Fermentation optimization process for high-quality koji pieces based on time series clustering

Abstract

The present application relates to a kind of high-quality koji block fermentation optimization process based on time series clustering, comprising the following steps: obtaining koji history data, including koji initial state information when entering koji and koji koji-making process temperature data;The koji history data is preprocessed, and clustering is carried out based on preprocessed data, the clustering includes short time axis clustering and long time axis clustering, both clustering methods adopt K-Means clustering method, and the similarity of koji is quantified by combining DTW distance measurement;Extract key features in class sample, construct koji high-quality koji block guide interval and standard curve;In new round koji-making process, the closest class is matched, and the matching guide interval is found;If deviate from the matching guide interval in koji-making process, then dynamically adjust koji-making parameter.The present application extracts pattern using koji temperature change sequence, realizes the intelligent guidance to new round koji-making process, so as to improve the consistency and overall level of koji quality.

Description

Technical Field

[0001] This invention relates to an optimization process for high-quality koji block fermentation in koji rooms based on time series clustering, belonging to the fields of brewing engineering, intelligent manufacturing and artificial intelligence technology. Background Technology

[0002] In the brewing process, the koji-making process in the koji room is a crucial step, and its quality directly affects the fermentation quality and the final flavor of the product. Traditional koji-making processes rely excessively on the experience of operators for temperature and humidity control, severely lacking precise data analysis as a powerful guide. Most existing methods rely on single-point or static parameters for control, such as collecting temperature and humidity data from only a few locations within the koji room or analyzing environmental parameters at a few fixed time points. This approach has significant drawbacks, failing to comprehensively and dynamically reflect the real-time temperature evolution of different sections of the koji blocks throughout the entire koji-making process. This makes it difficult to deeply understand the complex biochemical reaction mechanisms involved in koji making, and even more difficult to accurately identify the fermentation patterns of high-quality koji blocks.

[0003] To address these issues, researchers attempted to introduce cluster analysis into the koji-making process, hoping to reveal the temperature variation patterns of koji blocks and uncover the fermentation characteristics of high-quality koji blocks through cluster analysis of large amounts of monitoring data. However, the commonly used K-means clustering algorithm has significant drawbacks. Firstly, the K-means algorithm cannot accurately and autonomously determine the K value for clustering, and the selection of the K value has a decisive impact on the clustering results; an unreasonable K value may lead to a significant deviation between the clustering results and the actual situation. Secondly, the algorithm is extremely sensitive to outliers in the dataset. Occasional environmental fluctuations or equipment malfunctions during the koji-making process can cause significant shifts in cluster centers, distorting the clustering results. Furthermore, when processing large-scale koji-making monitoring data, the K-means algorithm is prone to getting trapped in local optima. The clustering results are greatly affected by the initial point selection and the order of data input; different initial points may yield significantly different clustering results, and the results lack stability. Moreover, it cannot perform incremental calculations, making it difficult to effectively uncover and utilize the fermentation patterns of high-quality koji blocks, and thus hindering the provision of reliable data support and decision-making basis for optimizing the koji-making process.

[0004] Therefore, it is necessary to provide a new high-quality koji fermentation process in koji rooms to solve the above problems. Summary of the Invention

[0005] This invention provides a high-quality fermentation optimization process for koji blocks in koji-making rooms based on time series clustering. By utilizing the koji temperature change sequence extraction mode, it achieves intelligent guidance for a new round of koji-making in koji-making rooms, thereby improving the consistency and overall quality of koji blocks.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] A high-quality koji fermentation optimization process based on time series clustering specifically includes the following steps:

[0008] Step S1: Obtain historical data of the koji room, including initial state information when koji is added to the koji room and temperature data during the koji-making process.

[0009] Step S2: Preprocess the historical data of the music rooms and perform clustering based on the preprocessed data. The clustering includes short-time axis clustering and long-time axis clustering. Both clustering methods use the K-Means clustering method and combine DTW distance metric to quantify the similarity of the music rooms. The long-time axis clustering method performs preliminary clustering based on the initial state information of the music rooms when they enter the music room before implementation.

[0010] Step S3: Analyze the clustering results obtained in step S2, extract key features from the class samples, and construct the high-quality curvature block guidance interval and standard curve for the curvature room.

[0011] In step S4, during the new round of koji making, based on the initial information of the newly entered koji room, the closest class is matched in the model of step S2, and the matching guidance interval is found in the high-quality koji block guidance interval of step S3; if the koji making process deviates from the matching guidance interval, the koji making parameters are dynamically adjusted and the fermentation process continues.

[0012] Furthermore, in step S1, the method for obtaining temperature data during the koji-making process is to install several temperature sensors within the selected koji block sample and collect temperature data of the koji block sample during the fermentation cycle.

[0013] Furthermore, the initial state information when the music is placed in the music room is static data, which includes music block type, room temperature, room humidity, music room orientation, music room location, and month of music placement.

[0014] The temperature data includes data on the temperature change of the koji blocks over time during the fermentation cycle.

[0015] Furthermore, in step S2, the specific steps for clustering based on short-time-axis time series are as follows:

[0016] Step S211: Standardize the temperature data of the koji block samples during the koji-making process. The processing formula is as follows:

[0017]

[0018] In the formula, μ i For X i The average value, X iFor the temperature data of the curved block sample, σ i Let i be the standard deviation, i = 1, 2, ..., N;

[0019] Step S212: Select k time series as initial cluster centers. Calculate the distance (DTW) between each time series and all k cluster centers. Assign the corresponding time series to the nearest cluster center for dynamic time warping. The calculation formula is as follows:

[0020]

[0021] In the formula, W represents the alignment path; t k For X i The corresponding k-th time point, X i Let t be the time series corresponding to the i-th sample. l For X j The corresponding l-th time point, X j The time series corresponding to the j-th sample, where the i-th sample or the j-th sample refers to a complete music production process corresponding to each sample;

[0022] Step S213: Output the cluster label of each time series, and use the average sequence output of all sequences in each cluster as the new cluster center. Repeat the assignment and update.

[0023] Step S214: The clustering results based on the short time axis are used as the basis for constructing the high-quality cursive block guidance interval in step S3;

[0024] Furthermore, in step S2, the specific steps for clustering based on the time series along the long time axis are as follows:

[0025] Step S221: Based on the entry status, the historical data of the music room when the music is played is uniformly converted into numerical features to construct the entry status vector;

[0026] Step S222: Normalize the features of the transformed historical data;

[0027] Step S223: Use Euclidean distance for measurement, calculated using the following formula:

[0028]

[0029] In the formula, S represents the state of entering the song, S ik For sample S i The value of S in the k-th feature dimension jk For sample S j The value taken at the k-th feature dimension, where m is the total number of feature dimensions;

[0030] Step S224: Group the entry state and output the static subset S_group_i∈{1,...,M} to which the music room belongs, thus completing the static entry state clustering;

[0031] Step S225: The temperature time series set of the obtained static subset of the curing blocks is processed according to steps S212-S212 to obtain the clustering result based on the long time axis, which serves as the basis for constructing the high-quality curing block guidance interval in step S3.

[0032] Furthermore, in step S223, the entry state vector d represents the block type code, initial temperature and humidity, entry time, location code, or wind direction and speed.

[0033] Furthermore, based on the initial short-time axis or long-time axis clustering results, the clustering interval of the music room that is most similar to the initial state when the music room was entered in the history is used as the guidance interval of the target music room;

[0034] Furthermore, in step S4, if the temperature change curve deviates from the matching guidance range during the koji-making process, an early warning is issued, and adjustments are made to the degree of opening and closing of the koji room doors and windows and the koji-turning time to ensure that the temperature change during the koji-making process remains stable within the guidance range.

[0035] By employing the above technical solutions, the present invention has the following beneficial effects compared to the prior art:

[0036] 1. The high-quality koji block fermentation optimization process based on time series clustering provided by this invention can utilize historical high-quality koji block data to form a reusable "experience knowledge model".

[0037] 2. The high-quality koji block fermentation optimization process based on time series clustering provided by this invention offers a two-level clustering strategy for both short and long time axes, adapting to different analysis scenarios;

[0038] 3. The time-series clustering-based high-quality koji fermentation optimization process provided by this invention improves the controllability of the new koji-making process, reduces koji quality fluctuations, and facilitates intelligent management and stable quality control of the koji room. Attached Figure Description

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0040] Figure 1 This is a comparison chart of the clustering performance of KMeans clustering provided by this invention under two metric methods: Euclidean distance and DTW distance.

[0041] Figure 2 This is a preferred embodiment of the cluster number selection provided by the present invention. Detailed Implementation

[0042] The invention will now be described in further detail with reference to the accompanying drawings.

[0043] As described in the background section, the commonly used clustering algorithm in the prior art is K-means clustering analysis. This traditional clustering algorithm does not require pre-constructing a tree structure, has lower computational cost, is more stable in clustering convex clusters, and has a more intuitive geometric meaning, making the results easier to interpret. However, considering the problems of length differences, time misalignment, and morphological changes in the fermentation process of the koji blocks in this application, the traditional K-means framework alone is no longer sufficient to accurately cluster the temperature evolution law.

[0044] In a curing chamber, both curing blocks may experience a temperature change process of initial slow → moderately strong → slow again, but one reaches its peak several hours earlier than the other. To improve clustering accuracy, this study attempts to combine K-means with different measurement methods. Specifically, it considers the clustering effects of K-Means clustering under two measurement methods: Euclidean distance and DTW distance.

[0045] The clustering performance of K-Means clustering under two metrics, Euclidean distance and DTW distance, is compared. Figure 1 As shown, the Euclidean distance in 1a focuses more on the absolute difference in time steps, resulting in poor clustering performance for sequences with misaligned time axes. In contrast, the DTW distance in 1b aligns the shape differences of time series, making it easier to group samples with similar shapes but time shifts into the same category. Since the temperature data of the flute chamber is time-series pattern data, there are temperature change characteristics with time shifts but similar shapes among the flute blocks. Clearly, DTW (Dynamic Time Warping), as a non-linear alignment technique, is used to measure the similarity between two time series, allowing for flexible matching of time axes. Using the DTW distance can more accurately identify these similarly shaped temperature curves, thereby improving the matching degree and interpretability of cluster analysis to the actual process.

[0046] Therefore, as the biggest innovation of this application, the use of Kmeans+DTW clustering avoids, to a certain extent, the impact of different times of entering the early, middle and late stages of fermentation on the clustering effect. It can effectively align time series with different rates of change, identify the similarities between them, and is not affected by time axis misalignment. It will classify them into the same warming trend, which is more in line with the actual law.

[0047] The following describes the optimal fermentation process for high-quality koji blocks in a koji room based on time series clustering provided in this application, which specifically includes the following steps:

[0048] Step S1: Obtain historical data of the koji room, including the initial state information when koji is added to the koji room and the temperature data of the koji during the koji making process.

[0049] In traditional production models, product temperature detection mainly relies on manual probing (such as using bamboo skewers or metal rods) and the sensory experience of operators. While this method has accumulated some experience over a long period of practice, it has significant limitations: firstly, manual temperature probing cannot achieve numerical measurement, and the obtained temperature information is highly subjective and uncertain; secondly, due to the lack of a systematic temperature data recording mechanism, the temperature change process cannot be traced and quantitatively analyzed, making it difficult to provide reliable support for subsequent process optimization and intelligent control. To solve the above problems, the method for obtaining product temperature data during the koji-making process provided in this application involves deploying several temperature sensors within a selected koji block sample. This enables real-time monitoring, accurate acquisition, and automatic recording of product temperature, significantly improving the accuracy and completeness of temperature data.

[0050] Historical data regarding the entry of koji into the koji room refers to historical data from the koji room after the koji-making process has been completed. Ideally, this includes two key parts: static data and dynamic data on the temperature changes of the koji blocks over time. The former includes static characteristics such as koji block type, temperature upon entry, humidity upon entry, orientation of the koji room, location of the koji room, and month of entry. The latter is dynamic data on the temperature changes of the koji blocks over time during the koji-making cycle.

[0051] Step S2 involves preprocessing the historical data of the music rooms and performing clustering based on the preprocessed data. The clustering includes short-time axis clustering and long-time axis clustering. Both clustering methods use the K-Means clustering method and combine it with the DTW distance metric to quantify the similarity of the music rooms. The long-time axis clustering method performs preliminary clustering based on the initial state information of the music rooms when they enter the music room before implementation.

[0052] To improve data quality and ensure the effectiveness of the time series clustering model, this step first preprocesses the fermentation room data obtained in step S1 before clustering. This includes handling missing values: interpolation methods are used to fill in missing data to ensure the integrity of the time series; noise suppression: the original data is compressed at time resolution to one record per hour, which smooths the data and reduces noise; and handling inconsistent lengths: to ensure consistency in duration between samples, the shortest fermentation cycle is selected as the uniform truncation length, and it is ensured that the temperature change trend of each fermentation room remains unchanged within the truncated time period to avoid truncation from causing deviations in the clustering results.

[0053] After completing the above preprocessing, the K-Means time series clustering algorithm was used, and the Dynamic Time Warped Distance (DTW) was selected as the distance metric to establish a similarity model of the temperature change trajectory. The innovation here lies in using K-means + DTW clustering, which can effectively align time series with different rates of change, identify their similarities, and is unaffected by time axis misalignment. It will classify the identified time series into the same warming trend, which is more in line with actual patterns.

[0054] This step, a key and innovative part of the entire application, also includes dividing the clustering methods into two strategies: the first is a short-time-axis clustering strategy (coarse strategy), and the second is a long-time-axis clustering strategy (fine strategy). The applicability of these two strategies also differs. Regarding the short-time-axis clustering strategy, it usually considers three aspects: (1) the differences between initial conditions are not significant, (2) the need for rapid grouping, and resource constraints during model deployment, and (3) it is suitable for early model exploration, visualization template construction, and when there is no hierarchical control. Regarding the long-time-axis clustering strategy, it also considers three aspects: (1) the initial conditions have a significant impact on the fermentation process (such as wind speed, koji block type, etc.), (2) a more personalized temperature control strategy needs to be developed, and (3) it is suitable for large-scale historical data and production environments with high quality control requirements.

[0055] Of course, there are some differences in the selection of data before calculating the number of clusters. When using a shorter time axis, the data of music production in music production rooms in the same month of this year as the target music production room, as well as the same month of the same period last year and the same month, are selected before clustering all music production room data. When using a complete or longer time axis, the data of the same period and the most recent month of the target music production room in the past five years are selected, and music production rooms in the same category as the target music production room in the initial cluster are selected to reduce the number of music production rooms.

[0056] The short-time-axis clustering strategy mentioned above is characterized by its suitability for scenarios where rapid initial screening or differences in the state of fermentation entering the fermentation room do not significantly affect temperature control. The strategy does not distinguish the state of fermentation entering the fermentation room, but directly standardizes all historical fermentation room temperature data, applies the Kmeans+DTW distance method for time series clustering, and then uses the clustering results to construct a general temperature control reference interval.

[0057] The specific steps for clustering time series based on short time axes are as follows:

[0058] Step S211: Standardize the temperature data of the koji block samples during the koji-making process. The processing formula is as follows:

[0059]

[0060] In the formula, μ i For Xi The average value, X i For the temperature data of the curved block sample, σ i Let i be the standard deviation, i = 1, 2, ..., N;

[0061] Step S212: Select k time series as initial cluster centers. Calculate the distance (DTW) between each time series and all k cluster centers. Assign the corresponding time series to the nearest cluster center for dynamic time warping. The calculation formula is as follows:

[0062]

[0063] In the formula, W represents the alignment path; t k Let X be the k-th time point corresponding to Xi. i Let t be the time series corresponding to the i-th sample (here, one sample corresponds to a complete koji-making process, t) l Let Xj be the l-th time point, X j This is the time series corresponding to the j-th sample;

[0064] Step S213: Output the cluster label of each time series, and use the average sequence output of all sequences in each cluster as the new cluster center. Repeat the assignment and update.

[0065] Step S214: The clustering results based on the short time axis are used as the basis for constructing the high-quality cursive block guidance interval in step S3.

[0066] Example of the processing formula in step S211, if X i Defined as hourly temperature records, with an hourly temperature record length of T, then in step S212, input during clustering: {X i (t)}, i=1,2,…,N, Without considering the differences in state at the time of entry (such as type, initial temperature, orientation, etc.), the output is: the cluster label to which each time series belongs, such as C={c1,c2,...,cN},ci∈{1,...,K}, the output of each cluster center (representing a typical temperature rise curve) and the corresponding general temperature control reference interval (such as cluster center ± standard deviation) for each cluster.

[0067] Regarding the long-time axis clustering strategy, its difference from short-time axis clustering is that it requires preliminary clustering based on the initial state information when the music room is entered. The specific steps are as follows:

[0068] Step S221: Based on the entry status, the historical data of the music room when the music is played is uniformly converted into numerical features to construct the entry status vector;

[0069] Step S222: Normalize the features of the transformed historical data;

[0070] Step S223: Use Euclidean distance for measurement, calculated using the following formula:

[0071]

[0072] In the formula, S represents the state of entering the song, S ik For sample S i The value of S in the k-th feature dimension jk For sample S j The value taken at the k-th feature dimension, where m is the total number of feature dimensions;

[0073] Step S224: Group the entry state and output the static subset S_group_i∈{1,...,M} to which the music room belongs, thus completing the static entry state clustering;

[0074] Step S225: The temperature time series set of the obtained static subset of the curry blocks is processed according to steps S212-S212 to obtain the clustering result based on the long time axis, which serves as the basis for constructing the guidance interval for high-quality curry blocks in step S3.

[0075] Clearly, it consists of two stages. The first stage is static entry state clustering (feature vector similarity clustering). The input is the entry state vector of each sample. S represents the type of curved room, initial temperature and humidity, entry time, location, wind direction, and wind speed, etc. In step S222 of the first stage, features are normalized (0-1 or z-score), and distance measurement is performed, i.e., Euclidean distance in step S223. Finally, in step S224, a clustering method is used, such as K-Means, DBSCAN, or hierarchical clustering, to group the entry states S. Step S224 outputs the state cluster (static subset) to which each curved room sample belongs, where each subset represents a set of curved rooms with similar entry conditions.

[0076] The second stage is dynamic temperature curve clustering (time series + DTW). The input is the set of temperature time series within each static subset: {Xi(t)} within the same S_group. Clustering is performed using K-Means + DTW according to the "short time axis strategy" in steps S212-S212, applicable to each subset. The final output includes: cluster labels, heating modes, and differentiated temperature control ranges for different state clusters under each input state subset.

[0077] Whether in short-time or long-time clustering strategies, the issue of selecting the optimal number of clusters is involved. Elbow plots are preferred, as they are a classic visualization method used in cluster analysis to determine the optimal number of clusters (K value). They are particularly suitable for scenarios such as fermentation rooms (e.g., brewing fermentation rooms, microbial culture rooms) that require zoning management based on environmental data (temperature, humidity, oxygen concentration, etc.).

[0078] The following provides a preferred embodiment for selecting the number of clusters, such as... Figure 2 As shown in the figure, the X-axis represents the number of clusters k, and the Y-axis displays the sum of squared errors within each cluster (Inertia) for each k value. This value reflects the compactness of the data points within each cluster; the smaller the value, the better the clustering effect. As the value of k increases, the sum of squared errors within each cluster gradually decreases, but the rate of decrease gradually slows down and eventually flattens out, exhibiting a clear "elbow" shape.

[0079] As shown in the figure, the change from k=1 to k=2 is relatively large, and there is still a significant decrease from k=2 to k=3. However, the decrease from k=3 to k=4 and from k=4 to k=5 is significantly smaller, with almost no further benefit. Therefore, the optimal number of clusters is usually chosen at the elbow position, i.e., k=3, because after this point, the reduction in the sum of squared intra-cluster errors brought about by increasing the number of clusters becomes insignificant. To quantify the determination of the optimal number of clusters, the maximum curvature method can be used for calculation. The coordinates of the five points in the figure are (1, 37210.78), (2, 21588.20), (3, 10892.35), (4, 2511.52), and (5, 0). The calculation steps are as follows:

[0080] First, let the starting and ending points be A = (x1, y1) = (1, 37210.78); B = (x2, y2) = (5, 0);

[0081] The second step is to calculate the distance from each intermediate point P = (xi, yi) to AB:

[0082]

[0083] The third step is to calculate the distances between the three points k=2, k=3, and k=4 in sequence, and select the k value corresponding to the maximum distance, which is the inflection point.

[0084] Based on the calculations above, the optimal number of clusters determined by the maximum curvature method is k = 3, corresponding to a distance of 0.8291. This is the candidate point furthest from the line connecting the beginning and end, indicating that this is the turning point where the "elbow" is located. Therefore, recommending a cluster number of 3 is a reasonable choice.

[0085] Continue with step S3, analyze the clustering results obtained in step S2, extract key features from the class samples, and construct the high-quality curing block guidance interval for the curing room; regarding the extraction of the high-quality curing block guidance interval for the curing room, analyze the representative curing temperature change trend of each clustering result, and extract key features such as typical temperature range, gradient change rate, and peak period from the class samples to construct the interval.

[0086] In step S4, during the new round of music production, based on the initial information of the newly entered music room (including room temperature, season, and music block information), the model in step S2 is matched with the closest class, and the matching guidance interval is found in the high-quality music block guidance interval in step S3. The current music block temperature change is monitored in real time and matched with the historical high-quality music block pattern. If the music production process deviates from the matched guidance interval, control suggestions are triggered (such as opening and closing doors and windows, adjusting music turning time, etc.), thereby dynamically adjusting the music production parameters and improving the overall consistency of music block quality.

[0087] In summary, the time-series clustering-based high-quality koji fermentation optimization process provided in this application utilizes the koji temperature change sequence extraction mode to achieve intelligent guidance for a new round of koji production in the koji room, thereby improving the consistency and overall quality of koji blocks.

[0088] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0089] The meaning of "and / or" as used in this application includes situations where each exists alone or both exist simultaneously.

[0090] The term "connection" as used in this application can mean a direct connection between components or an indirect connection between components through other components.

[0091] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A time-series clustering-based optimization process for high-quality koji (fermentation blocks) fermentation in a koji room, characterized in that: Specifically, the following steps are included: Step S1: Obtain historical data of the koji room, including initial state information when koji is added to the koji room and temperature data during the koji-making process. Step S2: Preprocess the historical data of the music rooms and perform clustering based on the preprocessed data. The clustering includes short-time axis clustering and long-time axis clustering. Both clustering methods use the K-Means clustering method and combine DTW distance metric to quantify the similarity of the music rooms. The long-time axis clustering method performs preliminary clustering based on the initial state information of the music rooms when they enter the music room before implementation. The specific steps for clustering time series based on short time axes are as follows: Step S211: Standardize the temperature data of the koji block samples during the koji-making process. The processing formula is as follows: ， In the formula, for The average value, For block samples At any moment Temperature data of the product Standard deviation, , Represents the total number of block samples; Step S212, select Using 10 time series as the initial cluster centers, calculate the cluster centers for each time series and all the cluster centers. Distance between cluster centers The corresponding time series is assigned to the nearest cluster center for dynamic time warping, and the calculation formula is as follows: ， In the formula, To align with the complete set of paths, For the complete series One alignment path in for The corresponding number At a certain point in time, For the first Each sample at time... Temperature data of the product for The corresponding number At a certain point in time, For the first Each sample at time... The product temperature data, the first The first sample or the first Each sample refers to a complete koji-making process corresponding to one room; Step S213: Output the cluster label of each time series, and use the average sequence output of all sequences in each cluster as the new cluster center. Repeat the assignment and update. Step S214: The clustering results based on the short time axis are used as the basis for constructing the high-quality cursive block guidance interval in step S3; The specific steps for clustering time series based on a long time axis are as follows: Step S221: Based on the entry status, the historical data of the music room when the music is played is uniformly converted into numerical features to construct the entry status vector; Step S222: Normalize the features of the transformed historical data; Step S223: Use Euclidean distance for measurement, calculated using the following formula: ， In the formula, In the state of entering the music, For the sample In the The values ​​taken on each feature dimension For the sample In the The values ​​taken on each feature dimension The total number of dimensions of the features; Step S224: Group and output the entry state, output the static subset to which the music room belongs, and complete the static entry state clustering; Step S225: The temperature time series set of the obtained static subset of the koji blocks is used to obtain the clustering result based on the long time axis according to steps S212-S224, which serves as the basis for constructing the high-quality koji block guidance interval in step S3. Step S3: Analyze the clustering results obtained in step S2, extract key features from the class samples, and construct the high-quality curvature block guidance interval and standard curve for the curvature room. In step S4, during the new round of koji making, based on the initial information of the newly entered koji room, the closest class is matched in the model of step S2, and the matching guidance interval is found in the high-quality koji block guidance interval of step S3. If the koji making process deviates from the matching guidance interval, the koji making parameters are dynamically adjusted and the fermentation process continues.

2. The optimal fermentation process for high-quality koji blocks in a koji room based on time series clustering according to claim 1, characterized in that: In step S1, the method for obtaining temperature data during the koji-making process is to install several temperature sensors within the selected koji block sample and collect the temperature data of the koji block sample during the fermentation cycle.

3. The optimal fermentation process for high-quality koji blocks in a koji room based on time series clustering according to claim 2, characterized in that: The initial state information when the music is placed in the music room is static data, which includes music block type, room temperature, room humidity, music room orientation, music room location, and month of music placement; The temperature data includes data on the temperature change of the koji blocks over time during the fermentation cycle.

4. The optimal fermentation process for high-quality koji blocks in a koji room based on time series clustering according to claim 1, characterized in that: In step S223, the entry state vector , This represents the total number of dimensions of the state features.

5. The optimal fermentation process for high-quality koji blocks in a koji room based on time series clustering according to claim 1, characterized in that: Based on the initial short-time axis or long-time axis clustering results, the clustering interval of the music room that is most similar to the initial state when the music room was entered in the history is used as the guidance interval of the target music room.

6. The optimal fermentation process for high-quality koji blocks in a koji room based on time series clustering according to claim 1, characterized in that: In step S4, if the temperature change curve deviates from the matching guidance range during the koji-making process, an early warning will be issued, and adjustments will be made to the degree of opening and closing of the koji room doors and windows and the koji-turning time to ensure that the temperature change during the koji-making process remains stable within the guidance range.

Images