Equipment control method based on sludge anaerobic fermentation product analysis

Abstract

The invention provides an equipment control method based on sludge anaerobic fermentation product analysis, and relates to the technical field of sludge treatment.The method comprises the steps that initial feature information of current sludge before fermentation is obtained, and the initial feature information comprises the moisture content, volatile solids and / or the carbon-nitrogen ratio; acquiring the actual content of a fermentation product at a first time node according to preset key time nodes in the anaerobic fermentation process; inputting the sludge initial feature information into a preset machine learning model to obtain the predicted content of the fermentation product corresponding to the first time node; calculating a content difference value between the actual content and the predicted content of the product, and when the content difference value exceeds a preset threshold value, inputting the content difference value into a preset parameter adjustment model to obtain a corresponding parameter adjustment quantity; and adjusting the current fermentation parameters according to the parameter adjusting quantity. The problem that in the prior art, the fermentation state is difficult to monitor according to product information of sludge with different characteristics in the fermentation process, and consequently the parameter control precision of fermentation equipment is low is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sludge treatment, in particular to a device control method based on sludge anaerobic fermentation product analysis. BACKGROUND

[0002] In the field of sludge anaerobic fermentation treatment, converting organic matter in sludge into clean energy such as methane through microbial metabolism is an important technical path to realize sludge reduction and resource utilization. However, the sources of sludge are diverse (such as municipal wastewater treatment plants, industrial wastewater treatment stations, etc.), and the initial characteristics vary significantly. Fluctuations in key indicators such as moisture content, volatile solids content, and carbon-nitrogen ratio will directly affect the activity and metabolic pathways of fermentation microorganisms. In the prior art, the monitoring of the fermentation process relies mainly on manual sampling and detection at fixed time points or static monitoring of a single parameter, making it difficult to capture the product variation rules of sludge with different characteristics at each stage of fermentation in real time, resulting in a lag in determining whether the fermentation system is in a stable and efficient operating state.

[0003] Traditional sludge anaerobic fermentation device control methods are mostly based on experience and preset fixed parameters, or only use simple feedback adjustment based on a single environmental indicator (such as temperature and pH value), lacking in-depth analysis of the dynamic correlation between sludge initial characteristics and fermentation products. When the characteristics of sludge change, the fixed parameter control mode cannot adapt to the metabolic needs of microorganisms in a timely manner, which may lead to problems such as accumulation of volatile fatty acids, reduction of methane yield, and excessive hydrogen sulfide, affecting fermentation efficiency and product quality, and even causing unstable operation of the equipment and system collapse. Therefore, how to achieve precise parameter regulation based on sludge characteristics and product dynamic monitoring has become a key technical problem to improve the control accuracy and operation reliability of sludge anaerobic fermentation equipment. SUMMARY

[0004] Embodiments of the present application provide a device control method based on sludge anaerobic fermentation product analysis, aiming to solve the problem that the prior art cannot monitor the state of fermentation according to product information of sludge with different characteristics during the fermentation process, resulting in low control accuracy of fermentation equipment parameters.

[0005] To achieve the above-mentioned purpose, the present application provides a device control method based on sludge anaerobic fermentation product analysis, comprising the following steps: obtaining initial characteristic information of the current sludge before fermentation; the initial characteristics include moisture content, volatile solids, and / or carbon-nitrogen ratio; obtaining the actual content of the fermentation product at the first time node according to the key time nodes preset in the anaerobic fermentation process; inputting the initial characteristic information of the sludge into a preset machine learning model to obtain the predicted content of the fermentation product at the first time node; calculating the content difference between the actual content and the predicted content; When the content difference value exceeds a preset threshold value, input the content difference value into a preset parameter adjustment model to obtain a corresponding parameter adjustment amount; Adjust the current device parameter according to the parameter adjustment amount.

[0006] Further, the fermentation product includes methane, carbon dioxide, hydrogen sulfide, volatile fatty acid and / or ammonia nitrogen.

[0007] Further, the device parameter includes fermentation temperature, pH value, oxidation-reduction potential, substrate feed and / or stirring rate.

[0008] Further, the preset key time node includes a characteristic fermentation stage time point preset according to the sludge type, and the sludge type includes municipal sludge and / or industrial sludge.

[0009] Further, the preset machine learning model is configured as an MLP model, including the following steps: Collect initial characteristic information of different batches, and record actual content data of fermentation products of each batch of sludge at corresponding preset key time nodes in the anaerobic fermentation process; Divide the data set into a training set and a test set, the training set is used for model training, and the test set is used for evaluating the final performance of the model; Build an MLP model including an input layer, multiple hidden layers and an output layer, the input layer of the MLP model is the initial characteristic information of the sludge, and the output layer is the actual content of the corresponding fermentation product; Train the built MLP model using the training set data; Measure the difference between the predicted value and the actual value of the test set by the loss function, and continuously adjust the weight and bias of the model by the back propagation algorithm; Stop training when the loss function on the test set reaches the preset condition.

[0010] Further, the preset threshold value is set according to the type of fermentation product, wherein the methane content difference threshold value is ±5% to ±10%, the carbon dioxide content difference threshold value is ±8% to ±15%, the volatile fatty acid content difference threshold value is ±100 mg / L to ±300 mg / L, the hydrogen sulfide content difference threshold value is ±10 ppm to ±30 ppm, and the ammonia nitrogen content difference threshold value is ±50 mg / L to ±150 mg / L.

[0011] Further, the parameter adjustment model is a random forest model trained by historical period parameter adjustment experience data, wherein the random forest model takes the key time node and the content difference value as the independent variable, and the corresponding parameter adjustment amount as the dependent variable.

[0012] Further, when the content difference value does not exceed the preset threshold value, the current device parameter is kept unchanged, and the actual content of the fermentation product at the key time node is continuously acquired in time sequence.

[0013] The above technical solution has the following technical effects: By acquiring initial characteristic information of the current sludge before fermentation, including water content, volatile solid and / or carbon-nitrogen ratio, the actual content of the fermentation product at the first time node in the preset key time node in the anaerobic fermentation process is acquired, the initial characteristic information of the sludge is input into the preset machine learning model to obtain the predicted content of the fermentation product at the first time node, the content difference value between the actual content and the predicted content is calculated, when the content difference value exceeds the preset threshold value, the content difference value is input into the preset parameter adjustment model to obtain the corresponding parameter adjustment amount, and the current fermentation parameter is adjusted according to the parameter adjustment amount. The present application solves the problem that the prior art is difficult to monitor the fermentation state according to the product information of sludge with different characteristics in the fermentation process, resulting in low control precision of the fermentation device parameter. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A flowchart of a device control method based on sludge anaerobic fermentation product analysis according to an embodiment of the present application is shown in FIG. 1. DETAILED DESCRIPTION

[0015] To further illustrate the embodiments, the present application provides drawings. These drawings are part of the disclosure of the present application, mainly used to illustrate the embodiments, and can be used to explain the operating principle of the embodiments in conjunction with the related description of the specification. Those skilled in the art should understand other possible embodiments and advantages of the present application by referring to these contents. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0016] The present application will be further described in conjunction with the drawings and specific embodiments.

[0017] Figure 1 A flowchart of a device control method based on sludge anaerobic fermentation product analysis according to an embodiment of the present application is shown in FIG. 1. Figure 1 As shown in FIG. 1, the method of this embodiment includes the following steps: Acquiring initial characteristic information of the current sludge before fermentation, including water content, volatile solid and / or carbon-nitrogen ratio; In this embodiment, the core monitoring indicators for obtaining initial characteristic information of the sludge before fermentation include moisture content, volatile solids, and carbon-to-nitrogen ratio. Among them, moisture content directly reflects the ratio of water to solid matter in the sludge. Too high a moisture content will dilute the concentration of organic matter, reduce the efficiency of microbial metabolism, and increase equipment energy consumption; too low a moisture content may lead to poor substrate flowability, affecting mass transfer and mixing uniformity. Therefore, it is necessary to quickly measure the moisture content using a moisture sensor to provide data support for subsequent substrate pretreatment (such as dehydration adjustment).

[0018] Volatile solids (VFS) represent the total amount of organic matter in sludge that can be decomposed by microorganisms. It is a key indicator for assessing fermentation potential, and its content directly determines the theoretical yield of target products such as methane. Analysis using a VFS analyzer can clarify the organic matter richness of the sludge, helping to determine whether additional carbon sources are needed or the feed rate adjusted, thus avoiding low fermentation efficiency due to insufficient substrate components.

[0019] The carbon-to-nitrogen ratio (C / N) is a core parameter affecting the balance of the microbial community. Microorganisms in anaerobic fermentation require a specific ratio of carbon and nitrogen; an imbalance in the C / N ratio significantly inhibits the activity of methanogens. Insufficient carbon leads to limited microbial growth, while excess nitrogen can cause ammonia accumulation and abnormal pH levels in the system. Accurate measurement using a C / N ratio analyzer allows for early prediction of the nutrient balance during fermentation, providing a basis for substrate formulation.

[0020] The actual content of fermentation products at the first time point is obtained based on the preset key time points in the anaerobic fermentation process; in one specific implementation, the fermentation products include methane, carbon dioxide, hydrogen sulfide, volatile fatty acids and / or ammonia nitrogen.

[0021] In this embodiment, obtaining the actual content of fermentation products at preset key time nodes is the core link in the dynamic monitoring system's operation. Essentially, it involves understanding the metabolic patterns of microorganisms by tracking product changes. The setting of key time nodes needs to be combined with the characteristics of the fermentation cycle, typically covering the initial start-up phase (e.g., 2h, 6h), the stable phase (e.g., 12h, 24h), and the continuous fermentation stage (every 24h thereafter), or characteristic stages such as hydrolysis, acidification, and methanogenesis based on sludge type, ensuring the capture of key metabolic turning points. In one specific implementation, the preset key time nodes include characteristic fermentation stage time points preset according to sludge type, where sludge type includes municipal sludge and / or industrial sludge.

[0022] Among the monitored fermentation products, methane is the core target product, and its content directly reflects the activity of methanogens and fermentation efficiency. Carbon dioxide can help determine the degree of system acidification, and changes in the ratio of the two can quickly identify the risk of metabolic imbalance. Hydrogen sulfide is a typical inhibitory gas; excessive levels can poison microorganisms and corrode equipment, requiring real-time monitoring to ensure operational safety. Volatile fatty acids are precursors in the methanogenesis stage; excessive accumulation may trigger system acidification and collapse. Ammonia nitrogen is related to nitrogen source metabolism, and abnormal concentrations can disrupt pH stability.

[0023] The actual content is obtained in real time through online devices such as gas chromatograph and volatile fatty acid detector, and the data is synchronously transmitted to the control system to realize the visual tracking of product changes.

[0024] The initial characteristic information of the sludge is input into a preset machine learning model to obtain the predicted content of fermentation products at the corresponding first time point; In one specific implementation, the machine learning model is pre-configured as an MLP model, including the following steps: Collect initial characteristic information of different batches, and record the actual content data of fermentation products of each batch of sludge at the corresponding preset key time nodes during the anaerobic fermentation process; The dataset is divided into a training set and a test set. The training set is used for model training, and the test set is used to evaluate the final performance of the model. An MLP model is constructed, which includes an input layer, multiple hidden layers, and an output layer. The input layer of the MLP model is the initial feature information of the sludge, and the output layer is the actual content of the corresponding fermentation products. The built MLP model is trained using the training set data; The loss function measures the difference between the model's predictions and actual values ​​on the test set, and the backpropagation algorithm continuously adjusts the model's weights and biases. Training stops when the loss function on the test set reaches the preset condition.

[0025] In this embodiment, initial feature information is input into a preset machine learning model to obtain the predicted content of products. Essentially, this involves using data-driven modeling to uncover the intrinsic relationship between sludge characteristics and product changes. Taking the MLP (Multilayer Perceptron) model as an example, this process requires first constructing a high-quality dataset: collecting initial features such as moisture content, volatile solids, and carbon-to-nitrogen ratio of multiple batches of sludge, and simultaneously recording the actual content of products such as methane, carbon dioxide, and volatile fatty acids in each batch at key time points (such as 2h, 6h, and 24h after startup), forming a sample library covering different sludge types and fermentation stages.

[0026] After the dataset was proportionally divided into training and testing sets, an MLP network structure was constructed, comprising an input layer, hidden layers, and an output layer. Input layer nodes corresponded to the initial feature dimensions (e.g., 3 nodes for 3 core features), output layer nodes matched the predicted product types (e.g., 5 nodes for 5 products), and hidden layers implemented non-linear feature mapping through 2-3 layers of neurons. The number of layers and nodes was determined through experimental optimization. During the training phase, the training set data drove model learning. Loss functions such as mean squared error were used to quantify the deviation between predicted and actual values. Optimizers such as Adam, combined with backpropagation algorithms, iteratively adjusted the weights and bias parameters of each layer to continuously reduce the loss. Training stopped when the test set loss reached a preset threshold (e.g., it stabilized and generalization ability met the standard). The final model can accurately output the predicted product content of each key node based on the initial characteristics of new sludge, providing a scientific basis for subsequent deviation analysis and control decisions.

[0027] Calculate the content difference between the actual content and the predicted content; When the content difference exceeds the preset threshold, the content difference is input into the preset parameter adjustment model to obtain the corresponding parameter adjustment amount; In this embodiment, calculating the difference between the actual and predicted content of fermentation products is a crucial quantitative step in determining whether the anaerobic fermentation process deviates from the expected trajectory. Its core function is to identify system anomalies through data comparison. This difference directly reflects the degree of deviation between the actual fermentation state and the model prediction: if the difference is small and within a reasonable range, it indicates that the current sludge characteristics, microbial activity, and parameter settings are matched, and the system is operating stably; if the difference exceeds a preset threshold, it indicates metabolic imbalance, insufficient parameter adaptation, or interference factors, requiring the activation of a control mechanism.

[0028] When the content difference exceeds the standard, inputting it into a preset parameter adjustment model can transform the process from "anomaly identification" to "precise control." The parameter adjustment model has a built-in association rule base constructed based on microbial metabolic mechanisms and historical control data. It can quickly pinpoint the root cause of the problem based on the type of difference (e.g., low methane, high volatile fatty acids), its magnitude, and the characteristics of the corresponding products. For example, if the methane content is lower than the predicted value and the difference exceeds the limit, it may be related to insufficient temperature or pH deviation from the optimal range; excessive accumulation of volatile fatty acids may stem from insufficient stirring or excessively rapid substrate feeding. By matching the associations in the rule base, the model calculates targeted parameter adjustments, such as the temperature increase, the amount of pH adjusting agent added, or the stirring rate adjustment value. This provides specific operational guidelines for subsequent equipment parameter optimization, ensuring that the fermentation system promptly returns to a stable and efficient operating state.

[0029] In one specific implementation, the preset thresholds are set according to the type of fermentation product, wherein the threshold for difference in methane content is ±5%~±10%, the threshold for difference in carbon dioxide content is ±8%~±15%, the threshold for difference in volatile fatty acid content is ±100mg / L~±300mg / L, the threshold for difference in hydrogen sulfide content is ±10ppm~±30ppm, and the threshold for difference in ammonia nitrogen content is ±50mg / L~±150mg / L.

[0030] Adjust the current equipment parameters according to the parameter adjustment amount.

[0031] In one specific implementation, the equipment parameters include fermentation temperature, pH value, redox potential, substrate feed and / or stirring rate.

[0032] In this embodiment, adjusting the current equipment parameters according to the parameter adjustment amount is a key execution step in translating control decisions into actual operation. By precisely intervening in the equipment's operating status, dynamic optimization of the anaerobic fermentation process is achieved. The adjustment of equipment parameters must be closely combined with the specific adjustment amount output by the parameter adjustment model, targeting different product deviation problems.

[0033] For example, if the methane yield is lower than the predicted value and the difference exceeds the limit, and the model determines that it is related to insufficient temperature, the fermentation temperature can be adjusted from the current mesophilic range (30~38℃) to the optimal range using heating equipment, such as increasing it by 2~3℃ to enhance the activity of methanogens. If the accumulation of volatile fatty acids is caused by the pH value deviating from the suitable range of 6.5~8.0, the pH adjustment equipment can precisely add acid and alkali agents according to the adjustment amount to quickly restore the pH value to the stable range. For abnormal oxidation-reduction potential (ORP), the anaerobic environment of the system can be improved by adjusting the stirring rate or gas circulation flow rate of the stirring equipment. If the substrate feed rate is not matched with the sludge degradation capacity, the feed rate or frequency can be adjusted through the feeding equipment to avoid excessive load inhibiting microbial metabolism.

[0034] This process enables real-time adaptation of equipment parameters to the metabolic needs of microorganisms. By finely adjusting core parameters such as temperature, pH, and stirring rate, fermentation deviations are corrected in a timely manner, ensuring that the system operates in optimal condition, and ultimately improving the yield of the target product and the stability of equipment operation.

[0035] In one specific implementation, the parameter adjustment model is a random forest model trained with historical parameter adjustment experience data, where the random forest model uses key time nodes and content differences as independent variables, and the corresponding parameter adjustment amount as the dependent variable.

[0036] In the specific implementation of the parameter regulation model, a random forest model trained with historical parameter regulation experience data is adopted, which can fully leverage its ability to fit nonlinear relationships and its decision reliability. The model uses key time nodes and product content differences as core independent variables: key time nodes reflect the fermentation stage (such as the start-up period and the stationary period), and the differences in microbial metabolic characteristics at different stages determine the focus of regulation; the content difference quantifies the degree and direction of deviation between the actual product and the predicted value (such as low methane and high hydrogen sulfide), providing a direct basis for the direction of regulation.

[0037] During the model training phase, a large amount of historical control data needs to be collected, including the content differences of each batch of sludge at different key nodes, the corresponding output parameter adjustment amounts (such as temperature adjustment values ​​and pH adjuster addition amounts), and the feedback on the effects of the adjustments. By inputting this data into a random forest model, multiple decision trees are constructed for parallel learning. Each tree is trained independently based on a subset of samples and a subset of features, and finally, the optimal parameter adjustment amount is output through voting or averaging. This structure enables the model to effectively capture the complex relationship between content differences and equipment parameter adjustments. For example, if the volatile fatty acid difference exceeds the upper limit 6 hours after fermentation starts, the model can combine historical data to prioritize the output of a combined adjustment amount of increasing the stirring rate and fine-tuning the pH value.

[0038] Compared to traditional rule-based models, random forest models have stronger generalization capabilities, can adapt to changes in different sludge characteristics and fermentation scenarios, and can be dynamically optimized by continuously incorporating new control data to ensure that the output parameter adjustments are more accurate and better suited to actual fermentation needs, providing scientific decision support for the intelligent adjustment of equipment parameters.

[0039] In one specific implementation, when the content difference does not exceed a preset threshold, the current equipment parameters remain unchanged, and the actual content of fermentation products at key time nodes is obtained in chronological order.

[0040] In this embodiment, when the deviation between the actual product content and the predicted value is within a reasonable range, it indicates that the initial characteristics of the sludge, the metabolic state of the microorganisms, and the equipment parameter settings are well matched. No additional intervention is required to ensure that the system operates according to the expected trajectory and avoid new fluctuations caused by over-adjustment.

[0041] Continuous monitoring of product content at key time points allows for dynamic tracking of the fermentation process. Even when the current state is stable, as fermentation progresses (e.g., from the acidification stage to the methanogenesis stage), the metabolic demands of microorganisms may change, and sludge characteristics may undergo subtle adjustments due to the degradation process. By periodically collecting data on products such as methane and volatile fatty acids at preset key time points (e.g., 24h, 48h), potential deviation trends can be captured in a timely manner, providing continuous data support for subsequent regulation. This "monitoring when stable, intervening when abnormal" model ensures the stability of system operation and accumulates more samples through continuous data collection, providing a foundation for the optimization and iteration of machine learning models and parameter adjustment models, forming a closed-loop management mechanism of "monitoring-judgment-regulation-re-monitoring".

[0042] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A device control method based on the analysis of anaerobic fermentation products of sludge, characterized in that, Includes the following steps: Obtain the initial characteristic information of the current sludge before fermentation; the initial characteristics include moisture content, volatile solids and / or carbon-nitrogen ratio; The actual content of fermentation products at the first time point is obtained from the key time points preset in the anaerobic fermentation process. The initial characteristic information of the sludge is input into a preset machine learning model to obtain the predicted content of fermentation products at the corresponding first time point; Calculate the content difference between the actual content and the predicted content; When the content difference exceeds a preset threshold, the content difference is input into a preset parameter adjustment model to obtain the corresponding parameter adjustment amount; Adjust the current equipment parameters according to the parameter adjustment amount.

2. The equipment control method based on anaerobic fermentation product analysis of sludge according to claim 1, characterized in that, The fermentation products include methane, carbon dioxide, hydrogen sulfide, volatile fatty acids, and / or ammonia nitrogen.

3. The equipment control method based on anaerobic fermentation product analysis of sludge according to claim 1, characterized in that, The equipment parameters include fermentation temperature, pH value, redox potential, substrate feed and / or stirring rate.

4. The equipment control method based on anaerobic fermentation product analysis of sludge according to claim 1, characterized in that, The preset key time points include the characteristic fermentation stage time points preset according to the sludge type, which includes municipal sludge and / or industrial sludge.

5. The equipment control method based on anaerobic fermentation product analysis of sludge according to claim 1, characterized in that, The preset machine learning model is configured as an MLP model, including the following steps: Collect initial characteristic information of different batches, and record the actual content data of fermentation products of each batch of sludge at the corresponding preset key time nodes during the anaerobic fermentation process; The dataset is divided into a training set and a test set. The training set is used for model training, and the test set is used to evaluate the final performance of the model. An MLP model is constructed, which includes an input layer, multiple hidden layers, and an output layer. The input layer of the MLP model is the initial feature information of the sludge, and the output layer is the actual content of the corresponding fermentation products. The built MLP model is trained using the training set data; The loss function measures the difference between the model's predictions and actual values ​​on the test set, and the backpropagation algorithm continuously adjusts the model's weights and biases. Training stops when the loss function on the test set reaches the preset condition.

6. The equipment control method based on anaerobic fermentation product analysis of sludge according to claim 1, characterized in that, The preset thresholds are set according to the type of fermentation product, wherein the threshold for difference in methane content is ±5% to ±10%, the threshold for difference in carbon dioxide content is ±8% to ±15%, the threshold for difference in volatile fatty acid content is ±100mg / L to ±300mg / L, the threshold for difference in hydrogen sulfide content is ±10ppm to ±30ppm, and the threshold for difference in ammonia nitrogen content is ±50mg / L to ±150mg / L.

7. The equipment control method based on anaerobic fermentation product analysis of sludge according to claim 1, characterized in that, The parameter adjustment model is a random forest model trained using historical parameter adjustment experience data. The random forest model uses key time nodes and content differences as independent variables, and the corresponding parameter adjustment amounts as dependent variables.

8. The equipment control method based on anaerobic fermentation product analysis of sludge according to claim 1, characterized in that, When the content difference does not exceed the preset threshold, the current equipment parameters remain unchanged, and the actual content of fermentation products at key time nodes is obtained in chronological order.

Images