Intelligent laboratory and automatic laboratory sample distribution system

Abstract

The invention relates to the technical field of laboratory automation, in particular to an intelligent laboratory and an automatic laboratory sample distribution system, and provides the following scheme that dynamic optimization of experimental steps and parameters is achieved by combining experimental equipment control, real-time data analysis, machine learning optimization and automatic experimental method updating. The system can automatically acquire and update experimental method information, provides experimental scheme recommendation and intelligent scheduling experimental equipment by using deep learning and a self-attention mechanism, and adjusts experimental parameters based on real-time feedback, so that the automatic management level of a laboratory is improved, the experimental efficiency is optimized, and human intervention is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of laboratory automation, and in particular to an intelligent laboratory and an automatic distribution system for laboratory samples. Background Art

[0002] Current laboratory automation technology mainly relies on manual intervention and fixed experimental methods, which makes the experimental steps inflexible and inefficient, and difficult to adapt to complex and changing experimental needs. Existing experimental equipment control systems usually lack the ability to combine with real-time data feedback during the experiment, and cannot automatically optimize experimental parameters according to data changes in the experiment, which in turn affects the accuracy and efficiency of the experiment. In addition, the process of updating experimental methods usually relies on manual modification and manual reference, which is slow to update and prone to lag, resulting in insufficient timeliness and comprehensiveness of experimental method information. Therefore, there is an urgent need for a system that can automatically acquire and update experimental methods, intelligently optimize experimental processes, and efficiently dispatch experimental equipment to improve the automation level and experimental efficiency of the laboratory.

[0003] For example, the Chinese patent application with publication number CN118982216A discloses a management method and device for a smart laboratory based on artificial intelligence. The method realizes comprehensive monitoring of the laboratory environment and equipment by real-time collection of the laboratory's environmental data, equipment status and personnel operation behavior. The system efficiently identifies the received data type based on the association relationship, and analyzes and predicts it through AI algorithms to ensure the smooth progress of the experimental process. After that, the system intelligently dispatches resources according to experimental requirements and equipment availability, optimizes the allocation of experimental equipment, and monitors experimental equipment and personnel operations in real time during the experiment, provides feedback and handles abnormal situations in a timely manner to ensure the safety and effectiveness of the experiment. Through continuous machine learning and data accumulation, the invention can continuously optimize laboratory resource management, equipment scheduling and experimental scheduling, and achieve the beneficial technical effect of efficient management of the overall laboratory.

[0004] The above patents all have the problems raised by this background technology: they rely on manual intervention in updating experimental methods and lack intelligent and automated update mechanisms, resulting in low experimental efficiency and delayed updating of experimental plans. To solve the above problems, this application designs an intelligent laboratory and an automated laboratory sample distribution system. Summary of the invention

[0005] The technical problem to be solved by the present invention is to address the deficiencies of the prior art and provide an intelligent laboratory and an automated laboratory sample distribution system. By combining experimental equipment control, real-time data analysis, machine learning optimization, and automated experimental method updates, dynamic optimization of experimental steps and parameters is achieved. The system can automatically acquire and update experimental method information, use deep learning and self-attention mechanisms to provide experimental scheme recommendations, intelligently schedule experimental equipment, and adjust experimental parameters based on real-time feedback, thereby improving the level of laboratory automation management, optimizing experimental efficiency, and reducing human intervention, and has significant innovation and practical value.

[0006] To achieve the above object, the present invention provides the following technical solutions: An automated laboratory sample distribution system, the system comprising an experimental database, an experimental equipment control module, an intelligent optimization module and a database update module; The experimental database is used to store experimental method information; The experimental equipment control module is used to obtain sample information by scanning the barcode on the sample container, match the sample information with the experimental method information stored in the experimental database, and schedule the experimental equipment to execute the experimental steps according to the matching result; The intelligent optimization module is configured with a step optimization strategy, which adjusts the experimental parameters through a machine learning method based on data feedback during the experiment; The database update module is configured with an automatic update strategy, which is used to obtain scientific research reports from the network, parse the scientific research reports through natural language processing, determine whether there are updated contents, and store the updated experimental method information in the experimental database.

[0007] The intelligent optimization module is used to optimize the experimental parameters and steps according to the data collected in real time during the experiment. The intelligent optimization module specifically includes: A data analysis unit, used to collect relevant sensor data during the experiment, analyze the relevant sensor data, and extract data features; A parameter optimization unit, used to process the data features through a machine learning algorithm and output optimization parameters; A feedback adjustment unit, used for providing feedback to the experimental equipment control module according to the optimization parameters; The step optimization strategy includes feature extraction logic and feature analysis logic; The feature extraction logic is configured in the data analysis unit to simplify the sensor data and extract data features; The feature analysis logic is configured in the parameter optimization unit, and is used to construct a data analysis network, use the data features as input parameters of the data analysis network, analyze the input parameters through the data analysis network, and output optimized parameters.

[0008] The feature extraction logic includes: Performing tensor decomposition on the relevant sensor data, calculating a kurtosis value according to the decomposition result, and windowing the relevant sensor data according to the kurtosis value; The sensor data in each window is converted into a time-frequency domain signal by wavelet transform, and a high-frequency component and a low-frequency component of the time-frequency domain signal are obtained; Calculate the energy distribution of high-frequency components and low-frequency components respectively, analyze the instantaneous change of the energy distribution of high-frequency components, calculate the energy fluctuation rate, model the long-term trend of the energy distribution of low-frequency components, and calculate the frequency smoothing rate; The energy fluctuation rate and the frequency smoothing rate are correlated, and the correlation value is calculated. The priority of each window is calculated according to the correlation value, and the window data with the highest priority is output as the data feature. The calculation formula of the window priority is: , in, represents the window priority, T represents the total time length of the window, and t represents the unit time length of the window. Represents the time-frequency domain signal of the high-frequency component corresponding to the t-th time length of the window, represents the modulus of a complex number, represents the average energy of the high-frequency components in the window, f represents the frequency, Represents the time-frequency domain signal of the low-frequency component corresponding to the t-th time length of the window, It represents the overall smooth trend of the window, and k represents the kurtosis value corresponding to the window.

[0009] The data analysis network comprises: The feature fusion layer is used to fuse data features from different sources through the self-attention mechanism and calculate the fusion features; A processing layer, used for mapping the fused features into low-dimensional features through a multi-layer perceptron network, performing importance evaluation on the low-dimensional features according to a random forest algorithm, calculating information gain of the low-dimensional features, and identifying key features according to the information gain; Compare the key features with data of successful cases of historical experiments to analyze whether there are differences, and calculate feedback signals based on the differences; The output layer is used to adjust the experimental parameters and modify the experimental plan according to the feedback signal, and output the final experimental optimization configuration.

[0010] The database updating module comprises: Data acquisition unit, used to obtain the latest literature from designated scientific research websites, academic data and technical report platforms; The text parsing unit is used to parse the latest literature obtained through natural language processing and extract experimental method information; A database updating unit is used to determine whether the newly acquired experimental method information contains new content, and if there is a significant difference, update it to the experimental database; The automatic update strategy includes text parsing logic and content update logic; The text parsing logic is configured in the text parsing unit, and is used to identify and extract experimental methods, experimental steps and key parameters from the latest literature, and to normalize the extracted results; The content update logic is configured in the database update unit, and is used to compare the newly acquired experimental method information with the existing experimental method information in the experimental database to determine whether there is a significant change. If so, the newly acquired experimental method information is updated to the experimental database.

[0011] The text parsing logic includes: Preprocess the latest literature and convert it into structured text data; Constructing a keyword extraction network, taking the structured text data as input parameters of the keyword extraction network, processing the input parameters through the keyword extraction network, and outputting a keyword sequence; Analyze the keyword sequence according to the dependency parsing technology to construct a semantic relationship graph, wherein the semantic relationship graph includes nodes and edges, the nodes represent experimental entities, and the edges represent entity-related actions; The nodes and edges in the semantic relationship graph are traversed to extract the experimental methods, experimental steps and key parameters.

[0012] The keyword extraction network includes: The vector representation layer is used to obtain the segmented text of the input parameter, process the segmented text, and encode it in combination with the word position information to obtain the word vector sequence; The feature representation layer is used to extract the global features of the text context of the word vector sequence based on the bidirectional gated recurrent unit, perform sentence-level feature representation on the global features of the text context, and obtain deep semantic information; The feature classification layer is used to extract the core features of deep semantic information, calculate the confidence probability of the core feature category through the Softmax classifier, and take the experimental category corresponding to the highest confidence probability as the experimental category of the word vector corresponding to the core feature. The calculation formula of the confidence probability of the core feature category is: , Among them, P represents the confidence probability of the core feature category, Softmax[•] represents the classifier, W represents the classification matrix, M(•) represents the relationship matrix, S represents the core feature, A represents the diagonal matrix, represents the probability matrix of the core feature corresponding to the category, and b represents the bias; The output layer is used to score the core features according to the experimental category, obtain the score sequence, decode the score sequence, and output the keyword sequence.

[0013] The content update logic includes: Extracting features of the newly acquired experimental method information through a pre-trained model to generate a multi-dimensional feature vector; Calculate the cosine similarity between the experiment name vector in the multidimensional feature vector and the node entity name vector in the experiment database through a dynamic semantic matching algorithm, and generate a candidate method set according to the cosine similarity; Perform deep matching on the experimental steps, conditions and parameters in the candidate method set, and calculate the comprehensive similarity based on the matching results, wherein the deep matching includes experimental step matching, experimental condition matching and experimental parameter matching; The largest comprehensive similarity in the candidate method set is selected, and the comprehensive similarity is compared with the similarity threshold. If it is less than the similarity threshold, the newly acquired experimental method information is updated to the experimental database. If it is greater than or equal to the similarity threshold, the experimental database is not updated.

[0014] The system also includes a recommendation module, which is used to match the target experiment in the experiment database and provide an experimental plan, specifically including: The experimental target identification unit is used to identify the core elements of the experimental requirement input according to the experimental requirement input; An experimental method matching unit, used for matching the core elements with the stored experimental methods in the experimental database; The experimental method recommendation unit is used to output matching experimental methods.

[0015] Smart laboratory, the smart laboratory comprising: Laboratory hardware equipment, used to provide the physical platform required for the experiment, including laboratory benches, sample storage systems, laboratory sample automated distribution systems and related experimental equipment; Laboratory control center, used to manage and schedule experimental equipment, control experimental steps and equipment operation; Data acquisition and transmission equipment, used to collect experimental data in real time and transmit it to the data analysis platform; Data analysis platform, used to process experimental data and provide experimental optimization suggestions, including real-time parameter adjustment and experimental effect evaluation; Experimental plan recommendation equipment is used to provide the best matching experimental plan according to the experimental objectives; Laboratory management equipment, used to manage experimental tasks, resource scheduling and experimental progress tracking.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention can dynamically optimize experimental parameters according to real-time data in the experimental process through the combination of intelligent optimization module and database update module, and update experimental methods in real time through natural language processing technology, which significantly improves the automation level and experimental efficiency of the laboratory. In addition, the present invention can continuously adjust the optimization strategy according to experimental feedback, ensure the accuracy and efficiency of experimental results, reduce manual intervention and human errors, and improve the intelligence level and resource utilization efficiency of the laboratory. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 This is a module schematic diagram of the laboratory sample automatic distribution system according to Example 1 of the present invention; Figure 2 This is a schematic diagram of the optimization process of the enzymatic reaction experiment in Example 1 of the present invention; Figure 3 This is a data analysis network structure diagram of Example 1 of the present invention; Figure 4 This is a keyword extraction network structure diagram of Example 1 of the present invention; Figure 5 This is a schematic diagram of the smart laboratory according to Example 2 of the present invention. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0019] Example 1 See also Figure 1 , an embodiment provided by the present invention: a laboratory sample automatic distribution system, the system includes an experimental database, a recommendation module, an experimental equipment control module, an intelligent optimization module and a database update module; The experimental database is used to store experimental method information; The experimental database includes experimental methods, experimental steps, equipment configuration and material information. All data are stored in a structured manner and matched in vector form. Each experimental record has a corresponding unique identifier for quick retrieval and matching; The recommendation module is used to match the target experiment in the experiment database and provide an experimental plan; The experimental equipment control module is used to obtain sample information by scanning the barcode on the sample container, match the sample information with the experimental method information stored in the experimental database, and schedule the experimental equipment to execute the experimental steps according to the matching result; The intelligent optimization module is configured with a step optimization strategy, which adjusts the experimental parameters through a machine learning method based on data feedback during the experiment; The database update module is configured with an automatic update strategy, which is used to obtain scientific research reports from the network, parse the scientific research reports through natural language processing, determine whether there are updated contents, and store the updated experimental method information in the experimental database.

[0020] The intelligent optimization module is used to optimize the experimental parameters and steps according to the data collected in real time during the experiment. The intelligent optimization module specifically includes: A data analysis unit, used to collect relevant sensor data during the experiment, analyze the relevant sensor data, and extract data features; The unit monitors the key parameters of the experiment in real time by integrating multiple sensors (such as temperature sensor, pressure sensor, pH sensor, optical density sensor, etc.). The sensor data is continuously collected and transmitted to the data analysis unit. The collected raw data is first pre-processed, including denoising, standardization and normalization, to improve the accuracy and reliability of subsequent analysis.

[0021] A parameter optimization unit, used to process the data features through a machine learning algorithm and output optimization parameters; A feedback adjustment unit, used for providing feedback to the experimental equipment control module according to the optimization parameters; Based on the optimized parameter results, the unit will feed back the optimized experimental parameters to the experimental equipment control module and automatically adjust the working state of the experimental equipment. The feedback adjustment unit provides real-time dynamic adjustment during the experiment to ensure that the experiment is carried out according to the optimal solution. For example, if the temperature exceeds the preset range during the experiment, the feedback adjustment unit will pass the new temperature setting to the experimental equipment control module to adjust the output power of the heating equipment. Specifically, the system has a closed-loop control function, that is, at each experimental stage, the data analysis unit, parameter optimization unit and feedback adjustment unit work together to ensure continuous optimization of experimental parameters. Whenever the experimental equipment feeds back new sensor data, the intelligent optimization module will adjust the parameters in real time to adapt to the dynamic changes of the experiment, please refer to Figure 2 , a schematic diagram of the optimization process of an enzymatic reaction experiment according to an embodiment of the present invention, assuming that an enzymatic reaction experiment is being performed, the workflow of the intelligent optimization module is as follows: By monitoring temperature, pH value and reaction rate in real time, the system extracts features of these data, such as the changing trend of reaction rate, temperature fluctuation and pH change rate.

[0022] Through machine learning models, the system predicts the temperature, pH value, and reaction time that are most suitable for the current experiment to maximize enzyme activity and reaction efficiency.

[0023] The optimized parameters (e.g., adjusting the temperature from 37°C to 39°C) are passed to the experimental equipment control module, and the system automatically adjusts the heater to ensure that the experiment is continuously carried out under optimal conditions.

[0024] The step optimization strategy includes feature extraction logic and feature analysis logic; The feature extraction logic is configured in the data analysis unit to simplify the sensor data and extract data features; The feature analysis logic is configured in the parameter optimization unit, and is used to construct a data analysis network, use the data features as input parameters of the data analysis network, analyze the input parameters through the data analysis network, and output optimized parameters.

[0025] The feature extraction logic includes: Performing tensor decomposition on the relevant sensor data, calculating a kurtosis value according to the decomposition result, and windowing the relevant sensor data according to the kurtosis value; Sensor data is usually multidimensional time series data, covering multiple dimensions (such as temperature, pressure, reaction rate, etc.). In order to more effectively extract key information from these data, the tensor decomposition method is first used to reduce the dimension of the data and extract the potential structure. Tensor decomposition decomposes the original data matrix into multiple low-dimensional matrices, so that data features can be represented simultaneously in multiple dimensions, can process high-order data structures, and reveal the correlation and potential patterns between different dimensions; The kurtosis value is a statistic that measures the distribution of the signal. It reflects the sharpness or peakedness of the data distribution. For experimental data, a higher kurtosis value indicates that the signal has stronger "spikes" or "extreme fluctuations". These fluctuations usually represent abnormalities or sudden events in the signal, such as sudden environmental changes, drastic fluctuations during the experiment, etc. Abnormal fluctuations during the experiment are usually short-term and local. By changing the kurtosis value, it is possible to determine which time period in the signal has sudden abnormal fluctuations. During these time periods, the signal fluctuates greatly and the kurtosis value is high, so it is necessary to extract and analyze the data of these parts separately so that these abnormal areas can be focused on in subsequent processing. Windowing technology can adjust the analysis focus of the data according to the kurtosis value to avoid over-processing unimportant parts of the signal. In high kurtosis areas, windowing can help concentrate resources and computing power to optimize the detection efficiency of abnormal events; By performing multi-scale decomposition of the signal through wavelet transform, performing convolution operation on the signal using different wavelet basis functions, generating a time-frequency domain representation, converting the sensor data in each window into a time-frequency domain signal, and obtaining a high-frequency component and a low-frequency component of the time-frequency domain signal; Wavelet transform can effectively reveal the local frequency components of the signal at different time scales, and is particularly suitable for non-stationary signals (such as emergencies during the experiment, equipment fluctuations, etc.). The converted time-frequency domain signal can provide richer signal features, including high-frequency components (rapid changes) and low-frequency components (long-term trends). After the time-frequency domain signal is transformed by wavelet transform, high-frequency components and low-frequency components are obtained. The high-frequency component represents the rapid change of the signal, which is extracted from the high-frequency wavelet coefficients in the wavelet transform. The high-frequency component can reflect the sudden events or noise in the signal, such as equipment vibration, rapid temperature changes, etc., while the low-frequency component represents the stable change of the signal, which is extracted from the low-frequency wavelet coefficients in the wavelet transform. The low-frequency component helps to describe the long-term trend of the signal, such as the steady rise of temperature or the gradual change of the equipment.

[0026] Calculate the energy distribution of high-frequency components and low-frequency components respectively, analyze the instantaneous change of the energy distribution of high-frequency components, calculate the energy fluctuation rate, model the long-term trend of the energy distribution of low-frequency components, and calculate the frequency smoothing rate; Energy distribution is an important indicator that describes the intensity of a signal at different frequencies. In the time-frequency domain, the energy distribution of a signal is calculated by integrating the amplitudes of the high-frequency and low-frequency components. Energy calculation can reveal the intensity changes in different frequency bands in the signal, thereby providing a basis for subsequent anomaly identification, trend analysis, and signal processing. The rate of change of the energy of the high-frequency component, that is, the instantaneous frequency of the signal fluctuation, is calculated. Signals with high volatility usually mean sudden events or abnormal fluctuations during the experiment. The rate of change of the energy of the low-frequency component is calculated to reflect the long-term trend during the experiment. A low frequency smoothing rate indicates that the signal changes are relatively stable, and a signal with large changes may indicate a changing trend or equipment adjustment during the experiment. The energy fluctuation rate and the frequency smoothing rate are correlated and measured to calculate a correlation value. The priority of each window is calculated according to the correlation value, and the window data with the highest priority is output as a data feature.

[0027] By calculating the correlation between the energy fluctuation rate and the frequency smoothing rate, the relationship between abnormal fluctuations and stability during the experiment can be identified. The priority of the window data is determined according to the size of the correlation value. The window with a high priority indicates that the signal in the window contains more important experimental features.

[0028] The calculation formula for window priority is: , in, represents the window priority, T represents the total time length of the window, and t represents the unit time length of the window. Represents the time-frequency domain signal of the high-frequency component corresponding to the t-th time length of the window, represents the modulus of a complex number, represents the average energy of the high-frequency components in the window, f represents the frequency, Represents the time-frequency domain signal of the low-frequency component corresponding to the t-th time length of the window, It represents the overall smooth trend of the window, which is obtained by calculating the moving average of the signal in the window, and k represents the kurtosis value corresponding to the window; See also Figure 3 , a data analysis network structure diagram of an embodiment of the present invention, wherein the data analysis network comprises: The feature fusion layer is used to fuse data features from different sources through the self-attention mechanism and calculate the fusion features; In the laboratory sample automatic distribution system, there are many different sources of data, such as sensor data, experimental step data, environmental condition data, etc. These data have different dimensions and characteristics, and need to be fused to work together for experimental optimization and adjustment. Therefore, the self-attention mechanism is applied to the feature fusion layer to extract important information from multiple data sources and integrate this information to further analyze and optimize the experimental process; The core idea of ​​the self-attention mechanism is to assign an attention weight to each position of the input data. The size of the weight reflects the important relationship between this position and other positions. In experimental data fusion, the self-attention mechanism can dynamically adjust the weight of each feature according to the relationship between the features of different data sources, thereby realizing weighted fusion of information. The self-attention mechanism can adaptively adjust the weight of features according to the similarity between features, and then fuse the most informative features. Compared with the traditional weighted average method, the self-attention mechanism is more flexible in feature selection and can fully explore potential feature relationships; The calculation process of the self-attention mechanism is as follows: The data features from different sources are first organized into multiple input feature vectors. These vectors represent the data from each source in some form. For example, the data from the sensor may be preprocessed and converted into a vector containing multi-dimensional features such as temperature, humidity, and pressure, while the experimental steps and parameter data may be a vector containing experimental conditions, equipment type, etc. For each input feature vector, first calculate its similarity with other feature vectors to determine the degree of attention each input feature vector pays to other input feature vectors; Based on the similarity, the softmax operation is used to normalize the similarity of each pair of feature vectors to generate attention weights. Specifically, the weight can be understood as the ratio of the contribution of a feature to the similarity between the feature and other features and the sum of the similarities of all features. In this way, features with large weights are considered to be more important features and will be given more attention; Each input feature vector is weighted by the attention weight to calculate the final contribution of each feature in the current task. The weighted features will be used as the input of the fusion feature; Through weighted summation, the most representative features are extracted from data from different sources and fused into a new joint feature vector. The fused features contain the most useful information from all source data and can effectively support subsequent data processing and decision-making tasks; A processing layer, used for mapping the fused features into low-dimensional features through a multi-layer perceptron network, performing importance evaluation on the low-dimensional features according to a random forest algorithm, calculating information gain of the low-dimensional features, and identifying key features according to the information gain; The fused features are high-dimensional representations of the original features, which contain information from different feature sources. However, they are often of high dimension and have high redundancy, making them difficult to be used directly for further analysis. Therefore, it is necessary to map these fused features into low-dimensional features through a multi-layer perceptron to reduce redundancy, improve computational efficiency, and highlight important information in the data. The multi-layer perceptron network is a feedforward neural network consisting of multiple fully connected layers that can be used to extract low-dimensional representations from high-dimensional space. It performs nonlinear transformations on data by mapping layer by layer, and enhances nonlinear characteristics through activation functions to output low-dimensional features. The mapped low-dimensional features contain key data fused from multiple sources, but the contribution of these features still needs to be quantified in order to identify which features are most important to the experimental results. Therefore, the random forest algorithm is introduced to evaluate the importance of features. The random forest algorithm is a method of ensemble learning that makes predictions by building multiple decision trees. In feature importance evaluation, the random forest evaluates the importance of each feature by calculating the impact of the feature on the model error; Information gain is an indicator that measures how much "information" a feature can bring when dividing data. It indicates how much uncertainty or entropy can be reduced by selecting a certain feature for division. The greater the information gain, the more effectively the feature can help the task. Therefore, the criticality of each feature in experimental optimization can be evaluated based on its information gain. If the information gain of a feature is close to zero, it means that its contribution to the experimental prediction results is negligible, and it can be excluded to reduce the complexity of the model and improve the prediction performance of the model.

[0029] Compare the key features with data of successful cases of historical experiments to analyze whether there are differences, and calculate feedback signals based on the differences; Historical experimental success stories are usually stored in an experiment database and contain multiple experimental settings and results.

[0030] In the comparison process, key features refer to those variables that have the greatest impact on the experimental results during the experiment (such as experimental parameters, experimental conditions, sensor data, etc.). The data of historical experimental success cases include: experimental steps (such as temperature, pressure, time and other control parameters), experimental conditions (such as chemical concentration, sample volume, etc.), equipment configuration (such as equipment model, working status, etc.) and experimental results (such as reaction rate, product concentration, etc.); Compare the key features extracted from the current experiment with the corresponding features in the historical experiment cases one by one. By comparing the features, you can evaluate the similarity between the current experiment and the historical experiment. If the similarity is high, it means that the settings of the current experiment are consistent with the settings of the historical successful experiment, and you can continue to execute according to the parameters and steps of the historical experiment. If the similarity is low, it means that the current experiment may be different from the historical successful case, and you need to adjust the current experiment configuration according to the historical experiment; Based on the results of the difference analysis, a feedback signal is calculated to adjust the experimental parameters. The feedback signal is a numerical value that indicates the degree of difference between the experimental settings and historical successful cases. The feedback signal is used to determine whether the experimental parameters need to be optimized; The output layer is used to adjust the experimental parameters and modify the experimental plan according to the feedback signal, and output the final experimental optimization configuration.

[0031] The database updating module comprises: Data acquisition unit, used to obtain the latest literature from designated scientific research websites, academic data and technical report platforms; The data acquisition unit obtains the latest literature through designated scientific research websites, academic data and technical report platforms. These documents usually contain the experimental methods, steps and newly discovered technologies of researchers, and can provide the latest experimental method information for the experimental database; During the update process, the acquired timestamp and document ID are recorded to avoid repeated acquisition of processed documents, ensure that only new documents are collected, and perform preliminary cleaning on the collected document content to remove irrelevant web page information (such as advertisements, navigation bars, etc.) and retain only the core content of the document; The text parsing unit is used to parse the latest literature obtained through natural language processing and extract experimental method information; The documents acquired by the data acquisition unit are usually unstructured texts. The text parsing unit converts them into structured data through natural language processing (NLP) technology to extract relevant contents such as experimental method information, experimental steps, equipment, materials, etc. A database updating unit is used to determine whether the newly acquired experimental method information contains new content, and if there is a significant difference, update it to the experimental database; Through the introduction of natural language processing technology, high-quality experimental method information can be extracted from massive amounts of unstructured text and standardized to ensure that the information in the database is accurate and comprehensive; The automatic update strategy includes text parsing logic and content update logic; The text parsing logic is configured in the text parsing unit, and is used to identify and extract experimental methods, experimental steps and key parameters from the latest literature, and to normalize the extracted results; The content update logic is configured in the database update unit, and is used to compare the newly acquired experimental method information with the existing experimental method information in the experimental database to determine whether there is a significant change. If so, the newly acquired experimental method information is updated to the experimental database.

[0032] The text parsing logic includes: Preprocess the latest literature and convert it into structured text data; The goal of preprocessing is to convert the natural language text in the literature into a format that is easy for computers to analyze and process, thereby providing high-quality data input for subsequent keyword extraction, semantic analysis, and relationship building. Specifically, it includes: For scientific research documents, there are usually some contents that are not directly related to the experiment (such as charts, conclusions, discussion sections, etc.). Regular expressions and rule engines are used to remove irrelevant parts. If there are duplicate parts or multiple versions of the same document in the obtained documents, the duplicates are checked and deleted through the hash algorithm; Ensure that all document texts use a unified character encoding to avoid garbled characters caused by inconsistent encoding. In scientific research documents, abbreviations and terms are commonly used. Through the term library, abbreviations are expanded to full names and all punctuation marks in the text are unified, such as replacing Chinese commas with English commas and unifying the format of English quotation marks. Constructing a keyword extraction network, taking the structured text data as input parameters of the keyword extraction network, processing the input parameters through the keyword extraction network, and outputting a keyword sequence; The core task of the keyword extraction network is to automatically identify keywords related to experimental methods from structured text data. These keywords usually include experimental steps, operating conditions, experimental materials, equipment names, etc. The key to keyword extraction is how to effectively extract the core information that best represents the experimental method from complex text. By extracting keyword sequences, the system can effectively match experimental method information based on existing databases. When keywords in new documents match keywords in existing databases, the system can quickly determine whether the document provides new experimental methods or steps, thereby achieving automatic updates; Analyze the keyword sequence according to the dependency parsing technology to construct a semantic relationship graph, wherein the semantic relationship graph includes nodes and edges, the nodes represent experimental entities, and the edges represent entity-related actions; Dependency syntactic analysis can identify the dependency relationship between each word in a keyword sequence, that is, which word is the modifier or action object of which word. Common dependency relationships include "subject-predicate relationship", "verb-object relationship", "modification relationship", etc. For example, in "mix reagent A with reagent B", "mix" is a verb, "reagent A" and "reagent B" are its objects, forming a "verb-object relationship"; Based on the results of dependency parsing, a dependency tree is constructed, where each word is a node of the tree, and the dependencies between words are the edges of the tree. The root node of the tree is usually the predicate of the sentence (such as a verb), while other nodes are the subject, object or modifier of the verb. By analyzing the dependency tree, the relationship between experimental entities is extracted. For example: Node 1 (entity): Reagent A, Node 2 (entity): Reagent B, Edge (relationship): and, Action: Mix; Convert all identified entities and their relationships into a graph, where each node represents an experimental entity (such as "reagent A", "heat"), and each edge represents a relationship or action between entities (such as "with", "heat to"). Each node of the graph may contain one or more attributes (such as the name, quantity, equipment type, etc. of the experimental material); The nodes and edges in the semantic relationship graph are traversed to extract the experimental methods, experimental steps and key parameters.

[0033] By traversing the verb nodes in the graph (usually corresponding to the operation steps of the experiment, such as "heating", "mixing") and their connected noun nodes (such as "reagent A", "reactor"), the various steps in the experiment can be identified. For example, "heating reagent A to 100°C" can be extracted as an experimental step, where "heating" is the operation, "reagent A" is the operation object, and "100°C" is the condition; By traversing the entity nodes in the graph (e.g., “reagent A”, “reactor”), key information such as materials and equipment used in the experiment can be identified; By using the quantitative information related to the experimental steps and entities in dependency parsing (such as "temperature" and "concentration"), key parameters in the experiment can be extracted, such as "temperature = 100°C"; Integrate the extracted experimental steps, entities and key parameters to form a complete experimental description; See also Figure 4 , a keyword extraction network structure diagram of an embodiment of the present invention, wherein the keyword extraction network comprises: The vector representation layer is used to convert each word into a word vector containing semantic information and position information by segmenting, word segmentation and position encoding the text, thereby improving the accuracy and robustness of keyword extraction; Segment the input text, that is, divide the entire document into small pieces (such as sentences, paragraphs, etc.). Usually, these small pieces are divided into sentences or according to the natural structure of the document. The purpose of segmentation is to convert long text into parts with a certain degree of semantic independence, which helps to improve the efficiency and accuracy of subsequent model processing; Each text block after segmentation needs to be further segmented, that is, the words (or subwords) in each block need to be extracted. Chinese text usually needs to be segmented by a segmentation tool, while English text can be segmented directly based on spaces; Each word or subword after segmentation will be converted into a vector representation of a fixed dimension through a pre-trained word vector model. Each word vector represents the semantic information of the word, so that similar words have similar vector representations. At the same time, in order to capture the position information of the word in the sentence, each word is positionally encoded. Position encoding helps the model understand the order and grammatical structure of words, which is especially important when processing long texts. Considering the relative position relationship between words, a relative position encoding method is designed to enhance the model's understanding of sentence structure by calculating the relative position between words, concatenate the word vector and the position vector, and output a word vector sequence; The feature representation layer is used to extract the global features of the text context of the word vector sequence based on the bidirectional gated recurrent unit, perform sentence-level feature representation on the global features of the text context, and obtain deep semantic information; The bidirectional gated recurrent unit adopts a bidirectional long short-term memory neural network, which includes a forward long short-term memory neural network and a backward long short-term memory neural network, respectively obtaining the forward and backward information of the word vector sequence, and merging the forward and backward information into the output at the same time. The hidden state of each word contains the local grammatical information of the word and the semantic role of the word in the context, and the sentence-level feature representation further aggregates this information to form the global features of the text. Deep semantic information includes not only the surface grammatical structure of the text, but also the potential relationship between words in the text and the impact of the context on each word; The feature classification layer is used to extract the core features of deep semantic information, calculate the confidence probability of the core feature category through the Softmax classifier, and take the experimental category corresponding to the highest confidence probability as the experimental category of the word vector corresponding to the core feature. The calculation formula of the confidence probability of the core feature category is: , Among them, P represents the confidence probability of the core feature category, Softmax[•] represents the classifier, W represents the classification matrix, M(•) represents the relationship matrix, S represents the core feature, A represents the diagonal matrix, represents the probability matrix of the core feature corresponding to the category, and b represents the bias; The output layer is used to score the core features according to the experimental category, obtain the score sequence, and then globally evaluate the score sequence through the conditional random field, considering the relationship between adjacent labels in the sequence, and judging whether the transfer score matrix meets the global optimality. If it does not meet the global optimality, the transfer score matrix is ​​recalculated. If it meets the global optimality, the label score matrix is ​​adjusted through the transfer score matrix to make the label score matrix more in line with the law of sequence dependence; The optimal sequence is obtained by using the Viterbi algorithm based on adjacent labels. The Viterbi algorithm calculates the scores of all possible paths from the beginning of the sequence to the current time step, records the best path of each step, that is, the label sequence with the highest score, and after calculating the best path score of the last step, backtracks to gradually determine the optimal label of each time step, generates the optimal sequence, and decodes the optimal sequence to obtain the keyword sequence.

[0034] The content update logic includes: Extracting features of the newly acquired experimental method information through a pre-trained model to generate a multi-dimensional feature vector; First, for the newly acquired experimental method information, the pre-trained model is used to extract features from the experimental method text. The core task of this step is to convert the original experimental text into a vector with multi-dimensional features to facilitate subsequent calculations and comparisons; Calculate the cosine similarity between the experiment name vector in the multidimensional feature vector and the node entity name vector in the experiment database through a dynamic semantic matching algorithm, and generate a candidate method set according to the cosine similarity; By calculating the cosine similarity, we can effectively determine the similarity between the newly acquired experimental method information and the existing experimental method information in terms of name. If the similarity is high, it means that the experimental methods are roughly the same and enter the candidate method set; Perform deep matching on the experimental steps, conditions and parameters in the candidate method set, and calculate the comprehensive similarity based on the matching results, wherein the deep matching includes experimental step matching, experimental condition matching and experimental parameter matching; Experimental step matching is used to compare the newly acquired experimental steps with the experimental steps in the database one by one, and the similarity between them can be calculated through the word vector model. If the contents of the experimental steps are consistent or similar, the matching score is high; Experimental condition matching is used to compare experimental conditions (such as temperature, time, pressure, etc.). Condition matching can be calculated by the similarity of numerical data. If the numerical difference is small, the matching score is higher; Deep matching combines the matching results of multiple dimensions to more accurately determine whether a newly acquired experimental method is the same or similar to an existing method, thus avoiding the errors caused by simple name matching; The largest comprehensive similarity in the candidate method set is selected, and the comprehensive similarity is compared with the similarity threshold. If it is less than the similarity threshold, the newly acquired experimental method information is updated to the experimental database. If it is greater than or equal to the similarity threshold, the experimental database is not updated.

[0035] The recommendation module includes: The experimental target identification unit is used to identify the core elements of the experimental requirement input according to the experimental requirement input; Segment the experimental demand input according to the adaptive semantic segmentation model, perform part-of-speech tagging on the segmentation results through the syntax tree generation algorithm, obtain candidate word sequences, and vectorize the candidate word sequences through the word embedding model; The subject-verb-object core structure and modifying components in the candidate word sequence are identified through a tree structure based on dependency relationships, and the experimental demand input is divided into a multi-layer semantic structure. Performing temporal modeling on the multi-layer semantic structure according to the long short-term memory network, analyzing the front-end dependency relationship between the models, calculating the relevance matrix according to the front-end dependency relationship, performing feature extraction on the relevance matrix through convolution operation, and outputting the core elements; An experimental method matching unit, used for matching the core elements with the stored experimental methods in the experimental database; The experimental method recommendation unit is used to output matching experimental methods.

[0036] Example 2 See also Figure 5 The present invention provides an embodiment: a smart laboratory, the smart laboratory comprising: Laboratory hardware equipment, used to provide the physical platform required for the experiment, including laboratory benches, sample storage systems, laboratory sample automated distribution systems and related experimental equipment; Laboratory control center, used to manage and schedule experimental equipment, control experimental steps and equipment operation; Data acquisition and transmission equipment, used to collect experimental data in real time and transmit it to the data analysis platform; Data analysis platform, used to process experimental data and provide experimental optimization suggestions, including real-time parameter adjustment and experimental effect evaluation; Experimental plan recommendation equipment is used to provide the best matching experimental plan according to the experimental objectives; Laboratory management equipment, used to manage experimental tasks, resource scheduling and experimental progress tracking.

[0037] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. Laboratory sample automated distribution system, characterized in that: The system includes an experimental database, an experimental equipment control module, an intelligent optimization module and a database update module; The experimental database is used to store experimental method information; The experimental equipment control module is used to obtain sample information by scanning the barcode on the sample container, match the sample information with the experimental method information stored in the experimental database, and schedule the experimental equipment to execute the experimental steps according to the matching result; The intelligent optimization module is configured with a step optimization strategy, which adjusts the experimental parameters through a machine learning method based on data feedback during the experiment; The database update module is configured with an automatic update strategy, which is used to obtain scientific research reports from the network, parse the scientific research reports through natural language processing, determine whether there are updated contents, and store the updated experimental method information in the experimental database.

2. The laboratory sample automatic distribution system according to claim 1, characterized in that: The intelligent optimization module is used to optimize the experimental parameters and steps according to the data collected in real time during the experiment. The intelligent optimization module specifically includes: A data analysis unit, used to collect relevant sensor data during the experiment, analyze the relevant sensor data, and extract data features; A parameter optimization unit, used to process the data features through a machine learning algorithm and output optimization parameters; A feedback adjustment unit, used for providing feedback to the experimental equipment control module according to the optimization parameters; The step optimization strategy includes feature extraction logic and feature analysis logic; The feature extraction logic is configured in the data analysis unit to simplify the sensor data and extract data features; The feature analysis logic is configured in the parameter optimization unit, and is used to construct a data analysis network, use the data features as input parameters of the data analysis network, analyze the input parameters through the data analysis network, and output optimized parameters.

3. The laboratory sample automatic distribution system according to claim 2, characterized in that: The feature extraction logic includes: Performing tensor decomposition on the relevant sensor data, calculating a kurtosis value according to the decomposition result, and windowing the relevant sensor data according to the kurtosis value; The sensor data in each window is converted into a time-frequency domain signal by wavelet transform, and a high-frequency component and a low-frequency component of the time-frequency domain signal are obtained; Calculate the energy distribution of high-frequency components and low-frequency components respectively, analyze the instantaneous change of the energy distribution of high-frequency components, calculate the energy fluctuation rate, model the long-term trend of the energy distribution of low-frequency components, and calculate the frequency smoothing rate; The energy fluctuation rate and the frequency smoothing rate are correlated, and the correlation value is calculated. The priority of each window is calculated according to the correlation value, and the window data with the highest priority is output as the data feature. The calculation formula of the window priority is: ， in, represents the window priority, T represents the total time length of the window, and t represents the unit time length of the window. Represents the time-frequency domain signal of the high-frequency component corresponding to the t-th time length of the window, represents the modulus of a complex number, represents the average energy of the high-frequency components in the window, f represents the frequency, Represents the time-frequency domain signal of the low-frequency component corresponding to the t-th time length of the window, It represents the overall smooth trend of the window, and k represents the kurtosis value corresponding to the window.

4. The laboratory sample automatic distribution system according to claim 2, characterized in that: The data analysis network comprises: The feature fusion layer is used to fuse data features from different sources through the self-attention mechanism and calculate the fusion features; A processing layer, used for mapping the fused features into low-dimensional features through a multi-layer perceptron network, performing importance evaluation on the low-dimensional features according to a random forest algorithm, calculating information gain of the low-dimensional features, and identifying key features according to the information gain; Compare the key features with data of successful cases of historical experiments to analyze whether there are differences, and calculate feedback signals based on the differences; The output layer is used to adjust the experimental parameters and modify the experimental plan according to the feedback signal, and output the final experimental optimization configuration.

5. The laboratory sample automatic distribution system according to claim 1, characterized in that: The database updating module comprises: Data acquisition unit, used to obtain the latest literature from designated scientific research websites, academic data and technical report platforms; The text parsing unit is used to parse the latest literature obtained through natural language processing and extract experimental method information; A database updating unit is used to determine whether the newly acquired experimental method information contains new content, and if there is a significant difference, update it to the experimental database; The automatic update strategy includes text parsing logic and content update logic; The text parsing logic is configured in the text parsing unit, and is used to identify and extract experimental methods, experimental steps and key parameters from the latest literature, and to normalize the extracted results; The content update logic is configured in the database update unit, and is used to compare the newly acquired experimental method information with the existing experimental method information in the experimental database to determine whether there is a significant change. If so, the newly acquired experimental method information is updated to the experimental database.

6. The laboratory sample automatic distribution system according to claim 5, characterized in that: The text parsing logic includes: Preprocess the latest literature and convert it into structured text data; Constructing a keyword extraction network, taking the structured text data as input parameters of the keyword extraction network, processing the input parameters through the keyword extraction network, and outputting a keyword sequence; Analyze the keyword sequence according to the dependency parsing technology to construct a semantic relationship graph, wherein the semantic relationship graph includes nodes and edges, the nodes represent experimental entities, and the edges represent entity-related actions; The nodes and edges in the semantic relationship graph are traversed to extract the experimental methods, experimental steps and key parameters.

7. The laboratory sample automatic distribution system according to claim 6, characterized in that: The keyword extraction network includes: The vector representation layer is used to obtain the segmented text of the input parameter, process the segmented text, and encode it in combination with the word position information to obtain the word vector sequence; The feature representation layer is used to extract the global features of the text context of the word vector sequence based on the bidirectional gated recurrent unit, perform sentence-level feature representation on the global features of the text context, and obtain deep semantic information; The feature classification layer is used to extract the core features of deep semantic information, calculate the confidence probability of the core feature category through the Softmax classifier, and take the experimental category corresponding to the highest confidence probability as the experimental category of the word vector corresponding to the core feature. The calculation formula of the confidence probability of the core feature category is: ， Among them, P represents the confidence probability of the core feature category, Softmax[•] represents the classifier, W represents the classification matrix, M(•) represents the relationship matrix, S represents the core feature, A represents the diagonal matrix, represents the probability matrix of the core feature corresponding to the category, and b represents the bias; The output layer is used to score the core features according to the experimental category, obtain the score sequence, decode the score sequence, and output the keyword sequence.

8. The laboratory sample automatic distribution system according to claim 5, characterized in that: The content update logic includes: Extracting features of the newly acquired experimental method information through a pre-trained model to generate a multi-dimensional feature vector; Calculate the cosine similarity between the experiment name vector in the multidimensional feature vector and the node entity name vector in the experiment database through a dynamic semantic matching algorithm, and generate a candidate method set according to the cosine similarity; Perform deep matching on the experimental steps, conditions and parameters in the candidate method set, and calculate the comprehensive similarity based on the matching results, wherein the deep matching includes experimental step matching, experimental condition matching and experimental parameter matching; The largest comprehensive similarity in the candidate method set is selected, and the comprehensive similarity is compared with the similarity threshold. If it is less than the similarity threshold, the newly acquired experimental method information is updated to the experimental database. If it is greater than or equal to the similarity threshold, the experimental database is not updated.

9. The laboratory sample automatic distribution system according to claim 1, characterized in that: The system also includes a recommendation module, which is used to match the target experiment in the experiment database and provide an experimental plan, specifically including: The experimental target identification unit is used to identify the core elements of the experimental requirement input according to the experimental requirement input; An experimental method matching unit, used for matching the core elements with the stored experimental methods in the experimental database; The experimental method recommendation unit is used to output matching experimental methods.

10. Smart laboratory, characterized by, The smart laboratory includes: Laboratory hardware equipment, used to provide the physical platform required for the experiment, including a laboratory table, a sample storage system, a laboratory sample automatic distribution system as described in any one of claims 1 to 9, and related experimental equipment; Laboratory control center, used to manage and schedule experimental equipment, control experimental steps and equipment operation; Data acquisition and transmission equipment, used to collect experimental data in real time and transmit it to the data analysis platform; Data analysis platform, used to process experimental data and provide experimental optimization suggestions, including real-time parameter adjustment and experimental effect evaluation; Experimental plan recommendation equipment is used to provide the best matching experimental plan according to the experimental objectives; Laboratory management equipment, used to manage experimental tasks, resource scheduling and experimental progress tracking.

Images