Alginic acid hydrogel microrobot pH response deformation dynamic modeling method, system and device based on convolutional neural network, medium and product

Abstract

The invention discloses an alginic acid hydrogel micro-robot pH response deformation dynamic modeling method, system and device based on a convolutional neural network, a medium and a product, and relates to the technical field of micro-robot deformation dynamic modeling. The method comprises the following steps: collecting a time sequence image and a pH sequence of micro-robot deformation, and tracking feature points to extract time sequence displacement data; preprocessing the collected data, including cleaning, anomaly detection, sequence alignment and standardization; constructing a one-dimensional convolutional neural network model, and predicting the time sequence displacement data by taking the pH sequence as input; and training and optimizing the model by using preprocessed data, and determining network structure parameters by adopting hyper-parameter optimization. According to the method, accurate dynamic modeling and prediction of the pH response deformation of the alginic acid hydrogel micro-robot are realized, and the behavior control precision of the alginic acid hydrogel micro-robot in a complex environment is improved.

Description

Technical Field

[0001] This application belongs to the field of intelligent material robotic control technology, specifically relating to a convolutional neural network-based method, system, device, medium, and product for modeling the pH-responsive dynamic deformation of an alginate hydrogel microrobot. This invention utilizes deep learning technology to achieve high-precision prediction and control of the microrobot's nonlinear dynamic deformation, with applications in minimally invasive medical treatment, targeted drug delivery, and environmental monitoring. Background Art

[0002] Alginate hydrogel microrobots are an emerging class of intelligent material robots. Due to their excellent biocompatibility, controllable deformation capabilities, and responsiveness to environmental stimuli (particularly pH), they exhibit great potential for application in areas such as targeted drug delivery, minimally invasive surgery, and environmental monitoring. Their unique pH-responsive deformation properties enable them to change their morphology in complex chemical environments by regulating the ambient or internal pH, enabling them to perform functions such as navigation, obstacle avoidance, and grasping or releasing payloads.

[0003] However, the pH-responsive deformation of alginate hydrogel microrobots is a highly complex dynamic process. The speed, amplitude, and response to the rate and pattern of pH changes in their deformation are nonlinear and may be affected by multiple factors such as material batch and environmental viscosity. Accurately predicting and controlling this dynamic deformation in real time is a key technical challenge in achieving high-precision operation of microrobots. Traditional modeling methods, such as those based on simplified physical models or empirical models, often find it difficult to capture and describe this complex nonlinear dynamic behavior, resulting in limited prediction accuracy and insufficient robustness to environmental changes. Existing control strategies also rely heavily on empirical trial and error or simple feedback mechanisms, which are difficult to cope with rapidly changing pH environments and tasks that require extremely high deformation accuracy. Therefore, there is an urgent need for a method that can dynamically model the pH-responsive deformation of alginate hydrogel microrobots with high precision to provide a basis for their reliable and precise control in complex microenvironments. Summary of the Invention

[0004] The present application aims to solve the problems of low precision and difficulty in capturing complex nonlinear dynamic behaviors in the existing methods mentioned in the background technology in the dynamic modeling of pH-responsive deformation of alginate hydrogel microrobots, and to provide a high-precision and robust convolutional neural network-based pH-responsive deformation dynamic modeling method, system, device, medium and product to achieve accurate prediction and control of microrobot deformation, thereby improving its operating performance and reliability in complex environments.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for dynamic modeling of pH-responsive deformation of an alginate hydrogel microrobot based on a convolutional neural network, comprising:

[0007] Raw data acquisition and feature extraction: High-speed microscopy is used to acquire time-series image data of the alginate hydrogel microrobot's deformation process, and a pH sensor is used to simultaneously acquire the corresponding pH change sequence. Feature point tracking is performed on the time-series image data to extract time-series displacement data of at least three non-collinear feature points preset on the microrobot.

[0008] Time series data preprocessing: performing data cleaning, anomaly detection, sequence alignment, and standardization on the time series displacement data and the pH change sequence to obtain a preprocessed time series data set; dividing the preprocessed time series data set into a training set, a validation set, and a test set, and ensuring the continuity of the time series;

[0009] Convolutional neural network model construction: A one-dimensional convolutional neural network model is constructed for processing time-series input and predicting time-series deformation responses. The model receives the pH change sequence as input and predicts the corresponding time-series displacement data. The model adopts a cascaded convolution module design, each module containing a parallel branch structure and using one-dimensional convolution kernels of different sizes to capture local time-series features at different time scales. The model also includes an input layer, a pooling layer, and a fully connected layer.

[0010] Model training and optimization: The convolutional neural network model is trained using the training set, and the optimal network structure and parameters are determined using the Bayesian hyperparameter optimization method. The network parameters are trained using an adaptive learning rate adjustment strategy (such as the AdamW optimizer combined with cosine annealing), and regularization techniques (such as Dropout and L2 weight decay) are used to prevent overfitting, ultimately obtaining the deformation prediction model.

[0011] In a second aspect, the present application provides a pH-responsive deformation dynamic modeling system for an alginate hydrogel microrobot, comprising:

[0012] Data acquisition and feature extraction module: used to obtain raw data through the high-speed microscopy imaging unit and the pH sensing unit, ensure data synchronization through the synchronization trigger unit, and perform feature point tracking to extract the time series displacement data;

[0013] Time series data preprocessing module: used to perform data cleaning, anomaly detection, sequence alignment (such as using dynamic time warping algorithm) and standardization on the time series displacement data and the pH change sequence;

[0014] Deep learning modeling module: used to build, train and optimize the one-dimensional convolutional neural network model, using a multi-scale one-dimensional convolutional neural network structure and Bayesian optimization algorithm;

[0015] Prediction and control module: used to receive the prediction results of the model and convert them into robot motion control instructions.

[0016] In a third aspect, the present application provides a device for dynamic modeling of pH-responsive deformation of an alginate hydrogel microrobot, comprising a memory and a processor, wherein the memory is used to store a program, and the processor is used to implement the method for dynamic modeling of pH-responsive deformation of an alginate hydrogel microrobot as described in claim 1 when executing the program.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for dynamic modeling of pH-responsive deformation of an alginate hydrogel microrobot as described in claim 1.

[0018] In the fifth aspect, the present application provides a pH-responsive deformation dynamic modeling product of an alginate hydrogel microrobot, comprising the method of claim 1, the system of claim 5, the device of claim 8, or one of the computer-readable storage media of claim 9; and an application module, wherein the application module is configured to predict or control the deformation behavior of the alginate hydrogel microrobot based on real-time pH environmental information and using the method, system, or device.

[0019] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0020] This application uses high-speed microscopy and synchronized data acquisition with a high-precision pH sensor to obtain high-resolution time-series images and precise pH data of the alginate hydrogel microrobot's deformation process. Using feature point tracking technology, the complex morphological changes are quantified into high-precision time-series displacement data.

[0021] Using a specially designed one-dimensional convolutional neural network model, especially using a multi-scale convolution kernel structure (such as parallel convolution kernels of different sizes), we can effectively capture the local and global characteristics of the pH change sequence at different time scales and understand the complex nonlinear relationship between pH change and deformation response.

[0022] Through advanced time series data preprocessing methods (such as dynamic time warping DTW), the precise temporal alignment of pH series and displacement series is ensured, which improves data quality and model learning efficiency.

[0023] The Bayesian optimization method is used to automatically optimize the network structure and parameters. Combined with adaptive learning rate adjustment and regularization strategies, the training effect of the model is significantly improved, and a deformation prediction model with high prediction accuracy and good generalization ability is obtained.

[0024] This dynamic modeling method provides an accurate and real-time prediction model basis for the subsequent microrobot deformation control, which helps to achieve more precise and robust morphological control, thereby improving the microrobot's operational capability and reliability in complex microenvironments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0026] Figure 1 A schematic flow chart of a method for dynamic modeling of pH-responsive deformation of an alginate hydrogel microrobot based on a convolutional neural network according to an embodiment of the present application;

[0027] Figure 2 Schematic diagram of the architecture of a convolutional neural network-based alginate hydrogel microrobot pH-responsive deformation dynamic modeling system provided in one embodiment of the present application;

[0028] Figure 3 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0030] The purpose of this application is to provide a method, system, device, medium and product for dynamic modeling of pH-responsive deformation of alginate hydrogel microrobots based on convolutional neural networks, aiming to solve the problem of low accuracy in dynamic modeling of pH-responsive deformation of alginate hydrogel microrobots.

[0031] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0032] In an exemplary embodiment, Figure 1 As shown, the pH-responsive deformation dynamic modeling method of an alginate hydrogel microrobot based on a convolutional neural network in this embodiment includes the following steps:

[0033] Step 1: Raw data collection and feature extraction.

[0034] Specifically, according to claim 1 and claim 2S2, time-series image data of the deformation process of the alginate hydrogel microrobot is obtained using a high-speed microscopic imaging device, wherein the spatial resolution of the high-speed microscopic imaging device is not less than 1920×1080 pixels and the temporal resolution is not less than 200 frames per second, so as to capture the fine deformation process of the microrobot. Simultaneously, according to claim 1 and claim 2S3, the corresponding pH change sequence is obtained synchronously using a micro pH sensor with a response time of less than 0.1 seconds and a measurement accuracy of ±0.05 pH. To ensure accurate synchronization of image data and pH data, according to claim 6S1, a synchronous trigger unit, such as a hardware-level trigger circuit, can be used to ensure that the data acquisition time difference between the imaging device and the pH sensor is less than 10ms. According to claim 2S1, the dynamic pH excitation mode can be designed according to actual application or modeling requirements, for example, including a step change mode (pH range 2.0-9.0) and a periodic pulse mode (pulse period 5-60 seconds). These modes can fully stimulate the dynamic deformation response of the microrobot and provide rich data for model training.

[0035] Feature point tracking is performed on the time-series image data to extract the time-series displacement data of at least three non-collinear feature points preset for the microrobot. The overall deformation of the microrobot can be quantified by the relative displacement of key feature points on its surface (for example, if there are specific marks or naturally formed protrusions / depressions). At least three non-collinear feature points are selected in order to be able to capture two-dimensional or three-dimensional deformations (such as elongation, bending, contraction, etc.), not just overall translation. According to claim 2 S4, the feature point tracking can adopt a motion estimation technology based on the optical flow method, which can robustly track the movement of pixels in the image sequence, thereby obtaining the displacement data of the feature points that change over time. According to claim 6 S2, the feature point tracking algorithm can support sub-pixel displacement detection to achieve a spatial resolution of 0.1 μm to ensure the fineness of the displacement data. The extracted displacement data is a time-series sequence that changes over time.

[0036] Step 2: Time series data preprocessing.

[0037] Specifically, according to claim 1, the time series displacement data and the pH change sequence are subjected to data cleaning, anomaly detection, sequence alignment and standardization to obtain a preprocessed time series data set. The original collected data may contain noise, outliers or time deviations due to asynchronous start / stop of the equipment. Data cleaning and anomaly detection are used to remove or correct these inaccurate data points. Sequence alignment is crucial because it needs to ensure that the pH change sequence and the corresponding deformation displacement sequence are precisely corresponding in time. According to claims 2S 5, a dynamic time warping (DTW) algorithm can be used for sequence alignment to minimize the alignment error (for example, less than 50ms), which is very important for establishing an accurate pH-deformation dynamic mapping relationship. Standardization processing (for example, scaling the data to a range of 0-1 or a mean of 0 and a variance of 1) helps to improve the training efficiency and performance of the neural network.

[0038] According to claim 1 and claim 2S6, the preprocessed time series dataset is divided into a training set, a validation set, and a test set. In order to evaluate the generalization ability of the model, it is necessary to divide the dataset into different subsets for training, validation, and final testing. The commonly used ratio is 70% for the training set, 10% for the validation set, and 20% for the test set. When dividing, it is necessary to preserve the continuity of the time series according to claim 2S6, and avoid splitting a single complete deformation process into different datasets to ensure that the model can learn continuous temporal dynamic relationships.

[0039] Step 3: Convolutional neural network model construction.

[0040] Specifically, according to claim 1, a one-dimensional convolutional neural network model is constructed for processing time-series input and predicting time-series deformation responses. The model receives the pH change sequence as input and predicts the corresponding time-series displacement data. Since pH changes and deformations are both time-series signals, one-dimensional convolutional neural networks are very suitable for processing such sequence data and can capture local temporal patterns and dependencies. According to claim 1, the model includes an input layer, at least three one-dimensional convolutional layers, a pooling layer, and a fully connected layer.

[0041] Furthermore, according to claim 3S1 and claim 7S2, the one-dimensional convolutional neural network model may include cascaded convolution modules, and the number of the convolution modules may vary between 3 and 6. According to claim 3S1 and claim 7S1, each of the convolution modules adopts a parallel branch structure, and the parallel branches contain one-dimensional convolution layers using one-dimensional convolution kernels of different sizes (for example, 3, 5, and 7). This multi-scale convolution kernel design can simultaneously capture the characteristics of the pH series at different time scales, such as the impact of short-term rapid pH fluctuations and the impact of long-term pH change trends, so as to more comprehensively understand the complex dynamic effects of pH on deformation. The calculation formula for one-dimensional convolution according to claim 3 is:

[0042]

[0043] where y i is the i-th element of the output sequence, x is the input sequence, w is the convolution kernel weight, b is the bias term, and k is the convolution kernel size;

[0044] According to claim 3S2, the excitation layer can adopt the Relu activation function, whose formula is f(z)=max(0,z). The Relu function helps to introduce nonlinearity and alleviate the gradient vanishing problem. The pooling layer (such as maximum pooling or average pooling) is used to reduce the dimension of the feature map, reduce the amount of calculation and enhance the robustness of the model. The fully connected layer receives the features after convolution and pooling processing, and maps them to the final prediction output, that is, the time series displacement data of the feature points in the future period of time. According to claim 3S3 and claim 7S2, the output dimension of the fully connected layer is positively correlated with the predicted length of the time series displacement data sequence, and the predicted length can range from 5 to 20 time steps. According to claim 7S2, the number of neurons in the fully connected layer can vary between 64 and 256.

[0045] Step 4: Model training and optimization.

[0046] Specifically, according to claim 1, the convolutional neural network model is trained using the training set. The training objective is to minimize the error between the displacement sequence predicted by the model and the actual observed displacement sequence. A commonly used loss function is the mean squared error (MSE). The training process updates the model's weights and biases through a backpropagation algorithm.

[0047] According to claim 1 and claim 4S3, a hyperparameter optimization method is used to determine the network structure and parameters. The selection of hyperparameters (such as the number of convolution kernels, network depth / number of modules, learning rate, regularization strength, sequence input length, predicted sequence length, etc.) is crucial to the performance of the model. According to claim 4S3 and claim 7S2, the present application can adopt a Bayesian hyperparameter search method. Bayesian optimization guides the search for a set of hyperparameters by constructing a probability model of the objective function (such as model performance on the validation set) (according to claim 6S3, a Gaussian process regression model can be used), thereby finding the optimal hyperparameter combination more efficiently. According to claim 7S2, the optimized hyperparameter range may include the number of one-dimensional convolution kernels (between 16 and 64), the number of cascaded convolution modules (between 3 and 6), the number of fully connected layer neurons (between 64 and 256), the sequence input length (between 30 and 100 time steps), and the predicted sequence length (between 5 and 20 time steps).

[0048] According to claim 1 and claim 4S1, the network parameters are trained by an adaptive learning rate adjustment strategy. The learning rate is a key parameter that affects the convergence speed and performance of model training. According to claim 4S1, an adaptive learning rate optimizer, such as AdamW, can automatically adjust the learning rate of each parameter according to the gradient of the parameter. Combined with a learning rate scheduling strategy, such as a cosine annealing strategy, a larger learning rate can be used to converge quickly in the early stage of training, and then the learning rate can be gradually reduced in the later stage of training to fine-tune the model parameters. According to claim 4S1, the learning rate can be dynamically adjusted in the range of 0.0001-0.001. In addition, according to claim 4S2, the use of regularization techniques (such as dropout, probability 0.4-0.6; L2 weight decay, strength 0.0001-0.001) can prevent the model from overfitting the training data and improve the model's generalization ability on unseen data.

[0049] After sufficient training and optimization, the deformation prediction model is finally obtained. This model can receive a historical pH sequence as input and predict the time-series displacement data of the micro-robot's feature points over a period of time in the future.

[0050] In an exemplary embodiment, the present application also provides a convolutional neural network-based alginate hydrogel microrobot pH response deformation dynamic modeling system, the structure of which is as follows: Figure 2 The system according to claim 5 includes a data acquisition and feature extraction module, a time series data preprocessing module, a deep learning modeling module and a prediction and control module.

[0051] The data acquisition and feature extraction module, as described in claim 5S1, includes a high-speed microscopic imaging unit, a pH sensing unit, and a synchronization trigger unit. The high-speed microscopic imaging unit is used to capture images, the pH sensing unit is used to obtain pH data, and the synchronization trigger unit is used to ensure synchronization between the two and perform preliminary data acquisition. The module also performs feature point tracking, for example, using an integrated or connected computing unit to run an optical flow tracking algorithm (as described in claim 2S4) to extract displacement data of feature points.

[0052] The time series data preprocessing module, as described in claim 5S2, includes a data processing unit equipped with an optical flow tracking algorithm, a dynamic time warping algorithm, and anomaly detection algorithm. This module is responsible for cleaning, anomaly detection, time alignment (as described in claim 2S5), and standardization of the collected raw displacement and pH data to prepare data for model training and prediction.

[0053] The deep learning modeling module, as described in claim 5S3, includes a multi-scale one-dimensional convolutional neural network (as described in claim 7S1, using one-dimensional convolution kernels of sizes 3, 5 and 7 to process the input sequence in parallel) and a Bayesian optimizer (as described in claim 6S3, configured with a Gaussian process regression model, used to search for the optimal network depth in a preset parameter search space, and as described in claim 7S2, automatically optimizes hyperparameters such as the number of convolution kernels, the number of cascaded convolution modules, the number of fully connected layer neurons, the sequence input length, and the length of the predicted time series displacement data sequence). The module is responsible for building, training and optimizing the convolutional neural network model. The computing unit can be a high-performance computer or a GPU cluster to support the complex calculations of the deep learning model. The multi-scale one-dimensional convolutional neural network is responsible for implementing the computational logic of the model, and the Bayesian optimizer is used to automatically find the optimal model hyperparameters.

[0054] The prediction and control module according to claim 5S4 includes an execution unit that converts the model prediction results into robot motion control instructions. Once the model training is completed, the module receives real-time or near real-time pH data and predicts the future deformation posture of the microrobot using the trained model. Then, based on the predicted deformation and the preset task objectives (such as achieving a specific shape, moving to a specific position, etc.), the prediction results are converted into specific control instructions (for example, changing the external pH, applying an external magnetic field, etc.) and sent to the microrobot or its drive system to achieve precise deformation control and task execution.

[0055] In an exemplary embodiment, the present application also provides a pH-responsive deformation dynamic modeling device for an alginate hydrogel microrobot, such as Figure 3As shown. The device according to claim 8 includes a memory and a processor. The memory is used to store a program, and the processor is used to implement the method according to claim 1 when executing the program. For example, the processor can be a central processing unit (CPU), a graphics processing unit (GPU) or other computing unit, and the memory can be a random access memory (RAM) and a non-volatile memory (such as a solid state drive, a hard disk, etc.). The program stored in the memory may include instructions for implementing data acquisition control, image processing, pH data processing, feature point tracking, data preprocessing algorithm, convolutional neural network model code, training algorithm, optimization algorithm, etc.

[0056] In an exemplary embodiment, the present application further provides a computer-readable storage medium, such as a non-transitory storage medium, having a computer program stored thereon. When executed by a processor, the program implements the method of claim 1. For example, the storage medium may be a USB flash drive, a mobile hard drive, an optical disk, or storage space on a network server.

[0057] In an exemplary embodiment, the present application also provides a product for pH-responsive deformation dynamic modeling of an alginate hydrogel microrobot. The product, according to claim 10, comprises one of the method according to claim 1, the system according to claim 5, the device according to claim 8, or the computer-readable storage medium according to claim 9; and an application module, wherein the application module is configured to predict or control the deformation behavior of the alginate hydrogel microrobot based on real-time pH environmental information using the method, system, or device. For example, the application module can be a software application or an embedded control system that receives real-time pH data of the microrobot's environment, inputs this data into a trained CNN model for deformation prediction, and then generates a control signal based on the prediction results and task requirements to drive an external pH control device or other drive device, thereby precisely controlling the shape of the microrobot.

[0058] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0059] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0060] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0061] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0062] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0063] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A convolutional neural network-based pH-responsive deformation dynamic modeling method for alginate hydrogel microrobots, characterized in that: The convolutional neural network-based pH-responsive deformation dynamic modeling method for an alginate hydrogel microrobot comprises: Raw data acquisition and feature extraction: High-speed microscopy is used to acquire time-series image data of the alginate hydrogel microrobot's deformation process, and a pH sensor is used to simultaneously acquire the corresponding pH change sequence. Feature point tracking is performed on the time-series image data to extract time-series displacement data of at least three non-collinear feature points preset on the microrobot. Time series data preprocessing: performing data cleaning, anomaly detection, sequence alignment, and standardization on the time series displacement data and the pH change sequence to obtain a preprocessed time series data set; dividing the preprocessed time series data set into a training set, a validation set, and a test set; Convolutional neural network model construction: A one-dimensional convolutional neural network model is constructed for processing time series input and predicting time series deformation response. The model receives the pH change sequence as input and predicts the corresponding time series displacement data. The model includes an input layer, at least three one-dimensional convolutional layers, a pooling layer, and a fully connected layer. Model training and optimization: The convolutional neural network model is trained using the training set, the network structure and parameters are determined using a hyperparameter optimization method, the network parameters are trained using an adaptive learning rate adjustment strategy, and finally the deformation prediction model is obtained.

2. The pH-responsive deformation dynamic modeling method of an alginate hydrogel microrobot according to claim 1, characterized in that: In the steps of raw data collection and feature extraction and time series data preprocessing: S1: The dynamic pH excitation mode includes a step change mode and a periodic pulse mode, wherein the pH range of the step change is 2.0-9.0, and the period of the pulse is 5-60 seconds; S2: The spatial resolution of the high-speed microscopic imaging device is not less than 1920×1080 pixels, and the temporal resolution is not less than 200 frames per second; S3: The response time of the pH sensor is less than 0.1 seconds, and the measurement accuracy reaches ±0.05pH; S4: The feature point tracking uses a motion estimation technology based on an optical flow method to extract two-dimensional or three-dimensional displacement data of the at least three non-collinear feature points; S5: The sequence alignment adopts a dynamic time warping algorithm, and the alignment error is less than 50ms; S6: The continuity of the time series is retained when dividing the data, and the ratio of the training set, validation set, and test set is 7:1:

2.

3. The pH-responsive deformation dynamic modeling method of an alginate hydrogel microrobot according to claim 1, characterized in that: In the convolutional neural network model construction steps: S1: The one-dimensional convolutional neural network model includes cascaded convolution modules, the number of which is 3-6; each convolution module adopts a parallel branch structure, and the parallel branches contain one-dimensional convolution layers using one-dimensional convolution kernels of different sizes (for example, 3, 5, and 7), which are used to extract local temporal features of the input sequence at different time scales. The formula is: where y i is the i-th element of the output sequence, x is the input sequence, w is the convolution kernel weight, b is the bias term, and k is the convolution kernel size; S2: The ReLU activation function formula used in the excitation layer is: f(z)=max(0,z); S3: The output dimension of the fully connected layer is positively correlated with the predicted length of the time series displacement data sequence, and the predicted length ranges from 5 to 20 time steps.

4. The pH-responsive deformation dynamic modeling method of an alginate hydrogel microrobot according to claim 1, characterized in that: In the model training and optimization steps: S1: The adaptive learning rate adjustment strategy adopts the AdamW optimizer, and the learning rate is dynamically adjusted in the range of 0.0001-0.001 according to the cosine annealing strategy; S2: The regularization technique includes random dropout with a dropout rate of 0.4-0.6 and weight decay with an L2 regularization strength of 0.0001-0.001; S3: The hyperparameter optimization method uses Bayesian hyperparameter search to optimize the number of convolution kernels, network depth, and sequence input length, where the input sequence length optimization range is 30-100 time steps.

5. A pH-responsive deformation dynamic modeling system for alginate hydrogel microrobots, characterized in that: include: S1: Data acquisition and feature extraction module: used to acquire raw data through a high-speed microscopic imaging unit and a pH sensing unit, and perform feature point tracking to extract the time-series displacement data; the module includes a high-speed microscopic imaging unit, a pH sensing unit, and a synchronous triggering unit; S2: Time series data preprocessing module: used to perform data cleaning, anomaly detection, sequence alignment and standardization on the time series displacement data and the pH change sequence; the module is equipped with a data processing unit for optical flow tracking algorithm, dynamic time warping algorithm and anomaly detection algorithm; S3: Deep learning modeling module: used to build, train and optimize the one-dimensional convolutional neural network model; the module includes a multi-scale one-dimensional convolutional neural network and a computing unit of the Bayesian optimizer; S4: Prediction and control module: used to receive the prediction results of the model and convert them into robot motion control instructions; the module includes an execution unit that converts the model prediction results into robot motion control instructions.

6. The pH-responsive deformation dynamic modeling system of alginate hydrogel microrobot according to claim 5, characterized in that: S1: The synchronization trigger unit adopts a hardware-level trigger circuit to ensure that the time synchronization error between the imaging device and the pH sensor is less than 10ms; S2: The feature point tracking algorithm supports sub-pixel displacement detection, and the spatial resolution of the displacement detection reaches 0.1 μm; S3: The Bayesian optimizer is configured with a Gaussian process regression model for searching for the optimal network depth in a preset parameter search space.

7. The pH-responsive deformation dynamic modeling system of an alginate hydrogel microrobot according to claim 5, characterized in that: S1: The deep learning modeling module is configured to construct a CNN model structure including a multi-scale one-dimensional convolutional layer design, using one-dimensional convolution kernels of sizes 3, 5, and 7 to process the input sequence in parallel; S2: The deep learning modeling module is configured to automatically optimize the hyperparameters of the model through a Bayesian optimization method based on Gaussian processes. The hyperparameters include the number of one-dimensional convolution kernels (between 16 and 64), the number of cascaded convolution modules (between 3 and 6), the number of fully connected layer neurons (between 64 and 256), the sequence input length (between 30 and 100), and the predicted length of the time series displacement data sequence (between 5 and 20).

8. A pH-responsive deformation dynamic modeling device for an alginate hydrogel microrobot, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store a program, and the processor is used to implement the pH-responsive deformation dynamic modeling method of the alginate hydrogel microrobot according to claim 1 when executing the program.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, and when the program is executed by the processor, the method for dynamic modeling of pH-responsive deformation of an alginate hydrogel microrobot according to claim 1 is implemented.

10. An alginate hydrogel microrobot pH-responsive deformation dynamic modeling product, characterized in that: The invention comprises one of the method according to claim 1, the system according to claim 5, the device according to claim 8 or the computer-readable storage medium according to claim 9; and an application module, wherein the application module is configured to predict or control the deformation behavior of the alginate hydrogel microrobot based on real-time pH environment information by using the method, system or device.

Images