Intelligent humidity control method and system for fermentation of bio-organic fertilizer

Abstract

The invention discloses an intelligent humidity control method and system for bio-organic fertilizer fermentation, and belongs to the field of humidity control. The method comprises the steps that a stacking area is divided into basic units, environmental parameters and material characteristic parameters of each unit are periodically obtained, and corresponding humidity regulation and control time delay and compensation coefficients are calculated; dynamically combining the basic units by using a clustering algorithm to obtain partitions, and constructing a coupling influence matrix by considering moisture migration between the partitions; an improved particle swarm algorithm is adopted, the inertia weight is adjusted in a self-adaptive mode according to the humidity regulation and control time delay of each partition, and the optimal fuzzy parameter of each partition is obtained; the preset PID parameter is corrected based on the coupling influence matrix and the humidity regulation and control time delay and compensation coefficient of each partition, the optimal PID parameter of each partition is obtained, the humidity control quantity of each partition is obtained through fuzzy PID control, the problems of regulation and control lag and uneven local humidity in existing humidity control are solved, and the humidity control efficiency is improved. And self-adaptive regulation and control and differential control are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of humidity control, in particular to an intelligent humidity control method and system for biological organic fertilizer fermentation. BACKGROUND

[0002] Biological organic fertilizer fermentation refers to the process of converting various organic waste into high-quality organic fertilizer by means of microbial metabolic activity. In biological organic fertilizer aerobic fermentation, humidity is a key parameter affecting microbial activity and fermentation effect. Too high humidity can easily cause material caking and air blockage, while too low humidity can inhibit microbial metabolism and slow down the fermentation process, directly affecting the quality of finished fertilizer.

[0003] The current mainstream humidity control scheme still has shortcomings: on the one hand, existing methods mostly rely on traditional PID control or fixed ventilation logic, and can only be adjusted passively based on real-time monitored humidity data, without solving the data lag problem caused by asynchronous monitoring and adjustment; on the other hand, existing schemes generally use global regulation mode, without considering the problem of local humidity unevenness caused by structural differences in different piles in the fermentation plant, resulting in low humidity control precision and slow response, which restricts fermentation efficiency and product stability. SUMMARY

[0004] The present application discloses an intelligent humidity control method and system for biological organic fertilizer fermentation, aiming to solve the problems of regulation lag and local humidity unevenness in existing humidity control, and realize adaptive humidity regulation and differentiated humidity control in different fermentation zones during biological organic fertilizer fermentation.

[0005] The technical solution for achieving the purpose of the present application is as follows:

[0006] On the one hand, the present application provides an intelligent humidity control method for biological organic fertilizer fermentation, comprising the following steps:

[0007] dividing the stacking area where the fermentation pile is located into units, periodically acquiring environmental parameters and material characteristic parameters of each unit, and calculating the corresponding humidity regulation time delay and compensation coefficient, wherein, is the total number of units;

[0008] Based on the humidity data in the environmental parameters of each unit, combined with the target humidity and humidity regulation time delay, the clustering algorithm is used to dynamically merge units to obtain zones, and the corresponding humidity data, humidity regulation time delay and compensation coefficient are determined according to the subordinate units in each zone, and the coupling influence matrix is constructed considering the water migration between zones, wherein, is the total number of zones;

[0009] A general fuzzy rule base is established, the fuzzy parameters of each partition are initialized, an improved particle swarm optimization algorithm is adopted, the inertia weight is adaptively adjusted according to the humidity regulation time delay of each partition, the genetic mutation idea is introduced to iteratively optimize the fuzzy parameters of each partition, and the optimal fuzzy parameters of each partition are obtained;

[0010] The preset PID parameters are corrected based on the coupling influence matrix and the humidity regulation time delay and the compensation coefficient of each partition, and the optimal PID parameters of each partition are obtained. According to the deviation and the change rate of the humidity data of each partition from the target humidity, the fuzzy PID control algorithm is adopted to obtain the control amount of each partition.

[0011] Further, the environmental parameters include humidity, temperature, oxygen concentration, and the material characteristic parameters include material moisture content and pH value.

[0012] Specifically, in order to solve the problem of local humidity unevenness in the fermentation process, the fermentation pool where the fermentation is located is divided into a basic monitoring unit according to the structural difference of the fermentation pile, the sensors are installed at appropriate positions of the fermentation pile in each monitoring unit, so as to monitor the environmental parameters and material characteristic parameters in real time. The humidity sensor, temperature sensor, zirconium oxide oxygen sensor, moisture sensor and in-situ pH value sensor are used to obtain the environmental parameters and material characteristic parameters at each sampling time. All the sensors are connected to the edge computing module, the collected data are encapsulated by time stamp, and the time alignment of each parameter at the same sampling time is realized by time stamp calibration.

[0013] Further, the collected data are preprocessed using criteria, the abnormal values in the continuous data of temperature, humidity and the like are eliminated, and the sliding average of the sampling values of the previous 5 times is used for replacement. Meanwhile, based on the historical sampling data, the noise variance of each parameter is calculated, the Kalman filtering algorithm is used to eliminate high-frequency noise, and all the parameter values are normalized to the interval [0, 1].

[0014] Further, the compensation coefficient is calculated by the following steps:

[0015] According to different fermentation stages, the target humidity interval of the current fermentation stage is obtained;

[0016] The deviation degree of the current humidity data from the target humidity interval is calculated;

[0017] The microbial activity is calculated according to the temperature change curve and the oxygen concentration change curve;

[0018] The compensation coefficient is obtained according to the deviation degree and the microbial activity.

[0019] Specifically, the fermentation stage is determined based on the acquired temperature data. A sliding window is used to monitor temperature changes during the fermentation stage, and fermentation is considered to have entered the fermentation stage when the temperature characteristic threshold is met for 3 consecutive hours. Set corresponding target humidity ranges for different fermentation stages. .

[0020] Furthermore, microbial activity is mainly determined by the synergistic effect of temperature and oxygen concentration changes. When the temperature is rising and the oxygen concentration is falling, microbial activity consumes oxygen and releases heat, resulting in higher microbial activity. Based on this pattern, microbial activity can be calculated from the average changes in temperature and oxygen concentration during the monitoring period.

[0021] Calculate the degree of deviation between the current humidity data and the target humidity range. Based on the degree of deviation and microbial activity Obtain the compensation coefficient This is used for subsequent humidity adjustment correction, and the formula is:

[0022] ,

[0023] in This represents the influence coefficient.

[0024] Furthermore, a time delay prediction model is established based on environmental parameters and material characteristic parameters to calculate the humidity control delay of each unit. Specifically, the model is trained based on a CNN-LSTM hybrid network architecture to obtain the time delay prediction model, and the humidity control delay of each unit is calculated. The time series of humidity, temperature, oxygen concentration, material moisture content and pH value aligned with the timestamp is used as multi-channel input. 1D convolution is used to extract the local dynamic features of each input, and then the LSTM time series dependency layer is used to capture the long-term dependency relationship of the feature sequence. A Dropout layer is added to reduce overfitting. Attention weights are assigned to each time step of the LSTM output, and the weighted sum is used to obtain the final feature vector. The humidity control delay prediction value is output through MLP.

[0025] Specifically, the training of the latency prediction model includes:

[0026] The historical data collected underwent outlier removal, noise reduction, and normalization. It was then divided into three mutually exclusive subsets in a 7:2:1 ratio: training, validation, and test sets. The model was then trained using a CNN-LSTM hybrid network architecture, employing mean squared error. As the main loss function, combined with the Adam optimizer to iteratively calculate and update the model weights, the model is evaluated using the validation set, and the early stopping strategy is used to continuously monitor the validation loss. When the validation set loss no longer decreases, training is stopped, and the trained time delay prediction model is solidified into the system for calculating the humidity regulation time delay of each unit.

[0027] Preferably, the time delay prediction model also includes an incremental learning mechanism. After the fermentation of each batch of bio-organic fertilizer is completed, the newly generated time series data-time delay label is used to fine-tune the model incrementally. The specific operation is to freeze the convolutional layer and only train the fully connected layer to adapt to changes in the long-term fermentation environment.

[0028] Preferably, the time delay prediction model also includes an early warning mechanism. When the humidity regulation time delay predicted by the time delay prediction model meets any of the following conditions, an early warning is triggered to avoid humidity out of control due to excessive regulation time delay:

[0029] The predicted humidity regulation time delay is greater than the system's preset absolute threshold warning;

[0030] The predicted humidity regulation time delay is more than 50% higher than the historical average humidity regulation time delay;

[0031] In the above cases, there may be equipment failure or system failure, which should be warned in time and manually intervened if necessary.

[0032] Further, a dynamic partitioning model is established based on the two-dimensional coordinates of the unit monitoring points, humidity, humidity deviation, and humidity regulation time delay. The improved DBSCAN algorithm is used to realize dynamic partitioning of the fermentation area, which specifically includes:

[0033] The two-dimensional coordinates of the monitoring unit , humidity , humidity deviation and humidity regulation time delay are input into the DBSCAN clustering algorithm. According to the different fermentation stages, the field radius function is adaptively adjusted, combined with the corresponding basic radius and minimum cluster point number of different fermentation stages to calculate the humidity standard deviation of different fermentation stages to obtain the adaptively adjusted field radius function , which reflects the uniformity of humidity distribution. When the humidity distribution is uniform, fewer large partitions are formed, and when the humidity fluctuation is large, more small partitions are formed.

[0034] The Euclidean distance metric in the traditional clustering algorithm is improved, and a weighted distance metric function that integrates spatial and state characteristics is designed, with the calculation formula being:

[0035] ,

[0036] wherein, is the spatial distance, expressed as the spatial distance between two unit monitoring points; is the humidity state distance, calculated from the humidity , humidity deviation between two units; is the humidity regulation time delay distance, the weight coefficient , , The sum of is 1, and the influence degree of each parameter is adjusted by adjusting the weight;

[0037] The clustering constraint condition is set as:

[0038] The standard deviation of the humidity deviation of all monitoring points in the same partition is not greater than the preset standard deviation threshold;

[0039] The coefficient of variation of the average humidity regulation time delay of the same partition is not greater than the preset coefficient threshold; The trigger condition is set as: when the average humidity deviation

[0040] in the partition exceeds the preset deviation value for three consecutive sampling periods, the dynamic partition update is triggered, and the dynamic partition model is restarted;

[0041] After the dynamic aggregation of the basic units is completed through the dynamic partition model, the humidity data, humidity regulation time delay and compensation coefficient of each partition are the mean values of the corresponding data of all units under them, representing the overall characteristics of the partition, and providing basic input for subsequent regulation.

[0042] Further, a coupling influence matrix is established, including the following steps:

[0043] For the partitions divided dynamically, the water migration influence factor of the region on the region is calculated, and the calculation formula is:

[0044] , ,

[0045] wherein, is the normalized center distance between the region and the region , is the number of monitoring points contained in the region , is the intensity coefficient, is the distance attenuation coefficient,​ The larger the area represents The stronger the influence on the moisture migration of the area is;

[0046] The moisture migration influence factor of the area is normalized to obtain , and a coupling influence matrix is constructed according to .

[0047] Further, a general fuzzy rule base table is established according to historical sampling data and humidity control data and expert experience, and an improved particle swarm algorithm and genetic algorithm fusion optimization strategy is adopted to iteratively search for optimal fuzzy parameters of each partition, including the following steps:

[0048] The input of the fuzzy PID controller of each partition is defined as the humidity deviation and the humidity change rate , and the output is the PID parameter correction amount: , , , so as to construct the input variable fuzzy subset and the output variable fuzzy subset, and obtain the general initial fuzzy rule base table, wherein the fuzzy parameters to be optimized of each partition include the fuzzy subset center value and width of the input humidity deviation , the humidity change rate , and the fuzzy subset center value and width of the output , , .

[0049] The humidity control accuracy and the control response speed are optimized as double targets to ensure that the optimization result adapts to the characteristics of the actual fermentation delay, the absolute value integral of the humidity control deviation is calculated according to the humidity data of each partition , and the average humidity control time delay of each partition is combined to obtain the fitness function, when the cumulative humidity deviation is small and the average humidity control time delay is short, the fitness is smaller, and the corresponding fuzzy parameters are better;

[0050] Initialize the particle swarm, generate the initial position of the PSO particle by uniform random sampling and boundary constraint, set the initial velocity, and dynamically adjust the inertia weight based on the average humidity control time delay of the partition :

[0051] ,

[0052] wherein and represent the maximum inertia weight and the minimum inertia weight respectively, and ​​The maximum and minimum values of the humidity control time delay in the historical iteration are set as the delay constraint.

[0053] The particle velocity is updated and the particle position is adjusted in combination with the dynamic inertia weight and the delay constraint.

[0054] When the particle position change rate is less than a preset number threshold or the iteration number is reached, the PSO iteration is terminated.

[0055] The particles after the PSO iteration are integrated into the GA initial population, the roulette wheel selection method is used, the fitness of each GA individual is calculated, and the selection probability is allocated according to the fitness, and a number of high-quality individuals are selected as parents to reserve high fitness genes.

[0056] The parents are randomly paired, a crossover point is selected to exchange parameter segments to generate offspring.

[0057] The individual parameters are randomly disturbed, and a low mutation probability is set to avoid destroying high-quality individuals.

[0058] The disturbed individuals are integrated with the remaining particles to form a new population, i.e. one GA operation is completed.

[0059] The PSO-GA iteration is repeated, and when the global optimal fitness changes less than a preset change threshold for 5 consecutive generations or the iteration number threshold is reached, the iteration is terminated, and the corresponding particle parameters are output as the optimal fuzzy parameters of the partition.

[0060] Further, the fuzzy PID control algorithm is used to obtain the control amount of each partition, specifically including using the Ziegler-Nichols time domain method to set the PID parameter reference value, setting the setting result as the PID parameter reference value of each partition , , , inputting the humidity deviation and the deviation change rate at the current time, and using a three-step correction method to correct the PID parameters, and the specific steps are:

[0061] Based on the partition humidity control time delay , the PID parameter reference value is corrected, and the PID parameter is corrected using linear correction to obtain the corrected control parameter , , ;

[0062] The PID control parameters are linearly corrected based on the compensation coefficient, and the PID control parameters are finally corrected based on the coupling influence matrix to realize the differential regulation of different partitions.

[0063] The PID parameters are fine-tuned in real time using fuzzy reasoning, specifically the input and Carrying out the fuzzification processing, and outputting the PID parameter correction amount through the optimal fuzzy parameter calculation of each partition 、 、 After the deblurring by the sampling gravity method, the PID parameters are calibrated to obtain final parameters, and the control amount is calculated based on the final parameters.

[0064] The control amount is essentially the strength of humidity adjustment, and when the control amount is positive, it represents the water supplement strength, and when the control amount is negative, it represents the dehumidification strength.

[0065] Optionally, the system further comprises a closed-loop feedback mechanism and an abnormal alarm mechanism.

[0066] On the other hand, the application also provides an intelligent humidity control system for bio-organic fertilizer fermentation, comprising a collection module, a dynamic partition module, a control optimization module and a regulation and control module.

[0067] The collection module periodically acquires the environmental parameters and material characteristic parameters of each unit in the stacking area, and calculates the humidity regulation time delay and compensation coefficient of each unit.

[0068] The dynamic partition module dynamically merges units based on the humidity data in the environmental parameters of each unit, combines the target humidity and the humidity regulation time delay, and uses a clustering algorithm to obtain partition, and determines the corresponding humidity data, humidity regulation time delay and compensation coefficient according to the subordinate units in each partition, and constructs a coupling influence matrix considering the moisture migration between partitions.

[0069] The control optimization module initializes the fuzzy parameters of each partition, uses an improved particle swarm algorithm, adaptively adjusts the inertia weight according to the humidity regulation time delay of each partition, introduces the genetic mutation idea to iteratively optimize the fuzzy parameters of each partition, and obtains the optimal fuzzy parameters of each partition.

[0070] The regulation and control module corrects the preset PID parameters, and obtains the control amount of each partition by using a fuzzy PID control algorithm according to the deviation and deviation change rate of the humidity data of each partition and the target humidity.

[0071] Compared with the prior art, the application has the following advantages:

[0072] 1. Differentiated humidity regulation of different regions is realized by dividing multiple basic units and dynamically aggregating units, and the phenomenon of mold growth or inactivation due to over-drying in local areas while the global humidity meets the standard in traditional global regulation is solved.

[0073] 2. Calculate humidity control time delay and compensation coefficient to correct fuzzy PID control parameters, and use improved particle swarm algorithm to adaptively adjust inertia weight according to partition humidity control time delay to obtain optimal fuzzy parameters of the partition, so that active prediction and adjustment of humidity are realized, and the problems of excessive or insufficient water caused by traditional control lag are avoided. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 It is an intelligent humidity control method flow chart for bio-organic fertilizer fermentation;

[0075] Figure 2 It is a humidity control flow chart based on fuzzy PID control in the application.

[0076] Figure 3 It is a humidity control flow chart based on fuzzy PID control in the application.

[0077] Figure 4 It is a humidity control flow chart based on fuzzy PID control in the application. DETAILED DESCRIPTION

[0078] The application will be further described in detail below in combination with the drawings and examples.

[0079] As shown in the drawings, Figure 1 The application discloses an intelligent humidity control method for bio-organic fertilizer fermentation, which comprises the following steps:

[0080] The stacking area where the fermentation pile is located is divided into units, environmental parameters and material characteristic parameters of each unit are periodically sampled and obtained, and corresponding humidity control time delay and compensation coefficient are calculated, wherein, is the total number of divided units;

[0081] Based on humidity data in the environmental parameters of each unit, target humidity and humidity control time delay are combined, clustering algorithm is used to dynamically merge units to obtain partitions, humidity data, humidity control time delay and compensation coefficient of the subordinate units in each partition are determined, and a coupling influence matrix is constructed by considering the moisture migration between the partitions, wherein, is the total number of partitions, and ;

[0082] A general fuzzy rule base is established, fuzzy parameters of each partition are initialized, an improved particle swarm optimization algorithm is used, inertia weight is adaptively adjusted according to the humidity control time delay of each partition, genetic variation idea is introduced to iteratively optimize the fuzzy parameters of each partition, and optimal fuzzy parameters of each partition are obtained.

[0083] The preset PID parameters are corrected based on the coupling influence matrix and the humidity regulation time delay and compensation coefficient of each subzone to obtain optimal PID parameters of each subzone, and a fuzzy PID control algorithm is used to obtain a control amount of each subzone according to the deviation and the deviation rate of the humidity data of each subzone from the target humidity.

[0084] The purpose of the embodiment is to realize intelligent humidity control of the bio-organic fertilizer suitable for tobacco in the fermentation process. The fermentation parameters in the fermentation process of the bio-organic fertilizer influence each other, have strong time sequence and coupling, so real-time monitoring of the fermentation environment parameters and material characteristic parameters in the fermentation process is very important for realizing accurate humidity control. The environment parameters to be monitored include humidity, temperature, and oxygen concentration, and the material characteristic parameters are material moisture content and pH value.

[0085] Specifically, in order to solve the problem of local humidity unevenness in the fermentation process, the fermentation pool in which the bio-organic fertilizer is fermented is divided into monitoring units according to the structural differences of the fermentation piles. The sensors are installed at appropriate positions of the fermentation piles in each monitoring unit to monitor the environment parameters and material characteristic parameters in real time. In the embodiment, humidity sensors, temperature sensors, zirconia oxygen sensors, moisture sensors, and in-situ pH value sensors are used to obtain the environment parameters and material characteristic parameters at each sampling time. The collected raw data are converted into electrical signal data, and all the sensors are connected to the edge computing module. The collected data are encapsulated by time stamp, and the time alignment of each parameter at the same sampling time is realized by time stamp calibration. In the embodiment, the sampling period is set to 30 minutes, and in other embodiments, the sampling period can be extended or shortened according to the actual situation, which is not limited in the application.

[0086] Further, the acquired data are preprocessed using criteria, and the abnormal values in the continuous data of temperature, humidity, etc. are removed and replaced by the sliding average of the previous 5 sampling values. Based on the historical sampling data, the noise variance of each parameter is calculated, the Kalman filtering algorithm is used to eliminate high-frequency noise, and all the parameter values are normalized to the interval [0, 1].

[0087] Further, the compensation coefficient is calculated including the following steps:

[0088] According to different fermentation stages, the current target humidity interval is obtained;

[0089] The deviation degree of the current humidity data from the target humidity interval is calculated;

[0090] The microbial activity is calculated according to the change curve of temperature and the change curve of oxygen concentration;

[0091] A compensation coefficient is obtained according to the deviation degree and the microbial activity.

[0092] In the embodiment, in order to ensure sufficient composting of the bio-organic fertilizer, the fermentation process needs to go through different fermentation stages such as a temperature rising stage, a high temperature stage, a temperature falling stage and a composting stage. The fermentation stages are mainly determined by monitoring temperature changes, and the fermentation stages are determined according to the obtained temperature change curve. Meanwhile, the ideal humidity range of the fermentation process dynamically changes with the fermentation process. Therefore, a sliding window is used to monitor the change of the fermentation stage. When the monitored temperature meets the temperature threshold value of the corresponding stage for 3 consecutive hours, it is determined that the fermentation enters the fermentation stage , and the corresponding target humidity range is set for different fermentation stages .

[0093] Specifically, the microbial activity is mainly determined by the synergistic effect of temperature and oxygen concentration in the fermentation environment. When the temperature is in the rising stage and the oxygen concentration is in the falling stage, the metabolic activity of microorganisms is vigorous, oxygen is consumed and heat is released. At this time, the microbial activity is high. According to this rule, in the embodiment, the calculation formula of the microbial activity in the fermentation process is defined as:

[0094] ,

[0095] Wherein, is the average value of the temperature change in a monitoring period, which reflects the temperature change trend and average change amplitude in the period; is the average value of the oxygen concentration change in a monitoring period, which reflects the average change of the oxygen concentration in the period; is the average temperature in the monitoring period. In the embodiment, the monitoring period corresponds to a sliding window with a size of 12 time steps, and the microbial activity is obtained by normalizing the microbial activity;

[0096] The deviation degree of the current humidity from the target humidity range is calculated, and the calculation formula is:

[0097] ,

[0098] When the microbial activity is high, the response to humidity deviation is usually more sensitive. When the microbial activity is low, the response to humidity deviation is relatively insensitive. Therefore, the compensation coefficient is calculated based on the humidity deviation and the microbial activity , and the formula is:

[0099] ​ ,

[0100] wherein, is the influence coefficient.

[0101] When the humidity of the fermentation pile in the monitoring unit deviates from the target humidity interval, the actuator fails to perform water replenishment or ventilation operation in time, but starts operation after a certain interval, then the interval time represents the regulation hysteresis of the actuator, and the time interval between the start of the humidity control instruction issued by the PLC and the start of the effective rise or fall of the humidity monitored by the humidity sensor represents the control hysteresis in the humidity regulation process, and in the embodiment, the humidity regulation time delay of the bio-organic fertilizer in the fermentation process is defined as the sum of the regulation hysteresis and the control hysteresis.

[0102] As Figure 2 mentioned, specifically, the humidity regulation time delay of each unit is calculated according to the establishment of the time delay prediction model based on the environmental parameters and the material characteristic parameters, and specifically, the time delay prediction model is obtained based on the CNN-LSTM hybrid network architecture for model training to calculate the humidity regulation time delay of each unit; the time-stamped humidity, temperature, oxygen concentration, material moisture content and pH value time series are used as multi-channel input, 1D convolution is used to extract the local dynamic features of each input, LSTM time sequence dependent layer is used to capture the long-time dependence of the feature sequence, and Dropout layer is added to reduce overfitting, attention weight is assigned to each time step of the LSTM output, and the final feature vector is obtained by weighted summation, and the humidity regulation time delay prediction value is output through MLP;

[0103] Specifically, the model training of the time delay prediction model includes:

[0104] The data collected in the historical period is subjected to outlier rejection, denoising and normalization processing, and is divided into three mutually exclusive subsets in the ratio of 7:2:1, which are training set, validation set and test set, and then the model is trained based on the CNN-LSTM hybrid network architecture, and the mean square error is used as the main loss function, and the calculation formula is:

[0105] ,

[0106] wherein, is the actual value of the i-th sample in the input sample set, is the prediction value of the i-th sample output by the CNN-LSTM network training, ​​The sample quantity is used; the mean square error is used to calculate and update the model weight iteratively combined with the Adam optimizer, the model is evaluated using the validation set, and the early stopping strategy is used to monitor the validation loss continuously, when the validation set loss no longer decreases, the training is stopped, and the trained time delay prediction model is solidified to the system for calculating the humidity regulation time delay of each unit.

[0107] Preferably, the time delay prediction model also includes an incremental learning mechanism, which uses the newly generated time series data-time delay label to fine-tune the model after the fermentation of each batch of bio-organic fertilizer is completed. The specific operation is to freeze the convolutional layer and only train the fully connected layer. After training, the updated model is saved to replace the original model to adapt to changes in the long-term fermentation environment.

[0108] Optionally, the time delay prediction model also includes an early warning mechanism, which triggers a warning when the humidity regulation time delay output by the time delay prediction model meets any of the following conditions to avoid humidity out of control due to long regulation time delay:

[0109] The predicted humidity regulation time delay is greater than the system preset absolute threshold warning;

[0110] The predicted humidity regulation time delay is more than 50% higher than the historical average humidity regulation time delay;

[0111] In the above cases, there may be equipment failure or system failure, which should be warned in time and manually intervened if necessary.

[0112] As shown in Figure 3 Further, considering the actual fermentation of bio-organic fertilizer in the fermentation tank, local humidity changes do not follow a simple pattern of the basic unit. By dynamically merging the basic unit, the complex spatial distribution and dynamic evolution characteristics of humidity changes in the fermentation process can be effectively captured. Therefore, a dynamic partitioning model is established based on the two-dimensional coordinates of the unit monitoring point, humidity, humidity deviation, and humidity regulation time delay. The improved DBSCAN algorithm is used to realize dynamic partitioning of the fermentation area, which specifically includes:

[0113] The two-dimensional coordinates of the monitoring unit , humidity , humidity deviation , and humidity regulation time delay are input into the DBSCAN clustering algorithm, and the field radius function is adaptively adjusted according to the different fermentation stages, the formula is:

[0114] ,

[0115] wherein is the basic radius set according to historical data, is the different fermentation stages The minimum number of cluster points set, The humidity standard deviation of different fermentation stages reflects the uniformity of humidity distribution. When the humidity distribution is uniform, fewer large partitions are formed, and when the humidity fluctuation is large, more small partitions are formed.

[0116] The Euclidean distance metric in the traditional clustering algorithm is improved, and a weighted distance metric function that integrates spatial and state characteristics is designed , the formula is:

[0117] ,

[0118] where, represents the spatial distance between the unit monitoring point and the unit monitoring point , which is calculated from the position of the unit monitoring point and the position of the unit monitoring point , and the calculation formula is:

[0119] ;

[0120] is the humidity state distance, which is calculated from the humidity and between the two unit monitoring points, the humidity deviation and , and the humidity deviation coefficient , and the calculation formula is:

[0121] ;

[0122] is the humidity control delay distance, and the calculation formula is: , where, are the temperature control delay of the unit monitoring point and the unit monitoring point , respectively, and the sum of the weight coefficients , , is 1. By adjusting the weight, the influence degree of each parameter is adjusted.

[0123] The clustering constraint condition is set as:

[0124] The humidity deviation standard deviation of all monitoring points in the same partition is not greater than the preset standard deviation threshold, and the calculation formula is:

[0125] ,

[0126] in, Indicates the number of monitoring points within the zone. for Average humidity at any given time for Time of the first Humidity within each zone;

[0127] Average humidity control delay in the same zone coefficient of variation Not greater than the preset coefficient threshold;

[0128] The trigger condition is set as follows:

[0129] When the average humidity deviation within the zone If the preset deviation value is exceeded within three consecutive sampling periods, dynamic partitioning update is triggered, and the dynamic partitioning model is restarted.

[0130] After the basic units are dynamically aggregated through the dynamic zoning model, the humidity data, humidity control delay and compensation coefficient of each zone are taken as the average of the corresponding data of all its subordinate units, so as to represent the overall characteristics of the zone and provide basic input for subsequent control.

[0131] Furthermore, establish the coupling influence matrix. This includes the following steps:

[0132] For dynamic partitioning Each partition, calculation area For the region Factors affecting water migration The calculation formula is:

[0133] ,

[0134] in, For the region With the region Standardized center distance, region For the region The factors influencing water migration decrease with increasing center distance. For the region Number of monitoring points included The strength coefficient, This is the distance attenuation coefficient, which is usually taken as a constant; Indicates the region Average humidity control delay For the region The average humidity deviation; The larger the value, the larger the area. For the region The stronger the impact of water migration;

[0135] The moisture migration influence factor is standardized:

[0136] ,

[0137] The value range is [0, 1], and the moisture migration influence factor after standardization is The coupling influence matrix is constructed .

[0138] Further, considering that the universal fuzzy parameters cannot adapt to the differences between different partitions, in order to accurately realize the humidity regulation and control requirements of each partition in subsequent fuzzy PID control, an improved particle swarm optimization algorithm is used for iterative optimization to obtain the optimal fuzzy parameters of each partition; Specifically, according to historical sampling data and humidity regulation and control data, as well as expert experience, a universal fuzzy rule library table is established, and an improved particle swarm optimization algorithm is used to search for the optimal fuzzy parameters of each partition, including the following steps:

[0139] The input of the fuzzy PID controller of each partition is defined as the humidity deviation and the humidity change rate , and the output is the PID parameter correction amount: , , , so as to construct the input variable fuzzy subset and the output variable fuzzy subset, and obtain the universal initial fuzzy rule library table, wherein the fuzzy parameters to be optimized of each partition include the fuzzy subset center value and width of the input humidity deviation , humidity change rate and the fuzzy subset center value and width of the output , , .

[0140] The humidity control accuracy and regulation and control response speed are optimized as double targets to ensure that the optimization result adapts to the characteristics of the actual fermentation delay, and the absolute value integral of the humidity control deviation is calculated according to the humidity data of each partition, and the calculation formula is:

[0141] ,

[0142] wherein represents the upper limit of the humidity control deviation time integral, and the average humidity regulation and control time delay of each partition is combined to obtain the fitness function :

[0143] ,

[0144] wherein, and As a weighting coefficient, when the cumulative humidity deviation is small and the average humidity control delay is short, the adaptability is smaller, corresponding to a better fuzzy parameter;

[0145] Initialize the particle swarm by generating the initial positions of PSO particles using uniform random sampling and boundary constraints, setting the initial velocities, and adjusting the time delay based on the partitioned average humidity. Dynamically adjust inertia weight :

[0146] ,

[0147] in, and These represent the maximum inertia weight and the minimum inertia weight, respectively. and The maximum and minimum values ​​of humidity control delay in historical iterations will be used to determine the values. Set as a delay constraint;

[0148] Combining dynamic inertia weights and delay constraints to update particle velocity and adjust particle position, the updated particle velocity is:

[0149] ,

[0150] in, This represents the velocity of the particle at time t. This indicates the position of the particle at time t. and These represent individual learning factors and group learning factors, respectively. The value is a random number between [0,1] to avoid the algorithm getting trapped in local optima. For particles The optimal position of itself at any given moment. For population The global optimal position at time t, if the predicted regulation delay value corresponding to the particle is greater than the delay constraint. Then, the speed will decrease.

[0151] The formula for updating the particle position is:

[0152] ,

[0153] When the particle position change rate for five consecutive generations Or, if the number of iterations reaches a preset threshold, terminate a single PSO iteration;

[0154] Initialize the GA population, integrate the best particles after PSO iteration into the initial GA population, use roulette wheel selection to calculate the fitness of each GA individual, allocate selection probabilities according to fitness, select several high-quality individuals as parents, and retain high-fitness genes.

[0155] Randomly pair up parent generations, select intersection points to exchange parameter fragments to generate offspring;

[0156] Randomly perturb individual parameters and set low mutation probabilities to avoid destroying high-quality individuals;

[0157] The mutated individuals are merged with the remaining particles to form a new population, thus completing one GA operation;

[0158] Repeat the PSO-GA iteration until the change in the global optimal fitness is less than the preset change threshold or the iteration count threshold is reached after 5 consecutive generations. Output the parameters of the corresponding particles as the optimal fuzzy parameters for the partition.

[0159] like Figure 4 As shown, further, a fuzzy PID control algorithm is used to obtain the control quantity for each partition. Specifically, this includes using the Ziegler-Nichols time-domain method to tune the PID parameter reference values, and setting the tuning results as the PID parameter reference values ​​for each partition. , , Enter the humidity deviation at the current time. and rate of change of deviation The PID parameters are corrected using a three-step correction method. The specific steps are as follows:

[0160] Based on the time delay of humidity control in different zones Correct the baseline values ​​of the PID parameters, and then use linear correction to adjust the PID parameters to obtain the corrected control parameters. , , ;

[0161] Based on compensation coefficient Linear correction is applied to the PID control parameters to obtain the partition. Corrected control parameters , , ;

[0162] Based on the coupling influence matrix The PID control parameters are then subjected to a final linear correction to obtain the partition. Final control parameters , , To achieve differentiated control over different zones, the modified formula is as follows:

[0163] ,

[0164] wherein, denotes a set of sub-zones adjacent to sub-zone , denotes the humidity bias of sub-zone j, is a sign function for determining , is a constant, and are correction coefficients, which fully consider the moisture migration effect between adjacent sub-zones based on the coupling influence matrix, and when the adjacent sub-zone j has high water supplement demand and strong coupling influence on sub-zone i, the control parameter of sub-zone i is enhanced;

[0165] The PID parameters are fine-tuned in real time using fuzzy reasoning, specifically, the inputs and are fuzzified, the output PID parameter correction amount , , is calculated through the optimal fuzzy parameters of each sub-zone, and after the fuzzy reasoning is solved by the sampling barycentric method, the final parameters are calibrated for the PID parameters, the humidity adjustment intensity is calculated based on the final parameters, and the control amount is output. The calculation formula of the control amount is:

[0166] ,

[0167] wherein, is a proportional coefficient, is an integral coefficient, is a differential coefficient, when is a positive value, it represents the water supplement intensity, and when is a negative value, it represents the humidity reduction intensity;

[0168] The control amount is output to the programmable logic controller (PLC) through the fuzzy PID controller, wherein the control amount is essentially the intensity of humidity adjustment, when the control amount is a positive value, it represents the water supplement intensity, and when the control amount is a negative value, it represents the humidity reduction intensity; the PLC drives the actuator to perform corresponding actions according to the preset program, for example, for the low-humidity water-deficient area, the PLC controls the spraying device / drip irrigation device to supplement water; for the high-humidity overflow area, the PLC controls the ventilator / aeration system to ventilate and accelerate air circulation to reduce humidity.

[0169] Preferably, a closed-loop feedback mechanism is further included to adaptively adjust the PID control, and after one-time adjustment, the deviation of the adjusted actual humidity data from the target humidity is compared to correct the fuzzy PID control parameters, so as to avoid long-term deviation, and specifically, the adjusted humidity data is continuously sampled, the regional average humidity is calculated, and then the steady-state deviation is calculated, if the steady-state deviation exceeds / is lower than the preset deviation range, the fuzzy PID parameter self-correction is triggered, the corrected PID control parameters are obtained, and are used for the next round of humidity adjustment.

[0170] Optionally, an abnormal state alarm mechanism is further included, when the partition humidity state exceeds the adjustable range, an alarm should be sent in time and emergency equipment should be triggered, when the real-time monitoring humidity of the high-humidity overflow area exceeds 50% of the target humidity range and the duration exceeds 10 minutes, or the real-time monitoring humidity of the low-humidity water shortage area is lower than 50% of the target humidity range and the duration exceeds 10 minutes, an alarm is immediately started and emergency equipment is triggered.

[0171] The application further provides an intelligent humidity control system for bio-organic fertilizer fermentation, comprising an acquisition module, a dynamic partition module, a control optimization module and a regulation and control module.

[0172] The acquisition module periodically acquires the environmental parameters and material characteristic parameters of each unit in the stacking area, and calculates the humidity regulation time delay and compensation coefficient of each unit.

[0173] The dynamic partition module dynamically merges units by using a clustering algorithm to obtain K partitions, and determines the corresponding humidity data, humidity regulation time delay and compensation coefficient according to the subordinate units in each partition, and constructs a coupling influence matrix by considering the moisture migration between the partitions.

[0174] The control optimization module initializes the fuzzy parameters of each partition, adopts an improved particle swarm algorithm, adaptively adjusts the inertia weight according to the humidity regulation time delay of each partition, introduces the genetic variation idea to iteratively optimize the fuzzy parameters of each partition, and obtains the optimal fuzzy parameters of each partition.

[0175] The regulation and control module corrects the preset PID parameters, adopts a fuzzy PID control algorithm to obtain the control amount of each partition according to the deviation and deviation change rate of the humidity data of each partition from the target humidity.

[0176] The above only describes the preferred embodiments of the application, and the protection scope of the application is not limited to the above-described embodiments, and any technical solution falling within the idea of the application falls within the protection scope of the application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the application are also considered to fall within the protection scope of the application.​

Claims

1. A smart humidity control method for bio-organic fertilizer fermentation, characterized in that, Includes the following steps: The stacking area is divided into Each unit periodically acquires environmental and material characteristic parameters, and calculates the corresponding humidity control delay and compensation coefficient. The total number of units; Based on humidity data from the environmental parameters of each unit, combined with the target humidity and humidity control delay, a clustering algorithm is used to dynamically merge units to obtain... The system is divided into several partitions, and based on the subordinate units within each partition, the corresponding humidity data, humidity control delay, and compensation coefficient are determined. A coupling effect matrix is ​​constructed considering moisture migration between partitions. Total number of partitions; The fuzzy parameters of each partition are initialized. An improved particle swarm optimization algorithm is used to adaptively adjust the inertia weight based on the humidity control delay of each partition. The concept of genetic mutation is introduced to iteratively optimize the fuzzy parameters of each partition to obtain the optimal fuzzy parameters of each partition. The preset PID parameters are corrected based on the coupling influence matrix, the humidity control delay and compensation coefficient of each zone, and the optimal PID parameters of each zone are obtained. Based on the deviation and rate of change of the humidity data of each zone from the target humidity, fuzzy PID control is used to obtain the control quantity of each zone.

2. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 1, characterized in that, The periodic acquisition of environmental and material characteristic parameters for each unit includes: environmental parameters such as humidity, temperature, and oxygen concentration; and material characteristic parameters such as moisture content and pH value. The environmental and material characteristic parameters are then processed using a 3D model. The criteria and Kalman filtering algorithm are used for preprocessing, and the environmental parameters and material property parameters at each sampling time are aligned by timestamps.

3. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 1, characterized in that, Calculate the compensation coefficient for the corresponding unit, including the following steps: Based on different fermentation stages, the current target humidity range is obtained; Calculate the degree of deviation between the current humidity data of the corresponding unit and the target humidity range; Microbial activity was calculated based on the temperature change curve and oxygen concentration change curve of the corresponding unit. The compensation coefficient is obtained based on the degree of deviation and microbial activity.

4. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 1, characterized in that, Calculating the humidity control delay for the corresponding unit includes the following steps: A hybrid network architecture based on CNN-LSTM was constructed, using historical sampling data and temperature, humidity, oxygen concentration, material moisture content, and pH value as multi-channel inputs. 1D convolution is used to extract local dynamic features of each input, and then LSTM temporal dependency layer is used to capture the dependency relationship of the feature sequence. Attention weights are assigned to each time step of the LSTM output, and the weighted sum is used to obtain the final feature vector. After model training, a time delay prediction model is obtained, and the output humidity control time delay is calculated.

5. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 4, characterized in that, The latency prediction model also includes an incremental learning mechanism and an early warning mechanism: After each batch of bio-organic fertilizer fermentation is completed, the time delay prediction model is incrementally fine-tuned using the newly generated time series data - time delay labels. Specifically, the convolutional layers are frozen and only the fully connected layers are trained. When the humidity control delay predicted by the time delay prediction model exceeds the preset threshold, an early warning is triggered.

6. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 1, characterized in that, Dynamically merge units using clustering algorithms to obtain The system is divided into several partitions, and based on the subordinate units within each partition, the corresponding humidity data, humidity control delay, and compensation coefficient are determined. A coupling influence matrix is ​​constructed considering moisture migration between partitions, including the following steps: A dynamic partitioning model is established based on the two-dimensional coordinates, humidity, humidity deviation and humidity control delay of each unit, and the improved DBSCAN algorithm is used to realize dynamic merging of units. The result obtained after dynamic merging units For each partition, the water migration influence factor of region i on region j is calculated using the standardized center distance between region i and region j and the number of units contained in region i. The water migration influence factor is then standardized to obtain the coupling influence matrix.

7. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 6, characterized in that, The establishment of the dynamic partitioning model includes the following steps: The two-dimensional coordinates of the unit monitoring points, humidity, humidity deviation, and humidity control delay are input into the DBSCAN clustering algorithm, with the following constraints set: The standard deviation of humidity at all monitoring points within the same zone shall not exceed the preset standard deviation threshold. The coefficient of variation of the average humidity control delay in the same zone is not greater than the preset coefficient threshold. Among them, the domain radius function is adaptively adjusted according to different fermentation stages, and a weighted distance metric function is designed based on spatial distance, humidity state distance, and humidity regulation delay distance, with the weight coefficients summed to 1; The trigger condition is set as follows: when the average humidity deviation within a partition exceeds the preset deviation value for three consecutive sampling periods, a dynamic partition update is triggered, and the dynamic partition model is restarted.

8. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 1, characterized in that, To obtain the optimal fuzzy parameters for each partition, the following steps are included: Establish a general fuzzy rule base and initialize the fuzzy parameters for each partition; The particle swarm is initialized and optimized with humidity control accuracy and regulation response speed as dual objectives. The integral of the absolute value of humidity control deviation is calculated based on the humidity data of each zone, and the fitness function is obtained by combining the average humidity regulation delay of each zone. The inertial weight is dynamically adjusted based on the average humidity regulation time delay to update particle velocity and adjust particle position. PSO-optimized particles are integrated into the initial GA population, and selection, crossover, and mutation operations are performed. The mutated individuals are then merged with the remaining particles to form a new population. Repeat the PSO-GA iteration until the termination condition is met, and output the parameters of the corresponding particles as the optimal fuzzy parameters for the partition.

9. The intelligent humidity control method for bio-organic fertilizer fermentation as described in claim 1, characterized in that, Obtaining the control values ​​for each partition involves the following steps: The baseline values ​​of PID parameters were obtained using the Ziegler-Nichols time-domain method. Based on the coupling effect matrix, and combined with the humidity control delay and compensation coefficient of each zone, the baseline value of the PID parameters is linearly corrected to obtain the optimal PID parameters for each zone. Input the humidity deviation and the rate of change of deviation at the current moment, calculate the PID parameter correction amount through the optimal fuzzy parameters of each zone, calibrate the PID parameters to obtain the final parameters, calculate the humidity regulation intensity based on the final parameters, and output the control quantity; When the control value is positive, it represents the water replenishment intensity; when the control value is negative, it represents the dehumidification intensity.

10. An intelligent humidity control system for bio-organic fertilizer fermentation, used to implement the intelligent humidity control method for bio-organic fertilizer fermentation as described in any one of claims 1-9, characterized in that, It includes a data acquisition module, a dynamic partitioning module, a control optimization module, and a regulation module; The acquisition module periodically acquires environmental parameters and material characteristic parameters of each unit in the stacking area, and calculates the humidity control delay and compensation coefficient of each unit; The dynamic partitioning module uses a clustering algorithm to dynamically merge units to obtain... The system is divided into several partitions, and the corresponding humidity data, humidity control delay and compensation coefficient are determined based on the subordinate units in each partition. A coupling influence matrix is ​​constructed considering the moisture migration between partitions. The control optimization module initializes the fuzzy parameters of each partition, adopts an improved particle swarm optimization algorithm, adaptively adjusts the inertia weight based on the humidity control delay of each partition, and introduces the idea of ​​genetic mutation to iteratively optimize the fuzzy parameters of each partition to obtain the optimal fuzzy parameters of each partition. The control module corrects the preset PID parameters and uses a fuzzy PID control algorithm to obtain the control quantity for each zone based on the deviation and rate of change of the humidity data of each zone from the target humidity.

Images