index is that the nutrient retention rate should reach 85%, and the predicted value of the nutrient retention rate under the current processing parameter group is 80%. The model determines the deviation weight of each process parameter for achieving the target quality index by analyzing the raw material attribute correlation features and the process parameter coupling features. For example, the deviation weight of the drying temperature is 0.3, indicating that the drying temperature has a greater impact on not reaching the target quality index.

[0020] Based on the deviation weight, the process parameter coupling features are dynamically corrected. The parameter influence factor of each process parameter is extracted from the deviation weight. For example, the parameter influence factor of the drying temperature is 0.3. According to this parameter influence factor, the weight adjustment coefficient of each process parameter edge in the second association graph is determined. Suppose the initial weight of the edge connecting the drying temperature and the tableting pressure is 0.5, and its weight adjustment coefficient is determined to be 1.2 according to the parameter influence factor (this is just a hypothetical calculation method, and the actual value is determined by the complex calculation mechanism inside the model). Based on this weight adjustment coefficient, the weights of the process parameter edges in the second association graph are dynamically scaled proportionally. For example, the weight of the edge connecting the drying temperature and the tableting pressure is scaled to 0.5 * 1.2 = 0.6. The cross-edge weight normalization process is performed on the scaled process parameter coupling features to make the sum of all process parameter edge weights consistent with that before scaling, generating the normalized process parameter coupling features. According to these normalized process parameter coupling features, the edge weight distribution in the second association graph is updated to generate the corrected process parameter coupling features.

[0021] Then, the corrected process parameter coupling features and the quality fluctuation mode features are input into the suggestion generation module of the pre-trained co-occurrence analysis model. The priority score of the raw material pretreatment optimization item, the adjustment amplitude threshold of the processing parameter adjustment item, and the monitoring frequency of the quality monitoring strengthening item are calculated. For example, the priority score of the optimization item of screening wheat germ in the raw material pretreatment to reduce the impurity ratio is calculated to be 0.8, indicating that this optimization item is relatively important; the adjustment amplitude threshold of the tableting pressure in the processing parameter adjustment item is ±5 kgf, that is, the tableting pressure can be adjusted up and down by 5 kgf based on the current value; the monitoring frequency of the nutrients in the quality monitoring strengthening item is increased from once every 2 hours to once every 1 hour. According to these priority scores, adjustment amplitude thresholds, and monitoring frequencies, a process optimization suggestion set is generated.

[0022] Step S140: Feed back the set of process optimization suggestions to the wheat germ processing control system to trigger the processing parameter adjustment operation.

[0023] In this embodiment, after the set of process optimization suggestions is generated, it can be fed back to the wheat germ processing control system. For the moisture control suggestion in the raw material pretreatment optimization item, assume that the duration control of the cleaning equipment in the previous processing was based on fixed parameter settings, and now according to the optimization suggestion, the moisture needs to be controlled more precisely. If the optimization suggestion is to reduce the moisture content of wheat germ to 10%, and it is calculated that the duration of the cleaning equipment needs to be increased from the original 30 minutes to 35 minutes, then convert this moisture control suggestion into a duration control instruction for the cleaning equipment, such as "Set the duration of the cleaning equipment to 35 minutes".

[0024] For the tablet pressing pressure suggestion in the processing parameter adjustment item, convert it into a pressure gradient adjustment parameter for the tablet press. If the previous pressure setting of the tablet press was 100 kgf, and now according to the optimization suggestion, the adjustment amplitude threshold of the tablet pressing pressure is ±5 kgf, then send a pressure gradient adjustment parameter like "Adjust the tablet pressing pressure to 105 kgf" to the tablet press.

[0025] For the detection frequency suggestion in the quality monitoring enhancement item, convert it into a sampling interval configuration for the quality inspection instrument. If the previous detection of nutritional components was once every 2 hours, and now according to the optimization suggestion, it needs to be increased to once every 1 hour, then send a sampling interval configuration like "Set the sampling interval of nutritional components to 1 hour" to the quality inspection instrument.

[0026] Then issue these duration control instructions, pressure gradient adjustment parameters, and sampling interval configurations to the wheat germ processing control system, and receive the adjustment confirmation signal returned by the wheat germ processing control system to complete the closed-loop control. For example, after receiving these instructions, the wheat germ processing control system adjusts the corresponding parameters of the cleaning equipment, tablet press, and quality inspection instrument according to the instructions. After the adjustment is completed, it returns a confirmation signal of "Adjustment completed" to the sending end, indicating that the processing parameter adjustment operation has been successfully executed, and the entire system enters a new processing state and operates according to the optimized parameters.

[0027] Based on the above steps, in the embodiments of the present application, by integrating historical processing parameters and quality evaluation indicators of multiple batches, and on this basis, through co-occurrence feature extraction processing, the potential associations among raw material attributes, process parameters, and quality fluctuations are made explicit in the form of a feature set. The pre-trained co-occurrence analysis model can intelligently identify the key factors affecting the processing effect through dynamic optimization analysis processing, and generate a targeted set of process optimization suggestions. This set of process optimization suggestions covers the entire process from raw material pretreatment to processing parameter adjustment and then to quality monitoring enhancement, ensuring the systematicness and synergy of the optimization measures. Finally, by seamlessly feeding back the set of optimization suggestions to the wheat germ processing control system, real-time adjustment of processing parameters and continuous improvement of the process are achieved, significantly improving the processing efficiency and product quality of wheat germ.

[0028] In a possible implementation manner, step S120 includes: Step S121, extracting the raw material attribute subset and the process parameter subset of each processing batch from the raw material data set. The raw material attribute subset includes the moisture content, protein content, and impurity ratio of wheat germ, and the process parameter subset includes the cleaning duration, drying temperature, tabletting pressure, and storage environment humidity.

[0029] In this embodiment, taking a certain processing batch as an example, in the raw material attribute subset, the moisture content of wheat germ is detected to be 12%, the protein content is 25%, and the impurity ratio is 3%. In the process parameter subset, the cleaning duration is set to 30 minutes, the drying temperature is set to 50 degrees Celsius, the tabletting pressure is set to 100 kilogram-force, and the storage environment humidity is 40%. This is the content of the raw material attribute subset and the process parameter subset completely extracted from the data of this processing batch.

[0030] Step S122, performing synchronous sliding window sampling on the raw material attribute subset and the process parameter subset to obtain process parameter sequences of multiple processing stages.

[0031] In this embodiment, it is assumed that the sliding window length is 5 processing time units and the step size is 1 processing time unit. Starting from the beginning of processing, the first sliding window covers the first 5 processing time units. Within this window, the mean value of the raw material attributes is calculated. For the moisture content, if the moisture contents within these 5 time units are 12.1%, 11.9%, 12.0%, 11.8%, and 12.2% respectively, add these values and divide by 5 to obtain a moisture content mean value of 12.0%. Similarly, if the protein contents are 24.9%, 25.1%, 25.0%, 24.8%, and 25.2% respectively, the mean value is 25.0%, and if the impurity ratios are 2.9%, 3.1%, 3.0%, 3.2%, and 2.8% respectively, the mean value is 3.0%. At the same time, obtain the maximum value of the process parameters. The cleaning duration within this window is 30 minutes for all, so the maximum value is 30 minutes; the drying temperatures are 50 degrees Celsius, 50 degrees Celsius, 50 degrees Celsius, 50 degrees Celsius, and 50 degrees Celsius respectively, and the maximum value is 50 degrees Celsius; the tableting pressures are 100 kgf, 100 kgf, 100 kgf, 100 kgf, and 100 kgf respectively, and the maximum value is 100 kgf; the storage environment humidity is 40%, 40%, 40%, 40%, and 40% respectively, and the maximum value is 40%. In this way, the process parameter sequence of the first processing stage is obtained. As processing progresses, continuously slide the window according to the step size to obtain the process parameter sequences of multiple processing stages.

[0032] Step S123, construct a first association graph based on the co-occurrence frequency of the raw material attributes and process parameters in the process parameter sequence, and extract the raw material attribute association features from the first association graph.

[0033] For example, count the synchronous occurrence of the moisture content and drying temperature within multiple processing stages. Suppose after statistics, in a total of 100 processing stages, there are 60 processing stages where the moisture content is in the range of 11% - 13% and the drying temperature is in the range of 48 - 52 degrees Celsius. When calculating the co-occurrence intensity weight, divide the number of synchronous occurrences, 60, by the total number of processing stages, 100, to obtain a co-occurrence intensity weight of 0.6. Calculate the co-occurrence intensity weights of the protein content and tableting pressure, impurity ratio and cleaning duration, etc. in the same way. Construct a first association graph with the raw material attributes as nodes, process parameters as edges, and co-occurrence intensity weights as edge weights, and extract the raw material attribute association features from this first association graph.

[0034] Step S124, construct a second association graph based on the synchronous change trend between different process parameters in the process parameter sequence, and extract the process parameter coupling features from the second association graph.

[0035] For example, examine the changing trends of the drying temperature and the tableting pressure at each processing stage. If the drying temperature rises from 48 degrees Celsius to 52 degrees Celsius and the tableting pressure rises from 98 kgf to 102 kgf during this process, calculate the similarity of the changing trends of the two. First, calculate the change in the drying temperature as 52 - 48 = 4 degrees Celsius, and the change in the tableting pressure as 102 - 98 = 4 kgf. Assume a reference change amount of 5 (this 5 is a reference value set according to the processing technology requirements and experience). For the drying temperature, the change ratio is 4 divided by 5, which equals 0.8. For the tableting pressure, the change ratio is also 4 divided by 5, which equals 0.8. Since the change ratios of the two are the same, their changing trend similarity is relatively high, and determine their coupling strength. Construct a second association graph with process parameters as nodes and coupling strength as the edge weight, and then extract the process parameter coupling characteristics from this second association graph.

[0036] Step S125, perform pattern matching on the change ranges of the quality evaluation indicators at multiple processing stages, and extract the quality fluctuation pattern characteristics.

[0037] For example, the nutrient retention rate is used as a quality evaluation indicator. It is 80% at the initial stage of processing, becomes 78% after 5 processing stages, and then becomes 75% after another 5 processing stages. It can be seen that its change range gradually decreases. The appearance color score is 8 points at the beginning and becomes 7 points after several processing stages, also showing a downward trend. By analyzing the change ranges of these quality evaluation indicators at different processing stages, quality fluctuation pattern characteristics such as the continuous decrease of the nutrient retention rate and the gradual decrease of the appearance color score are summarized.

[0038] In a possible implementation manner, step S130 includes: Step S131, input the raw material attribute association characteristics and the process parameter coupling characteristics into the parameter optimization module of the pre-trained co-occurrence analysis model, and parse to obtain the deviation weight between the current processing parameter group and the target quality indicator.

[0039] Specifically, taking the previously mentioned wheat germ processing batch as an example, the target quality index is set such that the retention rate of nutritional components reaches 85%, while the actual retention rate of nutritional components under the current processing parameter set is detected to be 80%. In terms of the associated characteristics of raw material attributes, there is a certain correlation between the moisture content and the drying temperature. In terms of the coupling characteristics of process parameters, there are coupling relationships such as those between the drying temperature and the tableting pressure. These data are input into the parameter optimization module. The model analyzes the deviation weights through its internal complex calculation mechanism. Assuming that the model analyzes the relationships between process parameters such as the drying temperature and the tableting pressure and the retention rate of nutritional components, and finds that the drying temperature has a greater impact on the non - compliance of the retention rate of nutritional components. It may be calculated that the deviation weight of the drying temperature is 0.3, which means that the contribution degree of the drying temperature to the deviation between the current processing parameter set and the target quality index is 0.3; the deviation weight of the tableting pressure is calculated to be 0.15, indicating that its impact on non - compliance is relatively small.

[0040] Step S132: Dynamically correct the coupling characteristics of the process parameters based on the deviation weights to generate corrected coupling characteristics of the process parameters.

[0041] In a possible implementation manner, step S132 includes: Step S1321: Extract the parameter influence factors of each process parameter from the deviation weights.

[0042] Step S1322: Determine the weight adjustment coefficients of each process parameter connection edge in the second association graph according to the parameter influence factors.

[0043] Step S1323: Dynamically scale the weights of the process parameter connection edges in the second association graph based on the weight adjustment coefficients to generate scaled coupling characteristics of the process parameters.

[0044] For example, extract the deviation weight of the drying temperature, which is 0.3, as a parameter influence factor. Determine the weight adjustment coefficient for each process parameter edge in the second correlation graph based on this parameter influence factor. Assume that the initial weight of the edge between the drying temperature and the tableting pressure in the second correlation graph is 0.5. Since the parameter influence factor of the drying temperature is 0.3, according to a calculation logic preset by the enterprise (this calculation logic is summarized based on a large number of experiments and processing experiences), the calculated weight adjustment coefficient is 1.2. Dynamically scale the weights of the process parameter edges in the second correlation graph based on this weight adjustment coefficient. For example, scale the weight of the edge between the drying temperature and the tableting pressure to 0.5 multiplied by 1.2, which is equal to 0.6. For other process parameter edges, calculate and scale them in the same way. For example, if the initial weight of the edge between the cleaning duration and the storage environment humidity is 0.4, and the analyzed deviation weight of the cleaning duration is 0.1, and the corresponding weight adjustment coefficient is 1.05, then the scaled weight of the process parameter edge is 0.4 multiplied by 1.05, which is equal to 0.42.

[0045] Step S1324, perform cross-edge weight normalization on the scaled process parameter coupling features so that the sum of all process parameter edge weights is the same as before scaling, and generate normalized process parameter coupling features.

[0046] Step S1325, update the edge weight distribution in the second correlation graph according to the normalized process parameter coupling features to generate the corrected process parameter coupling features.

[0047] In this embodiment, calculate the sum of the weights of all scaled process parameter edges. Assume that after the previous calculation, the sum of the scaled weights of all edges is 1.2 (this is just an assumed value), and the sum of the weights before scaling is 1. Calculate the normalization factor for each edge weight as 1 divided by 1.2, which is approximately equal to 0.833. For the edge weight of 0.6 between the drying temperature and the tableting pressure, after normalization, it is 0.6 multiplied by 0.833, which is approximately equal to 0.5; for the edge weight of 0.42 between the cleaning duration and the storage environment humidity, after normalization, it is 0.42 multiplied by 0.833, which is approximately equal to 0.35. Update the edge weight distribution in the second correlation graph according to these normalized process parameter coupling features to generate the corrected process parameter coupling features.

[0048] Step S133, input the corrected process parameter coupling features and the quality fluctuation mode features into the suggestion generation module of the pre-trained co-occurrence analysis model, and calculate the priority score of the raw material pretreatment optimization item, the adjustment amplitude threshold of the processing parameter adjustment item, and the monitoring frequency of the quality monitoring strengthening item.

[0049] For the raw material pretreatment optimization items, such as the optimization item of screening the impurity ratio of wheat germ, the model calculates based on the coupled characteristics of the corrected process parameters and the quality fluctuation mode characteristics. Suppose the model analyzes from the previous processing data that the change in the impurity ratio has a great impact on the quality fluctuation. Especially when the impurity ratio is higher than 3%, the retention rate of nutrients will decrease significantly. In the current batch, the impurity ratio is 3%. After complex calculations (this kind of calculation involves the analysis of a large amount of historical data and the operation of internal algorithms), the model obtains a priority score of 0.8 for the optimization item of screening wheat germ to reduce the impurity ratio, indicating that this optimization item is relatively important.

[0050] For the processing parameter adjustment items, taking the tabletting pressure as an example, the model calculates the adjustment amplitude threshold according to the input data. Suppose that under normal processing conditions, the ideal range of the tabletting pressure is between 95 - 105 kgf to better ensure the product quality. The current tabletting pressure is set at 100 kgf. Considering various factors such as the change in the drying temperature and the moisture content of the raw materials, after detailed analysis and calculation (this kind of calculation involves the weighing of the mutual relationships between various factors in the processing process), the model obtains an adjustment amplitude threshold of ±5 kgf for the tabletting pressure, that is, the tabletting pressure can be adjusted up and down by 5 kgf based on the current value.

[0051] For the quality monitoring strengthening items, taking the retention rate of nutrients as an example. Since the retention rate of nutrients is a key indicator of product quality and it has been found to fluctuate greatly in the previous processing. The model calculates based on the quality fluctuation mode characteristics, such as the downward trend of the retention rate of nutrients during the processing, and the coupled characteristics of the corrected process parameters, and obtains that it is necessary to strengthen the monitoring of the retention rate of nutrients. After calculation, the original monitoring frequency of once every 2 hours is increased to once every 1 hour.

[0052] Step S134, generate the process optimization recommendation set according to the priority score, the adjustment amplitude threshold, and the monitoring frequency.

[0053] For example, in this embodiment, in the process optimization recommendation set, it includes the relevant information of the raw material pretreatment optimization item, such as the priority score of 0.8 for the impurity ratio optimization; the information of the processing parameter adjustment item, such as the adjustment amplitude threshold of the tabletting pressure is ±5 kgf; the information of the quality monitoring strengthening item, such as the monitoring frequency of the retention rate of nutrients is once every 1 hour.

[0054] In a possible implementation manner, the pre-trained co-occurrence analysis model is trained through the following steps: Step S210, obtain the historical wheat germ processing data set, and the historical wheat germ processing data set includes the raw material attribute training set, the process parameter training set, and the quality index training set.

[0055] For example, it is possible to trace back numerous past wheat germ processing batches. For the raw material attribute training set, data such as the moisture content, protein content, and impurity ratio of wheat germ in each batch are collected. For example, in a certain batch, the moisture content is 12%, the protein content is 25%, and the impurity ratio is 3%, and other batches also have their corresponding values. The process parameter training set covers information such as the cleaning duration, drying temperature, tablet pressing pressure, and storage environment humidity. For example, in a certain processing batch, the cleaning duration is 30 minutes, the drying temperature is 50 degrees Celsius, the tablet pressing pressure is 100 kilogram-force, and the storage environment humidity is 40%. The quality index training set includes the nutrient retention rate, appearance color score, microbial index, etc. For example, in a certain batch, the nutrient retention rate is 80%, and the appearance color score is 8 points, etc.

[0056] Step S220: Perform cross-batch alignment processing on the raw material attribute training set and the process parameter training set to generate an aligned training data group.

[0057] Since there may be differences in processing time, equipment status, etc. among different batches, these data need to be aligned according to certain rules. For example, taking a key operation point in the processing flow as a reference, the data of each batch are adjusted to ensure that the data at the same position are comparable among different batches. For example, in all batches, taking the start of the drying operation as an alignment point, the raw material attribute data and process parameter data before that are sorted out so that the data of each batch before this time point can correspond to each other, thereby generating an aligned training data group.

[0058] Step S230: Perform sliding window sampling on the aligned training data group to generate process parameter training sequences for multiple training stages.

[0059] In a possible implementation manner, step S230 includes: Step S231: Set the length and step size of the sliding window according to the operation cycle of the processing equipment.

[0060] Suppose that a complete operation cycle of a processing device is 10 time units, the length of a sliding window is set to 5 time units, and the step size is 1 time unit. Multiple sliding windows are generated based on the length and step size of the sliding window. The mean value of raw material attributes, the maximum value of process parameters, and the variance of quality indicators are extracted as window features within each sliding window. For example, within a window, if the moisture contents are 11.8%, 12.0%, 12.2%, 11.9%, and 12.1% respectively, add these values and divide by 5 to obtain a moisture content mean of 12.0%. For process parameters, such as the drying temperature within this window being 49 degrees Celsius, 50 degrees Celsius, 50 degrees Celsius, 50 degrees Celsius, and 51 degrees Celsius respectively, the maximum value is 51 degrees Celsius. For quality indicators, such as the retention rate of nutritional components, calculate its variance (when calculating the variance, first calculate the square of the difference between each data and the mean, and then find the average of these squares). In this way, the window features of this sliding window are obtained.

[0061] Step S232: Generate multiple sliding windows based on the length and step of the sliding window, and extract the mean value of raw material attributes, the maximum value of process parameters, and the variance of quality indicators as window features within each sliding window.

[0062] Step S233: Construct a process parameter training sequence based on the change rate between adjacent window features, where each process parameter training sequence contains the feature change trajectories of N consecutive windows.

[0063] For example, for the moisture content, if the mean value of the previous window is 12.0% and the mean value of the next window is 11.8%, then the change rate of the moisture content is (11.8 - 12.0) divided by 12.0 (this division calculation is to obtain the relative change rate), resulting in -0.0167. Calculate the change rate for other raw material attributes, process parameters, and quality indicators in the same way. Each process parameter training sequence contains the feature change trajectories of N consecutive windows. Suppose N is 5, then continuously calculate the change rates between 5 consecutive windows to construct the process parameter training sequence.

[0064] Step S234: Perform noise filtering processing on the process parameter training sequence to remove abnormal sequences that exceed the preset fluctuation threshold.

[0065] The preset fluctuation threshold is set according to the long-term processing experience of the enterprise and the requirements for product quality stability. For example, for the change rate of moisture content, the preset fluctuation threshold is set to ±0.05. If in a certain process parameter training sequence, the change rate of moisture content is 0.1, exceeding the preset fluctuation threshold, then this process parameter training sequence may be caused by abnormal factors such as equipment failure and measurement error, and this sequence will be excluded. The change rates of other raw material attributes, process parameters, and quality indicators are also checked and abnormal sequences are excluded in the same way.

[0066] Step S240: Based on the co-occurrence pattern of raw material attributes and process parameters in the process parameter training sequence, construct a first association graph for training, and extract the associated features of raw material attributes for training.

[0067] Step S250: Based on the coupling relationship between process parameters in the process parameter training sequence, construct a second association graph for training, and extract the coupling features of process parameters for training.

[0068] Step S260: Perform fluctuation pattern annotation on the quality index training set to generate the quality fluctuation pattern features for training.

[0069] For example, for the quality index of nutrient retention rate, observe its changes in each processing batch. If in a certain batch, the nutrient retention rate is 80% at the beginning of processing, and as processing progresses, it becomes 78% after several stages and then 75% after several more stages, this continuously decreasing pattern is labeled as a fluctuation pattern. The decreasing pattern of the appearance color score from 8 points to 7 points and then to 6 points is also labeled. These labeled patterns are used as the quality fluctuation pattern features for training.

[0070] Step S270: Input the associated features of raw material attributes for training, the coupling features of process parameters for training, and the quality fluctuation pattern features for training into the initial model to generate predicted process optimization suggestions, and train the co-occurrence analysis model based on the difference between the predicted process optimization suggestions and the labeled optimization suggestions.

[0071] In a possible implementation, step S270 includes: Step S271: Input the associated features of raw material attributes for training into the parameter optimization module of the initial model, and parse to obtain the deviation weights for training.

[0072] For example, in the raw material attribute correlation features for training, there are correlation relationships such as the moisture content and the drying temperature, and the protein content and the tableting pressure. Suppose that when constructing the first correlation map for training previously, the co-occurrence intensity weight of the moisture content and the drying temperature is 0.6, and the co-occurrence intensity weight of the protein content and the tableting pressure is 0.5, etc. After these correlation relationship data are input into the parameter optimization module of the initial model, a series of complex calculation mechanisms are used inside the model to analyze and obtain the deviation weights for training. Taking the drying temperature as an example, suppose the model analyzes and obtains that the deviation weight for training of the drying temperature is 0.3 based on the correlation relationship between the moisture content and the drying temperature, and the comprehensive influence of many other raw material attribute correlation relationships on the final product quality indicators (such as the nutrient retention rate, the appearance color score, etc.), indicating that the degree of deviation influence of the drying temperature on the target quality indicator during the entire processing process is 0.3; similarly, the deviation weight for training of the tableting pressure is calculated to be 0.15, etc.

[0073] Step S272: Based on the deviation weights for training, perform weighted correction on the process parameter coupling features for training to generate corrected process parameter coupling features for training.

[0074] Among the deviation weights for training obtained previously, for example, the deviation weight for training of the drying temperature is 0.3, and the deviation weight for training of the tableting pressure is 0.15. For the coupling relationship between the drying temperature and the tableting pressure involved in the process parameter coupling features for training, suppose that before weighted correction, the edge weight of the connection between the drying temperature and the tableting pressure in the second correlation map is 0.5. When performing weighted correction, the edge weight is corrected according to the deviation weight for training of the drying temperature, which is 0.3. The specific calculation is to multiply the original edge weight 0.5 by (1 + 0.3) to get 0.65 (the calculation logic here is based on a way set inside the model to adjust the coupling feature weight according to the deviation weight, aiming to highlight or weaken the influence of the coupling relationship between certain process parameters on the final result according to the deviation weight). Perform weighted correction on other process parameter coupling relationships in the same way to generate corrected process parameter coupling features for training.

[0075] Step S273: Concatenate the corrected process parameter coupling features for training with the quality fluctuation mode features for training to generate concatenated features for training.

[0076] For example, the coupling feature of the training correction process parameters contains information such as the coupling relationship weights between the corrected drying temperature and tablet pressing pressure, cleaning duration and storage environment humidity, etc. obtained through the previous calculations. The quality fluctuation mode feature for training contains information such as the fluctuation mode of the nutrient retention rate during the processing (such as first decreasing and then increasing), the fluctuation mode of the appearance color score, etc. These features are spliced together in a certain order to form a new spliced feature for training, which contains both the information on the relationship adjustment between process parameters and the mode information of quality fluctuations.

[0077] Step S274: Invoke the suggestion generation module of the initial model to parse the spliced feature for training, and generate predicted raw material pretreatment optimization items, predicted processing parameter adjustment items, and predicted quality monitoring enhancement items.

[0078] After receiving the spliced feature for training, the suggestion generation module of the initial model analyzes it according to the internal algorithm logic. For the predicted raw material pretreatment optimization item, assume that the model analyzes and obtains the priority score of the optimization item of the impurity screening ratio of wheat germ based on the data in the spliced feature for training. The model may consider the impact of the impurity ratio on product quality (such as too high an impurity ratio may affect the nutrient retention rate and appearance color score, etc.), as well as various factors such as the coupling relationship of process parameters and the quality fluctuation mode. After calculation (this calculation process involves the comprehensive analysis and weighing of each data element in the spliced feature), the priority score of the optimization item of screening wheat germ to reduce the impurity ratio is 0.8, indicating that this optimization item is relatively important. For the predicted processing parameter adjustment item, taking the tablet pressing pressure as an example, the model considers various factors such as the change in drying temperature and the moisture content of the raw material (these factors are reflected by the coupling relationship of process parameters and the quality fluctuation mode in the spliced feature for training). After detailed analysis and calculation, the adjustment amplitude threshold of the tablet pressing pressure is obtained. Assume that in normal processing, the ideal range of the tablet pressing pressure is between 95 - 105 kgf to better ensure product quality. The model calculates the adjustment amplitude threshold of the tablet pressing pressure to be ±5 kgf based on the current process parameter state and quality fluctuation trend, that is, the tablet pressing pressure can be adjusted up and down by 5 kgf based on the current value. For the predicted quality monitoring enhancement item, taking the nutrient retention rate as an example, since the nutrient retention rate is a key indicator of product quality and it has been found to fluctuate greatly during the previous processing (known from the quality fluctuation mode feature for training), the model calculates that it is necessary to strengthen the monitoring of the nutrient retention rate. After calculation, the original monitoring frequency of once every 2 hours is increased to once every 1 hour.

[0079] Step S275: Calculate the first error between the predicted raw material pretreatment optimization item and the labeled raw material pretreatment optimization item, the second error between the predicted processing parameter adjustment item and the labeled processing parameter adjustment item, and the third error between the predicted quality monitoring enhancement item and the labeled quality monitoring enhancement item.

[0080] For example, for the predicted priority score of the impurity ratio optimization in the predicted raw material pretreatment optimization item, it is predicted to be 0.8, while the actual labeled priority score is 0.9. Then the first error is 0.8 - 0.9 = -0.1. For the predicted adjustment range threshold of the tablet pressing pressure in the predicted processing parameter adjustment item, it is predicted to be ±5 kgf, while the labeled adjustment range threshold is ±4 kgf. When calculating the second error, first calculate the errors of the upper and lower limits respectively. The upper limit error is 5 - 4 = 1 kgf, and the lower limit error is -5 - (-4) = -1 kgf (here, calculating the errors of the upper and lower limits is to comprehensively evaluate the error situation of the adjustment range threshold). For the predicted monitoring frequency of the nutrient retention rate in the predicted quality monitoring enhancement item, it is predicted to be once every 1 hour, while the labeled one is once every 1.5 hours. When calculating the third error, since the monitoring frequency is a time interval, the error needs to be calculated according to a certain conversion method (here, it is assumed that the time interval is converted to the detection times per minute. Once every 1 hour means 1 / 60 times per minute, and once every 1.5 hours means 1 / 90 times per minute. The third error is 1 / 60 - 1 / 90 = 1 / 180 times per minute).

[0081] Step S276: Construct a multi-task loss function according to the first error, the second error, and the third error. The multi-task loss function includes a mean square error term and a cross-entropy error term.

[0082] For the mean squared error term, calculate the squares of the first error, the second error, and the third error, and then find the average. Taking the first error of -0.1 as an example, its square is 0.01; for the upper and lower limit errors of 1 kgf and -1 kgf for the tablet pressing pressure adjustment amplitude threshold, their squares are 1 and 1 respectively; for the third error of 1 / 180 times per minute, its square is (1 / 180)². Add these squared values and divide by 3 (since there are three error terms) to obtain the value of the mean squared error term. For the cross-entropy error term, since it involves prediction optimization terms (such as the priority score of the raw material pretreatment optimization term, the adjustment amplitude threshold of the processing parameter adjustment term, the monitoring frequency of the quality monitoring enhancement term, etc.), it is necessary to convert the predicted values and the labeled values into probability distributions according to specific classifications or numerical ranges (this conversion process is carried out according to pre-set rules, for example, mapping the priority score to a probability between 0 and 1, and converting the adjustment amplitude threshold into a probability according to its relative position within a reasonable range, etc.), and then calculate the difference between the predicted probability distribution and the labeled probability distribution according to the cross-entropy formula to obtain the value of the cross-entropy error term. Combining the mean squared error term and the cross-entropy error term with a certain weight (this weight is pre-set according to the requirements of the model and the degree of emphasis on different error terms) constructs the multi-task loss function.

[0083] Step S277, adjust the parameters of the initial model through the backpropagation algorithm until the multi-task loss function converges to obtain a pre-trained co-occurrence analysis model.

[0084] In this embodiment, the backpropagation algorithm starts from the multi-task loss function and gradually adjusts each parameter of the initial model according to the derivative of the loss function with respect to the model parameters (this derivative is calculated through mathematical principles such as the chain rule and is calculated based on the complex internal structure of the model and the multi-task loss function constructed previously). For example, for the parameters related to the raw material attribute correlation features, the process parameter coupling features, and the quality fluctuation pattern feature processing related parameters, etc., adjust them in the direction of the derivative with a certain learning rate (this learning rate is a pre-set numerical value that controls the step size of parameter adjustment). Continuously repeat this process, recalculate the value of the multi-task loss function after each adjustment until the value of the multi-task loss function no longer decreases significantly (i.e., converges). At this time, the obtained model is the pre-trained co-occurrence analysis model, and this co-occurrence analysis model can better generate reasonable process optimization suggestions according to the input features.

[0085] For example, in a possible implementation manner, step S233 includes: Step S2331, arrange multiple window features in chronological order according to the step size of the sliding window to generate adjacent window feature pairs, and the adjacent window feature pairs include the current window feature and the next window feature.

[0086] Assume that the previously set sliding window step size is 1 processing time unit. According to this sliding window step size, arrange each window feature in the order of processing time. Each adjacent window feature pair includes the current window feature and the next window feature. For example, the first window feature includes an average moisture content of 12.0%, an average protein content of 25.0%, an average impurity ratio of 3.0%, a maximum cleaning duration of 30 minutes, a maximum drying temperature of 50 degrees Celsius, a maximum tabletting pressure of 100 kgf, and a maximum storage environment humidity of 40%. The corresponding parameters of the next window feature are 11.8%, 24.8%, 3.1%, 30 minutes, 51 degrees Celsius, 101 kgf, and 40% respectively, thus forming an adjacent window feature pair.

[0087] Step S2332: Calculate the element-by-element difference between the current window feature and the next window feature in each adjacent window feature pair to determine the change rate vector between adjacent windows. The change rate vector includes the moisture content change rate, protein content change rate, impurity ratio change rate, cleaning duration change rate, drying temperature change rate, tabletting pressure change rate, and storage environment humidity change rate.

[0088] For example, for the moisture content, the change rate is (11.8 - 12.0) ÷ 12.0 = -0.0167; for the protein content, the change rate is (24.8 - 25.0) ÷ 25.0 = -0.008; for the impurity ratio, the change rate is (3.1 - 3.0) ÷ 3.0 = 0.0333; for the cleaning duration, since it is 30 minutes for both, the change rate is (30 - 30) ÷ 30 = 0; for the drying temperature, the change rate is (51 - 50) ÷ 50 = 0.02; for the tabletting pressure, the change rate is (101 - 100) ÷ 100 = 0.01; for the storage environment humidity, the change rate is (40 - 40) ÷ 40 = 0. In this way, a change rate vector including the moisture content change rate, protein content change rate, impurity ratio change rate, cleaning duration change rate, drying temperature change rate, tabletting pressure change rate, and storage environment humidity change rate is obtained.

[0089] Step S2333: Compare each change rate in the change rate vector with a preset change rate threshold, mark the elements whose absolute value of the change rate exceeds the preset change rate threshold as significant change items, and generate a status coding sequence including the significant change items.

[0090] For example, in a possible implementation, step S2333 includes: Step S2333-1, jointly analyze the moisture content change rate and the drying temperature change rate in the change rate vector. If the moisture content change rate is negative and the drying temperature change rate is positive, generate a first joint status code.

[0091] Step S2333-2, jointly analyze the protein content change rate and the tablet pressing pressure change rate in the change rate vector. If the protein content change rate exceeds the protein change rate threshold and the tablet pressing pressure change rate synchronously exceeds the pressure change rate threshold, generate a second joint status code.

[0092] Step S2333-3, jointly analyze the impurity ratio change rate and the cleaning duration change rate in the change rate vector. If the impurity ratio change rate is positive and the cleaning duration change rate does not exceed the cleaning duration change rate threshold, generate a third joint status code.

[0093] Step S2333-4, combine the first joint status code, the second joint status code, and the third joint status code in chronological order to generate the status code sequence including significant change items.

[0094] Specifically, assume that the preset moisture content change rate threshold is ±0.05, the protein content change rate threshold is ±0.03, the impurity ratio change rate threshold is ±0.04, the cleaning duration change rate threshold is ±0.1, the drying temperature change rate threshold is ±0.05, and the tablet pressing pressure change rate threshold is ±0.04, and the storage environment humidity change rate threshold is ±0.05. In the above change rate vector, the absolute value of the impurity ratio change rate of 0.0333 exceeds the impurity ratio change rate threshold of ±0.04, and the drying temperature change rate of 0.02 exceeds the drying temperature change rate threshold of ±0.05. Therefore, the impurity ratio change rate and the drying temperature change rate are marked as significant change items.

[0095] When performing joint analysis to generate a state encoding sequence containing significant change items, joint analysis is carried out on the moisture content change rate and the drying temperature change rate in the change rate vector. If the moisture content change rate is negative and the drying temperature change rate is positive, for example, the moisture content change rate calculated previously is -0.0167 which is negative, and the drying temperature change rate is 0.02 which is positive, then the first joint state encoding is generated. For the joint analysis of the protein content change rate and the tableting pressure change rate, if the protein content change rate exceeds the protein change rate threshold and the tableting pressure change rate synchronously exceeds the pressure change rate threshold, here the protein content change rate of -0.008 does not exceed the protein change rate threshold of ±0.03, so the corresponding joint state encoding is not generated. For the joint analysis of the impurity ratio change rate and the cleaning duration change rate in the change rate vector, if the impurity ratio change rate is positive and the cleaning duration change rate does not exceed the cleaning duration change rate threshold, the impurity ratio change rate calculated previously is 0.0333 which is positive, and the cleaning duration change rate is 0 which does not exceed the cleaning duration change rate threshold of ±0.1, then the third joint state encoding is generated. The first joint state encoding, the second joint state encoding (not generated here, assumed to be empty), and the third joint state encoding are combined in chronological order to generate a state encoding sequence containing significant change items.

[0096] Step S2334: Merge the same significant change items that continuously appear in the state encoding sequence to generate a merged state encoding sequence, and insert a timestamp interval in the merged state encoding sequence to retain the time dependence of the window feature.

[0097] For example, in a possible implementation manner, step S2334 includes: Step S2334-1: Statistically calculate the start timestamp and end timestamp of each merged segment in the merged state encoding sequence, and calculate the duration length of the merged segment.

[0098] Step S2334-2: Determine the number of inserted timestamp intervals according to the proportional relationship between the duration length and the step size of the sliding window. The number of timestamp intervals is equal to the integer value obtained by rounding down the duration length divided by the step size of the sliding window.

[0099] Step S2334-3: Insert the timestamp interval at the start position of each merged segment in the merged state encoding sequence, and record the duration length of the merged segment and the corresponding window feature index in the timestamp interval.

[0100] Suppose that three identical significant change items (here, assume they are significant change items related to the impurity ratio change rate) appear consecutively in the state encoding sequence. These three identical significant change items are merged into one. The start timestamp and end timestamp of each merged segment in the merged state encoding sequence are counted, and the duration length of the merged segment is calculated. For example, a merged segment starts at the 3rd processing time unit and ends at the 5th processing time unit, and the duration length is 3 processing time units. The number of inserted timestamp intervals is determined according to the proportional relationship between the duration length and the step size of the sliding window. Suppose the step size of the sliding window is 1 processing time unit and the duration length is 3 processing time units, then the number of inserted timestamp intervals is 3÷1 = 3 (rounded down). Timestamp intervals are inserted at the start position of each merged segment in the merged state encoding sequence, and the duration length of the merged segment, which is 3 processing time units, and the corresponding window feature index (assume it starts from the 3rd window) are recorded in the timestamp intervals.

[0101] Step S2335: Generate a candidate process parameter training sequence according to the arrangement order of the significant change items in the merged state encoding sequence, and fill the time periods in the candidate process parameter training sequence that do not contain significant change items with default stationary encodings.

[0102] For example, in a possible implementation, the default stationary encoding is generated through the following steps: A. Extract the window features of the candidate process parameter training sequences that are not marked with significant change items to generate a set of stationary window features.

[0103] B. Calculate the mean values of the moisture content, protein content, and impurity ratio in the set of stationary window features to generate the mean values of the stationary raw material attributes.

[0104] C. Calculate the variances of the cleaning duration, drying temperature, and tablet pressing pressure in the set of stationary window features to generate the variances of the stationary process parameters.

[0105] D. Combine the mean values of the stationary raw material attributes and the variances of the stationary process parameters into the default stationary encoding.

[0106] Step S2336: Perform cross-sequence alignment verification on the candidate process parameter training sequence filled with the default stationary encoding and the original window features, and remove the abnormal encoding segments in the candidate process parameter training sequence that are inconsistent with the trend of the original window features to generate the final process parameter training sequence.

[0107] For example, for a time period that does not contain significant change items, default stationary coding needs to be filled. First, extract the window features of the candidate process parameter training sequences that are not marked with significant change items to generate a set of stationary window features. For example, within certain time periods, the moisture content, protein content, and impurity ratio do not change significantly. Calculate the mean values of the moisture content, protein content, and impurity ratio in these window features. Suppose the moisture contents in these window features are 12.0%, 11.9%, and 12.1% respectively, and the mean value is 12.0%; the protein contents are 25.0%, 24.9%, and 25.1% respectively, and the mean value is 25.0%; the impurity ratios are 3.0%, 3.1%, and 2.9% respectively, and the mean value is 3.0% to generate the mean values of the stationary raw material attributes. Calculate the variances of the cleaning duration, drying temperature, and tableting pressure in the set of stationary window features. Suppose the cleaning durations are 30 minutes, 30 minutes, and 30 minutes respectively, and the calculated variance is 0; the drying temperatures are 50 degrees Celsius, 50 degrees Celsius, and 50 degrees Celsius respectively, and the variance is 0; the tableting pressures are 100 kgf, 100 kgf, and 100 kgf respectively, and the variance is 0 to generate the variances of the stationary process parameters. Combine the mean values of the stationary raw material attributes and the variances of the stationary process parameters into the default stationary coding, such as (12.0%, 25.0%, 3.0%, 0, 0, 0), and fill it into the time periods in the candidate process parameter training sequences that do not contain significant change items.

[0108] For example, in a possible implementation manner, step S2336 includes: Step S2336-1, extract the original window features corresponding to each coding segment in the candidate process parameter training sequence to generate a subset of coding segment window features.

[0109] Step S2336-2, perform trend fitting on the subset of coding segment window features to generate a coding segment trend line, and calculate the included angle between the coding segment trend line and the change direction of the coding segment in the candidate process parameter training sequence.

[0110] For example, in a possible implementation manner, step S2336-2 includes: Step S2336-21, perform weighted summation on the moisture content, protein content, and impurity ratio in the subset of coding segment window features to generate a raw material attribute trend index.

[0111] Step S2336-22, perform normalization processing on the cleaning duration, drying temperature, and tableting pressure in the subset of coding segment window features to generate a process parameter trend index.

[0112] Step S2336-23, generate the coding segment trend line based on the linear combination of the raw material attribute trend index and the process parameter trend index.

[0113] Step S2336-3, if the included angle exceeds a preset included angle threshold, determine that the coded segment is an abnormal coded segment, and replace the abnormal coded segment with the replacement coded segment to generate the final process parameter training sequence.

[0114] For example, for the original window features corresponding to a coded segment, it includes data such as moisture content, protein content, impurity ratio, cleaning duration, drying temperature, tableting pressure, and storage environment humidity. Perform trend fitting on the coded segment window feature subset to generate a coded segment trend line. Specifically, perform weighted summation on the moisture content, protein content, and impurity ratio in the coded segment window feature subset. Assume that the weight of the moisture content is 0.3, the weight of the protein content is 0.4, and the weight of the impurity ratio is 0.3. If the moisture content is 12.0%, the protein content is 25.0%, and the impurity ratio is 3.0%, the weighted summation is 0.3×12.0% + 0.4×25.0% + 0.3×3.0% = 13.5%, generating a raw material attribute trend index. Normalize the cleaning duration, drying temperature, and tableting pressure in the coded segment window feature subset. Assume that the cleaning duration is 30 minutes, the drying temperature is 50 degrees Celsius, and the tableting pressure is 100 kgf. After normalization (here, the normalization is based on a specific algorithm, such as mapping the values to the 0 - 1 interval), a process parameter trend index is obtained. Based on the linear combination of the raw material attribute trend index and the process parameter trend index, assume that the linear combination coefficients are 0.6 and 0.4, and generate a coded segment trend line. Calculate the included angle between the coded segment trend line and the change direction of the coded segment in the candidate process parameter training sequence. If the included angle exceeds the preset included angle threshold (assume the preset included angle threshold is 30 degrees), determine that the coded segment is an abnormal coded segment, and replace the abnormal coded segment with a replacement coded segment (here, it can be a coded segment estimated based on adjacent normal coded segments) to generate the final process parameter training sequence. Through such a series of operations, a process parameter training sequence that accurately reflects the change relationship of the wheat germ processing process parameters is finally obtained.

[0115] In a possible implementation manner, step S240 includes: Step S241, count the number of synchronous occurrences of the moisture content and the drying temperature, the number of synchronous occurrences of the protein content and the tableting pressure, and the number of synchronous occurrences of the impurity ratio and the cleaning duration in each process parameter training sequence.

[0116] For example, among 100 process parameter training sequences, for the synchronous occurrence of moisture content and drying temperature, each sequence is checked one by one. Assume that when the moisture content is in the range of 11% - 13%, and at the same time the drying temperature is in the range of 48 - 52 degrees Celsius, it is regarded as a synchronous occurrence. After careful statistics, it is found that 60 process parameter training sequences meet this situation, that is, the synchronous occurrence times of moisture content and drying temperature are 60 times. For protein content and tableting pressure, when the protein content is in the range of 24% - 26% and the tableting pressure is in the range of 98 - 102 kgf, it is regarded as a synchronous occurrence. After statistics, 40 sequences among these 100 process parameter training sequences meet the requirement, that is, the synchronous occurrence times are 40 times. For impurity ratio and cleaning duration, if the impurity ratio is in the range of 2% - 4% and the cleaning duration is in the range of 28 - 32 minutes, it is regarded as a synchronous occurrence. The statistics show that there are 30 synchronous occurrences among 100 process parameter training sequences.

[0117] Step S242, calculate the co-occurrence intensity weight of the raw material attribute and the process parameter according to the synchronous occurrence times.

[0118] When calculating the co-occurrence intensity weight, taking moisture content and drying temperature as an example, the co-occurrence intensity weight is equal to the synchronous occurrence times divided by the total number of process parameter training sequences. That is, the co-occurrence intensity weight of moisture content and drying temperature is 60÷100 = 0.6. According to the same calculation method, the co-occurrence intensity weight of protein content and tableting pressure is 40÷100 = 0.4, and the co-occurrence intensity weight of impurity ratio and cleaning duration is 30÷100 = 0.3.

[0119] Step S243, construct the first association graph for training with the raw material attribute as the node, the process parameter as the edge, and the co-occurrence intensity weight as the edge weight.

[0120] For example, take moisture content, protein content, and impurity ratio as nodes, drying temperature, tableting pressure, and cleaning duration as edges, and their respective co-occurrence intensity weights as edge weights. For example, the edge weight from the moisture content node to the drying temperature node is 0.6, indicating the co-occurrence intensity between moisture content and drying temperature; the edge weight from the protein content node to the tableting pressure node is 0.4; the edge weight from the impurity ratio node to the cleaning duration node is 0.3. Thus, the first association graph for training is constructed, and this first association graph for training intuitively reflects the co-occurrence relationship and its intensity between the raw material attribute and the process parameter.

[0121] And, step S250 includes: Step S251, calculate the change trend similarity of every two process parameters within the same processing stage, and the change trend similarity is obtained by matching with the dynamic time warping algorithm.

[0122] Taking the drying temperature and the tableting pressure as examples, their numerical changes in each identical processing stage are investigated. Suppose that in 10 consecutive processing stages, the numerical values of the drying temperature are 48, 49, 50, 51, 52, 51, 50, 49, 48, 47 degrees Celsius respectively, and the numerical values of the tableting pressure are 98, 99, 100, 101, 102, 101, 100, 99, 98, 97 kilogram-force respectively.

[0123] On this basis, the dynamic time warping algorithm combines the distortion of the two sequences on the time axis to find the optimal matching path. During the calculation process, the changes in the drying temperature and the tableting pressure at different time points will be compared. For example, when the drying temperature rises from 48 degrees Celsius to 49 degrees Celsius, the tableting pressure rises from 98 kilogram-force to 99 kilogram-force. This synchronous upward trend is an embodiment of similarity. After complex calculations by the dynamic time warping algorithm (this calculation process involves operations such as calculating the distance between each element in the two sequences and searching for paths, aiming to find the matching method that minimizes the overall difference between the two sequences), it is obtained that the similarity of the change trends of the drying temperature and the tableting pressure in these 10 processing stages is 0.8 (here, 0.8 is a value representing the degree of similarity calculated according to the algorithm, ranging from 0 to 1, and the closer to 1, the higher the similarity).

[0124] Calculate the similarity of the change trends between other process parameters in the same way. For example, the similarity of the change trends of the cleaning duration and the storage environment humidity in the same processing stage is calculated to be 0.6.

[0125] Step S252, determine the coupling strength between process parameters according to the similarity of the change trends.

[0126] Taking the drying temperature and the tableting pressure as examples, since their similarity of the change trends is 0.8, according to a mapping relationship preset by the enterprise (this mapping relationship is summarized based on a large amount of processing data and experience. For example, the coupling strength corresponding to the similarity of the change trends between 0.7 and 1 is high), it is determined that the coupling strength between the drying temperature and the tableting pressure is high. For the cleaning duration and the storage environment humidity, since the similarity of the change trends is 0.6, according to the mapping relationship, its coupling strength is determined to be medium.

[0127] Step S253, construct the second associated graph for training with process parameters as nodes and coupling strength as the edge weight.

[0128] For example, process parameters such as drying temperature, tablet pressing pressure, cleaning duration, and storage environment humidity are used as nodes, and the coupling strength between them is used as the edge weight. For example, the edge weight from the drying temperature node to the tablet pressing pressure node is high (which can be represented by a numerical value here. Assume that a high coupling strength corresponds to 0.8), and the edge weight from the cleaning duration node to the storage environment humidity node is medium (assume that a medium coupling strength corresponds to 0.5). Thus, a second association graph for training is constructed, and this second association graph for training reflects the coupling relationship and its strength between process parameters.

[0129] In a possible implementation manner, step S140 includes: Step S141, converting the moisture control suggestion in the raw material pretreatment optimization item into a duration control instruction for the cleaning equipment.

[0130] Assume that in the raw material pretreatment optimization item, after analysis, it is obtained that in order to achieve better processing effects and product quality, the moisture content of wheat germ needs to be controlled at 10%. The current moisture content of wheat germ is 12%. According to the processing data and experience accumulated by the enterprise for a long time, it is known that for every 1% reduction in moisture content, the duration of the cleaning equipment needs to be increased by 5 minutes (this data is summarized based on a large number of previous tests and actual processing situations). Then, to reduce the moisture content from 12% to 10%, a reduction of 2% is required, so the duration of the cleaning equipment needs to be increased by 2×5 = 10 minutes. If the original duration of the cleaning equipment is set at 30 minutes, then now the moisture control suggestion needs to be converted into a duration control instruction for the cleaning equipment, that is, "adjust the duration of the cleaning equipment to 40 minutes".

[0131] Step S142, converting the tablet pressing pressure suggestion in the processing parameter adjustment item into a pressure gradient adjustment parameter for the tablet press.

[0132] In the processing parameter adjustment item, for example, the optimization suggestion for the tablet pressing pressure is that according to the current processing situation and the target product quality requirements, the tablet pressing pressure needs to be adjusted to 105 kgf. The current pressure setting of the tablet press is 100 kgf, then the tablet pressing pressure suggestion needs to be converted into a pressure gradient adjustment parameter for the tablet press. Here, it is clear that the pressure of the tablet press needs to be increased from 100 kgf to 105 kgf, so the generated pressure gradient adjustment parameter is "increase the pressure of the tablet press by 5 kgf". This adjustment parameter is obtained based on a comprehensive consideration of various factors such as raw material characteristics, process requirements, and quality objectives in the entire processing process, ensuring that after adjusting the tablet pressing pressure, product quality indicators such as improving the retention rate of nutritional components and improving the appearance color can be improved.

[0133] Step S143, converting the detection frequency suggestion in the quality monitoring strengthening item into a sampling interval configuration for the quality detection instrument.

[0134] For example, for the key quality indicator of the nutrient retention rate in the quality monitoring enhancement items, according to the current processing status and quality fluctuation conditions, it is recommended to increase the detection frequency from once every 2 hours to once every 1 hour. For the quality detection instrument, this requires converting this detection frequency recommendation into a sampling interval configuration. Originally, the detection was carried out once every 2 hours, that is, the sampling interval was 120 minutes. Now, it is adjusted to once every 1 hour, so the sampling interval becomes 60 minutes. Therefore, the generated sampling interval configuration is "set the sampling interval for the nutrient retention rate detection to 60 minutes". Such an adjustment can monitor the product quality more timely, so as to quickly respond when quality problems occur and adjust the processing parameters.

[0135] Step S144, send the duration control instruction, the pressure gradient adjustment parameter, and the sampling interval configuration to the wheat germ processing control system, and receive the adjustment confirmation signal returned by the wheat germ processing control system to complete the closed-loop control.

[0136] For example, instructions such as "adjust the duration of the cleaning equipment to 40 minutes", "increase the pressure of the tablet press by 5 kgf", and "set the sampling interval for the nutrient retention rate detection to 60 minutes" can be sent to the wheat germ processing control system. After receiving these instructions, the wheat germ processing control system adjusts the corresponding parameters of the cleaning equipment, the tablet press, and the quality detection instrument according to the instructions. After receiving the duration adjustment instruction, the cleaning equipment adjusts the cleaning duration to 40 minutes; after receiving the pressure gradient adjustment parameter, the tablet press increases the pressure to 105 kgf; after receiving the sampling interval configuration, the quality detection instrument sets the sampling interval for the nutrient retention rate detection to 60 minutes. After completing these parameter adjustments, the wheat germ processing control system will return an adjustment confirmation signal, such as "all parameter adjustments have been completed according to the instructions". Once receiving this adjustment confirmation signal, it means that the entire processing parameter adjustment operation has been successfully executed, and the whole system enters a new processing state and operates according to the optimized parameters, thus realizing the closed-loop control process from the process optimization suggestion to the actual processing parameter adjustment, ensuring the efficiency of the wheat germ processing process and the stability of the product quality.

[0137] Figure 2 The hardware structure diagram of the processing service system 100 provided by the embodiment of the present application for implementing the above-mentioned method for generating wheat germ processing process optimization suggestions based on co-occurrence analysis is shown, as Figure 2 shown, the processing service system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0138] In one possible design, the processing service system 100 can be a single server or a server group. The server group can be centralized or distributed (e.g., the processing service system 100 can be a distributed system). In some embodiments, the processing service system 100 can be local or remote. For example, the processing service system 100 can access information and / or data stored in the machine-readable storage medium 120 via a network. As another example, the processing service system 100 can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the processing service system 100 can be implemented on a processing service system. By way of example only, the processing service system can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any aggregation thereof.

[0139] The machine-readable storage medium 120 can store data and / or instructions. In some embodiments, the machine-readable storage medium 120 can store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 can store the data and / or instructions that the processing service system 100 uses to execute or use to complete the exemplary methods described in this application.

[0140] In a specific implementation process, one or more processors 110 execute the computer-executable instructions stored in the machine-readable storage medium 120, so that the processors 110 can execute the method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis in the above method embodiments. The processors 110, the machine-readable storage medium 120, and the communication unit 140 are connected through a bus 130, and the processors 110 can be used to control the transceiver actions of the communication unit 140.

[0141] For the specific implementation process of the processors 110, reference can be made to the various method embodiments executed by the above processing service system 100. Their implementation principles and technical effects are similar, and will not be elaborated here in this embodiment.

[0142] In addition, an embodiment of the present application also provides a readable storage medium, in which computer-executable instructions are set. When a processor runs the computer-executable instructions, the method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis as described above is implemented.

[0143] It should be noted that, in order to simplify the description of the present application disclosure and thus help the understanding of one or more invention embodiments, in the previous description of the embodiments of the present application, sometimes multiple features are merged into one embodiment, drawing, or description thereof. Similarly, it should be noted that, in order to simplify the description of the present application disclosure and thus help the understanding of one or more invention embodiments, in the previous description of the embodiments of the present application, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. A method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis, characterized in that, The method includes: Obtaining a raw material data set of wheat germ of a target batch, where the raw material data set includes historical processing parameter groups of multiple processing batches and corresponding quality evaluation indicators; Performing co-occurrence feature extraction processing on the raw material data set to obtain a processing co-occurrence feature set for each processing batch, where the processing co-occurrence feature set includes raw material attribute association features, process parameter coupling features, and quality fluctuation pattern features; Invoking a pre-trained co-occurrence analysis model to perform dynamic optimization and analysis processing on the processing co-occurrence feature set to generate a process optimization suggestion set for the target batch, where the process optimization suggestion set includes raw material pretreatment optimization items, processing parameter adjustment items, and quality monitoring strengthening items; Feeding back the process optimization suggestion set to the wheat germ processing control system to trigger processing parameter adjustment operations.

2. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 1, characterized in that, The performing co-occurrence feature extraction processing on the raw material data set to obtain a processing co-occurrence feature set for each processing batch includes: Extracting a raw material attribute subset and a process parameter subset for each processing batch from the raw material data set, where the raw material attribute subset includes the moisture content, protein content, and impurity ratio of wheat germ, and the process parameter subset includes the cleaning duration, drying temperature, tableting pressure, and storage environment humidity; Performing synchronous sliding window sampling on the raw material attribute subset and the process parameter subset to obtain process parameter sequences for multiple processing stages; Constructing a first association graph based on the co-occurrence frequency of raw material attributes and process parameters in the process parameter sequence, and extracting the raw material attribute association features from the first association graph; Constructing a second association graph based on the synchronous change trend between different process parameters in the process parameter sequence, and extracting the process parameter coupling features from the second association graph; Performing pattern matching on the change range of the quality evaluation indicators in multiple processing stages to extract the quality fluctuation pattern features.

3. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 2, characterized in that, The invoking a pre-trained co-occurrence analysis model to perform dynamic optimization and analysis processing on the processing co-occurrence feature set to generate a process optimization suggestion set for the target batch includes: Inputting the raw material attribute association features and the process parameter coupling features into the parameter optimization module of the pre-trained co-occurrence analysis model to parse the deviation weight between the current processing parameter group and the target quality index; Dynamically correcting the process parameter coupling features based on the deviation weight to generate corrected process parameter coupling features; Inputting the corrected process parameter coupling features and the quality fluctuation pattern features into the suggestion generation module of the pre-trained co-occurrence analysis model to calculate the priority score of the raw material pretreatment optimization item, the adjustment amplitude threshold of the processing parameter adjustment item, and the monitoring frequency of the quality monitoring strengthening item; Generating the process optimization suggestion set according to the priority score, the adjustment amplitude threshold, and the monitoring frequency.

4. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 3, characterized in that, The dynamically correcting the process parameter coupling features based on the deviation weight to generate corrected process parameter coupling features includes: Extracting the parameter influence factor of each process parameter from the deviation weight; Determine the weight adjustment coefficient of each process parameter edge in the second association graph according to the parameter influence factor; Dynamically scale the weights of the process parameter edges in the second association graph based on the weight adjustment coefficient to generate a scaled process parameter coupling feature; Perform cross-edge weight normalization on the scaled process parameter coupling feature so that the sum of the weights of all process parameter edges is the same as before scaling, and generate a normalized process parameter coupling feature; Update the edge weight distribution in the second association graph according to the normalized process parameter coupling feature to generate the corrected process parameter coupling feature.

5. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 1, wherein, The pre-trained co-occurrence analysis model is obtained through the following steps: Obtain a historical wheat germ processing dataset, which includes a raw material attribute training set, a process parameter training set, and a quality index training set; Perform cross-batch alignment processing on the raw material attribute training set and the process parameter training set to generate an aligned training data group; Perform sliding window sampling on the aligned training data group to generate process parameter training sequences for multiple training stages; Construct a first association graph for training based on the co-occurrence pattern between raw material attributes and process parameters in the process parameter training sequence, and extract the associated features of raw material attributes for training; Construct a second association graph for training based on the coupling relationship between process parameters in the process parameter training sequence, and extract the coupling features of process parameters for training; Perform fluctuation pattern annotation on the quality index training set to generate quality fluctuation pattern features for training; Input the associated features of raw material attributes for training, the coupling features of process parameters for training, and the quality fluctuation pattern features for training into the initial model to generate predicted process optimization suggestions, and train the co-occurrence analysis model based on the difference between the predicted process optimization suggestions and the annotated optimization suggestions.

6. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 5, characterized in that, Perform sliding window sampling on the aligned training data group to generate process parameter training sequences for multiple training stages, including: Set the length and step size of the sliding window according to the operation cycle of the processing equipment; Generate multiple sliding windows based on the length and step of the sliding window, and extract the mean value of raw material attributes, the maximum value of process parameters, and the variance of quality indicators within each sliding window as window features; Construct a process parameter training sequence based on the change rate between adjacent window features, where each process parameter training sequence contains the feature change trajectories of N consecutive windows; Perform noise filtering on the process parameter training sequence to remove abnormal sequences that exceed the preset fluctuation threshold.

7. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 5, characterized in that The construction of the first association graph for training based on the co-occurrence pattern between raw material attributes and process parameters in the process parameter training sequence includes: Count the number of synchronous occurrences of moisture content and drying temperature, protein content and tableting pressure, and impurity ratio and cleaning duration in each process parameter training sequence; Calculate the co-occurrence intensity weight between raw material attributes and process parameters according to the number of synchronous occurrences; Construct the first association graph for training with raw material attributes as nodes, process parameters as edges, and co-occurrence intensity weight as edge weights; Moreover, constructing a second association graph for training based on the coupling relationship between process parameters in the process parameter training sequence includes: Calculating the similarity of the change trends between every two process parameters within the same processing stage, where the similarity of the change trends is obtained by matching through the dynamic time warping algorithm; Determining the coupling strength between process parameters according to the similarity of the change trends; Constructing the second association graph for training with process parameters as nodes and the coupling strength as the edge weight.

8. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 5, characterized in that, Inputting the training raw material attribute association features, training process parameter coupling features, and training quality fluctuation pattern features into the initial model to generate predicted process optimization suggestions, including: Inputting the training raw material attribute association features into the parameter optimization module of the initial model and parsing to obtain the training deviation weights; Performing weighted correction on the training process parameter coupling features based on the training deviation weights to generate training corrected process parameter coupling features; Concatenating the training corrected process parameter coupling features and the training quality fluctuation pattern features to generate training concatenated features; Invoking the suggestion generation module of the initial model to parse the training concatenated features to generate predicted raw material pretreatment optimization items, predicted processing parameter adjustment items, and predicted quality monitoring strengthening items; Moreover, training the co-occurrence analysis model based on the difference between the predicted process optimization suggestions and the labeled optimization suggestions includes: Calculating the first error between the predicted raw material pretreatment optimization items and the labeled raw material pretreatment optimization items, the second error between the predicted processing parameter adjustment items and the labeled processing parameter adjustment items, and the third error between the predicted quality monitoring strengthening items and the labeled quality monitoring strengthening items; Constructing a multi-task loss function according to the first error, the second error, and the third error, where the multi-task loss function includes a mean square error term and a cross-entropy error term; Adjusting the parameters of the initial model through the backpropagation algorithm until the multi-task loss function converges to obtain a pre-trained co-occurrence analysis model.

9. The method for generating optimization suggestions for wheat germ processing technology based on co-occurrence analysis according to claim 1, characterized in that Feeding back the process optimization suggestion set to the wheat germ processing control system to trigger a processing parameter adjustment operation, including: Converting the moisture control suggestion in the raw material pretreatment optimization item into a duration control instruction for the cleaning equipment; Converting the tablet pressing pressure suggestion in the processing parameter adjustment item into a pressure gradient adjustment parameter for the tablet press; Converting the detection frequency suggestion in the quality monitoring strengthening item into a sampling interval configuration for the quality inspection instrument; Issuing the duration control instruction, the pressure gradient adjustment parameter, and the sampling interval configuration to the wheat germ processing control system and receiving an adjustment confirmation signal returned by the wheat germ processing control system to complete the closed-loop control.

10. A processing service system, characterized in that, The processing service system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions, or codes, and the processor is used to run the programs, instructions, or codes in the memory to implement the method for generating wheat germ processing process optimization suggestions based on co-occurrence analysis according to any one of claims 1-9 above.

Images