Construction of online viable count monitoring model based on artificial neural network
By constructing a model based on backpropagation neural network, using multiple parameters during the fermentation process to monitor the number of live bacteria in the online, the problem of time-consuming and labor-consuming and errors in the detection of live bacteria in the existing technology is solved, and the problem of large errors is realized. High accuracy and high efficiency online monitoring effect.
Patent Information
- Application Number
- CN202310533863.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-12
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing technology cannot effectively monitor the number of live bacteria during microbial fermentation. Common offline detection methods are time-consuming and labor-intensive, and cannot distinguish between dead cells and live cells. The error is large and it cannot guide the regulation of the fermentation process in real time.
Using the method based on the backpropagation neural network, an artificial neural network model is constructed using parameters such as fermentation time, fermentation broth capacitance value, conductivity, online pH value and initial substrate concentration, and the neural network is trained through the error backpropagation algorithm to achieve the activity. Online monitoring of bacteria count.
It realizes fast and accurate online monitoring of the number of live bacteria, has a low error rate and a fit degree of more than 0.95, saving time, manpower and material resources, and is suitable for microbial fermentation processes that require pH regulation.
Smart Images

Figure CN118942548A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of artificial neural network models in machine learning, and specifically relates to the construction of an online monitoring model for viable bacteria count based on an artificial neural network. Background Art
[0002] Fermentation process detection is of great significance for process optimization and regulation, especially online monitoring of key parameters can provide a basis for fermentation process regulation. The total viable count can reflect the growth status of the bacteria and is an important parameter in the fermentation process, but it can generally only be detected offline. The commonly used offline biomass detection method has the disadvantages of being time-consuming and labor-intensive, unable to distinguish between dead cells and living cells, and having large errors. It also consumes a lot of manpower, material resources and time. At present, online live cell sensors are widely used in the fermentation process. They can realize online monitoring of the number of live cells through real-time monitoring of the online capacitance value of the fermentation liquid.
[0003] With the rapid development of artificial intelligence (AI) technology, the intelligence of microbial fermentation process is becoming a hot field. Microbial fermentation data has multidimensionality and complex implicit relationships, which can be processed by AI technology based on artificial neural network (ANN). Therefore, the development of ANN-based soft sensors has become a new technology in microbial fermentation research. Soft sensors can predict complex parameters in microbial fermentation process based on the relationship between multiple related factors such as substrate concentration, pH, online parameters, etc. The back propagation (BP) neural network structure is one of the most widely used neural network structures at present. It is good at capturing the nonlinear relationship between multivariate input and single variable output. The participation of BP neural network can realize simple online measurement parameters to predict complex parameters measured offline, and has been widely used in microbial fermentation process.
[0004] Therefore, there is an urgent need to provide a method for constructing an online monitoring model for viable bacteria count based on artificial neural networks. By online monitoring of fermentation time, capacitance and conductivity of the fermentation liquid, online pH value and initial substrate concentration, online monitoring of the key indicator of viable bacteria count can be achieved, which can effectively guide similar fermentation processes. Summary of the invention
[0005] The purpose of the present invention is to provide a method for constructing an online monitoring model for viable bacteria count based on an artificial neural network, which utilizes the acquired fermentation data set, constructs and trains an artificial neural network model by one-to-one correspondence between the parameters of each batch of data set and the viable bacteria count, and finally realizes the online monitoring of the viable bacteria count.
[0006] To achieve the above object, the present invention adopts the following technical solutions.
[0007] The present invention provides a method for constructing an online monitoring model of viable bacteria count based on an artificial neural network, the method comprising:
[0008] (1) Normalization of 12 batch fermentation data sets;
[0009] (2) Select 11 batches of data sets, call the “feedforwardnet” function to build a feedforward neural network with a certain hidden layer size, and use the backpropagation training function, where “trainlm” is selected as the training function, and the weights and bias values are updated according to the Levenberg-Marquardt optimization method;
[0010] (3) Use the “configure” function to configure the size of the input and output layers, and train the neural network using the “train” function with default parameters. The training process continues until the error rate reaches the lowest possible level.
[0011] (4) Determine the number of hidden layer nodes. Currently, there is no theoretical method to determine the number of hidden layer nodes. It is generally determined through empirical formulas and trial and error methods.
[0012] (5) Select a batch of independent data sets and call the "perform" function to test the neural network;
[0013] (6) Use the tested artificial neural network to perform online monitoring of the number of viable bacteria.
[0014] The artificial neural network model uses parameters from 12 batches of fermentation data as input values, and corresponds the parameters of each batch of data sets to the number of viable bacteria one by one, thereby realizing online monitoring of the number of viable bacteria.
[0015] The neural network is trained using an error back propagation algorithm.
[0016] The neural network training is performed by forward transmission of neurons in the order of input layer, hidden layer and output layer, which is described by the following formula:
[0017]
[0018]
[0019]
[0020] After the forward transmission of the neural network training, the reverse transmission from the output layer to the hidden layer is carried out. Its function is to transmit the error between the actual output and the expected output, and based on this, the weight (w) in formula (1) and the bias (w0) in formula (2) are continuously adjusted to reduce the error.
[0021] The neural network is tested, and the number of layers and nodes determined in step (4) are 5 and 600 respectively. Figure 3 .
[0022] The neural network training and testing divides the acquired data set into two parts, 11 batches as normal training samples and 1 batch as test samples for model verification.
[0023] Before the training of the neural network begins, the data set is first normalized by MATLB software and then input, and the number of viable bacteria and the error rate are calculated based on the output.
[0024] The error rate of the training process of the neural network is evaluated by the mean square error (MSE), and the formula is as follows:
[0025]
[0026] Beneficial effects of the present invention:
[0027] 1. The present invention develops a method for constructing an online monitoring model of viable bacteria based on artificial neural network, and selects fermentation time, online capacitance value, online conductivity, online pH value and initial substrate concentration (total sugar, NH4 + and Ca 2+ ) make the BPNN model have lower MSE value and higher R 2 The fitting degree can reach above 0.95, which can effectively predict the number of viable bacteria quickly and accurately.
[0028] 2. The present invention can be used for similar fermentation processes where some microbial fermentation processes require alkali supplementation.
[0029] 3. The present invention provides a certain reference value for the application of artificial intelligence to the microbial fermentation process. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 Schematic diagram of the structure of an online monitoring model for viable bacteria count based on artificial neural networks: Back propagation neural network is a machine learning model that uses the gradient descent method to calculate the minimum square of the network error. It is mainly composed of an input layer, a hidden layer, and an output layer. Each layer contains multiple neurons (nodes), and each neuron has an activation function.
[0031] Figure 2 The fitting results of the online monitoring viable count model based on artificial neural network: the viable count predicted by the model is well consistent with the experimental data, R 2 Reach above 0.95.
[0032] Figure 3 The optimal number of layers and nodes for the online monitoring model of viable bacteria count: When the number of layers and nodes reached 5 and 600, respectively, the largest R 2 value and the minimum MSE value.
[0033] Figure 4 Training, verification and testing of the online monitoring model for viable bacteria count: Whether it is the training, testing or verification data set, the model prediction data is very close to the experimental results, proving that this online monitoring model for viable bacteria count has been successfully developed. DETAILED DESCRIPTION
[0034] The following is a detailed description of the construction of an online monitoring model for viable bacteria count based on an artificial neural network according to the present invention with reference to examples.
[0035] (1) Obtain fermentation process parameters and normalize the data using MATLAB software.
[0036] (2) The back propagation neural network was constructed using the corresponding functional modules in MATLAB2019b software. First, 11 batches of fermentation data were selected, and the “feedforwardnet” function was called to construct a feedforward neural network with a certain hidden layer size. The back propagation training function was used, in which “trainlm” was selected as the training function. The weights and bias values were updated according to the Levenberg-Marquardt optimization method. This method has the advantages of both the gradient descent method and the Gauss-Newton method, and can also achieve rapid convergence.
[0037] (3) Secondly, the sizes of the input and output layers are configured using the “configure” function, and the neural network is trained using the “train” function with default parameters until the error rate reaches the lowest possible level (manually defined).
[0038] (4) Determine the number of hidden layer nodes. The selection of the number of hidden layer nodes is one of the important factors affecting the performance of the BP network model. However, there is currently no theoretical method to determine the number of hidden layer nodes. It is generally determined through empirical formulas and trial and error methods. In order to ensure sufficiently high simulation performance and generalization ability, it is usually necessary to select as few nodes as possible while meeting the accuracy requirements. The number of hidden layer nodes is determined by trial and error, that is, through experiments, the error between the model prediction value and the actual measurement value is used as the standard to select the appropriate number of nodes. The mean square error (MSE) is used to evaluate the model error;
[0039] (5) Then, a separate batch of fermentation data sets that was not used for training was selected, and the “perform” function was called to test a neural network to determine the fit and error between the model prediction value and the experimental true value, thus completing the construction of the neural network model.
[0040] (6) Finally, the tested neural network was used to monitor the number of viable bacteria online. The data was normalized and then input into the tested neural network model to calculate R online. 2 and MSE value, the number of live bacteria in the fermentation process can also be obtained in real time.
[0041] A method for constructing an online monitoring model for viable bacteria count based on artificial neural networks provides a theoretical basis for online monitoring of viable bacteria count in microbial fermentation processes, and is applicable to all microbial fermentation processes that require pH control. The innovation of the present invention is that an online monitoring model for viable bacteria count can be constructed using the obtained microbial fermentation data set, and compared with common offline measurement methods, it is more accurate and faster, saving time, manpower and material resources. The online monitoring fit of viable bacteria count reached above 0.95, and the number of hidden layers and nodes of the model were determined to be 6 and 500, respectively. Results are shown in Figure 2 , Figure 3 , Figure 4 .
[0042] The above description is only an example of the implementation mode of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. Construction of an online monitoring model for viable bacteria count based on artificial neural network, characterized in that: The parameters of the fermentation data set are obtained as input values; an artificial neural network model capable of online monitoring of the number of viable bacteria is constructed and trained and the accuracy of the artificial neural network model is calculated; and the number of viable bacteria is monitored online using the constructed model.
2. The construction of an online monitoring model for viable bacteria count based on an artificial neural network according to claim 1, characterized in that: The input values correspond one-to-one to the number of viable bacteria in each batch of data sets, and the input values include fermentation time, online capacitance value, online conductivity, online pH value and initial substrate concentration.
3. The construction of an online monitoring model for viable bacteria count based on an artificial neural network according to claim 2, characterized in that: The initial substrate concentration includes the initial total sugar concentration, the initial NH4 + Concentration and initial Ca 2+ concentration.
4. The construction of an online monitoring model for viable bacteria count based on an artificial neural network according to claim 1, characterized in that: The method of constructing and training an artificial neural network model capable of online monitoring of the number of viable bacteria and calculating the accuracy of the artificial neural network model includes: performing artificial neural network training based on seven input values and the number of viable bacteria corresponding to each batch of data sets.
5. The construction of an online monitoring model for viable bacteria count based on an artificial neural network according to claim 1 or 4, characterized in that: The artificial neural network training adopts the error back propagation algorithm to perform training and determine the number of nodes and layers of the artificial neural network model.
6. The construction of an online monitoring model for viable bacteria count based on an artificial neural network according to claims 1-5, characterized in that: The viable bacteria count is: the viable bacteria count at each time point is measured offline using a plate viable bacteria count (CFU) method.
7. The construction of an online monitoring model for viable bacteria count based on an artificial neural network according to claim 1, characterized in that: The specific steps include: (1) Normalization of 12 batch fermentation data sets; (2) Select 11 batches of data sets, call the "feedforwardnet" function to build a feedforward neural network with a certain hidden layer size, and use the backpropagation training function, where "trainlm" is selected as the training function, and the weights and bias values are updated according to the Levenberg-Marquardt optimization method; (3) Use the "configure" function to configure the size of the input and output layers, and train the neural network using the "train" function with default parameters. The training process continues until the error rate reaches the lowest possible level. (4) Determine the number of hidden layer nodes. Currently, there is no theoretical method to determine the number of hidden layer nodes. It is generally determined through empirical formulas and trial and error methods. (5) Select a batch of independent data sets and call the "perform" function to test the neural network; (6) Use the tested artificial neural network to perform online monitoring of the number of viable bacteria.
8. The construction of an online monitoring model for viable bacteria count based on an artificial neural network according to claim 1, characterized in that: The artificial neural network model ultimately achieves accurate online monitoring of viable bacteria counts and can be applied to a variety of microbial fermentation processes that require pH regulation.
Citation Information
Cited By
Method for producing L-carnitine soybean meal by using neural network to optimize fermentation conditions
CN121415878A
