An experimental stage automatic identification method and system based on artificial intelligence

Abstract

This invention provides an artificial intelligence-based method and system for automatic identification of experimental stages, relating to the fields of experimental automation and artificial intelligence. The method includes: acquiring multidimensional state data during the experimental process; the multidimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operating state data, experimental process images, video data, and microscopic imaging and detection data; performing joint modeling and time series analysis on the multidimensional state data sequence to extract state feature data reflecting the evolution of the experimental process; and based on the state feature data, identifying abnormal states and atypical stages through multimodal state feature joint analysis, analysis of changes in trends and rates over time, determination of stage evolution trajectories and stage boundaries, historical experimental stage pattern matching and similarity analysis, and identification of abnormal states and atypical stages. This invention improves the accuracy, standardization, and automation level of experimental stage identification.

Description

Technical Field

[0001] This invention relates to the fields of experimental automation and artificial intelligence technology, and in particular to an automatic identification method and system for experimental stages based on artificial intelligence. Background Technology

[0002] When conducting nanomaterial preparation experiments on an automated experimental platform for material synthesis, experimenters need to complete experimental operations through devices such as temperature control, stirring, and feeding. At the same time, they rely on imaging, particle size detection, and various sensors to collect multi-dimensional data such as the morphology of the experimental object, ambient temperature and pressure, and the operating speed of the device. The experimental process will show obvious semantic stages such as feeding preparation, reaction change, and stable convergence. Moreover, the evolution of each stage is affected by multiple factors such as temperature, stirring rate, and feeding amount, and there are clear conditional dependencies and evolutionary correlations between stages.

[0003] Existing experimental automation systems can only execute operations according to preset fixed procedures and rely on fixed thresholds of single physical parameters for state judgment. They cannot perform joint analysis of multi-dimensional experimental state data to perceive the overall experimental state, nor can they automatically identify the specific stage at the experimental semantic level. In this nanomaterial preparation experiment, it is impossible to judge the start and progress of the reaction stage based on the coordinated changes in particle size, temperature fluctuations, and the operating status of the stirring device. Furthermore, it is impossible to provide input for experimental decisions such as adjusting the feeding rate and optimizing temperature control parameters in a structured form based on the stage identification results. The experimenters have to manually analyze data, judge the experimental stage, and manually adjust the operating parameters, which increases the cost of human intervention and makes it difficult to adapt to uncertain state changes during the experiment, resulting in low levels of automation and intelligence. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that existing experimental automation systems have difficulty in jointly analyzing multi-dimensional state data during the experimental process, in automatically identifying the semantic stage and stage boundary of the current experiment, and in outputting standardized structured stage identification results.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, an automatic identification method for experimental phases based on artificial intelligence, the method comprising: Acquire multidimensional state data during the experiment; multidimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operation state data, experimental process images, video data, microscopic imaging and detection data; By jointly modeling and analyzing the multidimensional state data sequence, state feature data reflecting the evolution of the experimental process can be extracted. Based on state feature data, state features are analyzed through multimodal state feature joint analysis, time-dimensional trend and rate of change analysis, stage evolution trajectory and stage boundary determination, historical experimental stage pattern matching and similarity analysis, and identification of abnormal states and atypical stages, and the analysis results are obtained. Based on the analysis results, the current experimental stage is automatically identified, and an experimental stage identification result is generated. The experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, the change or reaction stage, the stable or convergence stage, the abnormal stage, and the termination stage. Furthermore, joint modeling and time series analysis are performed on the multidimensional state data sequence to extract state feature data reflecting the evolution of the experimental process, including: The acquired multidimensional state data is preprocessed to obtain standardized multidimensional time series data. Temporal features of each dimension are extracted from the standardized multidimensional time series data. These temporal features include statistical features, rate of change features, and frequency domain features. The extracted temporal features from various dimensions are fused in a multimodal manner to construct a multidimensional temporal state vector; A time series analysis is performed on the multidimensional time series state vector to calculate the state characteristics that reflect the evolution of the experimental process. The state characteristics include the trend characteristics, the rate of change characteristics, and the stability characteristics. The trend characteristics are obtained by fitting the slope of the time series data, the rate of change characteristics are obtained by calculating the change in adjacent time moments, and the stability characteristics are obtained by calculating the variance within the sliding window.

[0006] Furthermore, based on state feature data, the state features are analyzed through multimodal state feature joint analysis, analysis of change trends and rates of change over time, determination of stage evolution trajectories and stage boundaries, historical experimental stage pattern matching and similarity analysis, and identification of abnormal states and atypical stages. The analysis results include: Based on state characteristics, a multi-dimensional joint analysis of change trend characteristics, change rate characteristics, and stability characteristics is conducted to obtain the evolution direction and evolution intensity of the experimental state. Based on the evolution direction and intensity, the change trend characteristics and change rate characteristics are continuously tracked in the time dimension to identify the inflection points and abrupt change points of the state characteristics as candidate boundary points for stage evolution. Based on the boundary candidate points, and combining the characteristics of change trend and stability, the evolution trajectory of the experimental state in the feature space is constructed, and the curvature and direction changes of the evolution trajectory are analyzed to obtain the stage interval determination result of the current state. The current experimental evolution trajectory is compared with the evolution trajectory of the marked stages in the historical experiment. If the similarity exceeds the preset threshold, the stage interval is corrected by referring to the division boundary of the historical stage to obtain the corrected current stage judgment result. During the similarity comparison process, if the similarity between the current experimental evolution trajectory and the evolution trajectory of all historical normal stages is lower than the preset threshold, and the change pattern of the state characteristics deviates from the preset normal evolution range, then the current state is identified as an atypical stage, and the identification result of the atypical stage is obtained. By combining the results of the phase interval determination and the results of the atypical phase identification, the specific phase of the experiment is finally determined automatically.

[0007] Furthermore, based on the analysis results, the current experimental stage is automatically identified, and experimental stage identification results are generated, including: Based on the current specific stage, determine the corresponding stage identifier; based on one or more of the consistency of various analysis results, similarity comparison results, and clarity of evolution trajectory, determine the confidence level of the current stage identification; based on the boundary candidate points, determine the entry time point and exit time point of the current stage to obtain the stage boundary information; By combining stage identifiers, stage confidence levels, and stage boundary information, a structured experimental stage identification result is obtained.

[0008] Furthermore, the experimental phase is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation phase, the change or reaction phase, the stabilization or convergence phase, the anomaly phase, and the termination phase, including: Based on the state characteristics, determine whether the conditions for the initial preparation stage are met. The initial preparation stage corresponds to the state interval from the start of the experiment to the stable state. The state characteristics are characterized by a trend of change close to zero and a rate of change lower than a preset first threshold. If the conditions are met, it is determined that the experiment is currently in the initial preparation stage. If the conditions for the initial preparation stage are not met, then it is determined whether the conditions for the reaction stage are met. The reaction stage corresponds to the state range in which the state continues to change after the experimental conditions are triggered. The state characteristics are that the absolute value of the change trend characteristics exceeds the preset second threshold, and the change rate characteristics are continuously higher than the preset third threshold. If the conditions are met, it is determined that the current state is in the change or reaction stage. If the conditions for the reaction stage are not met, then it is determined whether the conditions for the convergence stage are met. The convergence stage corresponds to a state range where the state parameters tend to be stable. The state characteristics are that the trend of change is close to zero, the rate of change is lower than the preset fourth threshold, and the stability is lower than the preset fifth threshold. If these conditions are met, it is determined that the current stage is the convergence stage. If the conditions for the convergence phase are not met, then the conditions for the abnormal phase are determined. The abnormal phase corresponds to a state range in which the state change deviates from the expected evolution trajectory. The state characteristics are that the similarity between the evolution trajectory and the historical normal phase evolution trajectory is lower than the preset sixth threshold, and the change pattern cannot match any known phase pattern. If the conditions are met, then it is determined that the current stage is abnormal. If the conditions for an abnormal phase are not met, the current state is determined to be in the termination phase. The termination phase corresponds to the state interval where the experiment ends. The state characteristics are that the experimental operation instructions have been executed and all state parameters remain stable within a preset time.

[0009] Furthermore, based on the analysis results, the current experimental stage is automatically identified, generating experimental stage identification results. An experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, change or reaction stage, stabilization or convergence stage, anomaly stage, and termination stage, among others. Based on the current experimental semantic stage, extract the corresponding stage identifier; obtain the confidence value of the current stage from the experimental stage identification results; and organize the stage entry boundary information and stage exit boundary information according to the boundary candidate points and stage entry and exit time points. The stage identifier, stage confidence level, and stage entry and exit boundary information are encapsulated according to a preset data structure to obtain a standardized structured data packet; The structured data packets are output in real time through the output interface for external modules to use.

[0010] Secondly, an artificial intelligence-based automatic identification system for experimental phases includes: The acquisition module is used to acquire multidimensional state data during the experiment. The multidimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operation state data, experimental process images, video data, and microscopic imaging and detection data. The extraction module is used to perform joint modeling and time series analysis on multidimensional state data sequences to extract state feature data that reflects the evolution of the experimental process. The analysis module is used to analyze state features based on state feature data, through multimodal state feature joint analysis, analysis of change trends and rates of change in the time dimension, determination of stage evolution trajectory and stage boundaries, pattern matching and similarity analysis of historical experimental stages, and identification of abnormal states and atypical stages, to obtain analysis results. The identification module is used to automatically identify the current experimental stage based on the analysis results and generate experimental stage identification results. The experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, change or reaction stage, stable or convergence stage, abnormal stage and termination stage. Thirdly, a computing device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0011] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0012] The above-described solution of the present invention has at least the following beneficial effects: This research employs multi-dimensional state data acquisition, multi-modal feature joint modeling, and time series analysis techniques. Combined with a multi-dimensional stage identification method based on the continuous evolution of the experimental process and a standardized structured result output scheme, it systematically overcomes the technical problems of existing automated experimental systems that rely solely on preset procedures or fixed thresholds for judgment, cannot automatically perceive the overall experimental state and semantic-level stages, struggle to provide structured input for experimental path planning, and are highly dependent on human experience. This achieves automatic perception of the experimental process and accurate identification of semantic-level stages. By outputting structured stage identification results containing stage identifiers, stage confidence levels, and stage boundary information, the standardization and callability of experimental stage identification results are improved. Furthermore, it provides a unified foundation of stage state information for subsequent external systems, reduces the cost of manual analysis and judgment, and enhances the automation and intelligence level of experimental stage identification. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating an artificial intelligence-based automatic identification method for experimental stages provided by an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of an artificial intelligence-based automatic identification system for experimental stages provided by an embodiment of the present invention. Detailed Implementation

[0015] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0016] like Figure 1 As shown, an embodiment of the present invention proposes an automatic identification method for experimental stages based on artificial intelligence, the method comprising the following steps: Step 1: Acquire multidimensional state data during the experiment; multidimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operation state data, experimental process images, video data, microscopic imaging and detection data. Step 2: Perform joint modeling and time series analysis on the multidimensional state data sequence to extract state feature data that reflects the evolution of the experimental process; Step 3: Based on the state feature data, analyze the state features through multimodal state feature joint analysis, time dimension change trend and change rate analysis, stage evolution trajectory and stage boundary determination, historical experimental stage pattern matching and similarity analysis, and identification of abnormal states and atypical stages, and obtain the analysis results. Step 4: Based on the analysis results, automatically identify the current experimental stage and generate experimental stage identification results; the experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, change or reaction stage, stable or convergence stage, abnormal stage and termination stage. In this embodiment of the invention, by employing multi-dimensional state data acquisition, multi-modal feature joint modeling, and time series analysis, combined with a multi-dimensional stage identification method based on the continuity of experimental process evolution and a standardized structured result output scheme, the technical problems of existing experimental automation systems—which rely solely on preset processes or fixed thresholds for judgment, cannot automatically perceive the overall experimental state and semantic-level stages, struggle to provide structured input for experimental path planning, and are highly dependent on human experience—are overcome. This achieves automatic perception of the experimental process and accurate identification of semantic-level stages, providing reliable structured state input for dynamic experimental path planning, supporting the automated execution of complex multi-step experiments, reducing human intervention, and building the perception foundation for an intelligent automated experimental platform. It allows the selection and switching of experimental operations to be synchronized in real time with the evolution of experimental stages, effectively improving the automation and intelligence level of the experimental process.

[0017] In a preferred embodiment of the present invention, step 1 above may include: Step 1: Acquire multi-dimensional state data during the experiment. This multi-dimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operating status data, experimental process images, video data, and microscopic imaging and detection data. Specifically, this includes: establishing a multi-source data acquisition hardware interface system, unifying the sensor group, imaging equipment, detection instruments, and experimental device control module in the experimental scenario into a unified network, establishing communication connections between each device and the data acquisition terminal, ensuring that the data acquired by each device can be transmitted to the acquisition terminal in real time without loss, and simultaneously calibrating the acquisition clocks of each device, calculating the clock deviation value of each device, and unifying the acquisition time of all devices to the system reference clock. The calibration method involves adding the clock deviation value to the local acquisition time of each device to match the reference clock time, ensuring the synchronization of timestamps for multi-source data.

[0018] Establish rules for collecting multidimensional state data. Based on the type of experiment and the evolution rate of the experimental process, determine whether the data collection mode is continuous collection or collection at preset time intervals. If collection is at preset time intervals, first, statistically analyze the state change cycles of each stage in the experimental history, and take the least common divisor of the change cycles of each stage as the basic collection interval. Then, based on the identification requirements of key stages of the experiment, adjust the collection interval of key stages to 0.5 times the basic collection interval, and the collection interval of non-key stages to 2 times the basic collection interval. Clarify the collection frequency and duration of data for each dimension to ensure that the collected dataset can completely cover the state evolution of the experimental process. Collect state data of experimental objects. Through dedicated detection instruments and analysis equipment, collect physical feature data such as structure, morphology, color, position, and phase transition of experimental objects in real time. Quantify visual feature data such as morphology and color, and characterize the phase transition process through the numerical changes of feature parameters. At the same time, record the collected state data of experimental objects in real time and mark the corresponding collection timestamp and experimental operation node.

[0019] The experimental environment status data is collected by using sensors such as temperature sensors, pressure sensors, humidity sensors, atmosphere monitors, and field strength detectors deployed within the experimental setting. These sensors collect data on temperature, pressure, humidity, atmosphere composition, and field strength. For the atmosphere composition data, the volume percentage of each component is calculated by dividing the volume of a single component by the total volume of gas in the experimental environment. For the field strength data, values ​​from different spatial locations are collected and the average value is calculated by dividing the sum of the field strength values ​​at each location by the number of locations collected, forming a multi-dimensional experimental environment status dataset. The experimental apparatus operation status data is also collected by using the control system and operation monitoring module of the experimental apparatus to collect data on the rotation speed, power, flow rate, and execution status of various experimental devices such as stirring devices, feeding devices, reaction vessels, and temperature control devices. For the flow rate data, the amount delivered per unit time is calculated by dividing the total mass of the delivered medium by the delivery time. The execution status is quantified and encoded, converting the device's running, paused, and fault states into standardized values, thus achieving quantitative collection of the device's operating status.

[0020] The experiment involved collecting images and video data. High-definition industrial cameras and high-speed video equipment were deployed within the experimental setting. Images and videos of the overall experimental process and key reaction areas were captured at preset angles and frequencies. The camera's frame rate and resolution were set to ensure clear capture of changes in the experimental state. Each frame and video segment was labeled with a timestamp to achieve time matching with other data dimensions. Microscopic imaging and detection data were also collected. Microscopic detection equipment such as scanning electron microscopes, transmission electron microscopes, and laser particle size analyzers were used to detect and image the microstructure, particle size distribution, and micromorphology of the experimental objects. Microscopic images and quantitative detection data were acquired. The average particle size and variance were calculated from the particle size distribution data. The average value was calculated using the following method: The distribution variance is calculated by multiplying the values ​​of each particle size by their corresponding proportions. The variance is calculated by multiplying the square of the difference between each particle size value and the average value by the sum of the corresponding proportions. This allows for precise acquisition of the microscopic state of the experimental objects. Preprocessing and integration of multi-source data are then performed. Outliers are removed from the raw data collected in each dimension, based on a value exceeding three standard deviations of the normal fluctuation range for that dimension. Missing data is filled in by averaging the two adjacent valid data points before and after the missing time point, calculated as the sum of the values ​​of the two adjacent valid data points divided by 2. After data cleaning, the data in each dimension are aligned and integrated according to the timestamps. All dimension data under the same timestamp are associated to form a structured experimental multidimensional state data sequence, completing the acquisition process of multidimensional state data for the entire experiment.

[0021] In this embodiment of the invention, a standardized and synchronized multi-source data acquisition method is used to overcome the limitations of existing single and fragmented experimental data acquisition. This enables the complete, accurate, and synchronized acquisition of experimental state data across all dimensions, solving the technical problem in existing technologies where only a single parameter or a small number of dimensions of data are collected, thus failing to comprehensively reflect the overall experimental state. Simultaneously, real-time preprocessing and timestamp alignment of the acquired data effectively avoids the impact of missing, abnormal, or asynchronous data on subsequent analysis. This provides a high-quality, structured data source for joint modeling and time series analysis of multi-dimensional state data, ensuring the accuracy and reliability of subsequent experimental stage identification and laying a data foundation for analyzing the evolution of the experimental process from an experimental semantic perspective.

[0022] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1 involves preprocessing the acquired multidimensional state data to obtain standardized multidimensional time-series data. This includes: firstly, performing data cleaning on the multidimensional state data, screening outliers and missing values ​​in the experimental object state data, experimental environment state data, experimental device operation state data, experimental process image and video data, and microscopic imaging and detection data. Outliers are determined when their values ​​exceed three times the normal fluctuation range of the corresponding dimension. These outliers are then removed. For missing values, linear interpolation is used for completion. The completion calculation method is to take the sum of the values ​​of the two adjacent valid data points before and after the missing time point and divide it by two to obtain the completed data at the missing position. After data cleaning, the raw data for each dimension are standardized to eliminate analytical biases caused by differences in units and magnitudes of data in different dimensions. The standardization calculation method is to subtract the average value of the corresponding dimension data from a single raw data point, and then divide the result by the standard deviation of the corresponding dimension data. The standardization calculation is performed on the cleaned data of all dimensions in turn, while retaining the timestamp information of each dimension data. All standardized dimension data are then arranged in order according to timestamps to form standardized multidimensional time series data.

[0023] Step 2.2: Extract time series features for each dimension from the standardized multidimensional time series data. Time series features include statistical features, rate of change features, and frequency domain features. Specifically, based on the standardized multidimensional time series data, extract three types of time series features sequentially according to each dimension. First, extract statistical features. For each dimension of time series data, calculate the mean, median, maximum, minimum, variance, and standard deviation of the dataset. The mean is the sum of all data values ​​in that dimension divided by the total number of data points. The median is the value in the middle position after arranging the data in that dimension in order of size. The variance is the sum of the squares of the differences between each data point in that dimension and the mean divided by the total number of data points. The standard deviation is the arithmetic square root of the variance. This completes the extraction and quantification of the statistical features for that dimension. Next, the rate of change feature is extracted. Using timestamps as the order, for standardized data at two adjacent moments in this dimension, the difference between the data at the later moment and the data at the previous moment is calculated. This difference is then divided by the value at the previous moment to obtain the instantaneous rate of change at adjacent moments. This process is repeated for all adjacent moments in this dimension. Simultaneously, the average of all instantaneous rates of change within a preset time window in this dimension is calculated as the average rate of change within that time window, thus completing the rate of change feature extraction. Finally, frequency domain features are extracted. A Fast Fourier Transform is performed on the standardized time-series data for this dimension to convert the time-domain data into frequency-domain data. The dominant frequency, frequency energy, and spectral bandwidth in the frequency-domain data are calculated. The dominant frequency is the frequency value with the largest amplitude in the frequency-domain data, the frequency energy is the sum of the squares of the amplitudes corresponding to all frequencies, and the spectral bandwidth is the frequency range in the frequency-domain data with an amplitude greater than half the amplitude of the dominant frequency, thus completing the frequency domain feature extraction. The above method is used to extract the three types of time-series features for each dimension of the standardized multi-dimensional time-series data, forming the time-series feature set corresponding to each dimension.

[0024] Step 2.3 involves multimodal fusion of the extracted time-series features from various dimensions to construct a multidimensional time-series state vector. Specifically, this includes: first, normalizing the extracted statistical features, rate of change features, and frequency domain features from each dimension, mapping the values ​​of all features to the interval between zero and one. The normalization calculation method is to subtract the minimum value of a feature class from a single feature value, and then divide the result by the difference between the maximum and minimum values ​​of that feature class. This ensures that features of different types and dimensions are comparable in numerical magnitude. After feature normalization, multimodal fusion is performed using feature concatenation, with the timestamp as the core association criterion, to fuse the statistical features and rate of change features of each dimension within the same time window. The normalized values ​​of rate features and frequency domain features are arranged sequentially according to a preset dimension. First, the three types of time-series feature values ​​of the first dimension are arranged, and then the three types of time-series feature values ​​of all subsequent dimensions are arranged in sequence to form a one-dimensional feature sequence. For each time window, the features are spliced ​​and fused in the above manner. The fused feature sequences of all time windows are combined in chronological order to construct a multi-dimensional time-series state vector that can comprehensively represent the experimental state. The dimension of this vector is the sum of the number of the three types of time-series features of all experimental data dimensions. Each element of the vector corresponds to the normalized value of the time-series feature of a specific time window, a specific dimension, and a specific type, thereby realizing the joint modeling of multi-dimensional state data.

[0025] Step 2.4: Conduct time series analysis on the multi-dimensional time series state vector to calculate the state features reflecting the evolution of the experimental process. The state features include change trend features, change rate features, and stability features. The change trend features are obtained by fitting the slope of the time series data. The change rate features are obtained by calculating the change amount between adjacent moments. The stability features are obtained by calculating the variance within a sliding window, specifically as follows: Based on the constructed multi-dimensional time series state vector, conduct time series analysis on the vector data in chronological order, and calculate the three types of core state features in sequence. Calculate the change trend features. Taking time as the horizontal axis and the eigenvalue of the multi-dimensional time series state vector as the vertical axis, perform a unary linear regression fit on the vector eigenvalues within a preset time range to obtain the slope of the fitted line. This slope is the change trend feature within this time range. A positive slope indicates that the experimental state shows a positive evolution trend, a negative slope indicates that the experimental state shows a reverse evolution trend, and the larger the absolute value of the slope, the more significant the evolution trend. Then calculate the change rate features. In chronological order of timestamps, calculate the difference between the eigenvalues of the multi-dimensional time series state vectors corresponding to two adjacent time windows. This difference is the change rate feature between adjacent moments. The larger the absolute value of the difference, the faster the change rate of the experimental state within this time period. Calculate the change rate features of all adjacent time windows in sequence to form a continuous change rate feature sequence. Finally, calculate the stability features. Set a sliding time window with a fixed length, slide the sliding window along the change rate feature sequence in chronological order, and calculate the variance of all change rate feature values within each sliding window. This variance is the stability feature corresponding to the sliding window. The calculation method of variance is the sum of the squares of the differences between each change rate feature value within the window and the average change rate within the window divided by the total amount of data within the window. The larger the variance, the worse the stability of the experimental state within this time period, and the smaller the variance, the better the stability of the experimental state. Through the above calculations, obtain the change trend features, change rate features, and stability features covering the entire experimental process, complete the time series analysis of the multi-dimensional state data, and extract the state feature data reflecting the evolution of the experimental process.

[0026] In this embodiment of the invention, standardized preprocessing eliminates the differences in dimensions and magnitudes of multidimensional state data, solving the problem of analysis result deviation caused by inconsistent data features in existing technologies, and laying a high-quality data foundation for subsequent feature extraction and analysis. Through multi-dimensional and multi-type time-series feature extraction, the time-domain and frequency-domain features of experimental state data are comprehensively explored, breaking through the limitation of existing technologies that only analyze single data features, and can more accurately capture subtle changes in experimental states. By constructing multi-dimensional time-series state vectors through multi-modal feature fusion, joint modeling of multi-dimensional experimental state data is realized, solving the technical problem in existing technologies where data of each dimension is analyzed independently and cannot reflect the overall experimental state. Through targeted time-series analysis and calculation, three types of core state features are obtained, realizing the quantitative representation of the evolution law of the experimental process, providing a scientific and effective feature basis for accurate identification of experimental stages, changing the status quo in existing technologies where the lack of quantitative evolution features leads to reliance on manual experience judgment for experimental stage identification, and improving the objectivity and accuracy of experimental stage identification.

[0027] In a preferred embodiment of the present invention, step 3 above may include: Step 3.1: Based on the state characteristics, conduct a multi-dimensional joint analysis of the trend characteristics, rate of change characteristics, and stability characteristics to obtain the evolution direction and intensity of the experimental state. Specifically, this includes: first, quantifying and calibrating the trend characteristics, rate of change characteristics, and stability characteristics; using the positive or negative value of the slope of the trend characteristics as the basis for determining the evolution direction, with a positive slope indicating positive evolution, a negative slope indicating negative evolution, and a slope of zero indicating no obvious evolution direction; using the absolute value of the rate of change characteristics as the basic reference for the speed of evolution, and the variance value of the stability characteristics as the basic reference for the stability of evolution. Based on this, a multi-dimensional joint analysis is conducted to calculate the evolution intensity of the experimental state. The evolution intensity is calculated by multiplying the absolute value of the rate of change characteristic by the evolution direction weight, and then dividing by the variance value of the stability characteristic. The evolution direction weight is preset according to the experimental type, with the weights for forward and reverse evolution both set to 1, and the weight for no obvious evolution direction set to 0. The larger the evolution intensity value obtained by this calculation, the more significant the evolution of the experimental state, and the smaller the value, the weaker the evolution of the experimental state. Combining the calibration results of the evolution direction and the calculation results of the evolution intensity, the evolution direction and evolution intensity of the experimental state are finally determined.

[0028] Step 3.2: Based on the evolution direction and intensity, continuously track the trend and rate of change characteristics over time to identify inflection points and abrupt change points of the state characteristics as candidate boundary points for stage evolution. Specifically, this includes: continuously numerically tracking and monitoring the trend and rate of change characteristics along the time axis, following the chronological order of the experimental process; setting thresholds for inflection points and abrupt change points, where the inflection point threshold is the threshold for the change in the slope of the trend characteristic, and the abrupt change point threshold is the threshold for the numerical abrupt change in the rate of change characteristic. During the tracking process, calculate the trend of change within two adjacent time windows. The difference in the slope of the potential characteristic is used to identify the time point as an inflection point of the trend characteristic if the absolute value of the difference exceeds the preset inflection point judgment threshold and the evolution intensity at that position increases. The difference in the numerical value of the rate of change characteristic within two adjacent time windows is calculated. If the absolute value of the difference exceeds the preset mutation point judgment threshold and the evolution direction at that position does not shift continuously, the time point is identified as a mutation point of the rate of change characteristic. All identified inflection points and mutation points are summarized, and after removing duplicate time points, the remaining time points are used as candidate boundary points for the evolution of the experimental stage, providing a time dimension reference for the determination of the stage interval.

[0029] Step 3.3: Based on the boundary candidate points, and combining the trend and stability features, construct the evolution trajectory of the experimental state in the feature space, and analyze the curvature and direction changes of the evolution trajectory to obtain the stage interval determination result of the current state. Specifically, this includes: constructing a two-dimensional experimental state feature space with the trend features as the abscissa and the stability features as the ordinate; mapping the values ​​of the trend and stability features of each time window in the entire experimental process to this feature space to form feature points; connecting the feature points in the feature space in chronological order based on the boundary candidate points, and marking the nodes at the boundary candidate point positions to construct the complete evolution trajectory of the experimental state in the feature space; and analyzing the curvature and direction changes of this evolution trajectory. The curvature value of each node on the evolutionary trajectory is calculated. The curvature value is calculated by dividing the angle of change of the tangent direction of the evolutionary trajectory at that node by the feature space distance between that node and the next node, and then multiplying by the angle correction coefficient, which is preset to 1. The larger the curvature value, the more drastic the trajectory change at that node. At the same time, the overall directional change of the evolutionary trajectory is tracked to determine whether the trajectory shows a trend of continuous rise, continuous fall, steady fluctuation, or disordered change. Combined with the semantic definition of the experimental stage, the corresponding experimental stage interval is matched for different trajectory regions in the feature space. Based on the position of the feature point of the current experimental state in the evolutionary trajectory, the trajectory curvature and directional change characteristics of the region, the corresponding experimental stage interval is matched to obtain the stage interval determination result of the current state.

[0030] Step 3.4: Compare the current experimental evolution trajectory with the evolution trajectories of marked stages in historical experiments. If the similarity exceeds a preset threshold, the determined stage interval is corrected by referring to the historical stage division boundaries to obtain the corrected current stage determination result. Specifically, this includes: retrieving evolution trajectory data of stage-marked historical experiments to construct a historical experimental evolution trajectory database. This database contains the evolution trajectory of each historical experiment, the corresponding stage division boundaries, and stage identifiers. A trajectory similarity algorithm is used to align and calculate the similarity between the current experimental evolution trajectory and the historical experimental evolution trajectory one by one. The similarity is calculated as the reciprocal of the sum of the squared distances between corresponding feature points of the current experimental evolution trajectory and the historical experimental evolution trajectory in the feature space, multiplied by... Using a trajectory length matching coefficient, calculated as the shortest length of the current trajectory divided by the longest length of a historical trajectory, a similarity value closer to 1 indicates higher similarity between the two trajectories. A preset threshold for trajectory similarity is set. If the similarity between the current experimental trajectory and a historical experimental trajectory exceeds this threshold, the two trajectories are considered similar. The stage division boundary corresponding to that historical experiment is retrieved, and combined with the candidate boundary points of the current experiment, the stage interval determination result is corrected. The entry and exit time points of the stage are adjusted to make the current stage interval division more consistent with the verification pattern of the historical experiment, resulting in a corrected current stage determination result. If no historical experimental trajectory with a similarity exceeding the preset threshold is found, the stage interval determination result is directly used.

[0031] Step 3.5: During the similarity comparison process, if the similarity between the current experimental evolution trajectory and the evolution trajectories of all historical normal stages is lower than a preset threshold, and the change pattern of the state characteristics deviates from the preset normal evolution range, then the current state is identified as an atypical stage, and the identification result of the atypical stage is obtained. Specifically, this includes: based on the similarity comparison, checking the similarity results between the current experimental evolution trajectory and the evolution trajectories of all historical normal stages in the historical experimental evolution trajectory database; if all similarity results are lower than the preset threshold, then entering the normal evolution range determination stage, the preset normal evolution range of the experimental state is defined as the reasonable range of values ​​for the trend characteristics, rate of change characteristics, and stability characteristics, which is statistically obtained based on a large amount of historical normal experimental state characteristic data; comparing the real-time values ​​of the trend characteristics, rate of change characteristics, and stability characteristics of the current experiment with the preset normal evolution range; if the value of any characteristic continuously exceeds the normal evolution range, and the deviation does not recover in a short time, then it is determined that the change pattern of the current experimental state characteristics deviates from the preset normal evolution range. If both criteria are met, the current experimental state is identified as an atypical stage, and the start time, feature deviation type, and degree of deviation of the atypical stage are recorded to form the identification result of the atypical stage. If either criterion is not met, the atypical stage identification is not performed, and the previous stage interval determination result is continued to be used.

[0032] Step 3.6: Combining the stage interval determination results and the atypical stage identification results, the specific stage of the experiment is automatically determined. This includes: first, validating the stage interval determination results and the atypical stage identification results by checking whether the time nodes corresponding to the results match the current experimental time window and whether there are any abnormal missing features. If the atypical stage identification result is valid, meaning the current experiment is identified as an atypical stage, then the atypical stage is directly taken as the specific stage of the experiment. If the atypical stage identification result is invalid, meaning no atypical stage is identified, then the stage interval determination result is used as the core, combined with the semantic definition of the experimental stage and the feature determination conditions, to finally confirm the result, clarify the specific stage identifier corresponding to the current experimental state. After the determination is completed, the final result is recorded, and the core feature data and evolutionary trajectory features of the determination basis are marked to obtain complete feature analysis results, providing a direct basis for the formal identification of the experimental stage.

[0033] In this embodiment of the invention, by conducting multi-dimensional joint analysis of experimental state characteristics, the limitations of single-parameter analysis in existing technologies are overcome, enabling quantitative determination of the direction and intensity of experimental state evolution and solving the technical problem of not being able to accurately grasp the overall changes in experimental states. By tracking and identifying boundary candidate points through time-dimensional feature tracking, precise time references are provided for the division of experimental stages, avoiding the problem of ambiguous stage division boundaries in existing technologies. By constructing feature space evolution trajectories and analyzing their curvature and direction changes, the evolution process of experimental states is visualized and quantified, enabling scientific determination of experimental stage intervals. By comparing similarity with historical experimental trajectories and correcting stage intervals, combined with the verification patterns of historical experiments, the accuracy and reliability of stage determination are improved. By accurately identifying atypical stages, the shortcomings of existing technologies in identifying abnormal evolution states are compensated, enabling comprehensive determination of experimental states. Finally, through comprehensive determination of multiple results, the specific stage of the experiment is obtained, providing a scientific, accurate, and comprehensive analytical basis for the formal identification of experimental stages, changing the current situation where existing technologies rely on human experience to judge experimental stages, and improving the objectivity and intelligence level of experimental stage identification.

[0034] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1: Based on the current specific stage, determine the corresponding stage identifier; determine the confidence level of the current stage identification based on one or more of the consistency of various analysis results, similarity comparison results, and clarity of evolutionary trajectory; determine the entry and exit time points of the current stage based on boundary candidate points to obtain stage boundary information, specifically including: determining the stage identifier, pre-configuring a unique and fixed stage identifier for each specific stage at the experimental semantic level, which corresponds one-to-one with the initial preparation stage, change or reaction stage, convergence stage, abnormal stage, termination stage, and atypical stage, based on the final judgment of the current stage. In the specific stages of the experiment, the corresponding preset stage identifiers are directly matched and retrieved to determine the stage identifiers. Each stage identifier maintains a unique association with the specific experimental stage, with no duplication or omission. The confidence level of the current stage identification is comprehensively calculated. The confidence level calculation focuses on three core indicators: the consistency of various analysis results, the similarity value of similarity comparison, and the clarity of the evolutionary trajectory. Weighting coefficients are preset for each of these three indicators, based on extensive experimental verification results. The weighting coefficient for the consistency of various analysis results is set to 0.4, the weighting coefficient for the similarity value of similarity comparison is set to 0.3, and the weighting coefficient for the clarity of the evolutionary trajectory is set to... Set to 0.3, with the sum of the three weighting coefficients being 1, each indicator is quantified and assigned a value. The consistency of each analysis result is assigned a value based on the matching degree of the stage interval judgment result, the atypical stage identification result, and the feature data analysis result: a perfect match is assigned a value of 1, a basic match is assigned a value of 0.8, a partial match is assigned a value of 0.5, and no match is assigned a value of 0. The similarity value for similarity comparison is directly used as the calculated value, with a value range of 0 to 1. The clarity of the evolutionary trajectory is assigned a value based on the continuity of the trajectory and the completeness of the feature points: a continuous trajectory with no missing feature points is assigned a value of 1, and a basically continuous trajectory with a few missing feature points is assigned a value of 1. 0.7 is assigned a value of 0.4 for discontinuous trajectories with partially missing feature points, and 0 for broken trajectories with a large number of missing feature points. The specific calculation method for confidence is as follows: the consistency value of each analysis result is multiplied by its weight coefficient, plus the similarity value of the similarity comparison is multiplied by its weight coefficient, plus the clarity value of the evolution trajectory is multiplied by its weight coefficient. This calculation yields a confidence value between 0 and 1. The closer the value is to 1, the higher the reliability of the identification result at the current stage. In some implementations, weight coefficients can be set for each indicator and weighted calculation can be performed. The above quantification values ​​are exemplary assignment methods and do not constitute a limitation.

[0035] Finally, the stage boundary information is determined to obtain the stage entry time point and exit time point. Based on the stage evolution boundary candidate points as the core basis, and combined with the evolution characteristics of the current specific stage, the time point that matches the current stage entry characteristics is selected from the boundary candidate points and determined as the current stage entry time point. For the current stage exit time point, if the experiment is still in the current stage and no stage switch has occurred, the latest experimental data acquisition time point is used as the temporary marker of the exit time point. At the same time, the evolution of the experimental state is continuously monitored. After the boundary candidate point of stage switch is identified, the point is updated as the official exit time point. If the experiment has completed the current stage and entered the next stage, the boundary candidate point of stage switch is directly determined as the current stage exit time point. The determined stage entry time point and exit time point are integrated to form complete stage boundary information.

[0036] Step 4.2 combines the stage identifier, stage confidence level, and stage boundary information to obtain a structured experimental stage identification result. Specifically, this includes: verifying the stage identifier, stage confidence level, and stage boundary information; checking whether the stage identifier matches the current specific stage; verifying whether there are data errors in the calculation process of the stage confidence level; and verifying whether the time points in the stage boundary information are within the experimental time range and conform to the experimental evolution law. If any verification anomalies are found, return to the previous step to reconfirm the relevant information. If no anomalies are found, proceed to the subsequent information combination stage. Standardize the format of the three types of information that have passed verification. The stage identifier is written according to the preset coding rules to ensure its uniqueness and readability. The stage confidence level is standardized to two decimal places, and the calculation basis of the confidence level is indicated. The stage boundary information clearly indicates the entry and exit time points. For the time of exit, if it is a temporary marker, a clear temporary marker is added to distinguish between the official time point and the temporary time point. The three types of information are combined and encapsulated according to the preset structured data format. The structured data format is based on the principle of clearly presenting the core information of the experimental stage identification and facilitating the dynamic planning system of the experimental path to parse and read it. The stage identifier is used as the core identifier item of the structured result, the stage confidence level is used as the reliability assessment item, and the stage boundary information is used as the time feature item. The three items of information are arranged and combined in a fixed order of core identifier item, reliability assessment item, and time feature item. At the same time, a corresponding field name is added to each item of information to ensure that the fields correspond one-to-one with the information content, without misalignment or confusion. After the combination and encapsulation are completed, a complete structured experimental stage identification result is obtained. This result can present all the core information of the experimental stage identification in a complete, clear and standardized manner.

[0037] In this embodiment of the invention, by configuring a unique stage identifier for each experimental stage, standardized identification of experimental stages is achieved, solving the problems of vague descriptions and lack of unified identifiers in existing technologies, facilitating rapid identification and parsing by subsequent systems. By using a multi-indicator weighted comprehensive calculation of stage confidence, the limitations of existing technologies in lacking stage identification reliability assessment are overcome, objectively and quantitatively reflecting the reliability of stage identification results and providing a credible reference for experimental decision-making. By accurately screening boundary candidate points to determine stage entry and exit time points, the time boundaries of experimental stages are clarified, solving the problem of ambiguous stage division time in existing technologies, and accurately characterizing the evolution time range of experimental stages. Through the verification, organization, and structured combination of various identification information, standardized, complete, and easily parsed experimental stage identification results are formed, solving the technical deficiency of existing technologies in providing structured input for dynamic planning of experimental paths, providing a standardized and structured core basis for dynamic planning of experimental paths and dynamic adjustment of experimental operations, while further reducing the dependence of experimental stage identification on human experience and improving the standardization and intelligence level of experimental stage identification.

[0038] In a preferred embodiment of the present invention, step 4 above may include: Step 4.3: Based on the state characteristics, determine whether the conditions for the initial preparation stage are met. The initial preparation stage corresponds to the state interval from the start of the experiment to the stabilization of conditions. The state characteristics are characterized by a trend feature close to zero and a rate of change feature lower than a preset first threshold. If met, it is determined that the current stage is the initial preparation stage. Specifically, this includes: retrieving the real-time trend feature and rate of change feature of the experimental process, and retrieving the preset first threshold. This threshold is a critical value of the rate of change feature obtained by statistical analysis of initial preparation stage data from a large number of similar experiments. It is the maximum reasonable value of the rate of change during the experiment from start to stabilization. The trend feature is numerically verified to determine whether it is close to zero. The criterion is that the absolute value of the slope of the trend feature is less than the preset trend critical value. This critical value is a reasonable fluctuation range where the slope value approaches zero, determined by the stability requirements of the experimental state evolution. The rate of change feature is numerically compared to determine whether it is lower than the preset first threshold, i.e., the value of the real-time rate of change feature is less than the specific value of the first threshold. If the trend feature meets the criteria of being close to zero and the rate of change feature simultaneously meets the condition of being below the first threshold, then the current experiment is determined to meet the conditions of the initial preparation stage, and the experiment is currently in the initial preparation stage; if any feature does not meet the corresponding condition, then the current experiment is determined not to meet the conditions of the initial preparation stage, and the reaction stage determination process begins.

[0039] Step 4.4: If the initial preparation stage conditions are not met, then determine whether the reaction stage conditions are met. The reaction stage corresponds to the state interval where the state continues to change after the experimental conditions are triggered. The state characteristics are manifested by the absolute value of the change trend characteristics exceeding the preset second threshold, and the change rate characteristics continuously exceeding the preset third threshold. If met, then determine that the current state is in the change or reaction stage. Specifically, after determining that the initial preparation stage conditions are not met, retrieve the real-time change trend characteristics and change rate characteristics of the experimental process, and simultaneously retrieve the preset second and third thresholds. The second threshold is the critical absolute value of the change trend characteristics, which is the minimum value of the trend characteristics for the experimental state to enter the continuous change stage. The third threshold is the critical value of the change rate characteristics, which is the minimum value of the rate characteristics for the continuous change of the experimental state. Both thresholds are statistically set according to the evolution law of the reaction stage of similar experiments. Calculate the absolute value of the change trend characteristics by directly taking the positive value of the slope value of the change trend characteristics, and then comparing the absolute value with the second threshold to determine whether it exceeds the second threshold, that is, the specific value of the absolute value of the change trend characteristics greater than the second threshold. Next, the rate of change characteristics are monitored for continuous time windows. The values ​​of the rate of change characteristics within a preset continuous time range are statistically analyzed to determine whether they are consistently higher than the third threshold. The criterion for determination is that all values ​​of the rate of change characteristics within the continuous time window are greater than the specific value of the third threshold, and there is no moment when they are lower than the threshold. If the absolute value of the trend characteristic meets the condition of exceeding the second threshold, and the rate of change characteristic also meets the condition of consistently being higher than the third threshold, then the current experiment is determined to meet the conditions of the change or reaction stage, and the experiment is determined to be in the change or reaction stage. If any characteristic does not meet the corresponding condition, then the current experiment is determined not to meet the conditions of the change or reaction stage, and the convergence stage determination process begins.

[0040] Step 4.5: If the conditions for the reaction stage are not met, then determine whether the conditions for the convergence stage are met. The convergence stage corresponds to a state interval where the state parameters tend to be stable. The state characteristics are that the trend characteristic approaches zero, the rate of change characteristic is lower than the preset fourth threshold, and the stability characteristic is lower than the preset fifth threshold. If these conditions are met, then it is determined that the current stage is the convergence stage. Specifically, after determining that the conditions for the reaction stage are not met, the real-time trend characteristic, rate of change characteristic, and stability characteristic of the experimental process are retrieved. At the same time, the preset fourth and fifth thresholds are retrieved. The fourth threshold is the critical value of the rate of change characteristic, which is the maximum value of the rate characteristic when the experimental state tends to be stable. The fifth threshold is the critical value of the stability characteristic, which is the minimum value of the stability characteristic when the experimental state reaches convergence. The two thresholds are set according to the statistical data of the stable convergence stage of similar experiments. The trend characteristic is numerically verified to determine whether it approaches zero. The judgment criteria are consistent, that is, the absolute value of the slope of the trend characteristic is less than the preset trend critical value. Next, the rate of change feature is compared with the fourth threshold to determine if it is below the fourth threshold, i.e., the value of the real-time rate of change feature is less than the specific value of the fourth threshold. Finally, the stability feature is compared with the fifth threshold to determine if it is below the fifth threshold, i.e., the variance value of the real-time stability feature is greater than the specific value of the fifth threshold. The larger the variance value, the worse the stability of the experimental state in that time period. If the trend feature meets the criterion of approaching zero, the rate of change feature meets the condition of being below the fourth threshold, and the stability feature also meets the condition of being below the fifth threshold, then the current experiment is comprehensively judged to meet the conditions of the stable or convergent stage, and the experiment is determined to be in the stable or convergent stage. If any feature does not meet the corresponding condition, the current experiment is judged not to meet the conditions of the convergent stage, and the abnormal stage judgment process is entered.

[0041] Step 4.6: If the convergence phase conditions are not met, then determine whether the abnormal phase conditions are met. The abnormal phase corresponds to a state range where the state change deviates from the expected evolution trajectory. The state characteristics are that the similarity between the evolution trajectory and the historical normal phase evolution trajectory is lower than the preset sixth threshold, and the change pattern cannot match any known phase pattern. If the conditions are met, then it is determined that the current stage is abnormal. Specifically, after determining that the stable or convergence phase conditions are not met, retrieve the similarity results between the current experimental evolution trajectory and the historical normal phase evolution trajectory, and simultaneously retrieve the preset sixth threshold. This threshold is the critical value of trajectory similarity and is the minimum similarity value for determining whether the experimental state evolution is normal. It is statistically set based on the trajectory matching rules of historical normal experiments. Compare the similarity results with the sixth threshold to determine whether it is lower than the sixth threshold. That is, the similarity values ​​between the current experimental evolution trajectory and all historical normal phase evolution trajectories are less than the specific value of the sixth threshold, and no historical trajectory has a similarity value that reaches the threshold. Next, the current experimental state characteristic change pattern is matched and verified. The real-time change patterns of change trend characteristics, change rate characteristics, and stability characteristics are matched one by one with the characteristic change patterns of all known stages in the historical experiments to determine whether they can match any known stage pattern. The judgment criteria are that the consistency between the pattern, amplitude, and trend of the characteristic change and the known stage pattern reaches the preset matching standard. If the evolution trajectory similarity result meets the condition of being below the sixth threshold, and the state characteristic change pattern also meets the condition of not being able to match any known stage pattern, then the current experiment is judged to meet the conditions of an abnormal stage, and the experiment is determined to be in an abnormal stage. If any condition is not met, the current experiment is judged not to meet the conditions of an abnormal stage, and the termination stage judgment process is entered.

[0042] Step 4.7: If the conditions for an abnormal stage are not met, the current stage is determined to be the termination stage. The termination stage corresponds to the state interval of the experiment's end. The state characteristics are that the experimental operation instructions have been executed and all state parameters remain stable within a preset time. Specifically, after determining that the conditions for an abnormal stage are not met, the execution status of the operation instructions in the experimental control system is retrieved first to check whether all preset experimental operation instructions have been executed. The judgment standard is that all operation instructions set in the experimental process, such as feeding, reaction, stirring, and detection, have been executed and there are no unfinished instructions to be executed. Real-time data of state parameters in all dimensions during the experiment are retrieved, and the values ​​of all state parameters within the preset stable monitoring time range are verified to determine whether they remain stable within that time range. The judgment standard is that the fluctuation range of the values ​​of all state parameters within that time range is less than the preset parameter stability threshold, and there is no obvious increase, decrease, or sudden change. The experimental process has ended, and all state parameters simultaneously meet the condition of remaining stable within the preset time. Therefore, the current experiment is determined to be in the termination stage.

[0043] In this embodiment of the invention, by setting clear and quantifiable feature judgment conditions for each experimental semantic stage, the limitations of existing technologies that rely on subjective judgment of experimental stages based on human experience are overcome, achieving standardized and quantifiable judgment of experimental stages and improving the objectivity and accuracy of stage judgment. By judging in sequence according to the logic of initial preparation, change or reaction, stability or convergence, anomaly, and termination, a progressive stage judgment system is formed, ensuring that experimental stage judgment is complete and without repetition, solving the problems of chaotic stage judgment logic and ambiguous boundaries in existing technologies. By combining the judgment method of comparing feature values ​​with preset thresholds, matching evolution patterns with historical data, and verifying operation instructions with parameter status, the system achieves comprehensive judgment of each stage. The accurate capture of core evolutionary features can effectively distinguish the state differences of different experimental stages and avoid stage misjudgment. By setting dual judgment conditions of trajectory similarity and pattern matching separately for abnormal stages, the accuracy of abnormal stage identification is improved, which solves the technical defects of existing technologies that cannot identify abnormal experimental states in a timely and accurate manner. The overall judgment process relies on the core feature data of the experimental state and objective judgment standards, completely eliminating the dependence on human experience and realizing fully automatic intelligent judgment of experimental stages. This provides accurate and reliable stage basis for the generation of subsequent structured recognition results, and at the same time provides clear experimental semantic stage reference for dynamic planning of experimental paths, supporting the efficient operation of the intelligent automated experimental platform.

[0044] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1: Based on the current experimental semantic stage, extract the corresponding stage identifier; obtain the confidence value of the current stage from the experimental stage identification results; based on the boundary candidate points and stage entry and exit time points, organize the stage entry boundary information and stage exit boundary information, specifically including: carrying out the stage identifier extraction work, according to the final determined current experimental semantic stage, extract the stage identifier that uniquely corresponds to it from the preset experimental stage identifier library. The stage identifier library has fixed and unique identifiers configured for the initial preparation stage, change or reaction stage, stable or convergence stage, abnormal stage, and termination stage respectively. The extraction process strictly matches the correspondence between the experimental semantic stage and the identifier to ensure that there are no mismatches or omissions in the identifiers. Directly retrieve the confidence value of the current stage from the experimental stage identification results. This value is a value between 0 and 1 obtained by weighted calculation of multiple indicators. The retrieval process retains the confidence value of the current stage. The original precision of the numerical values ​​is preserved without additional rounding to ensure that the confidence scores accurately reflect the reliability of the stage identification results. Finally, the stage entry and exit boundary information is organized. Based on the stage evolution boundary candidate points and combined with the determined current stage entry and exit time points, the boundary information is standardized. The stage entry boundary information clearly marks the specific time point of stage entry, the core state feature data corresponding to that time point, and the matching basis of the boundary candidate points. The stage exit boundary information clearly marks the specific time point of stage exit. If it is a temporary exit time point for which the stage switch has not been completed, a temporary identifier and the evolution monitoring status of the current state must be marked simultaneously. At the same time, the time points corresponding to the entry and exit boundary information are converted to the time format consistent with the experimental data collection to ensure the uniformity and readability of the time information. After completion, complete and standardized stage boundary information is formed.

[0045] Step 5.2: Encapsulate the stage identifier, stage confidence level, and stage entry and exit boundary information according to a preset data structure to obtain a standardized structured data package. Specifically, this includes: retrieving a preset structured data encapsulation template, which is designed to meet the data parsing requirements of the experimental path dynamic planning system. The template specifies the data field names, field order, data types, and data precision requirements. The core fields include the stage identifier field, stage confidence level field, stage entry boundary information field, and stage exit boundary information field. Each field corresponds to the matching information type and data format. Fill the extracted stage identifier, the obtained confidence level value, and the organized stage entry and exit boundary information into the corresponding positions according to the field requirements of the preset template, and perform format validation on the filled information. The process involves standardization, where stage identifiers are written according to the character format required by the template, stage confidence scores retain a specified number of decimal places according to the template's preset precision requirements, and stage boundary information is organized according to the template's text format, including time points, feature data, and matching criteria. This ensures that the information content and data format of each field fully comply with the template requirements. After completing the information entry and standardization, the correlation of each field's information is verified. This involves checking whether the experimental semantic stages corresponding to the stage identifiers and boundary information are consistent, and whether the correlation between the confidence scores and the stage identification results matches. After verification, all field information is uniformly encapsulated according to the overall structure of the preset template to form a standardized structured data package. This data package can be directly recognized and parsed by the experimental path dynamic planning system without format compatibility issues.

[0046] Step 5.3 involves outputting structured data packets in real time via the output interface for external modules to use. Specifically, this includes: debugging the link between the automatic identification system and the dynamic planning of the experimental path during the experimental phase to ensure stable communication and data transmission without delay or loss; confirming that the interface's transmission protocol and data parsing rules match the format of the structured data packets to meet real-time data transmission requirements; importing standardized structured data packets into the output interface's transmission queue; and transmitting them in real time according to the state evolution rhythm of the experimental process. If the experimental state is in a continuous evolution process, when stage boundary information is updated or confidence values ​​are adjusted, the output interface will synchronously update the structured data packets in the transmission queue, achieving dynamic real-time output of data packets. If the experiment is in a fixed stage and the state does not change significantly... The output interface will repeatedly transmit data packets at preset time intervals to ensure that the experimental path dynamic planning system can continuously obtain the latest experimental stage identification results. During the data packet transmission process, a transmission check code is added to each structured data packet. The check code is calculated by summing the core feature values ​​of each field in the data packet and then taking the remainder of the summation result with a fixed number of bits. After receiving the data packet, the experimental path dynamic planning system verifies the data integrity by calculating the check code. If the verification is correct, the data packet is used as input for subsequent experimental path dynamic planning. If the verification finds that the data is missing or incorrect, a retransmission instruction is fed back through the interface. After receiving the instruction, the output interface immediately retransmits the corresponding structured data packet to ensure that the experimental path dynamic planning system can obtain complete and accurate experimental stage identification structured data.

[0047] In this embodiment of the invention, by standardizing the extraction and organization of stage identifiers, confidence scores, and stage boundary information, the integrity and accuracy of core information for experimental stage identification are ensured. This solves the problem of scattered experimental stage-related information and lack of unified organization standards in existing technologies, laying a standardized information foundation for structured encapsulation. By standardizing the encapsulation of various types of information according to a preset data structure, a structured data package that can be directly parsed by dynamic planning of experimental paths is formed. This overcomes the technical limitations of existing technologies where experimental stage identification results lack standardized output formats and cannot be effectively integrated with planning, achieving data compatibility and interoperability between different systems. The real-time output of structured data packages is completed through an output interface with a verification mechanism, ensuring the stability, real-time performance, and integrity of data transmission. It also enables dynamic updates of data packages based on the evolution of experimental states, solving the problems of delayed transmission of experimental state information and easy data loss in existing technologies. This allows the dynamic planning system of experimental paths to continuously obtain the latest and most accurate experimental stage identification results, promoting the automation and intelligence of the experimental process, reducing the degree of human involvement in data transmission and format conversion, and improving the overall operating efficiency of the intelligent automated experimental platform.

[0048] like Figure 2As shown, embodiments of the present invention also provide an artificial intelligence-based automatic identification system for experimental phases, comprising: The acquisition module is used to acquire multidimensional state data during the experiment. The multidimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operation state data, experimental process images, video data, and microscopic imaging and detection data. The extraction module is used to perform joint modeling and time series analysis on multidimensional state data sequences to extract state feature data that reflects the evolution of the experimental process. The analysis module is used to analyze state features based on state feature data, through multimodal state feature joint analysis, analysis of change trends and rates of change in the time dimension, determination of stage evolution trajectory and stage boundaries, pattern matching and similarity analysis of historical experimental stages, and identification of abnormal states and atypical stages, to obtain analysis results. The identification module is used to automatically identify the current experimental stage based on the analysis results and generate experimental stage identification results. The experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, change or reaction stage, stable or convergence stage, abnormal stage and termination stage. The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An automatic identification method for experimental stages based on artificial intelligence, characterized in that, The method includes: Acquire multidimensional state data during the experiment; multidimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operation state data, experimental process images, video data, microscopic imaging and detection data; By jointly modeling and analyzing the multidimensional state data sequence, state feature data reflecting the evolution of the experimental process can be extracted. Based on state feature data, state features are analyzed through multimodal state feature joint analysis, time-dimensional trend and rate of change analysis, stage evolution trajectory and stage boundary determination, historical experimental stage pattern matching and similarity analysis, and identification of abnormal states and atypical stages, and the analysis results are obtained. Based on the analysis results, the current experimental stage is automatically identified, and an experimental stage identification result is generated. The experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, the change or reaction stage, the stable or convergence stage, the abnormal stage, and the termination stage.

2. The automatic identification method for experimental stages based on artificial intelligence according to claim 1, characterized in that, Joint modeling and time series analysis of multidimensional state data sequences were performed to extract state feature data reflecting the evolution of the experimental process, including: The acquired multidimensional state data is preprocessed to obtain standardized multidimensional time series data. Temporal features of each dimension are extracted from the standardized multidimensional time series data. These temporal features include statistical features, rate of change features, and frequency domain features. The extracted temporal features from various dimensions are fused in a multimodal manner to construct a multidimensional temporal state vector; A time series analysis is performed on the multidimensional time series state vector to calculate the state characteristics that reflect the evolution of the experimental process. The state characteristics include the trend characteristics, the rate of change characteristics, and the stability characteristics. The trend characteristics are obtained by fitting the slope of the time series data, the rate of change characteristics are obtained by calculating the change in adjacent time moments, and the stability characteristics are obtained by calculating the variance within the sliding window.

3. The automatic identification method for experimental stages based on artificial intelligence according to claim 2, characterized in that, Based on state feature data, the state features are analyzed through multimodal state feature joint analysis, time-dimensional trend and rate of change analysis, stage evolution trajectory and stage boundary determination, historical experimental stage pattern matching and similarity analysis, and identification of abnormal states and atypical stages. The analysis results include: Based on state characteristics, a multi-dimensional joint analysis of change trend characteristics, change rate characteristics, and stability characteristics is conducted to obtain the evolution direction and evolution intensity of the experimental state. Based on the evolution direction and intensity, the change trend characteristics and change rate characteristics are continuously tracked in the time dimension to identify the inflection points and abrupt change points of the state characteristics as candidate boundary points for stage evolution. Based on the boundary candidate points, and combining the characteristics of change trend and stability, the evolution trajectory of the experimental state in the feature space is constructed, and the curvature and direction changes of the evolution trajectory are analyzed to obtain the stage interval determination result of the current state. The current experimental evolution trajectory is compared with the evolution trajectory of the marked stages in the historical experiment. If the similarity exceeds the preset threshold, the stage interval is corrected by referring to the division boundary of the historical stage to obtain the corrected current stage judgment result. During the similarity comparison process, if the similarity between the current experimental evolution trajectory and the evolution trajectory of all historical normal stages is lower than the preset threshold, and the change pattern of the state characteristics deviates from the preset normal evolution range, then the current state is identified as an atypical stage, and the identification result of the atypical stage is obtained. By combining the results of the phase interval determination and the results of the atypical phase identification, the specific phase of the experiment is finally determined automatically.

4. The automatic identification method for experimental stages based on artificial intelligence according to claim 3, characterized in that, Based on the analysis results, the current experimental stage is automatically identified, and experimental stage identification results are generated, including: Based on the current specific stage, determine the corresponding stage identifier; based on one or more of the consistency of various analysis results, similarity comparison results, and clarity of evolution trajectory, determine the confidence level of the current stage identification; based on the boundary candidate points, determine the entry time point and exit time point of the current stage to obtain the stage boundary information; By combining stage identifiers, stage confidence levels, and stage boundary information, a structured experimental stage identification result is obtained.

5. The automatic identification method for experimental stages based on artificial intelligence according to claim 4, characterized in that, The experimental phase is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation phase, the change or reaction phase, the stabilization or convergence phase, the anomaly phase, and the termination phase, including: Based on the state characteristics, determine whether the conditions for the initial preparation stage are met. The initial preparation stage corresponds to the state interval from the start of the experiment to the stable state. The state characteristics are characterized by a trend of change close to zero and a rate of change lower than a preset first threshold. If the conditions are met, it is determined that the experiment is currently in the initial preparation stage. If the conditions for the initial preparation stage are not met, then it is determined whether the conditions for the reaction stage are met. The reaction stage corresponds to the state range in which the state continues to change after the experimental conditions are triggered. The state characteristics are that the absolute value of the change trend characteristics exceeds the preset second threshold, and the change rate characteristics are continuously higher than the preset third threshold. If the conditions are met, it is determined that the current state is in the change or reaction stage. If the conditions for the reaction stage are not met, then it is determined whether the conditions for the convergence stage are met. The convergence stage corresponds to a state range where the state parameters tend to be stable. The state characteristics are that the trend of change is close to zero, the rate of change is lower than the preset fourth threshold, and the stability is lower than the preset fifth threshold. If these conditions are met, it is determined that the current stage is the convergence stage. If the conditions for the convergence phase are not met, then the conditions for the abnormal phase are determined. The abnormal phase corresponds to a state range in which the state change deviates from the expected evolution trajectory. The state characteristics are that the similarity between the evolution trajectory and the historical normal phase evolution trajectory is lower than the preset sixth threshold, and the change pattern cannot match any known phase pattern. If the conditions are met, then it is determined that the current stage is abnormal. If the conditions for an abnormal phase are not met, the current state is determined to be in the termination phase. The termination phase corresponds to the state interval where the experiment ends. The state characteristics are that the experimental operation instructions have been executed and all state parameters remain stable within a preset time.

6. The automatic identification method for experimental stages based on artificial intelligence according to claim 5, characterized in that, Based on the analysis results, the current experimental stage is automatically identified, and an experimental stage identification result is generated. An experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, the change or reaction stage, the stabilization or convergence stage, the anomaly stage, and the termination stage, including: Based on the current experimental semantic stage, extract the corresponding stage identifier; obtain the confidence value of the current stage from the experimental stage identification results; and organize the stage entry boundary information and stage exit boundary information according to the boundary candidate points and stage entry and exit time points. The stage identifier, stage confidence level, and stage entry and exit boundary information are encapsulated according to a preset data structure to obtain a standardized structured data packet; The structured data packets are output in real time through the output interface for external modules to use.

7. An artificial intelligence-based automatic identification system for experimental stages, the system implementing the method as described in any one of claims 1 to 6, characterized in that, include: The acquisition module is used to acquire multidimensional state data during the experiment. The multidimensional state data includes one or more of the following: experimental object state data, experimental environment state data, experimental device operation state data, experimental process images, video data, and microscopic imaging and detection data. The extraction module is used to perform joint modeling and time series analysis on multidimensional state data sequences to extract state feature data that reflects the evolution of the experimental process. The analysis module is used to analyze state features based on state feature data, through multimodal state feature joint analysis, analysis of change trends and rates of change in the time dimension, determination of stage evolution trajectory and stage boundaries, pattern matching and similarity analysis of historical experimental stages, and identification of abnormal states and atypical stages, to obtain analysis results. The identification module is used to automatically identify the current experimental stage based on the analysis results and generate experimental stage identification results. The experimental stage is a phased abstraction of the experimental process at the experimental semantic level, including but not limited to the initial preparation stage, change or reaction stage, stable or convergence stage, abnormal stage and termination stage.

8. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images