A neural network-based sludge bulking prediction method and system
By using a neural network-based sludge bulking prediction method, which trains the neural network with activated sludge images and videos, the problem of predicting sludge bulking from process parameters and water quality indicators in existing technologies is solved, and accurate and timely sludge bulking prediction and prevention are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NINGBO POLYTECHNIC
- Filing Date
- 2021-12-17
- Publication Date
- 2026-07-14
AI Technical Summary
Existing technologies struggle to predict sludge bulking from limited process parameters and water quality indicators, and human observation requires operators with sufficient experience.
A neural network-based approach was adopted, which trained the neural network using images and videos of activated sludge. The artificial neural network, consisting of a sequence input layer, a sequence folding layer, a convolutional layer, a sequence unfolding layer, a flattening layer, and a long short-term memory layer, was used to predict the probability of sludge bulking, thus replacing human observation.
It enables timely prediction of sludge bulking, avoids the inaccuracy and lag of human observation, improves the accuracy and timeliness of prediction, and prevents sludge bulking from occurring.
Smart Images
Figure CN114444766B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of sludge bulking in wastewater treatment processes, and more particularly to a sludge bulking prediction method and system based on neural networks. Background Technology
[0002] Activated sludge bulking is a common abnormal phenomenon in activated sludge processes. It refers to the deterioration of the settling performance of activated sludge due to changes in certain factors, resulting in sludge loss with the effluent from the settling tank. Once sludge bulking occurs, it can lead to effluent exceeding discharge standards. If no control measures are taken, continued sludge loss will drastically reduce the microbial biomass in the aeration tank, failing to meet the needs of oxidizing and decomposing pollutants. It is generally believed that the settling performance of activated sludge is optimal when the SVI value is around 100. When the SVI value exceeds 150, it indicates that the activated sludge is about to or has already entered a bulking state, and control measures should be taken immediately.
[0003] Once sludge bulking occurs, it generally takes 1 to 4 weeks to recover. During this period, there is usually nowhere to store the wastewater exceeding the standards, so it is discharged in excess, causing pollution. Therefore, predicting sludge bulking and eliminating it in its infancy is of great significance.
[0004] Previous patents and methods for predicting sludge bulking reported in the literature involve collecting several wastewater treatment operation parameters and water quality variables, constructing expert systems or neural networks to predict sludge bulking. These studies used input parameters such as dissolved oxygen (DO), pH, chemical oxygen demand (COD), total nitrogen (TN), or floc diameter. However, the causes of sludge bulking are highly complex, and current understanding is inconsistent. In actual production reports, both excessively low and excessively high loads can cause bulking; both low and high dissolved oxygen levels can cause bulking; completely mixed aeration tanks and plug-flow aeration tanks can both experience bulking; low C:N ratios and high C:N ratios can both cause bulking, and so on. In short, it is difficult to predict sludge bulking from limited process parameters and water quality indicators.
[0005] Sludge bulking can currently be divided into two main categories: filamentous bulking and non-filamentous bulking. Filamentous bulking is caused by the excessive proliferation of filamentous bacteria in activated sludge flocs, while non-filamentous bulking refers to sludge bulking caused by abnormal physiological activity of the bacteria within the flocs themselves. If the number of filamentous bacteria in the activated sludge is too low, large flocs cannot form, resulting in poor settling performance; if filamentous bacteria proliferate excessively, filamentous sludge bulking occurs. Non-filamentous bulking is caused by abnormal physiological activity of the bacteria within the flocs themselves, leading to deterioration of activated sludge settling performance, or by the presence of large amounts of toxic substances in the influent, preventing the bacteria from secreting sufficient viscous substances to form flocs and causing bulking.
[0006] In addition, based on wastewater treatment practices and the causes of sludge bulking, sludge microscopic examination can help determine the cause of bulking and predict impending bulking in a timely manner, but this requires the operator to have extensive experience.
[0007] This invention uses neural networks to analyze activated sludge images, replacing human visual observation, to predict the probability of impending bulking and take timely measures to prevent sludge bulking. Summary of the Invention
[0008] To address the challenges of predicting sludge bulking from limited process parameters and water quality indicators, and the need for highly experienced operators for manual observation, this invention proposes a neural network-based sludge bulking prediction method. The neural network's layer structure includes a sequence input layer, a sequence folding layer, a convolutional layer, a sequence unfolding layer, a flattening layer, a long short-term memory artificial neural network, and an output layer. The sludge bulking prediction method includes the following steps:
[0009] S01: Obtain training samples, which are several activated sludge images and videos, and the training label is the SVI index of the activated sludge images and videos after a preset time period.
[0010] S02: Train a neural network using training samples with the SVI index after a preset time period as the training label. The trained neural network is a predictive neural network that can predict the SVI index of activated sludge after a preset time period using the layer structure of the neural network through activated sludge images and videos.
[0011] S03: By inputting activated sludge images and videos into a predictive neural network, the SVI index of activated sludge after a preset time period is predicted, and the sludge bulking probability is obtained based on the SVI index.
[0012] The specific steps for predicting the SVI index in step S03 include:
[0013] S31: Obtain activated sludge image videos of a preset duration through the sequence input layer, and store the activated sludge image videos as a continuous activated sludge image sequence at preset time intervals;
[0014] S32: Convert the activated sludge image sequence into an activated sludge image array through a sequence folding layer; the activated sludge image array is composed of each sequence image in the activated sludge image sequence arranged in a preset order;
[0015] S33: Extracting image features from an activated sludge image array using a convolutional layer;
[0016] S34: The image features of each sequence image in the activated sludge image array are stored as an image sequence through the sequence unfolding layer, and the corresponding image features are sorted according to the time sequence of each sequence image through the flattening layer to obtain the feature image sequence.
[0017] S35: Obtain image change sequences using feature image sequences based on long short-term memory artificial neural networks;
[0018] S36: Based on the image change sequence, output the SVI index of activated sludge after a preset time period using the output layer.
[0019] Furthermore, each image in the activated sludge image sequence in steps S01 and S31 is a sequence of images magnified 20 times.
[0020] This invention also proposes a sludge bulking prediction system based on a neural network. The layer structure of the neural network includes a sequence input layer, a sequence folding layer, a convolutional layer, a sequence unfolding layer, a flattening layer, a long short-term memory artificial neural network, and an output layer. The sludge bulking prediction system includes:
[0021] The sample acquisition module is used to acquire training samples, which are several activated sludge images and videos, and the training label is the SVI index of the activated sludge images and videos after a preset time period.
[0022] The training module is used to train a neural network using training samples with the SVI index after a preset time period as the training label. The trained neural network is a predictive neural network that can predict the SVI index of activated sludge after a preset time period using the layer structure of the neural network through activated sludge images and videos.
[0023] The prediction module is used to predict the SVI index of activated sludge after a preset time period by inputting activated sludge images and videos into the prediction neural network, and to obtain the sludge bulking probability based on the SVI index.
[0024] The prediction module includes:
[0025] A sequence input layer unit is used to acquire activated sludge image videos of a preset duration through the sequence input layer. The activated sludge image videos are stored as a continuous sequence of activated sludge images at preset time intervals.
[0026] A sequence folding layer unit is used to convert an activated sludge image sequence into an activated sludge image array through a sequence folding layer; the activated sludge image array is composed of each sequence image in the activated sludge image sequence arranged in a preset order;
[0027] Convolutional layer units are used to extract image features from an array of activated sludge images through convolutional layers.
[0028] The feature image sequence unit is used to store the image features of each sequence image in the activated sludge image array as an image sequence through the sequence unfolding layer, and sort the corresponding image features of each sequence image in the chronological order through the flattening layer to obtain the image feature sequence.
[0029] The image change sequence unit is used to obtain the image change sequence based on the image feature sequence using a long short-term memory artificial neural network.
[0030] The output unit is used to output the SVI index of activated sludge after a preset time period based on the image change sequence using the output layer.
[0031] Furthermore, the sludge bulking prediction system also includes:
[0032] The alarm module is used to issue an alarm when the probability of sludge bulking exceeds a preset threshold.
[0033] Compared with the prior art, the present invention has at least the following beneficial effects:
[0034] (1) The present invention uses activated sludge images and videos as training samples, and uses the SVI index corresponding to the activated sludge images and videos after a preset time period as training labels to train the neural network. The trained neural network can predict the activated sludge SVI index after a preset time period through the activated sludge images and videos using the layer structure of the neural network, and obtain the probability of sludge bulking based on the SVI index. It solves the current problem that it is difficult to predict sludge bulking from limited process parameters and water quality indicators, and that human observation requires operators to have sufficient experience.
[0035] (2) This invention uses neural network analysis of activated sludge images and videos to predict the probability of impending bulking, and takes timely measures based on the bulking probability to prevent sludge bulking. Moreover, this method replaces human observation and avoids the problems of inaccuracy and timeliness of human observation.
[0036] (3) Since activated sludge changes very quickly after sampling, some organisms will die soon. This invention can analyze activated sludge images and videos in real time, which solves this problem.
[0037] (4) The present invention sorts the corresponding image features of each sequence image in chronological order by using a flattened layer to obtain a feature image sequence. It not only analyzes the features of the image itself, but also reflects the drifting and settling characteristics of sludge flocs and the movement characteristics of lower organisms in sludge through the feature image sequence, so its prediction is more accurate. Attached Figure Description
[0038] Figure 1This is a flowchart of a sludge bulking prediction method based on neural networks.
[0039] Figure 2 This is a flowchart of a method for predicting sludge bulking based on neural networks.
[0040] Figure 3 This is a system structure diagram of a sludge bulking prediction system based on neural networks. Detailed Implementation
[0041] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.
[0042] Example 1
[0043] To address the current challenges of predicting sludge bulking from limited process parameters and water quality indicators, and the requirement for operators with sufficient experience for manual observation, such as... Figure 1 and Figure 2 As shown, this invention proposes a sludge bulking prediction method based on a neural network. The layer structure of the neural network includes a sequence input layer, a sequence folding layer, a convolutional layer, a sequence unfolding layer, a flattening layer, a long short-term memory artificial neural network, and an output layer. The sludge bulking prediction method includes the following steps:
[0044] S01: Obtain training samples, which are several activated sludge images and videos, and the training label is the SVI index of the activated sludge images and videos after a preset time period.
[0045] S02: Train a neural network using training samples with the SVI index after a preset time period as the training label. The trained neural network is a predictive neural network that can predict the SVI index of activated sludge after a preset time period using the layer structure of the neural network through activated sludge images and videos.
[0046] During training, when the SVI index of activated sludge is greater than 200, it is set as sludge bulking has occurred. In addition, in this embodiment, the SVI index of activated sludge after 4 hours is used as the training label to train the neural network.
[0047] S03: By inputting activated sludge images and videos into a predictive neural network, the SVI index of activated sludge after a preset time period is predicted, and the sludge bulking probability is obtained based on the SVI index.
[0048] This invention uses neural network analysis of activated sludge images and videos to predict the probability of impending bulking, and can take timely measures based on the bulking probability to prevent sludge bulking. Furthermore, this method replaces human observation, avoiding the problems of inaccuracy and timeliness of human observation.
[0049] The specific steps for predicting the SVI index in step S03 include:
[0050] S31: Obtain activated sludge image videos of a preset duration through the sequence input layer, and store the activated sludge image videos as a continuous activated sludge image sequence at preset time intervals;
[0051] Specifically, the continuous activated sludge image sequence at preset time intervals is composed of sequential images arranged in chronological order.
[0052] It should be noted that the activated sludge images and videos in this embodiment are acquired (captured) by a camera with image magnification function installed on the wastewater treatment process device. Specifically, the start time of the preset time period is the acquisition (capture) time of the activated sludge images and videos.
[0053] Because activated sludge changes very rapidly after sampling, some organisms die quickly. This invention solves this technical problem by acquiring real-time images and videos of activated sludge via a camera and transmitting them to a neural network to predict the probability of sludge expansion online and in real time.
[0054] S32: Convert the activated sludge image sequence into an activated sludge image array through a sequence folding layer; the activated sludge image array is composed of each sequence image in the activated sludge image sequence arranged in a preset order;
[0055] S33: Extracting image features from an activated sludge image array using a convolutional layer;
[0056] S34: The image features of each sequence image in the activated sludge image array are stored as an image sequence through the sequence unfolding layer, and the corresponding image features are sorted according to the time sequence of each sequence image through the flattening layer to obtain the feature image sequence.
[0057] This invention sorts the corresponding image features of each sequence of images in chronological order by using a flattened layer to obtain a feature image sequence. It not only analyzes the features of the images themselves, but also sorts the corresponding image features in chronological order to obtain the features of activated sludge changing over time (feature image sequence), thus making its prediction more accurate.
[0058] S35: Obtain image change sequences using feature image sequences based on long short-term memory artificial neural networks;
[0059] S36: Based on the image change sequence, output the SVI index of activated sludge after a preset time period using the output layer.
[0060] In this embodiment, the prediction effect is best when the image is magnified by about 20 times as input into the neural network. Therefore, each image in the activated sludge image sequence in step S01 and step S31 is a sequence image magnified by 20 times.
[0061] This invention uses activated sludge images and videos as training samples, and uses the SVI index corresponding to the activated sludge images and videos after a preset time period as training labels to train a neural network. The trained neural network can predict the activated sludge SVI index after the preset time period using the layer structure of the neural network through activated sludge images and videos, and derive the probability of sludge bulking based on the SVI index. This solves the current problems of difficulty in predicting sludge bulking from limited process parameters and water quality indicators, and the need for operators to have sufficient experience for manual observation.
[0062] Example 2
[0063] like Figure 3 As shown, this invention also proposes a sludge bulking prediction system based on a neural network. The layer structure of the neural network includes a sequence input layer, a sequence folding layer, a convolutional layer, a sequence unfolding layer, a flattening layer, a long short-term memory artificial neural network, and an output layer. The sludge bulking prediction system includes:
[0064] The sample acquisition module is used to acquire training samples, which are several activated sludge images and videos, and the training label is the SVI index of the activated sludge images and videos after a preset time period.
[0065] The training module is used to train a neural network using training samples with the SVI index after a preset time period as the training label. The trained neural network is a predictive neural network that can predict the SVI index of activated sludge after a preset time period using the layer structure of the neural network through activated sludge images and videos.
[0066] The prediction module is used to predict the SVI index of activated sludge after a preset time period by inputting activated sludge images and videos into the prediction neural network, and to obtain the sludge bulking probability based on the SVI index.
[0067] The prediction module includes:
[0068] A sequence input layer unit is used to acquire activated sludge image videos of a preset duration through the sequence input layer. The activated sludge image videos are stored as a continuous sequence of activated sludge images at preset time intervals.
[0069] A sequence folding layer unit is used to convert an activated sludge image sequence into an activated sludge image array through a sequence folding layer; the activated sludge image array is composed of each sequence image in the activated sludge image sequence arranged in a preset order;
[0070] Convolutional layer units are used to extract image features from an array of activated sludge images through convolutional layers.
[0071] The feature image sequence unit is used to store the image features of each sequence image in the activated sludge image array as an image sequence through the sequence unfolding layer, and sort the corresponding image features of each sequence image in the chronological order through the flattening layer to obtain the image feature sequence.
[0072] The image change sequence unit is used to obtain the image change sequence based on the image feature sequence using a long short-term memory artificial neural network.
[0073] The output unit is used to output the SVI index of activated sludge after a preset time period based on the image change sequence using the output layer.
[0074] The sludge bulking prediction system also includes:
[0075] The alarm module is used to issue an alarm when the probability of sludge bulking exceeds a preset threshold.
[0076] This system can operate online, provide real-time predictions, and can be provided as a complete monitoring system to wastewater treatment plants. It can be integrated with automated systems for activated sludge wastewater treatment. When sludge bulking is predicted, it can automatically control aeration rates, add chemical agents, and adjust pH levels to prevent sludge bulking.
[0077] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.
[0078] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0079] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0080] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.
Claims
1. A sludge bulking prediction method based on neural networks, characterized in that, The neural network's layer structure includes a sequence input layer, a sequence folding layer, a convolutional layer, a sequence unfolding layer, a flattening layer, a long short-term memory artificial neural network, and an output layer. The sludge bulking prediction method includes the following steps: S01: Obtain training samples, which are several activated sludge images and videos, and the training label is the SVI index of the activated sludge images and videos after a preset time period. S02: The neural network is trained using training samples and training labels. The trained neural network is used to predict the SVI index of activated sludge after a preset time period. S03: By inputting activated sludge images and videos into a predictive neural network, the SVI index of activated sludge after a preset time period is predicted, and the sludge bulking probability is obtained based on the SVI index. The specific steps for predicting the SVI index in step S03 include: S31: Obtain activated sludge image videos of a preset duration through the sequence input layer, and store the activated sludge image videos as a continuous activated sludge image sequence at preset time intervals; S32: Convert the activated sludge image sequence into an activated sludge image array through a sequence folding layer; the activated sludge image array is composed of each sequence image in the activated sludge image sequence arranged in a preset order; S33: Extracting image features from an activated sludge image array using a convolutional layer; S34: The image features of each sequence image in the activated sludge image array are stored as an image sequence through the sequence unfolding layer, and the corresponding image features are sorted according to the time sequence of each sequence image through the flattening layer to obtain the feature image sequence. S35: Obtain image change sequences using feature image sequences based on long short-term memory artificial neural networks; S36: Based on the image change sequence, output the SVI index of activated sludge after a preset time period using the output layer.
2. The sludge bulking prediction method based on neural networks according to claim 1, characterized in that, Each image in the activated sludge image sequence in step S31 is a sequence of images magnified 20 times.
3. A sludge bulking prediction system based on neural networks, characterized in that, The neural network's layer structure includes a sequence input layer, a sequence folding layer, a convolutional layer, a sequence unfolding layer, a flattening layer, a long short-term memory artificial neural network, and an output layer. The sludge bulking prediction system includes: The sample acquisition module is used to acquire training samples, which are several activated sludge images and videos, and the training label is the SVI index of the activated sludge images and videos after a preset time period. The training module is used to train the neural network using training samples and training labels. The trained neural network is a predictive neural network used to predict the SVI index of activated sludge after a preset time period. The prediction module is used to predict the SVI index of activated sludge after a preset time period by inputting activated sludge images and videos into the prediction neural network, and to obtain the sludge bulking probability based on the SVI index. The prediction module includes: A sequence input layer unit is used to acquire activated sludge image videos of a preset duration through the sequence input layer. The activated sludge image videos are stored as a continuous sequence of activated sludge images at preset time intervals. A sequence folding layer unit is used to convert an activated sludge image sequence into an activated sludge image array through a sequence folding layer; the activated sludge image array is composed of each sequence image in the activated sludge image sequence arranged in a preset order; Convolutional layer units are used to extract image features from an array of activated sludge images through convolutional layers. The feature image sequence unit is used to store the image features of each sequence image in the activated sludge image array as an image sequence through the sequence unfolding layer, and sort the corresponding image features of each sequence image in the chronological order through the flattening layer to obtain the image feature sequence. The image change sequence unit is used to obtain the image change sequence based on the image feature sequence using a long short-term memory artificial neural network. The output unit is used to output the SVI index of activated sludge after a preset time period based on the image change sequence using the output layer.
4. The sludge bulking prediction system based on a neural network according to claim 3, characterized in that, The sludge bulking prediction system also includes: The alarm module is used to issue an alarm when the probability of sludge bulking is greater than a preset threshold probability.