Spectral adaptive incremental learning modeling method and system
Through the spectral adaptive incremental learning modeling method, the number of hidden layer nodes is adaptively updated and parameters are randomly allocated, which solves the problems of slow convergence speed and redundant structure of the neural network model, and improves the performance and efficiency of the spectral analysis model.
Patent Information
- Application Number
- CN202411208040.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-11-26
AI Technical Summary
In the existing spectral analysis, the neural network model has problems with slow convergence speed and redundant structure in determining the number of hidden layer nodes, resulting in large interference in parameter settings and affecting model performance.
The spectral adaptive incremental learning modeling method is adopted, and the number of hidden layer nodes is adaptively updated through the supervised learning mechanism based on prediction error constraints during the iteration process, and weights and bias parameters are randomly allocated to avoid the generation of redundant network structures.
It achieves rapid convergence and reduces parameter calculation costs, improves the performance of spectral qualitative and quantitative analysis models, and reduces interference from human parameter settings.
Smart Images

Figure CN119026057A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of spectral data analysis and modeling, and in particular to a spectral adaptive incremental learning modeling method and system. Background Art
[0002] Spectral analysis technologies such as UV-Vis, near-infrared, mid-infrared and Raman have the advantages of fast, accurate and non-destructive testing. Currently, spectral analysis technology has been widely used in many industries such as petrochemicals, biomedicine, and semiconductor manufacturing. The main content of spectral analysis is to establish stable and reliable qualitative and quantitative models based on material spectral data and material physicochemical property data.
[0003] Artificial neural network is an important technology for building spectral analysis models. Artificial neural network can approximate any linear and nonlinear functions, but it takes a lot of time to determine the width and depth of the hidden layer. The random configuration network based on a single hidden layer can incrementally determine the number of hidden layer nodes with a supervision mechanism, which can reduce the interference caused by artificial parameter settings. However, the existing point increment or block increment mechanism is prone to problems such as slow convergence speed or redundant neural network structure. Summary of the invention
[0004] A brief summary of one or more aspects is given below to provide a basic understanding of these aspects. This summary is not an exhaustive overview of all conceived aspects, and is neither intended to identify the key or critical elements of all aspects nor to define the scope of any or all aspects. Its only purpose is to give some concepts of one or more aspects in a simplified form as a prelude to a more detailed description that will be given later.
[0005] The purpose of the present invention is to solve the above problems and provide a spectral adaptive incremental learning modeling method and system, which can adaptively expand the number of hidden layer nodes of the neural network during the learning process, thereby reducing the interference of artificial parameter settings and avoiding the generation of redundant network structures.
[0006] The technical solution of the present invention is: the present invention discloses a spectral adaptive incremental learning modeling method, the method comprising:
[0007] Step S1, obtaining a batch of samples, obtaining sample spectral data and physical and chemical property data, and constructing a spectral analysis data set;
[0008] Step S2, preprocessing the acquired spectral data to remove abnormal samples and spectral wavelengths containing obvious noise;
[0009] Step S3, randomly dividing the analysis data set into a training set and a test set;
[0010] Step S4, constructing a spectral adaptive incremental learning model using the training set;
[0011] Step S5: Apply the spectral adaptive incremental learning model obtained in step S4 to perform predictive analysis on the test set.
[0012] According to one embodiment of the spectral adaptive incremental learning modeling method of the present invention, in step S1, the physicochemical property data is one or more physicochemical indicators related to the analysis purpose; the sample spectrum includes ultraviolet-visible spectrum, near-infrared spectrum, mid-infrared spectrum, Raman spectrum and fluorescence spectrum; the sample spectral data includes the following categories: for ultraviolet-visible spectrum, near-infrared spectrum and mid-infrared spectrum, the spectral data includes wavelength or wavenumber, and corresponding absorbance; for Raman spectrum, the spectral data includes Raman shift and corresponding light intensity; for fluorescence spectrum, the spectral data includes wavelength or wavenumber, and corresponding light intensity.
[0013] According to one embodiment of the spectral adaptive incremental learning modeling method of the present invention, in step S2, the preprocessing includes abnormal sample processing and strong noise wavelength processing, the abnormal sample processing is to discard abnormal data samples, the abnormal data samples include sample data whose sample absorbance deviates from the sample average spectrum due to the experimental environment; the strong noise spectral wavelength processing is to discard wavelengths containing obvious noise at both ends.
[0014] According to an embodiment of the spectral adaptive incremental learning modeling method of the present invention, in step S3, randomly dividing the analysis data set into a training set and a test set further includes the following steps:
[0015] Step S31, using the spectral absorbance value as input data and the sample physicochemical property value as output data;
[0016] Step S32: randomly divide the analysis data into a training set and a test set.
[0017] According to an embodiment of the spectral adaptive incremental learning modeling method of the present invention, in step S4, constructing a spectral adaptive incremental learning model using sample spectra and physicochemical property values in a training set further includes the following steps:
[0018] Step S41, initializing the model tolerance, the maximum number of iterations, the hidden layer node batch increase block size, i.e., the incremental block, and the supervised learning mechanism parameters based on the prediction error constraint;
[0019] Step S42, constructing a neural network with spectral data as input, property data as output, and the number of incremental block variables equal to the number of hidden layers, wherein the weights and bias parameters between the input layer and the hidden layer are randomly allocated from a fixed interval;
[0020] Step S43, calculating the hidden layer output weight and model prediction error;
[0021] Step S44: Based on the supervised learning mechanism of prediction error constraints, adaptively update the incremental block size of hidden layer nodes in the iterative learning process, update the number of hidden layer nodes and re-randomly allocate weights and bias parameters between the input layer and the hidden layer;
[0022] Step S45, calculating the output weight and model prediction error of the hidden layer of the incremental network model obtained in step S44;
[0023] Step S46: If the prediction error is less than the model allowable error, stop the iteration to obtain the final model, otherwise loop through steps S44-S45 until the maximum number of iterations is reached.
[0024] The present invention also discloses a spectral adaptive incremental learning modeling system, the system comprising:
[0025] The analysis data set construction module obtains a batch of samples, obtains the sample spectral data and physical and chemical property data, and constructs the spectral analysis data set;
[0026] The data preprocessing module preprocesses the acquired spectral data to remove abnormal samples and spectral wavelengths containing obvious noise;
[0027] The data set partitioning module randomly divides the analysis data set into training set and test set;
[0028] The model building module uses the training set to build a spectral adaptive incremental learning model;
[0029] The prediction analysis module applies the model obtained by the model building module to perform prediction analysis on the test set.
[0030] The present invention also discloses a computer-readable storage medium storing program instructions executable by a processor to implement the steps of the spectral adaptive incremental learning modeling method as described above.
[0031] The present invention also discloses a computer device, which includes a memory, a processor, and program instructions stored in the memory and executable by the processor, wherein the processor executes the program instructions to implement the steps of the spectral adaptive incremental learning modeling method as described above.
[0032] Compared with the prior art, the present invention has the following beneficial effects: the present invention can adaptively determine the incremental block size according to the iterative process error, avoid generating redundant network structures while quickly converging, and reduce the interference caused by artificial parameter settings, which is conducive to improving the performance of spectral qualitative and quantitative analysis models. Compared with the existing spectral modeling method based on neural networks, the present invention can determine the number of hidden layer nodes of the neural network model in an adaptive incremental learning manner, effectively reducing the parameter calculation cost, which is of great significance for improving the efficiency of spectral modeling. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] The above features and advantages of the present invention can be better understood after reading the detailed description of the embodiments of the present disclosure in conjunction with the following drawings. In the drawings, the components are not necessarily drawn to scale, and components with similar related properties or features may have the same or similar reference numerals.
[0034] Figure 1 A flow chart of an embodiment of the spectral adaptive incremental learning modeling method of the present invention is shown.
[0035] Figure 2 A schematic diagram showing an embodiment of the spectral adaptive incremental learning modeling system of the present invention is shown.
[0036] Figure 3 A network structure diagram of a spectral adaptive incremental learning modeling method is shown.
[0037] Figure 4 It is an error convergence curve diagram of an example of the spectral adaptive incremental learning modeling method of the present invention. DETAILED DESCRIPTION
[0038] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Note that the aspects described below in conjunction with the accompanying drawings and specific embodiments are only exemplary and should not be construed as limiting the scope of protection of the present invention in any way.
[0039] Figure 1 FIG. 1 shows a flow chart of an embodiment of the spectrum adaptive incremental learning modeling method of the present invention. Figure 1 , the implementation steps of the method of this embodiment are described in detail as follows.
[0040] Step S1, obtaining a batch of samples, obtaining sample spectral data and physical and chemical property data, and constructing a spectral analysis data set.
[0041] In step S1, the data set {X, y} includes the spectral data X and the physicochemical property data y of the sample, wherein the sample spectral data includes the wavelength (or wave number) and the corresponding absorbance value, and the physicochemical property data is one or more physicochemical indicators related to the analysis purpose. The sample spectrum includes ultraviolet-visible spectrum, near infrared spectrum, mid-infrared spectrum, Raman spectrum and fluorescence spectrum.
[0042] Correspondingly, in step S1, the sample spectrum data specifically includes the following categories:
[0043] For UV-visible, near-infrared, and mid-infrared spectra, the spectral data include wavelength (or wave number) and corresponding absorbance;
[0044] For Raman spectroscopy, the spectral data include Raman shift and corresponding light intensity;
[0045] For fluorescence spectroscopy, spectral data includes wavelength (or wave number) and corresponding light intensity.
[0046] Step S2: pre-process the acquired spectral data to remove abnormal samples and spectral wavelengths containing obvious noise.
[0047] Preprocessing includes abnormal sample processing and strong noise wavelength processing. The abnormal sample processing specifically discards abnormal data samples, and the abnormal data samples include sample data whose sample absorbance deviates from the sample average spectrum due to the experimental environment; the strong noise spectrum wavelength processing specifically discards wavelengths containing obvious noise at both ends.
[0048] Step S3: randomly divide the analysis data set {X, y} into a training set and a test set.
[0049] In step S3, randomly dividing the analysis data set {X, y} into a training set and a test set specifically includes the following steps:
[0050] Step S31, using the spectral absorbance value X as input data and the sample physicochemical property value y as output data;
[0051] Step S32: randomly divide the analysis data into a training set and a test set.
[0052] Step S4: construct a spectral adaptive incremental learning model using the training set.
[0053] In step S4, constructing a spectral adaptive incremental learning model using sample spectra and physicochemical property values in the training set specifically includes the following steps:
[0054] Step S41, initializing the model allowable error ε, the maximum number of iterations K, the batch increase block size Δ1 (incremental block) of the hidden layer nodes, and the supervised learning mechanism parameter r based on the prediction error constraint;
[0055] Step S42, constructing a neural network with spectral data X as input, property data y as output, and the number of incremental block variables being the number of hidden layers Δ1, wherein the weight w and bias parameter b between the input layer and the hidden layer are randomly assigned from a fixed interval [-λ, λ];
[0056] For a given training set data {x i ,y i}, x i ={x i,1 ,…,x i,M}∈R M is the sample spectrum data, y i ={y i,1 ,…,y i,m}∈R m To materialize the attribute data, the hidden layer output H of the randomly configured network with the initial block size Δ1 as the number of hidden layer nodes can be expressed as:
[0057]
[0058] Where g(·) is the Sigmoid activation function, Enter the weight parameters for the hidden layer, Input bias parameters for the hidden layer, T represents the vector or matrix transpose operation.
[0059] Step S43: Calculate the hidden layer output weight and model prediction error e1;
[0060]
[0061] In the formula, H is the hidden layer output matrix, y is the materialized attribute vector, (·) -1 represents the inverse calculation, e1 represents the first prediction error, and the number of iterations k=1.
[0062] Step S44: Based on the supervised learning mechanism of prediction error constraints, adaptively update the incremental block size of hidden layer nodes in the iterative learning process, update the number of hidden layer nodes and re-randomly allocate weights and bias parameters between the input layer and the hidden layer;
[0063] Based on the supervised learning mechanism parameters 0<r<1 and b g , for a non-negative real-valued sequence μ L Satisfy 0≤μ L ≤(1-r), then:
[0064] δ k,q =(1-r-μ L )||e k,q || 2,q=1,2,…,m
[0065] If we add a random basis function Given under the following inequality mechanism:
[0066]
[0067] Then there is In the formula, Δ' k represents the k-th iteration incremental block size, L k represents the number of hidden layer nodes at the kth iteration, Represents the hidden layer output corresponding to the incremental block.
[0068] For the incremental block Δ during the iteration k The size of can be established as follows:
[0069]
[0070] L k =L k-1 +Δ' k
[0071] In the formula, Δ' k represents the incremental block size of the kth iteration, Δ1 is the batch increase block size of the hidden layer nodes, and e k-1 It represents the prediction error of the prediction model obtained by the k-1th incremental learning. When k = 1, e is defined k-1 =e1, Indicates the rounding operation. For the input layer and the L k The weights w and bias parameters b between the hidden layers of nodes are randomly assigned from a fixed interval [-λ,λ].
[0072] Step S45, calculating the output weight and model prediction error of the hidden layer of the incremental network model obtained in step S44;
[0073] For L k Incremental network output of nodes Its output weight It can be obtained by formula (2), and the corresponding prediction error can be expressed as:
[0074]
[0075] In the formula, e Lk Indicates content L k The prediction error of the incremental network of nodes, H is the hidden layer output matrix, y is the materialized attribute vector, Output weights for the hidden layer.
[0076] Step S46: If the prediction error e LkIf the value is less than the model tolerance ε, the iteration is stopped to obtain the final model; otherwise, steps S44-S45 are calculated repeatedly until the maximum number of iterations K is reached.
[0077] Step S5: Apply the model obtained in step S4 to perform predictive analysis on the test set.
[0078] In step S5, the spectral adaptive incremental learning model obtained in step S4 is used to perform prediction analysis on the test set, and the model performance comparison index is the determination coefficient R 2 And the root mean square error RMSE:
[0079]
[0080]
[0081] In the formula, y i,real represents the true value of the sample, Represents the average value of the sample true value, y i,pred represents the sample prediction value obtained based on the constructed spectral adaptive incremental learning model, N t Represents the number of prediction samples.
[0082] Figure 2 The principle of an embodiment of the spectrum adaptive incremental learning modeling system of the present invention is shown. Figure 2 The system of this embodiment includes: an analysis data set construction module, a data preprocessing module, a data set partitioning module, a model construction module, and a prediction analysis module.
[0083] The analysis data set construction module obtains a batch of samples, obtains sample spectral data and physicochemical property data, and constructs a spectral analysis data set; the physicochemical property data is one or more physicochemical indicators related to the analysis purpose; the sample spectrum includes ultraviolet-visible spectrum, near-infrared spectrum, mid-infrared spectrum, Raman spectrum and fluorescence spectrum; the sample spectral data includes the following categories: for ultraviolet-visible spectrum, near-infrared spectrum and mid-infrared spectrum, the spectral data includes wavelength or wave number and corresponding absorbance; for Raman spectrum, the spectral data includes Raman shift and corresponding light intensity; for fluorescence spectrum, the spectral data includes wavelength or wave number and corresponding light intensity.
[0084] The data preprocessing module preprocesses the acquired spectral data to remove abnormal samples and spectral wavelengths containing obvious noise; the preprocessing includes abnormal sample processing and strong noise wavelength processing, the abnormal sample processing is to discard abnormal data samples, the abnormal data samples include sample data whose sample absorbance deviates from the sample average spectrum due to the experimental environment; the strong noise spectral wavelength processing is to discard wavelengths containing obvious noise at both ends.
[0085] The data set partitioning module randomly partitions the analysis data set into a training set and a test set; the randomly partitioning of the analysis data set into a training set and a test set further comprises the following steps:
[0086] The spectral absorbance value is used as input data, and the sample physical and chemical property value is used as output data;
[0087] The analysis data is randomly divided into training and test sets.
[0088] The model building module uses the training set to build a spectral adaptive incremental learning model; using the training set to build the adaptive incremental learning model further includes the following steps:
[0089] Initialize the model's allowable error, the maximum number of iterations, the size of the hidden layer node batch increase block, i.e., the incremental block, and the parameters of the supervised learning mechanism based on the prediction error constraint;
[0090] A neural network is constructed with spectral data as input, property data as output, and the number of incremental block variables equal to the number of hidden layers, where the weights and bias parameters between the input layer and the hidden layer are randomly assigned from a fixed interval;
[0091] Calculate hidden layer output weights and model prediction errors;
[0092] Based on the supervised learning mechanism of prediction error constraints, the incremental block size of hidden layer nodes is adaptively updated during iterative learning, the number of hidden layer nodes is updated, and the weights and bias parameters between the input layer and the hidden layer are randomly redistributed;
[0093] Calculate the output weights and model prediction errors of the hidden layer of the incremental network model obtained in the previous step;
[0094] If the prediction error is less than the model allowable error, the iteration is stopped to obtain the final model, otherwise the above steps are calculated repeatedly until the maximum number of iterations is reached.
[0095] The prediction analysis module applies the model obtained by the model building module to perform prediction analysis on the test set.
[0096] The present invention also discloses a computer-readable storage medium, which stores program instructions executable by a processor to implement the following Figure 1 The steps of the spectrally adaptive incremental learning modeling method are shown.
[0097] The present invention also discloses a computer device, which comprises a memory, a processor and program instructions stored in the memory and executable by the processor, wherein the processor executes the program instructions to implement the following Figure 1 The steps of the spectrally adaptive incremental learning modeling method are shown.
[0098] The present invention is described below by taking diesel cetane number measurement as an example, and the specific steps include:
[0099] Step 1: Obtain 381 representative diesel samples and use a near-infrared spectrometer to collect spectral data for the collected samples. The near-infrared spectrum of the diesel samples is measured by inserting the probe into each sample bottle, and the cetane number of the diesel samples is measured according to relevant national or industry test standards. The final spectrum wavelength range is 900-1550nm, and the resolution is 4.72nm.
[0100] Step 2: Eliminate the abnormal spectrum of diesel and remove the wavelengths with obvious noise at both ends.
[0101] Step 3: Randomly divide the analysis data set consisting of sample spectrum numbers and cetane numbers into a training set and a test set, where the ratio of the training set to the test set is 7:3.
[0102] Step 4: Construct a spectral adaptive incremental learning model to predict diesel cetane number, such as Figure 3 In this example, λ=1, the incremental block size is initialized to Δ0=3, and the number of nodes in the final single hidden layer is 15. Figure 4 As shown in the figure, compared with the existing point-increment random configuration network (SCN-P) method and block-increment random configuration network (SCN-B) method, the proposed method can converge to the set model tolerance error at a faster speed.
[0103] Step 5: Use the prediction model established in step 4 to perform prediction analysis on the test samples. The corresponding results are shown in Table 1. Compared with the existing partial least squares (PLS) method, point incremental random configuration network (SCN-P) method and block incremental random configuration network (SCN-B) method, the proposed method effectively improves the model accuracy. 2 The average value of the regression line is 0.9716 and the RMSE is 0.2368.
[0104] Table 1 Experimental results analysis table
[0105]
[0106] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.
[0107] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.
[0108] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.
[0109] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.
[0110] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, each function may be stored on or transmitted by a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one place to another. Storage media may be any available medium that can be accessed by a computer. As an example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer. Any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, a server, or other remote source using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, wherein disk often reproduces data magnetically, while disc reproduces data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.
[0111] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to those skilled in the art, and the universal principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but should be granted the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A spectrum adaptive incremental learning modeling method, characterized in that: Methods include: Step S1, obtaining a batch of samples, obtaining sample spectral data and physical and chemical property data, and constructing a spectral analysis data set; Step S2, preprocessing the acquired spectral data to remove abnormal samples and spectral wavelengths containing obvious noise; Step S3, randomly dividing the analysis data set into a training set and a test set; Step S4, constructing a spectral adaptive incremental learning model using the training set; Step S5: Apply the spectral adaptive incremental learning model obtained in step S4 to perform predictive analysis on the test set.
2. The spectral adaptive incremental learning modeling method according to claim 1, characterized in that: In step S1, the physicochemical property data is one or more physicochemical indicators related to the analysis purpose; the sample spectrum includes ultraviolet-visible spectrum, near-infrared spectrum, mid-infrared spectrum, Raman spectrum and fluorescence spectrum; the sample spectrum data includes the following categories: for ultraviolet-visible spectrum, near-infrared spectrum and mid-infrared spectrum, the spectrum data includes wavelength or wave number and corresponding absorbance; for Raman spectrum, the spectrum data includes Raman shift and corresponding light intensity; for fluorescence spectrum, the spectrum data includes wavelength or wave number and corresponding light intensity.
3. The spectral adaptive incremental learning modeling method according to claim 1, characterized in that: In step S2, the preprocessing includes abnormal sample processing and strong noise wavelength processing. The abnormal sample processing is to discard abnormal data samples, and the abnormal data samples include sample data whose sample absorbance deviates from the sample average spectrum due to the experimental environment; the strong noise spectrum wavelength processing is to discard wavelengths containing obvious noise at both ends.
4. The spectral adaptive incremental learning modeling method according to claim 1, characterized in that: In step S3, randomly dividing the analysis data set into a training set and a test set further comprises the following steps: Step S31, using the spectral absorbance value as input data and the sample physicochemical property value as output data; Step S32: randomly divide the analysis data into a training set and a test set.
5. The spectral adaptive incremental learning modeling method according to claim 1, characterized in that: In step S4, constructing a spectral adaptive incremental learning model using sample spectra and physicochemical property values in the training set further includes the following steps: Step S41, initializing the model tolerance, the maximum number of iterations, the hidden layer node batch increase block size, i.e., the incremental block, and the supervised learning mechanism parameters based on the prediction error constraint; Step S42, constructing a neural network with spectral data as input, property data as output, and the number of incremental block variables equal to the number of hidden layers, wherein the weights and bias parameters between the input layer and the hidden layer are randomly allocated from a fixed interval; Step S43, calculating the hidden layer output weight and model prediction error; Step S44: Based on the supervised learning mechanism of prediction error constraints, adaptively update the incremental block size of hidden layer nodes in the iterative learning process, update the number of hidden layer nodes and re-randomly allocate weights and bias parameters between the input layer and the hidden layer; Step S45, calculating the output weight and model prediction error of the hidden layer of the incremental network model obtained in step S44; Step S46: If the prediction error is less than the model allowable error, stop the iteration to obtain the final model, otherwise loop through steps S44-S45 until the maximum number of iterations is reached.
6. A spectral adaptive incremental learning modeling system, characterized in that: The system includes: The analysis data set construction module obtains a batch of samples, obtains the sample spectral data and physical and chemical property data, and constructs the spectral analysis data set; The data preprocessing module preprocesses the acquired spectral data to remove abnormal samples and spectral wavelengths containing obvious noise; The data set partitioning module randomly divides the analysis data set into training set and test set; The model building module uses the training set to build a spectral adaptive incremental learning model; The prediction analysis module applies the model obtained by the model building module to perform prediction analysis on the test set.
7. A computer-readable storage medium storing program instructions executable by a processor to implement the steps of the spectral adaptive incremental learning modeling method as described in any one of claims 1 to 5.
8. A computer device comprising a memory, a processor, and program instructions stored in the memory and executable by the processor, wherein the processor executes the program instructions to implement the steps of the spectral adaptive incremental learning modeling method as described in any one of claims 1 to 5.
Citation Information
Cited By
Fluorescence immunochromatography test strip rapid reading and analyzing method based on deep learning and application of fluorescence immunochromatography test strip rapid reading and analyzing method
CN120104977A
Quality parameter detection model construction method based on incremental learning driving
CN120932777A
A Method for Constructing a Quality Parameter Detection Model Based on Incremental Learning
CN120932777B
