A chemical laboratory whole-process management method based on data analysis processing

Abstract

This invention discloses a method for full-process management of chemical laboratories based on data analysis and processing, relating to the fields of chemical laboratory management and data analysis and processing technology. The method includes acquiring a first sequence of operational actions and corresponding first resource consumption data for the entire chemical laboratory process; denoising the first sequence of operational actions and the first resource consumption data using a preset filtering algorithm to obtain a second sequence of operational actions and second resource consumption data. This method for full-process management of chemical laboratories based on data analysis and processing can improve the precision of resource consumption analysis and the accuracy of anomaly location in the entire chemical laboratory process, reduce misjudgments caused by relying on manual experience for process optimization, and is conducive to improving the utilization rate of experimental resources, reducing R&D costs, and providing objective data basis for process evaluation and optimization management of chemical laboratories.

Description

Technical Field

[0001] This invention relates to the field of chemical laboratory management and data analysis and processing technology, specifically to a method for full-process management of chemical laboratories based on data analysis and processing. Background Technology

[0002] In the research, verification, and pilot-scale exploration processes of chemical laboratories, the experimental procedures typically include multiple continuous operational steps such as raw material preparation, weighing and addition, temperature rise and reaction, temperature control, condensation and reflux, separation and purification, washing and drying, detection and analysis, and waste liquid treatment. Each step involves not only the coordinated control of experimental personnel, equipment, and operating conditions, but also the continuous consumption of various resources such as water, electricity, gas, chemical raw materials, solvents, additives, and purification reagents. Therefore, the scientific nature of the entire process management in chemical laboratories directly affects experimental efficiency, research and development costs, resource utilization, and the stability of experimental results.

[0003] Existing process management methods in chemical engineering laboratories typically focus on statistical evaluations of final product yield, total experimental time, overall energy consumption, or comprehensive cost. While these methods can reflect the overall results after the experiment, their evaluation basis is mostly summarized data, lacking detailed analysis of the relationship between changes in resource consumption and various operational actions during the experiment. Especially when the experimental time is long, the number of operational steps is large, and the types of resources are complex, relying solely on the final results or total data is insufficient to accurately reflect the actual consumption status and resource utilization efficiency in different operational stages. In actual chemical engineering experiments, various resource consumptions usually do not occur independently. For example, changes in heating power may affect the subsequent amount of condensate used, adjustments to reaction time may affect gas consumption or by-product generation, and changes in separation conditions may further increase the consumption of solvents, detergents, or purification reagents. Therefore, there is a clear coupling relationship between different operational actions and resource consumption, and this relationship may be transmitted and amplified as the experiment progresses and the experimental space changes. However, existing management methods often fail to accurately identify the cross-stage transmission of these resource consumption issues. On the one hand, the operational records and resource consumption data collected in the laboratory may be affected by factors such as equipment fluctuations, manual recording errors, sensor noise, or inconsistent sampling frequencies, leading to unstable data quality. On the other hand, the influence relationships between different operational steps have time lags and are often hidden. Resource waste caused by a certain preceding operational action may not be immediately apparent, but may manifest as abnormally high resource consumption in subsequent steps such as condensation, separation, washing, or waste liquid treatment. Due to the lack of comprehensive analysis of operational actions, resource consumption, link relationships, and abnormal transmission behaviors throughout the entire chemical laboratory process, managers can usually only identify outcome problems such as high total energy consumption, abnormal reagent consumption, or cost overruns in a particular experimental batch, but find it difficult to determine which specific operational action caused the abnormal consumption, or to pinpoint the critical link where resource waste occurred. This problem leads to process optimization relying on experience-based judgment, easily attributing the cause of anomalies to the end link, while ignoring the preceding operations that truly trigger the abnormal transmission of resources. Summary of the Invention

[0004] The purpose of this invention is to provide a method for full-process management of chemical laboratories based on data analysis and processing, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for full-process management of a chemical laboratory based on data analysis and processing, comprising: S1, acquiring a first operation sequence and corresponding first resource consumption data for the entire process of a chemical laboratory; denoising the first operation sequence and first resource consumption data using a preset filtering algorithm to obtain a second operation sequence and second resource consumption data; S2, constructing an action-resource coupling tensor based on the second operation sequence and second resource consumption data and performing time-delay attention decoupling to obtain consumption fluctuation characteristics including time and space dimensions; S3, constructing a multi-level directed graph structure based on the consumption fluctuation characteristics for node mapping, and determining the link transmission relationship matrix between nodes of different operation links; S4, based on the link transmission relationship matrix... S5. Calculate the consumption changes between adjacent nodes and normalize them to obtain a quantified transmission effect numerical sequence; S6. Obtain the transmission effect numerical sequence. If a specific value in the transmission effect numerical sequence is greater than a preset threshold, it is determined that there is abnormal transmission behavior, and an abnormal transmission identifier set is obtained; S7. Using the abnormal transmission identifier set, the isolated forest algorithm is used to perform abnormal distribution detection and cluster analysis to determine the coordinates of the waste location node that caused the abnormal transmission behavior; S8. Based on the coordinates of the waste location node, the specific actions in the corresponding second operation action sequence are traced back to determine the target operation action that caused the abnormal resource transmission, and based on the target operation action, a full-process management result of the chemical laboratory is generated, including the cause of abnormal resource consumption, the location of the waste link, the process evaluation result, and the optimization management strategy.

[0006] Preferably, step S1 includes acquiring a first operation sequence and first resource consumption data. The first operation sequence includes reagent adding and stirring actions, and the first resource consumption data includes reagent usage and power consumption. The first operation sequence and the first resource consumption data are aligned using timestamps to obtain an initial sequence. Based on the initial sequence, a mean filtering algorithm is used to calculate the average value of reagent usage and power consumption within a preset time window. The deviation between the actual value and the average value in the initial sequence is calculated. If the deviation exceeds a preset threshold, the sequence node corresponding to the actual value is determined to have isolated data. Isolated data in the initial sequence is removed using a mean filtering algorithm, and the isolated data is smoothly replaced using the average value to obtain a denoised sequence. Continuous action features and real consumption features are extracted from the denoised sequence, and a second operation sequence and second resource consumption data are determined based on the action features and consumption features.

[0007] Preferably, step S2 includes acquiring a second operation action sequence and second resource consumption data; extracting the action frequency of the second operation action sequence and the consumable usage of the second resource consumption data; constructing an action-resource coupling tensor based on the action frequency and consumable usage; performing spatial mapping on the action-resource coupling tensor; extracting the node association relationship between action nodes and resource nodes within the action-resource coupling tensor; obtaining a spatial distribution matrix based on the node association relationship; acquiring trajectory offset data from the second operation action sequence; assigning weights to the spatial distribution matrix based on the trajectory offset data to obtain a weighted spatial matrix; using a self-attention mechanism to decouple the weighted spatial matrix by time delay in the time dimension; extracting the period comparison difference of the weighted spatial matrix in different time periods; determining a dimension splitting vector based on the period comparison difference; performing peak extraction on the weighted spatial matrix based on the dimension splitting vector to obtain the consumption peak value; if the consumption peak value is greater than a preset threshold, generating consumption fluctuation features including time and spatial dimensions based on the consumption peak value.

[0008] Preferably, step S3 includes: acquiring a first consumption fluctuation feature; extracting fluctuation amplitude based on the first consumption fluctuation feature; mapping the fluctuation amplitude to a preset fluctuation amplitude range for hierarchical division to obtain multiple fluctuation levels; using the fluctuation levels as nodes; extracting the pointing relationship between nodes based on the time dimension in the first consumption fluctuation feature; constructing a directed graph based on the nodes and the pointing relationship; extracting operation link nodes from the directed graph; determining the transmission path based on the pointing relationship between the operation link nodes; performing node mapping on the transmission path to obtain node mapping relationship; calculating the path weight based on the node mapping relationship and the transmission frequency on the transmission path; acquiring the path weight; if the path weight is greater than a preset threshold, extracting the hierarchical association strength between operation link nodes based on the path weight; inputting the hierarchical association strength into a preset efficiency evaluation function for calculation to determine the transmission efficiency between operation link nodes; performing link transmission analysis on the operation link nodes through the transmission efficiency to obtain a transmission efficiency vector; constructing a relationship matrix based on the transmission efficiency vector; and concatenating the relationship matrix to obtain a link transmission relationship matrix.

[0009] Preferably, step S4 includes obtaining a link transmission relationship matrix, extracting adjacent link nodes corresponding to non-zero elements in the link transmission relationship matrix, obtaining the node load of adjacent link nodes during service transmission, calculating the node load difference between adjacent link nodes to obtain the consumption change; obtaining the consumption change, extracting extreme values ​​for all consumption changes to determine the maximum and minimum values, calculating the difference between the maximum and minimum values ​​to obtain the range value; determining whether the range value is greater than zero, if the range value is greater than zero, subtracting the minimum value from the consumption change to obtain the change difference, dividing the change difference by the range value to obtain the normalized value; obtaining the normalized value, combining the normalized values ​​according to the arrangement order of adjacent link nodes in the link transmission relationship matrix to obtain a transmission effect numerical sequence.

[0010] Preferably, step S5 includes obtaining a transmission effect numerical sequence, extracting the transmission effect values ​​corresponding to each node in the transmission effect numerical sequence, obtaining a preset transmission effect threshold, calculating the absolute value of the difference between the transmission effect value and the preset transmission effect threshold to obtain the deviation magnitude; obtaining the deviation magnitude, and if the deviation magnitude is greater than a preset deviation tolerance threshold, determining that the node has abnormal transmission behavior, extracting the node identifier with abnormal transmission behavior and the corresponding associated link identifier, and combining them to generate an abnormal transmission identifier set.

[0011] Preferably, step S6 includes obtaining an anomaly propagation identifier set, extracting resource consumption and state deviation from the anomaly propagation identifier set, combining resource consumption and state deviation to construct an anomaly feature matrix; using the isolated forest algorithm to detect anomaly distribution in the anomaly feature matrix, constructing multiple isolated trees, calculating the average path length of each data point in the anomaly feature matrix in the multiple isolated trees, obtaining an anomaly score based on the average path length, and if the anomaly score is greater than a preset threshold, determining the corresponding data point as an anomaly feature point, thus obtaining an anomaly feature point set.

[0012] Preferably, step S6 further includes performing cluster analysis on the set of abnormal feature points using the mean clustering algorithm, calculating the Euclidean distance between each abnormal feature point in the set of abnormal feature points, dividing the set of abnormal feature points into at least one abnormal cluster based on the Euclidean distance, calculating the mean of all abnormal feature points in at least one abnormal cluster, obtaining the coordinates of the center point of the abnormal cluster; obtaining the network topology diagram corresponding to the coordinates of the center point of the abnormal cluster, finding adjacent nodes connected to the coordinates of the center point of the abnormal cluster in the network topology diagram, comparing the resource consumption of adjacent nodes, and determining the adjacent node with the largest resource consumption as the waste location node coordinate that triggered the abnormal transmission behavior.

[0013] Preferably, step S7 includes obtaining a coordinate mapping matrix based on the coordinates of the waste location node, extracting the container transfer trajectory through the coordinate mapping matrix, calculating the transfer probability of the container transfer trajectory using a hidden Markov model, performing action sequence inversion based on the transfer probability to obtain the specific actions in the operation action sequence; extracting reagent consumption deviation and reaction temperature fluctuation for the specific actions, and determining the specific action as the target operation action that triggers abnormal resource transfer if the reagent consumption deviation is greater than a preset threshold; extracting the abnormal transfer path of the target operation action, extracting the waste liquid discharge volume and instrument occupancy time based on the abnormal transfer path, calculating the ratio of waste liquid discharge volume to the instrument occupancy time to obtain the consumable consumption rate.

[0014] Preferably, step S7 further includes obtaining process evaluation indicators by combining the consumable consumption rate with preset evaluation weights, locking the location of the waste process in a preset laboratory topology map based on the process evaluation indicators, and matching the waste process location in a preset abnormal cause database to extract abnormal causes of resource consumption, generating an optimization management strategy based on the abnormal causes of resource consumption and the process evaluation indicators, and obtaining the full process management results of the chemical laboratory.

[0015] As can be seen from the above technical solution, the present invention has the following beneficial effects: This data analysis-based end-to-end management method for chemical engineering laboratories acquires the sequence of operational actions and corresponding resource consumption data for the entire laboratory process, and performs noise reduction to minimize the impact of equipment fluctuations, manual recording errors, and sampling anomalies on subsequent analysis results, thereby improving the accuracy and usability of experimental data. By constructing an action-resource coupling tensor and performing time-delay attention decoupling, it can identify the coupling relationships and delay effects between different operational actions and the consumption of resources such as water, electricity, gas, and reagents from both temporal and spatial dimensions, avoiding rough evaluations based solely on final yield, total energy consumption, or total cost. Furthermore, by constructing a multi-level directed graph structure and a link transmission relationship matrix, it can characterize the transmission paths of resource consumption fluctuations between different experimental operation links. This invention quantifies the changes in consumption between adjacent stages, enabling accurate identification of abnormal resource transmission behavior across stages. By determining the coordinates of waste location nodes through anomaly transmission identifier sets, the isolated forest algorithm, and cluster analysis, it can quickly locate the key stages that trigger abnormal resource transmission and further trace back to the specific target operation. This allows managers to clearly understand the cause of abnormal resource consumption, the location of the waste stage, and the corresponding optimization direction. Therefore, this invention can improve the precision and accuracy of resource consumption analysis throughout the entire process in chemical laboratories, reduce misjudgments caused by relying on manual experience for process optimization, improve the utilization rate of experimental resources, reduce R&D costs, and provide objective data for process evaluation and optimization management in chemical laboratories. Attached Figure Description

[0016] Figure 1 This is a signal transmission diagram of the present invention; Figure 2 This is a structural block diagram of the local terminal of an exemplary electronic device of the present invention; Figure 3 This is a structural block diagram of the network terminal of an exemplary electronic device of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: like Figure 1 As shown, this invention provides a technical solution: a method for full-process management of a chemical laboratory based on data analysis and processing, comprising: S1. Obtain the first operation sequence and the corresponding first resource consumption data of the entire process in the chemical laboratory. Use a preset filtering algorithm to denoise the first operation sequence and the first resource consumption data to obtain the second operation sequence and the second resource consumption data. S2. Based on the second operation action sequence and the second resource consumption data, construct the action resource coupling tensor and perform time-delay attention decoupling to obtain the consumption fluctuation characteristics that include time and space dimensions. S3. By consuming fluctuation characteristics, a multi-level directed graph structure is constructed to map nodes and determine the link transmission relationship matrix between nodes in different operation links. S4. Based on the link transmission relationship matrix, calculate the consumption change between adjacent link nodes and perform normalization to obtain the quantified transmission effect numerical sequence. S5. Obtain the numerical sequence of the transmission effect. If a specific value in the numerical sequence of the transmission effect is greater than a preset threshold, it is determined that there is abnormal transmission behavior and an abnormal transmission identifier set is obtained. S6. Using the set of abnormal propagation identifiers, the isolated forest algorithm is used to perform abnormal distribution detection and cluster analysis to determine the coordinates of the wasteful location nodes that cause abnormal propagation behavior; S7. Based on the coordinates of the waste location node, trace back to the specific actions in the corresponding second operation action sequence, determine the target operation action that caused the abnormal transmission of resources, and generate the full process management results of the chemical laboratory based on the target operation action, including the cause of abnormal resource consumption, the location of the waste link, the process evaluation results, and the optimization management strategy.

[0019] In this embodiment, the entire process of the chemical laboratory includes reagent collection, sample weighing, solution preparation, reaction control, heating and stirring, cooling and separation, sampling and testing, waste liquid collection, equipment cleaning, and experimental record archiving.

[0020] The first operation sequence is used to characterize the original operation behavior records formed during the experiment. The record fields include operation time, operator number, equipment number, experiment project number, action type, action duration, and action location. The first resource consumption data is used to characterize the resource usage corresponding to each operation action. The data fields include reagent usage, solvent usage, pure water usage, power consumption, gas consumption, consumable usage, equipment occupancy time, and waste liquid generation.

[0021] Furthermore, the system first performs noise reduction processing on the first operation action sequence and the first resource consumption data.

[0022] Specifically, for continuous resource consumption data, sliding window filtering, median filtering, exponential smoothing filtering, or Kalman filtering are used to process sudden spikes and short-term sampling fluctuations; for discrete operation action sequences, invalid actions, duplicate actions, and action records that do not conform to the experimental procedure are deleted according to the action time sequence, action duration, process logic constraints, and repeated recording rules; after noise reduction, a second operation action sequence and second resource consumption data are formed, so that subsequent resource consumption analysis is performed based on the cleaned data.

[0023] Furthermore, the system constructs an action-resource coupling tensor based on the second operation sequence and the second resource consumption data.

[0024] Specifically, the action-resource coupling tensor includes time dimension, operation action dimension, resource type dimension and spatial location dimension; through this tensor, a mapping relationship is established between experimental actions that occur within the same time period and the corresponding resources consumed.

[0025] For example, in the solution preparation process, the system writes the liquid addition action, liquid addition location, liquid addition duration, solvent consumption, and experimental batch number into a tensor structure, thereby forming a coupled expression between actions and resources.

[0026] In this embodiment, the abnormal transmission of resources in the chemical laboratory is not limited to the current action stage; abnormal operations in the previous stage can delay and affect the subsequent stage.

[0027] For example, exceeding the set value for reagent weighing will lead to changes in resource consumption in subsequent solution preparation, reaction processing, testing and calibration, and waste disposal; deviation of the reaction temperature from the set value will lead to subsequent re-inspection or rework operations; the system performs time-delay attention decoupling on the action resource coupling tensor, and by setting a first time delay window, a second time delay window, and a third time delay window, it assigns weights to the correlation strength between different operation actions and subsequent resource consumption, thereby distinguishing between immediate consumption, delayed consumption, and cross-stage cumulative consumption, and obtaining consumption fluctuation characteristics that include time and space dimensions.

[0028] Furthermore, the system maps the consumption fluctuation characteristics to a directed graph structure consisting of a process layer, a link layer, an action layer, and a resource layer.

[0029] The process layer represents the complete experimental task, the step layer represents experimental steps such as weighing, preparation, reaction, detection, and cleaning, the action layer represents specific actions such as adding materials, transferring, stirring, heating, sampling, and cleaning, and the resource layer represents resource objects such as reagents, consumables, electrical energy, gas, water, and waste liquid. The nodes in the directed graph represent operation step nodes or action resource combination nodes, and the directed edges represent the transmission direction of resource consumption effects between different steps.

[0030] In this embodiment, the system determines the link transmission relationship matrix between nodes in different operation links based on the temporal sequence, spatial flow, and resource dependency relationships between nodes.

[0031] Specifically, the link transmission relationship matrix is ​​used to characterize the intensity of resource consumption transmission between adjacent or related operation links. Subsequently, the system calculates the consumption change between adjacent link nodes based on the link transmission relationship matrix. The consumption change includes the difference in resource consumption, the consumption growth rate, the fluctuation amplitude, and the change intensity after combining the time delay weight. Since the units of measurement are different between different resource types, such as reagents measured in milliliters or grams, electrical energy measured in kilowatt-hours, and consumables measured in the number of pieces, the system normalizes the consumption change to make the consumption changes of different resource types comparable on a uniform scale, thus obtaining a quantified numerical sequence of transmission effects.

[0032] As a preferred embodiment, the system compares each value in the numerical sequence of the transmission effect with a preset threshold.

[0033] When a specific value exceeds a preset threshold, it indicates that the resource consumption change between the corresponding operation nodes exceeds the normal fluctuation range. The system determines that there is abnormal transmission behavior and generates an abnormal transmission identifier set. The abnormal transmission identifier set includes the abnormal edge number, the starting node, the ending node, the abnormal resource type, the time of abnormal occurrence, the abnormal intensity value, and the corresponding experimental batch number.

[0034] Furthermore, the system employs the isolated forest algorithm to detect abnormal distributions in the set of anomaly propagation identifiers.

[0035] Specifically, the Isolation Forest algorithm isolates abnormal samples within a preset number of partitioning rounds by randomly dividing the feature space, thereby identifying abnormal nodes or edges that deviate from the normal resource consumption pattern. Subsequently, the system performs cluster analysis on the anomaly detection results, grouping anomaly identifiers that are similar in time, space, resource type, or operation action characteristics into the same anomaly cluster, thereby determining the coordinates of the waste location node that triggered the anomaly transmission behavior. The coordinates of the waste location node are represented as a combination of process number, link node number, action node number, resource type number, and time slice number.

[0036] As a preferred embodiment, the system traces back the specific actions in the second operation action sequence based on the coordinates of the waste location node to determine the target operation action that caused the abnormal transmission of resources.

[0037] For example, when the coordinates of the waste location node point to "solution preparation, solvent addition, or abnormal organic solvent consumption", the system traces the corresponding experimental record for the liquid addition equipment, operator number, reagent bottle number, liquid addition time, and actual liquid addition volume to determine whether the abnormality is caused by repeated liquid addition, metering equipment deviation, improper reagent use, or operation waiting time exceeding the set time.

[0038] Furthermore, the system ultimately generates full-process management results for the chemical laboratory based on the target operational actions. These results include the causes of abnormal resource consumption, the location of wasteful processes, process evaluation results, and optimization management strategies, which are used to guide the laboratory in resource control, equipment calibration, process optimization, and standardized management of personnel operations.

[0039] Example 2: S1 includes acquiring a first operation action sequence and first resource consumption data. The first operation action sequence includes reagent adding and stirring actions, and the first resource consumption data includes reagent usage and power consumption. The first operation action sequence and the first resource consumption data are aligned using timestamps to obtain an initial sequence. Based on the initial sequence, a mean filtering algorithm is used to calculate the average value of reagent usage and power consumption within a preset time window. The deviation between the actual value and the average value in the initial sequence is calculated. If the deviation exceeds a preset threshold, the sequence node corresponding to the actual value is determined to have isolated data. Isolated data in the initial sequence is removed using a mean filtering algorithm, and the isolated data is smoothly replaced using the average value to obtain a denoised sequence. Continuous action features and real consumption features are extracted from the denoised sequence, and a second operation action sequence and second resource consumption data are determined based on the action features and consumption features.

[0040] In this embodiment, the processing objects of S1 are the first operation action sequence and the first resource consumption data.

[0041] Specifically, the first sequence of operations includes reagent dripping and stirring. The reagent dripping action records the start time, end time, dripping device number, reagent number, dripping channel number, and dripping action identifier. The stirring action records the start time, end time, stirring device number, stirring speed setting, stirring duration, and stirring action identifier. The first resource consumption data includes reagent usage and power consumption. The reagent usage is obtained from the feedback value of the metering pump, flow meter, or weighing module of the dripping device. The power consumption is obtained from the power meter connected to the stirring device.

[0042] Furthermore, after obtaining the first operation sequence and the first resource consumption data, timestamp alignment is performed first.

[0043] Specifically, a unified time axis is established based on the sampling interval of the resource acquisition equipment; when the sampling intervals for reagent usage, power consumption, and action recording are inconsistent, the smallest sampling interval among the three is taken as the node interval of the unified time axis.

[0044] The system generates consecutive time nodes in chronological order and groups reagent addition, stirring, reagent usage, and power consumption that fall within the same time node range into the same sequence node. For cases where reagent addition occurs within the same sequence node, a reagent addition action identifier is written to that sequence node; similarly, for cases where stirring occurs within the same sequence node, a stirring action identifier is written to that sequence node. Simultaneously, the reagent usage and power consumption values ​​collected within the corresponding time range are written to that sequence node. After completing the above processing, an initial sequence arranged in chronological order is obtained.

[0045] Furthermore, after the initial sequence is formed, the mean filtering algorithm is used to process the reagent usage and power consumption respectively.

[0046] Specifically, the preset time window consists of 5 consecutive sampling nodes. For the current sequence node located in the middle of the sequence, the preset time window is formed by selecting the 2 sampling nodes before the current sequence node, the current sequence node, and the 2 sampling nodes after the current sequence node; if there are fewer than 2 sampling nodes at the beginning of the sequence, 5 valid sampling nodes are selected consecutively from the beginning of the sequence; if there are fewer than 2 sampling nodes at the end of the sequence, 5 valid sampling nodes are selected consecutively backward from the end of the sequence.

[0047] Therefore, each mean filtering calculation is based on 5 valid sampling nodes.

[0048] In this embodiment, when calculating the average reagent usage, the reagent usage values ​​corresponding to 5 sampling nodes within a preset time window are read, the 5 reagent usage values ​​are added together, and the sum is divided by 5 to obtain the average reagent usage value corresponding to the current sequence node; when calculating the average power consumption, the power consumption values ​​corresponding to 5 sampling nodes within the same preset time window are read, the 5 power consumption values ​​are added together, and the sum is divided by 5 to obtain the average power consumption value corresponding to the current sequence node.

[0049] Furthermore, after obtaining the average value, the deviation between the actual value and the average value in the current sequence node is calculated. For reagent usage, the actual value of reagent usage in the current sequence node is first taken, then the difference is calculated with the average value of reagent usage corresponding to that node, and the absolute value of the difference is taken to obtain the reagent usage deviation value. For energy consumption, the actual value of energy consumption in the current sequence node is first taken, then the difference is calculated with the average value of energy consumption corresponding to that node, and the absolute value of the difference is taken to obtain the energy consumption deviation value. After processing with absolute values, cases where the actual value is greater than the average value and cases where the actual value is less than the average value are both handled according to the same judgment rule.

[0050] Specifically, the preset threshold is determined based on the baseline experimental data.

[0051] The baseline experimental data consisted of 30 consecutive batches of historical data with no equipment alarm records, no manual correction records, and qualified experimental results under the same experimental conditions, including the same dropping equipment, the same stirring equipment, and the same sampling interval. For the reagent dosage threshold, the mean filtering calculation for each of the 30 batches of baseline experimental data at five sampling nodes was performed to obtain the average reagent dosage for each baseline sampling node. Then, the absolute deviation between the actual reagent dosage and the corresponding average value for each baseline sampling node was calculated. All absolute deviations were then summed and divided by the number of absolute deviations to obtain the average reagent dosage baseline deviation. Each absolute deviation was then subtracted from this average baseline deviation, and the resulting difference was squared. All squared results were summed and divided by the number of absolute deviations, and the square root of the result was taken to obtain the standard deviation of the reagent dosage baseline deviation. Finally, the average reagent dosage baseline deviation was added to three times the standard deviation of the reagent dosage baseline deviation to obtain the reagent dosage threshold.

[0052] Specifically, the energy consumption threshold is determined in the same way: First, mean filtering calculations are performed on 5 sampling nodes for each of the 30 batches of benchmark experimental data to obtain the average energy consumption value corresponding to each benchmark sampling node; then, the absolute deviation between the actual energy consumption value and the corresponding average value of each benchmark sampling node is calculated; then, the average value of all absolute deviations is calculated, and the standard deviation of the energy consumption benchmark deviation is calculated based on this; finally, the average value of the energy consumption benchmark deviation is added to 3 times the standard deviation of the energy consumption benchmark deviation to obtain the energy consumption threshold.

[0053] The above reagent usage thresholds and power consumption thresholds are stored together with the experimental item number, dropping device number, stirring device number, and sampling interval; when changing the experimental item, dropping device, stirring device, or sampling interval, 30 batches of baseline experimental data are reselected and the corresponding thresholds are regenerated.

[0054] In this embodiment, after the threshold is determined, isolated data is determined for each sequence node in the initial sequence.

[0055] Specifically, for the current sequence node, if the reagent usage deviation value is greater than the reagent usage threshold, the actual reagent usage value in the sequence node is determined to be isolated data; if the reagent usage deviation value is less than or equal to the reagent usage threshold, the actual reagent usage value is retained. For the current sequence node, if the power consumption deviation value is greater than the power consumption threshold, the actual power consumption value in the sequence node is determined to be isolated data; if the power consumption deviation value is less than or equal to the power consumption threshold, the actual power consumption value is retained.

[0056] This determination process is used to identify single-point abnormal records caused by instantaneous feedback errors from the dripping equipment, flow meter sampling jumps, power meter communication interruptions, transient disturbances during equipment startup, or manual recording errors.

[0057] Furthermore, after identifying isolated data, the initial sequence is eliminated and smoothed using a mean filtering algorithm.

[0058] Specifically, for the actual reagent usage values ​​identified as isolated data, the original reagent usage values ​​in the sequence node are deleted, and the average reagent usage value calculated within the preset time window corresponding to the node is written into the sequence node; for the actual energy consumption values ​​identified as isolated data, the original energy consumption values ​​in the sequence node are deleted, and the average energy consumption value calculated within the preset time window corresponding to the node is written into the sequence node; for reagent usage and energy consumption values ​​not identified as isolated data, the original values ​​remain unchanged; after processing all sequence nodes, a denoised sequence is obtained.

[0059] Subsequently, continuous action features and real consumption features are extracted from the denoised sequence.

[0060] For reagent adding actions, sequential nodes that are temporally adjacent and all have reagent adding action identifiers are merged into a single continuous adding process. The start time, end time, duration, number of adding nodes, and cumulative reagent consumption of this continuous adding process are recorded. The cumulative reagent consumption is obtained by adding the reagent consumption of each sequential node within the continuous adding process. For stirring actions, sequential nodes that are temporally adjacent and all have stirring action identifiers are merged into a single continuous stirring process. The start time, end time, duration, number of stirring nodes, and cumulative energy consumption of this continuous stirring process are recorded. The cumulative energy consumption is obtained by adding the energy consumption of each sequential node within the continuous stirring process.

[0061] Finally, the continuous dripping process and the continuous stirring process are arranged in chronological order to form the second operation sequence; the cumulative reagent consumption, cumulative power consumption, reagent consumption per unit time, and power consumption per unit time corresponding to each continuous action process are organized into the second resource consumption data; the reagent consumption per unit time is obtained by dividing the cumulative reagent consumption by the dripping duration, and the power consumption per unit time is obtained by dividing the cumulative power consumption by the stirring duration.

[0062] Through the above processing, the first operation sequence and the first resource consumption data are converted into a second operation sequence and the second resource consumption data after completion timestamp alignment, isolated data removal and smooth replacement.

[0063] Example 3: S2 includes acquiring the second operation action sequence and the second resource consumption data; extracting the action frequency of the second operation action sequence and the consumable consumption of the second resource consumption data; constructing an action-resource coupling tensor based on the action frequency and consumable consumption; performing spatial mapping on the action-resource coupling tensor; extracting the node association relationship between action nodes and resource nodes within the action-resource coupling tensor; obtaining a spatial distribution matrix based on the node association relationship; acquiring trajectory offset data from the second operation action sequence; weighting the spatial distribution matrix based on the trajectory offset data to obtain a weighted spatial matrix; using a self-attention mechanism to decouple the weighted spatial matrix by time delay in the time dimension; extracting the period comparison difference of the weighted spatial matrix in different time periods; determining the dimension splitting vector based on the period comparison difference; extracting the peak value of the weighted spatial matrix based on the dimension splitting vector; if the peak value of consumption is greater than a preset threshold, generating consumption fluctuation features including time and spatial dimensions based on the peak value of consumption.

[0064] In this embodiment, S2 is used to perform coupled analysis on the denoised action data and resource consumption data, and extract consumption fluctuation characteristics from them.

[0065] The second operation sequence is obtained from S1, and the recorded content includes the action name, action occurrence time, action end time, action duration, action occurrence location, equipment number, experimental stage number, and experimental batch number; the second resource consumption data has the same time base as the second operation sequence, and the recorded content includes the consumable name, consumable number, consumable usage, consumption occurrence time, consumption occurrence location, and corresponding experimental stage number.

[0066] Furthermore, when extracting the frequency of actions, the second operation sequence is first segmented according to the experimental stage number and time period.

[0067] Specifically, the time period consists of 10 consecutive sampling nodes, and the interval between sampling nodes follows the unified sampling interval after the timestamp alignment in S1. The number of times the same action name appears in each time period is accumulated to obtain the action frequency of the action in that time period. The same action name is determined by both the action identifier and the device number. Records with the same action identifier and the same device number are classified into the same action type. Thus, the reagent adding action, stirring action, sampling action, transfer action and cleaning action form corresponding frequency records.

[0068] Furthermore, when extracting the consumption of consumables, the second resource consumption data is collected according to the experimental stage number, consumable number, and time period.

[0069] For liquid consumables, the consumption amount is determined by the noise-reduced flow meter reading, metering pump output, or weighing module difference. For solid consumables, the consumption amount is determined by the barcode scanning record, equipment deployment record, or inventory deduction record. The consumption values ​​of consumables in the same experimental stage, with the same consumable number, and within the same time period are added together to obtain the cumulative consumption amount of the consumable within that time period. This cumulative result, together with the corresponding action frequency, serves as the basic data for the action resource coupling tensor.

[0070] Furthermore, when constructing the action resource coupling tensor, experimental batch, time period, spatial location, action type, and consumable type are used as data dimensions.

[0071] First, the experimental stage number is written in the order of the experimental procedure. Then, the frequency of the action is filled in the corresponding action type position, and the amount of consumables is filled in the corresponding consumable type position. For actions and consumables that occur within the same time period, a correspondence is established based on the action occurrence time, the consumable consumption time, and the experimental stage number. When the action occurrence time and the consumable consumption time fall within the same time period and belong to the same experimental stage, the frequency of the action and the amount of consumables are written into the same tensor unit. When an action occurs but there is no record of consumable consumption, the amount of consumables in the tensor unit is written as 0. When consumable consumption exists but no action record is matched, the consumable consumption is written into the resource node to be matched and associated with it in subsequent spatial mapping based on the temporal adjacency relationship and the stage belonging relationship.

[0072] Furthermore, after constructing the action resource coupling tensor, a space mapping is performed on the tensor.

[0073] The spatial mapping is based on the laboratory's planar coordinates and functional area numbers. The functional areas include reagent storage area, weighing area, dropping operation area, stirring and reaction area, sampling and testing area, cleaning area, and waste collection area. Each functional area is pre-configured with a spatial number and coordinate range. Action nodes are mapped to the corresponding spatial number according to the location where the action occurs, and resource nodes are mapped to the corresponding spatial number according to the location of consumables retrieval, consumables usage, or consumables consumption record. Action nodes include reagent dropping, stirring, sampling, transfer, and cleaning. Resource nodes include reagents, filter membranes, pipette tips, sample vials, cleaning solutions, and electrode consumables.

[0074] In this embodiment, when extracting the node association relationship between action nodes and resource nodes, three rules are followed: time consistency, spatial consistency, and process subordination.

[0075] Specifically, when the occurrence time of an action node and the consumption time of a resource node fall within the same time period and are mapped to the same spatial number, they are written as directly associated. When an action node occurs before a resource node, both belong to the same experimental stage, and the consumption time of the resource node does not exceed the end time of that experimental stage, they are written as continuing associated. When a resource node is a necessary input registered in the experimental procedure configuration table for that action, they are written as process associated; a process association is written between reagent addition and reagent usage, between sampling and sample vial usage, and between cleaning and cleaning solution usage.

[0076] Specifically, the spatial distribution matrix is ​​generated based on the above-mentioned node association relationships.

[0077] In this matrix, rows correspond to action nodes, and columns correspond to resource nodes. When an action node and a resource node are directly related, a 1 is written at the intersection. When there is a continuous relationship, the relationship value is written according to the number of sampling nodes between them: 0.8 for 1 sampling node, 0.6 for 2 sampling nodes, 0.4 for 3 sampling nodes, and 0 for more than 3 sampling nodes. When there is a process relationship, the process relationship value is written according to the experimental process configuration table: 1 for required inputs and 0.5 for auxiliary inputs. 0 is written when no such relationship exists. The above values ​​are written into the parameter table before the experimental project is activated and are bound to the experimental project number, experimental step number, and consumable number.

[0078] Furthermore, when acquiring trajectory offset data, the actual operation position of each action node is read from the second operation action sequence, and the standard operation position of the action node is read from the experimental process configuration table.

[0079] Specifically, the standard operating position is determined by the laboratory layout parameters, while the actual operating position is determined by the equipment positioning record, the experimental platform identification record, or the operating terminal check-in position. The calculation process of the offset distance is as follows: first, obtain the difference between the actual operating position and the standard operating position on the horizontal coordinate, and then obtain the difference between the two on the vertical coordinate. Subsequently, square the horizontal difference and the vertical difference respectively, add the two squared results, and then take the square root of the sum to obtain the trajectory offset distance of the action node relative to the standard operating position.

[0080] Furthermore, when assigning weights to the spatial distribution matrix based on trajectory offset data, the weights are first determined based on the offset distance.

[0081] Specifically, when the offset distance is 0, the weight is 1; when the offset distance is greater than 0 meters but not more than 1 meter, the weight is 1.1; when the offset distance is greater than 1 meter but not more than 3 meters, the weight is 1.3; when the offset distance is greater than 3 meters, the weight is 1.5. Then, the entire row of correlation values ​​corresponding to the action node in the spatial distribution matrix is ​​multiplied by the weight of the action node to obtain the weighted correlation values. After performing the same processing on all action nodes, the weighted spatial matrix is ​​obtained. This processing enables action nodes that deviate from the standard operating position to have corresponding spatial influence coefficients in subsequent time delay decoupling and peak extraction.

[0082] Furthermore, when performing time-delay decoupling in the time dimension, the weighted spatial matrix is ​​split into continuous matrix segments according to the time period.

[0083] Specifically, the matrix segment of the current time period is taken as the segment to be analyzed, and the matrix segments of the previous 1 time period, the previous 2 time periods, and the previous 3 time periods are read as time-lag reference segments respectively. For each row in the current matrix segment, the inner product operation is performed with the values ​​of the same row in the 3 time-lag reference segments to obtain 3 time-lag similarities. The 3 time-lag similarities are added together to obtain the total similarity. Then, each time-lag similarity is divided by the total similarity to obtain the attention weight of the corresponding time-lag period. When the total similarity is 0, the attention weight of each of the 3 time-lag periods is one-third.

[0084] In this embodiment, after obtaining the attention weights, the matrix segments of the first 1 time period, the first 2 time periods, and the first 3 time periods are multiplied by their respective attention weights, and then the three weighted time-delay reference segments are added together to obtain the historical influence segments.

[0085] Subsequently, the difference between the matrix segment of the current time period and the historical impact segment is calculated to obtain the time delay impact caused by historical action resource relationships in the current time period; then, the time delay impact is subtracted from the matrix segment of the current time period to obtain the matrix segment that completes the time delay decoupling; through this process, the resource consumption changes in the current time period can be separated from the delay impact of the previous 3 time periods.

[0086] In one possible embodiment, when extracting the period comparison difference, the original weighted spatial matrix segment of the current time period is compared item by item with the matrix segment after the time delay decoupling is completed.

[0087] Specifically, for each intersection of an action node and a resource node, the difference between the associated values ​​before and after decoupling is first calculated, and then the absolute value of the difference is taken to obtain the periodic difference value of the intersection. All periodic difference values ​​in the row containing the same action node are added together to obtain the action-side periodic comparison difference of the action node in the current time period. All periodic difference values ​​in the column containing the same resource node are added together to obtain the resource-side periodic comparison difference of the resource node in the current time period.

[0088] Specifically, the dimension splitting vector is determined based on the periodic comparison difference, and the splitting judgment value is generated by the benchmark experimental data. The benchmark experimental data is taken from 30 consecutive batches of experiments that passed the same experiment, the same spatial layout, the same equipment configuration, and the same sampling interval, and had no equipment alarm records.

[0089] The process involves first performing spatial mapping, trajectory offset weighting, and time-delay decoupling on the 30 batches of data, then calculating the periodic comparison differences in each batch. Next, all periodic comparison differences are summed and divided by the number of periodic comparison differences to obtain the baseline average difference. Then, each periodic comparison difference is subtracted from the baseline average difference, and the difference is squared. All squared results are summed and divided by the number of periodic comparison differences, and the square root of the result is taken to obtain the baseline standard deviation. Finally, the baseline average difference is added to twice the baseline standard deviation to obtain the splitting decision value. In the current experimental data, when the periodic comparison difference on the action side or the periodic comparison difference on the resource side is greater than the splitting decision value, the corresponding action node, resource node, and time period are marked as splitting objects in the dimension splitting vector; when they are less than or equal to the splitting decision value, they are marked as retained objects.

[0090] Furthermore, when performing peak extraction based on the dimension-splitting vector, only the action nodes, resource nodes, and time periods marked as split objects are processed.

[0091] Specifically, all weighted correlation values ​​of the corresponding action node and resource node within the time period are read, and the weighted correlation value with the largest value is selected as the candidate peak value. Then, the candidate peak values ​​of the same action node and the same resource node in the previous and next time periods are read, and the three candidate peak values ​​are compared. The candidate peak value with the largest value is retained as the consumption peak value. The time period, spatial number, action node and resource node corresponding to the consumption peak value are retained together for subsequent generation of consumption fluctuation characteristics.

[0092] As a preferred implementation, the peak consumption threshold is also determined from benchmark data.

[0093] Specifically, data from 30 consecutive batches of experiments that passed without any equipment alarms under the same experimental project, spatial layout, equipment configuration, and sampling interval were selected. Following the same tensor construction, spatial mapping, trajectory offset weighting, time delay decoupling, dimension splitting, and peak extraction process as the current experimental data, all baseline consumption peaks were obtained. All baseline consumption peaks were summed and divided by the number of peaks to obtain the average baseline peak value. The difference between each baseline consumption peak value and the average baseline peak value was then calculated. The squared differences were summed, and the sum was divided by the number of peaks. The square root of the sum was then taken to obtain the standard deviation of the baseline peak value. Finally, the average baseline peak value was added to three times the standard deviation of the baseline peak value to obtain the consumption peak value threshold.

[0094] This threshold is stored together with the experiment project number, spatial layout number, equipment configuration number, and sampling interval. When the experiment project, spatial layout, equipment configuration, or sampling interval changes, the corresponding threshold is regenerated.

[0095] Furthermore, when the peak consumption in the current experimental data is greater than the peak consumption threshold, it is determined that there is a consumption fluctuation in the action resource relationship corresponding to the peak consumption.

[0096] Subsequently, the time period number, action occurrence time, action node number, resource node number, standard operation position, actual operation position, spatial number, trajectory offset distance, and consumption peak value corresponding to the consumption peak are read, and the above content is organized into consumption fluctuation characteristics; the time dimension of the consumption fluctuation characteristics consists of action occurrence time, time period number, and time delay reference period, and the spatial dimension consists of standard operation position, actual operation position, spatial number, and trajectory offset distance.

[0097] Example 4: S3 includes: acquiring a first consumption fluctuation feature; extracting the fluctuation amplitude based on the first consumption fluctuation feature; mapping the fluctuation amplitude to a preset fluctuation amplitude range for hierarchical division to obtain multiple fluctuation levels; using the fluctuation levels as nodes; extracting the directional relationship between nodes based on the time dimension in the first consumption fluctuation feature; constructing a directed graph based on the nodes and the directional relationship; extracting operation link nodes from the directed graph; determining the transmission path based on the directional relationship between the operation link nodes; performing node mapping on the transmission path to obtain node mapping relationship; calculating the path weight based on the node mapping relationship and the transmission frequency on the transmission path; obtaining the path weight; if the path weight is greater than a preset threshold, extracting the hierarchical association strength between operation link nodes based on the path weight; inputting the hierarchical association strength into a preset efficiency evaluation function for calculation to determine the transmission efficiency between operation link nodes; performing link transmission analysis on the operation link nodes through the transmission efficiency to obtain a transmission efficiency vector; constructing a relationship matrix based on the transmission efficiency vector; and concatenating the relationship matrix to obtain a link transmission relationship matrix.

[0098] In this embodiment, S3 is used to convert the first consumption fluctuation feature into matrix data that can characterize the transmission relationship between operation links.

[0099] Specifically, the first consumption fluctuation feature is output by the preceding steps, and its data items include the experimental batch number, operation link number, action node number, resource node number, time cycle number, action occurrence time, spatial location number, consumption peak, baseline consumption value, and trajectory offset distance. All of the above data items are bound to the same experimental batch to ensure that subsequent hierarchical division, path extraction, and matrix construction have a definite data source.

[0100] Furthermore, after obtaining the first consumption fluctuation feature, the fluctuation amplitude is extracted first.

[0101] In specific processing, the peak consumption value in the current first consumption fluctuation feature is read, and the corresponding baseline consumption value is read. The baseline consumption value is determined by the data of 30 consecutive batches of experiments that are qualified and have no equipment alarm records under the same experimental project, the same operation link, the same resource node and the same spatial location. The determination process is as follows: add the consumption values ​​of the corresponding positions in the 30 batches of data, and then divide the sum by the number of batches involved in the calculation to obtain the baseline consumption value of that position.

[0102] Then, the current peak consumption value is subtracted from the baseline consumption value, and the absolute value of the subtraction result is taken. The resulting value is the fluctuation range of the current first consumption fluctuation characteristic.

[0103] In this embodiment, the fluctuation range is generated before the experimental project is started and stored together with the experimental project number, process configuration number and spatial layout number.

[0104] Specifically, when generating intervals, first select data from 30 consecutive batches of experiments that passed without any equipment alarm records under the same experimental project, process configuration, and spatial layout. Calculate the fluctuation range of each benchmark sample using the same method as the current data. Then, sum all the benchmark fluctuation ranges and divide the sum by the number of benchmark fluctuation ranges to obtain the average benchmark fluctuation range. After that, calculate the difference between each benchmark fluctuation range and the average value, square each difference, sum all the squared results, divide by the number of benchmark fluctuation ranges, and then take the square root of the result to obtain the standard deviation of the benchmark fluctuation range.

[0105] Specifically, the first fluctuation range is set to be no greater than the average value of the benchmark fluctuation range plus one standard deviation of the benchmark fluctuation range; the second fluctuation range is set to be greater than the upper limit of the first fluctuation range but no greater than the average value of the benchmark fluctuation range plus two standard deviations of the benchmark fluctuation range; the third fluctuation range is set to be greater than the upper limit of the second fluctuation range but no greater than the average value of the benchmark fluctuation range plus three standard deviations of the benchmark fluctuation range; and the fourth fluctuation range is set to be greater than the upper limit of the third fluctuation range.

[0106] Furthermore, during the hierarchical division, the current fluctuation range is matched sequentially with four fluctuation range intervals.

[0107] Among them, those falling into the first fluctuation range are identified as the first fluctuation level; those falling into the second fluctuation range are identified as the second fluctuation level; those falling into the third fluctuation range are identified as the third fluctuation level; and those falling into the fourth fluctuation range are identified as the fourth fluctuation level. Each fluctuation level is used as a node in a directed graph. The node content retains the experimental batch number, operation link number, action node number, resource node number, time period number, and spatial location number, so that the node can be mapped back to the original consumption fluctuation characteristics.

[0108] Furthermore, when extracting the pointing relationships between nodes, the time dimension in the first consumption fluctuation feature is used. For nodes within the same experimental batch, they are first sorted from front to back according to their time period numbers; if the time period numbers are the same, they are then sorted according to the time of action occurrence; if the time of action occurrence is the same, they are sorted according to the order of the operation steps in the experimental process configuration table; after sorting, the node that is sorted first points to the node that is sorted second; if two nodes correspond to adjacent operation steps, an inter-step pointing relationship is formed; if two nodes correspond to the same operation step, an intra-step pointing relationship is formed.

[0109] A directed graph is constructed from the nodes and their pointing relationships. The edges in the directed graph represent the transmission direction of consumption fluctuations as they occur over time and in the order of the process.

[0110] Furthermore, when extracting operation nodes from a directed graph, the operation node number carried by each node is read, and nodes with the same operation node number are merged into the same operation node.

[0111] During the merging process, the time period number, resource node number, and spatial location number of each fluctuation level node under the operation link node are retained; after the merging is completed, the nodes in the directed graph are organized into operation link nodes such as reagent collection, weighing, dropping, stirring, sampling, detection, cleaning, and waste collection.

[0112] As a preferred embodiment, the transmission path is determined based on the pointing relationship between the nodes of the operation process.

[0113] When one operation node directly points to another operation node, a direct transmission path is formed between them; when one operation node points to the termination operation node via an intermediate operation node, a continuous transmission path is formed according to the order of the directed edges; each transmission path records the starting operation node, the termination operation node, the intermediate operation nodes passed through, the fluctuation level corresponding to each node, and the occurrence time and spatial location number.

[0114] Furthermore, when performing node mapping on the transmission path, the standard operation sequence in the experimental process configuration table is read first, and then each operation node in the transmission path is mapped to the standard operation node.

[0115] When the node name, step number, and device number are all the same, a complete mapping is written; when the node name is the same but the step number changes due to batch process configuration adjustments, it is corrected based on the experiment project number, device number, and action occurrence time, and then written into the corrected mapping; nodes that cannot be mapped to standard operation steps are written into the unmapped node table and do not participate in path weight calculation; the node mapping relationship includes the transmission path number, the original node number, the standard operation step number, the mapping status, and the order of the node in the path.

[0116] In this embodiment, the path weight is determined by the node mapping relationship and the transmission frequency.

[0117] The transmission frequency refers to the number of times a transmission path with the same starting operation and the same ending operation occurs within the current statistical batch. After the transmission frequency statistics are completed, the mapping coefficient is determined according to the node mapping status: the coefficient of a fully mapped node is 1, the coefficient of a corrected mapped node is 0.8, and unmapped nodes are not included. The average of the node coefficients participating in the calculation in the same transmission path is used to obtain the mapping coefficient of the path. The transmission frequency is then multiplied by the mapping coefficient to obtain the path weight.

[0118] In this embodiment, the path weight threshold is determined based on the baseline transmission path data.

[0119] The baseline transfer path data consists of 30 consecutive batches of data from the same experimental project, with the same experimental process configuration and the same spatial layout, all of which passed the experiment without any equipment alarm records. The same directed graph construction, operation node extraction, transfer path determination, node mapping, and path weight calculation are performed on these 30 batches of data to obtain all baseline path weights. All baseline path weights are summed and divided by the number of weights to obtain the average baseline path weight. The difference between each baseline path weight and this average is then calculated, the squares of the differences are summed, the sum is divided by the number of weights, and the square root of the result is taken to obtain the standard deviation of the baseline path weight. The average baseline path weight is added to twice the standard deviation of the baseline path weight to obtain the path weight threshold.

[0120] Furthermore, when the path weight of the current transmission path is greater than the path weight threshold, the hierarchical association strength corresponding to that transmission path is extracted.

[0121] In specific processing, the fluctuation level index of the starting operation node and the fluctuation level index of the ending operation node of the transmission path are read, the difference between the two is calculated and the absolute value is taken; when the fluctuation level indexes are the same, the level difference is 1; then, the path weight is multiplied by the level difference to obtain the level association strength between the starting operation node and the ending operation node.

[0122] In this embodiment, the preset efficiency evaluation function is implemented through a defined arithmetic process.

[0123] Specifically, first, the occurrence time of the starting node and the occurrence time of the ending node of the transmission path are read, and the time length between them is calculated. When the time length is 0, the uniform sampling interval in claim 2 is used as the time length. Then, the number of directed edges contained in the transmission path is read, and the hierarchical association strength is divided by the number of directed edges to obtain the unilateral association strength. Subsequently, the unilateral association strength is divided by the time length to obtain the transmission efficiency corresponding to the transmission path. Through this calculation, the transmission efficiency can reflect the impact of fluctuation hierarchy changes, path length, and transmission time on the link transmission process.

[0124] Furthermore, when performing link transfer analysis, ordered node pairs are formed by the starting operation link node and the ending operation link node, and the transfer efficiency of each ordered node pair is written into the transfer efficiency vector according to the experimental procedure.

[0125] If there are more than two transmission paths for the same ordered node pair, the transmission efficiency of each transmission path is added together and then divided by the number of transmission paths to obtain the transmission efficiency of the ordered node pair; if there is no transmission path for an ordered node pair, 0 is written to the corresponding position.

[0126] Furthermore, when constructing the relation matrix, the starting operation node is used as the matrix row and the ending operation node is used as the matrix column, and the value in the transmission efficiency vector is written into the corresponding row and column intersection position.

[0127] To preserve the influence of the fluctuation level on the transmission relationship, four relational sub-matrices are formed for the 1st, 2nd, 3rd, and 4th fluctuation levels respectively; the four relational sub-matrices adopt the same operational step arrangement order. When concatenating the matrices, they are concatenated sequentially according to the 1st, 2nd, 3rd, and 4th fluctuation levels to obtain the step transmission relationship matrix.

[0128] Example 5: S4 includes obtaining the link transmission relationship matrix, extracting the adjacent link nodes corresponding to the non-zero elements in the link transmission relationship matrix, obtaining the node load of the adjacent link nodes in the service transmission process, calculating the node load difference between adjacent link nodes, and obtaining the consumption change; obtaining the consumption change, extracting the extreme values ​​of all consumption changes to determine the maximum and minimum values, calculating the difference between the maximum and minimum values, and obtaining the range value; determining whether the range value is greater than zero, if the range value is greater than zero, subtracting the minimum value from the consumption change to obtain the change difference, dividing the change difference by the range value to obtain the normalized value; obtaining the normalized value, combining the normalized values ​​according to the arrangement order of the adjacent link nodes in the link transmission relationship matrix, and obtaining the transmission effect numerical sequence.

[0129] In this embodiment, the rows of the link transmission relationship matrix correspond to the starting link of the transmission, and the columns correspond to the ending link of the transmission. The value at the intersection of the row and column is the transmission efficiency between the two corresponding operation links. When the value is 0, it means that no effective transmission has been formed between the two corresponding operation links; when the value is not 0, it means that there is a resource consumption transmission relationship between the two corresponding operation links.

[0130] Furthermore, when reading the relation matrix, the matrix elements are read one by one in the order of row number from front to back and column number within the same row from front to back.

[0131] Upon reading a non-zero element, the row and column numbers of the non-zero element are first recorded. Then, the experimental process configuration table is queried based on the row number to determine the starting node; the experimental process configuration table is queried based on the column number to determine the ending node. If the ending node is immediately following the starting node in the experimental process configuration table, the starting node and the ending node are considered adjacent nodes. If there are other operation steps between them, the non-zero element corresponds to a cross-step transfer record.

[0132] In this embodiment, the node load is obtained based on the resource carrying capacity data of adjacent nodes in the current experimental batch.

[0133] In specific processing, the reagent consumption, consumable consumption, power consumption, pure water consumption, and waste liquid generation within the corresponding time period of adjacent process nodes are read. The above resource data comes from the generated second resource consumption data and consumption fluctuation characteristics, and corresponds to the experimental batch number, process number, and time period number. In order to enable different resource types to enter the same load calculation process, each resource type is converted according to the pre-written conversion factor. The conversion factor for Class I reagents is 1, the conversion factor for Class II reagents is 0.6, the conversion factor for consumable consumption is 0.4, the conversion factor for power consumption is 0.3, the conversion factor for pure water consumption is 0.2, and the conversion factor for waste liquid generation is 0.8.

[0134] The aforementioned conversion factors are stored along with the experimental project parameter table and are entered by laboratory management personnel according to the resource classification table before the experimental project is put into use; when the resource classification table is changed, the corresponding conversion factors are updated synchronously.

[0135] Furthermore, when calculating the node load of a certain link node, first multiply the consumption value of each type of resource of the link node by the corresponding conversion factor to obtain the converted load of each resource type; then add up all the converted loads within the same link node to obtain the node load of the link node in the current service transmission process; if there is no consumption record for a certain resource type in the current link, the consumption value corresponding to the resource type is 0 and participates in the same summation process.

[0136] The node load obtained in this way is a value under a uniform scale, which is used to compare the changes in resource carrying capacity between adjacent nodes.

[0137] Furthermore, after obtaining the node load of adjacent links, the change in consumption is calculated.

[0138] Specifically, the node load of the terminating node and the node load of the starting node are read, the two are subtracted, and the absolute value of the subtraction result is taken to obtain the node load difference between adjacent nodes in the group; this node load difference is the consumption change; after using absolute value processing, the cases where the load of the previous node is greater than the load of the next node and the load of the next node is greater than the load of the previous node are all represented by non-negative values ​​to indicate their change magnitude.

[0139] Furthermore, after calculating the consumption changes of all adjacent nodes, extreme values ​​are extracted.

[0140] Specifically, the first consumption change is read first and used as both the initial maximum and minimum values. Then, subsequent values ​​are read according to the order in which the consumption changes were generated. If the currently read consumption change is greater than the recorded maximum value, the current value is replaced with the new maximum value; if the currently read consumption change is less than the recorded minimum value, the current value is replaced with the new minimum value. After all consumption changes have been read, the maximum and minimum consumption changes in the current experimental batch are obtained.

[0141] In this embodiment, the range value is determined by the maximum and minimum values. Specifically, the maximum value is subtracted from the minimum value, and the result is used as the range value. This range value is used to determine whether the consumption change within the current batch has a value difference and serves as the source of the divisor in the normalization calculation.

[0142] Furthermore, when the range value is greater than 0, normalization is performed on each consumption change.

[0143] During processing, the minimum value is first subtracted from the current change in consumption to obtain the change difference; then the change difference is divided by the range value to obtain the normalized value corresponding to the current change in consumption.

[0144] After the above processing, the normalized value corresponding to the minimum consumption change is 0, the normalized value corresponding to the maximum consumption change is 1, and the remaining consumption changes are converted to values ​​between 0 and 1 according to their position between the minimum and maximum values. This normalization process retains the relative change relationship between adjacent nodes, while eliminating the problem of incomparable values ​​caused by different units of measurement for reagents, consumables, electricity, pure water and waste liquid.

[0145] When the range value is equal to 0, it means that the consumption change of all adjacent nodes in the current batch is the same. At this time, the division calculation is not performed, and the normalized value corresponding to all adjacent nodes is written to 0. This process is used to avoid the division by 0, while keeping the calculation result that all adjacent nodes are at the same consumption change level.

[0146] Furthermore, after obtaining the normalized values, a numerical sequence of transmission effects is generated according to the arrangement order of adjacent nodes in the transmission relationship matrix.

[0147] Specifically, adjacent nodes are first arranged from front to back according to the matrix row number; for adjacent nodes with the same row number, they are then arranged from front to back according to the column number; subsequently, the normalized value corresponding to each group of adjacent nodes after sorting is read in sequence, and the normalized value is written into the corresponding position of the transmission effect numerical sequence; each value in the sequence is bound to the starting node number, ending node number, matrix row number, matrix column number, experimental batch number, and time period number.

[0148] Through the above processing, the non-zero transmission relationships in the link transmission relationship matrix are converted into a numerical sequence of transmission effects arranged in the order of the experimental procedure; this sequence not only retains the transmission direction between adjacent link nodes, but also uses normalized values ​​to represent the degree of consumption change between adjacent links.

[0149] Example 6: S5 includes obtaining a transmission effect numerical sequence, extracting the transmission effect values ​​corresponding to each node in the transmission effect numerical sequence, obtaining a preset transmission effect threshold, calculating the absolute value of the difference between the transmission effect value and the preset transmission effect threshold to obtain the deviation magnitude; obtaining the deviation magnitude, if the deviation magnitude is greater than a preset deviation tolerance threshold, determining that the node has abnormal transmission behavior, extracting the node identifier with abnormal transmission behavior and the corresponding associated link identifier, and combining them to generate an abnormal transmission identifier set.

[0150] In this embodiment, S5 is used to determine the anomaly of the transmission effect numerical sequence and form an anomaly transmission identifier set that can be used for subsequent anomaly distribution detection. The transmission effect numerical sequence is obtained from S4. Each sequence position in the sequence stores a transmission effect value and is simultaneously bound to the starting link node number, the ending link node number, the matrix row number, the matrix column number, the experimental batch number, the time period number, and the resource node number.

[0151] Furthermore, after obtaining the numerical sequence of the transmission effect, the data are read item by item in the order of the sequence positions from front to back.

[0152] Specifically, for each sequence bit read, the value of the transmission effect in that sequence bit is obtained first, and then the starting node number and the ending node number bound to that sequence bit are read. The starting node number corresponds to the front-end operation of the transmission relationship, and the ending node number corresponds to the back-end operation of the transmission relationship. The two constitute the node pair corresponding to the current transmission effect value. This node pair serves as the basic object for subsequent threshold matching and anomaly identification generation.

[0153] In this embodiment, the preset transmission effect threshold is generated before the experimental project is activated.

[0154] Specifically, during generation, 30 consecutive batches of data that passed the experiment without any equipment alarm records under the same experimental project, experimental process configuration, spatial layout, equipment configuration, and sampling interval are selected as benchmark data. These 30 batches of benchmark data are then subjected to sequential processing including operation denoising, action resource coupling, link transmission relationship matrix construction, and normalization, resulting in 30 sets of benchmark transmission effect numerical sequences. Subsequently, these sequences are grouped according to the starting link node number, ending link node number, resource node number, and time period number. All benchmark transmission effect values ​​within the same group are summed, and the sum is divided by the number of benchmark transmission effect values ​​within that group to obtain the transmission effect threshold of that node pair under the corresponding resource node and time period.

[0155] This threshold, along with the experiment project number, the starting node number, the ending node number, the resource node number, and the time period number, is written into the parameter table.

[0156] Furthermore, when performing anomaly detection in the current experimental batch, the corresponding transmission effect threshold is read from the parameter table based on the starting node number, ending node number, resource node number, and time period number bound to the current sequence bit.

[0157] If there is no completely corresponding threshold record in the parameter table, the threshold is read according to the same experimental item, the same starting node, and the same ending node. If there is still no corresponding record, the sequence bit will not be judged as abnormal and will be recorded as a baseline data item to be supplemented, so as to avoid making an unfounded judgment when the baseline data is missing.

[0158] Furthermore, after obtaining the current transmission effect value and the transmission effect threshold, the deviation magnitude is calculated.

[0159] In the calculation, the current transmission effect value is subtracted from the corresponding transmission effect threshold, and the absolute value of the subtraction result is taken to obtain the deviation magnitude of the current node pair. After the absolute value processing is used, the deviation magnitude is formed according to the same judgment criterion whether the current transmission effect value is greater than the transmission effect threshold or less than the transmission effect threshold, and the degree of abnormality is not canceled due to the opposite direction of deviation.

[0160] In this embodiment, the deviation tolerance threshold is generated based on the aforementioned 30 batches of benchmark data.

[0161] The specific processing is as follows: First, read the baseline transmission effect value and its corresponding transmission effect threshold for each group, and calculate the absolute deviation between the two; then, add up all the absolute deviation values ​​under the same starting node, the same ending node, the same resource node, and the same time period, and divide the sum by the number of absolute deviation values ​​to obtain the baseline deviation average; subsequently, calculate the difference between each absolute deviation value and the baseline deviation average, square each difference, add up all the squared results, divide by the number of absolute deviation values, and then take the square root of the result to obtain the baseline deviation standard deviation; finally, add the baseline deviation average to twice the baseline deviation standard deviation as the deviation tolerance threshold for that node under the corresponding resource node and time period.

[0162] The deviation tolerance threshold and the transmission effect threshold are stored using the same numbering relationship.

[0163] Furthermore, after calculating the deviation magnitude, the deviation magnitude is compared with the deviation tolerance threshold.

[0164] Specifically, if the deviation is greater than the deviation tolerance threshold, the node pair corresponding to the sequence position is determined to have abnormal transmission behavior; if the deviation is less than or equal to the deviation tolerance threshold, the node pair corresponding to the sequence position is determined not to have abnormal transmission behavior, and the next sequence position is read. The above judgments are executed item by item in the order of the transmission effect numerical sequence until the last position of the sequence is processed.

[0165] As a preferred embodiment, for node pairs that are determined to have abnormal transmission behavior, their node identifiers and associated link identifiers are extracted.

[0166] Specifically, the node identifier consists of the starting node number, the ending node number, the experimental batch number, the time period number, and the resource node number, which are used to indicate the location of the abnormal transmission. The associated link identifier consists of the matrix row number, the matrix column number, the transmission path number, the fluctuation level number, and the spatial location number, which are used to indicate the link source of the abnormal transmission in the link transmission relationship matrix and the directed graph. During extraction, the row and column positions in the link transmission relationship matrix are first looked up based on the position of the transmission effect value in the sequence, and then the transmission path number and fluctuation level number are read based on the row and column positions. Finally, the resource node number and spatial location number are combined to form a complete link record.

[0167] Furthermore, after all sequence bits have been processed, the node identifiers and corresponding associated link identifiers with abnormal transmission behavior are combined according to the original order in the transmission effect value sequence to generate an abnormal transmission identifier set. Each record in this set includes a node identifier, associated link identifier, current transmission effect value, transmission effect threshold, deviation magnitude, deviation tolerance threshold, and abnormal judgment result.

[0168] Example 7: S6 includes obtaining an anomaly propagation identifier set, extracting resource consumption and state deviation from the anomaly propagation identifier set, combining resource consumption and state deviation to construct an anomaly feature matrix; using the isolated forest algorithm to detect anomaly distribution in the anomaly feature matrix, constructing multiple isolated trees, calculating the average path length of each data point in the anomaly feature matrix in the multiple isolated trees, obtaining an anomaly score based on the average path length, and determining the corresponding data point as an anomaly feature point if the anomaly score is greater than a preset threshold, thus obtaining an anomaly feature point set; using the mean clustering algorithm to perform cluster analysis on the anomaly feature point set, calculating the Euclidean distance between each anomaly feature point in the anomaly feature point set, dividing the anomaly feature point set into at least one anomaly cluster based on the Euclidean distance, calculating the mean of all anomaly feature points in at least one anomaly cluster, and obtaining the coordinates of the anomaly cluster center point; obtaining the network topology diagram corresponding to the coordinates of the anomaly cluster center point, finding adjacent nodes connected to the coordinates of the anomaly cluster center point in the network topology diagram, comparing the resource consumption of adjacent nodes, and determining the adjacent node with the largest resource consumption as the waste location node coordinate that triggered the anomaly propagation behavior.

[0169] In this embodiment, S6 is used to perform feature processing on the abnormal transmission identifier set and determine the coordinates of the waste location node after abnormal distribution detection and cluster analysis; the abnormal transmission identifier set is output by S5, and each abnormal transmission identifier includes experimental batch number, time period number, starting link node number, ending link node number, resource node number, spatial location number, transmission effect value, transmission effect threshold, deviation magnitude and deviation tolerance threshold.

[0170] Furthermore, after obtaining the set of anomaly propagation identifiers, the resource consumption is extracted first.

[0171] Specifically, based on the experimental batch number, resource node number, spatial location number, and time period number in the anomaly transmission identifier, the second resource consumption data is retrieved, and the actual consumption record of the resource node within the corresponding time period is read. When the resource type is reagent, consumable, electricity, pure water, or waste liquid generation, it is first converted into a uniform load value according to the conversion factor corresponding to the resource type. The conversion factor is 1 for Class I reagents, 0.6 for Class II reagents, 0.4 for consumables, 0.3 for electricity, 0.2 for pure water, and 0.8 for waste liquid generation. If the same anomaly transmission identifier corresponds to more than two resource consumption records within the same time period, the converted resource consumption values ​​are added together to obtain the resource consumption amount for that anomaly transmission identifier.

[0172] In this embodiment, the state deviation is determined by the deviation magnitude and the deviation tolerance threshold.

[0173] Specifically, during processing, the deviation magnitude in the anomaly propagation identifier is read, followed by the corresponding deviation tolerance threshold. The deviation magnitude is then divided by the deviation tolerance threshold, and the result is used as the state deviation degree. When the deviation tolerance threshold is 0, the divisor is replaced with 0.0001 to avoid calculation interruption due to a division by zero; simultaneously, a threshold correction flag is written to the anomaly propagation identifier.

[0174] Furthermore, when constructing the anomaly feature matrix, each anomaly transmission identifier is treated as a data point, and resource consumption and state deviation are treated as two features of that data point; before writing into the matrix, resource consumption and state deviation are normalized respectively.

[0175] The specific processing method is as follows: First, find the maximum and minimum resource consumption values ​​in the current anomaly propagation identifier set. Then, subtract the minimum resource consumption value from the current resource consumption value, and divide the resulting difference by the difference between the maximum and minimum resource consumption values ​​to obtain the normalized resource consumption value. The state deviation is processed in the same way. If the maximum and minimum values ​​of a certain feature are equal, then all normalized values ​​in that feature column are written as 0. After completing the above processing, the anomaly feature matrix is ​​obtained.

[0176] Furthermore, when using the isolated forest algorithm for anomaly detection, the number of isolated trees is set to 100.

[0177] Specifically, 70% of the data points in the abnormal feature matrix are extracted as training samples for each isolated tree; when the number of extractions for 70% is less than 1, the number of extractions is 1; in order to ensure that the same experimental batch obtains consistent results under the same input, the extraction process and the splitting process both use pseudo-random seeds generated by the experimental batch number, tree number and time period number.

[0178] Furthermore, when constructing each isolated tree, a splitting feature is first selected from the resource consumption feature and the state deviation feature. Then, the maximum and minimum values ​​of the splitting feature in the current node sample are read, and a splitting value is generated between the maximum and minimum values.

[0179] In this process, when a data point's value on the splitting feature is less than the splitting value, it enters the left child node; when its value is greater than or equal to the splitting value, it enters the right child node. Each child node continues to be divided in the same way until the child node contains only one data point, or the tree depth reaches 8 levels. Through this process, 100 isolated trees form the division results of the data points in the anomaly feature matrix.

[0180] Furthermore, when calculating the average path length, each data point in the anomaly feature matrix is ​​sequentially input into 100 isolated trees.

[0181] For any given data point, starting from the root node of the isolated tree, the data point enters either the left or right child node based on its value in the splitting feature, until it reaches a leaf node. The number of edges traversed by the data point from the root node to the leaf node is recorded as its path length in the isolated tree. After traversing 100 isolated trees, the path lengths of the data point in the 100 isolated trees are added together and then divided by 100 to obtain the average path length of the data point.

[0182] As a preferred embodiment, the anomaly score is calculated based on the average path length.

[0183] Specifically, the path length correction value corresponding to the number of samples is first obtained. The process of obtaining the path length correction value is as follows: take the reciprocal of each integer from 1 to the number of samples minus 1 and sum them up. Multiply the sum by 2. Then divide twice the number of samples minus 1 by the number of samples. Finally, subtract the next result from the previous result to obtain the path length correction value.

[0184] Next, the average path length is divided by the path length correction value, the opposite of the quotient is taken, and an exponential operation is performed with base 2. The result is the anomaly score. When the average path length is 0, the anomaly score is 1. When the path length correction value is 0, the data point is written into the data table to be reviewed and is not included in the anomaly feature point judgment.

[0185] In this embodiment, the abnormal score threshold is determined by the baseline data.

[0186] Specifically, the baseline data is selected from 30 consecutive batches of data that passed the experiment without any equipment alarm records under the same experimental project, experimental process configuration, equipment configuration and spatial layout; the 30 batches of baseline data are used to construct an anomaly feature matrix in the same way, and the same isolated forest parameters are used to calculate the baseline anomaly score.

[0187] Subsequently, all baseline outlier scores are summed and divided by the number of baseline outliers to obtain the average baseline outlier score. Then, the difference between each baseline outlier score and this average is calculated, the squares of these differences are summed, the sum is divided by the number of baseline outliers, and the square root of the sum is taken to obtain the standard deviation of the baseline outlier score. Finally, the average baseline outlier score is added to three times the standard deviation of the baseline outlier score to obtain the outlier score threshold. If the calculated threshold is greater than 1, the outlier score threshold is set to 1.

[0188] Furthermore, the outlier score of the current data point is compared with the outlier score threshold.

[0189] Specifically, if the anomaly score is greater than the anomaly score threshold, the data point is identified as an anomalous feature point; if the anomaly score is less than or equal to the anomaly score threshold, the data point is not added to the anomalous feature point set. After all data points are evaluated, an anomalous feature point set is obtained; each anomalous feature point in this set retains its corresponding resource consumption, state deviation, node identifier, associated link identifier, spatial location number, and time period number.

[0190] In this embodiment, the mean clustering algorithm is used to perform cluster analysis on the set of abnormal feature points.

[0191] Specifically, the number of clusters is determined according to the number of outlier feature points: when there are 1 to 3 outlier feature points, the number of clusters is 1; when there are 4 to 10 outlier feature points, the number of clusters is 2; when there are 11 to 20 outlier feature points, the number of clusters is 3; when there are more than 20 outlier feature points, the number of clusters is 4. The initial centroids are selected according to the order of the outlier feature point set; when the number of clusters is 1, the first outlier feature point is selected; when the number of clusters is 2 or more, the corresponding outlier feature points are selected as the initial centroids according to the equal division of the set length.

[0192] Furthermore, when calculating the Euclidean distance, the resource consumption and state deviation of the two abnormal feature points are read.

[0193] First, the difference between the resource consumption of the two points is calculated and then squared. Next, the difference between the deviations between the two points is calculated and then squared. Then, the two squared results are added together and the square root of the sum is taken to obtain the Euclidean distance between the two abnormal feature points. For each abnormal feature point, the Euclidean distance between it and each initial center point is calculated, and the abnormal feature point is assigned to the abnormal cluster corresponding to the center point with the smallest Euclidean distance value.

[0194] Furthermore, after completing one partition, the coordinates of the center point of each anomalous cluster are recalculated.

[0195] Specifically, during the calculation, the resource consumption of all anomalous feature points within the anomaly cluster is read, summed, and divided by the number of anomalous feature points in the cluster to obtain the resource consumption coordinates of the center point. Then, the state deviation of all anomalous feature points within the cluster is read, summed, and divided by the number of anomalous feature points in the cluster to obtain the state deviation coordinates of the center point. Subsequently, the Euclidean distance is recalculated using the new center point coordinates, and the anomaly clusters are redefined. This process is executed continuously until the center point coordinates of the anomaly clusters obtained from two consecutive iterations are consistent, or the number of iterations reaches 20. After the iterations are completed, one or more anomaly clusters and their corresponding center point coordinates are obtained.

[0196] Furthermore, after obtaining the coordinates of the center point of the abnormal cluster, the corresponding network topology diagram is searched.

[0197] In the process, the abnormal feature point with the smallest Euclidean distance to the center point of the abnormal cluster is first found within the abnormal cluster, and this abnormal feature point is used as the central representative point. Then, based on the node identifier and associated link identifier of the central representative point, the corresponding directed graph, link transmission relationship matrix and spatial mapping result are read to generate a network topology diagram. The nodes of this network topology diagram include operation link nodes, action nodes, resource nodes and spatial location nodes, and the edges include link transmission edges, action resource association edges and spatial mapping edges.

[0198] In this embodiment, in the network topology diagram, the node corresponding to the central representative point is taken as the topology center node, and the adjacent nodes that are directly connected to the topology center node through one edge are found.

[0199] Specifically, adjacent nodes include preceding operation nodes, subsequent operation nodes, action nodes, resource nodes, and spatial location nodes. For each adjacent node, its resource consumption within the same experimental batch and the same time period is read; the resource consumption is still converted into a uniform load value according to the aforementioned resource conversion factor; when there are more than two resource consumption records for the same adjacent node, the converted resource consumption is added together.

[0200] Furthermore, after reading the resource consumption of adjacent nodes, the adjacent nodes are arranged in descending order of resource consumption. The adjacent node ranked first is identified as the wasteful location node that caused the abnormal transmission behavior. If the resource consumption of two or more adjacent nodes is equal, the adjacent node that is in the same time period as the topology center node is selected. If there are still equal results, the adjacent node that appears earlier in the experimental process configuration table is selected.

[0201] Finally, the operation link number, action node number, resource node number, spatial location number, and time period number of the wasteful positioning node are combined to form the coordinates of the wasteful positioning node.

[0202] Example 8: S7 includes obtaining a coordinate mapping matrix based on the coordinates of the waste location node, extracting the container transfer trajectory through the coordinate mapping matrix, calculating the transfer probability of the container transfer trajectory using a hidden Markov model, performing action sequence inversion based on the transfer probability to obtain the specific actions in the operation action sequence; extracting reagent consumption deviation and reaction temperature fluctuation for the specific actions, and determining the specific action as the target operation action that triggers abnormal resource transfer if the reagent consumption deviation is greater than a preset threshold; extracting the abnormal transfer path of the target operation action, extracting waste liquid discharge and instrument occupancy time based on the abnormal transfer path, calculating the ratio of waste liquid discharge to instrument occupancy time to obtain the consumable consumption rate; obtaining process evaluation indicators by combining the consumable consumption rate with preset evaluation weights, locking the link position in a preset laboratory topology map based on the process evaluation indicators to obtain the waste link position; matching the waste link position in a preset abnormal cause library to extract the abnormal resource consumption cause, generating an optimization management strategy based on the abnormal resource consumption cause and process evaluation indicators, and obtaining the full process management results of the chemical laboratory.

[0203] In this embodiment, S7 is used to trace back the specific actions in the second operation action sequence based on the coordinates of the waste positioning node, and generate the full process management results of the chemical laboratory accordingly.

[0204] Specifically, the coordinates of the waste positioning nodes include the operation link number, action node number, resource node number, spatial location number, and time cycle number; these coordinates are output by S6 and indexed with the abnormal transmission identifier set, the second operation action sequence, the second resource consumption data, the laboratory topology map, and the abnormal cause database.

[0205] Furthermore, when obtaining the coordinate mapping matrix based on the coordinates of the wasted positioning nodes, the spatial location number and time period number in the coordinates of the wasted positioning nodes are read first, and then records with the same spatial location number and time period number are searched in the coordinate mapping matrix.

[0206] Specifically, the coordinate mapping matrix consists of container number, equipment number, operation link number, action node number, spatial location number, horizontal coordinate, vertical coordinate, and time period number. After the target record is retrieved, the coordinate records of the same container number in the three time periods before and after the target time period are read and arranged from front to back according to the time period to form a data segment for trajectory extraction.

[0207] Furthermore, when extracting the container transfer trajectory through the coordinate mapping matrix, the spatial location number, horizontal coordinate, and vertical coordinate of the same container number within each time period are first obtained.

[0208] When the spatial position numbers of two adjacent time periods are the same, only the first position record is retained; when the spatial position numbers of two adjacent time periods are different, the container is recorded as moving from the previous spatial position to the next spatial position, and the corresponding departure time, arrival time, start coordinates, and end coordinates are written into the trajectory record. After processing all time periods, a container transfer trajectory consisting of the start position, the position passed through, the end position, and the corresponding time period is obtained. This container transfer trajectory is used to characterize the movement of the container between the reagent storage area, the dropping operation area, the stirring reaction area, the sampling and detection area, and the waste liquid collection area.

[0209] In this embodiment, when using a hidden Markov model to calculate the transfer probability of the container transfer trajectory, the actual operation actions are taken as hidden states, and the container position changes, equipment start-up and shutdown states, resource consumption records and time periods are taken as observation data.

[0210] Specifically, the hidden states include taking a container, transferring a container, adding reagents, starting stirring, adjusting the temperature, sampling and testing, discharging waste liquid, and cleaning the container. The transfer probability is determined by historical action data, which is selected from 30 consecutive batches of experiments that passed without any equipment alarm records under the same experimental project, the same process configuration, and the same equipment configuration. When making statistics, first count the number of times one operation is immediately followed by another operation, then count the total number of times the first operation appears as a preceding operation. Divide the number of the immediately following operations by the total number of preceding operations to obtain the transfer probability between the two operations.

[0211] Specifically, the observation matching probability in the Hidden Markov Model is also determined by the aforementioned 30 batches of historical action data.

[0212] In the process, the number of times a corresponding container position change, equipment status, and resource consumption record occur when a certain operation occurs is counted. Then, the total number of times the operation occurs is counted. The aforementioned count is divided by the total number of times the operation occurs to obtain the matching probability of the observed data corresponding to the operation. The initial operation probability is determined by counting the number of times the first operation occurs in each batch of 30 batches of historical operation data. The number of times a certain operation occurs as the first operation is divided by 30 to obtain the initial operation probability of the operation.

[0213] Furthermore, when performing action sequence inversion based on transition probabilities, the observation data is read segment by segment according to the time sequence of the container transition trajectory.

[0214] For the first trajectory segment, the initial action probability and observation matching probability of each candidate action are read and multiplied to obtain the path score of that candidate action. For the second and subsequent trajectory segments, the candidate path score retained from the previous trajectory segment is read, multiplied by the transition probability from the previous candidate action to the current candidate action, and then multiplied by the matching probability of the observation data corresponding to the current candidate action to obtain the candidate path score of the current trajectory segment. The candidate path with the highest path score in each trajectory segment is retained. When two candidate path scores are the same, the candidate path that matches the action order in the experimental procedure configuration table is retained. After processing the last trajectory segment, the action path with the highest path score is used as the inversion result, thereby obtaining the specific actions in the second action sequence.

[0215] As a preferred embodiment, for the specific actions obtained from the inversion, the deviation in reagent consumption and the fluctuation in reaction temperature are extracted.

[0216] Specifically, the process for obtaining reagent consumption deviation is as follows: first, read the actual reagent consumption in the second resource consumption data for the specific action; then, read the standard reagent dosage corresponding to the experimental item number, reagent number, and action node number in the experimental formula table; subtract the standard reagent dosage from the actual reagent consumption, and take the absolute value of the subtraction result to obtain the reagent consumption deviation; the process for obtaining reaction temperature fluctuation is as follows: read the temperature acquisition data during the occurrence of the specific action, extract the temperature value ranked first and the temperature value ranked last from the data, and subtract the temperature value ranked last from the temperature value ranked first to obtain the reaction temperature fluctuation.

[0217] In this embodiment, the reagent consumption deviation threshold is determined before the experiment is started.

[0218] When determining the specifics, select data from 30 consecutive batches of experiments that are qualified and have no equipment alarm records under the same experimental item, same reagent number, same dropping equipment, and same sampling interval, and calculate the reagent consumption deviation for the corresponding action in each batch.

[0219] Subsequently, all reagent consumption deviations are summed and divided by the number of reagent consumption deviations to obtain the average baseline reagent consumption deviation. Then, each reagent consumption deviation is subtracted from the average baseline reagent consumption deviation, and the difference is squared. All squared results are summed and divided by the number of reagent consumption deviations, and the square root of the result is taken to obtain the standard deviation of the baseline reagent consumption deviation. Finally, the average baseline reagent consumption deviation is added to three times the standard deviation of the baseline reagent consumption deviation to obtain the reagent consumption deviation threshold. This threshold is stored together with the experiment item number, reagent number, dispensing device number, and action node number.

[0220] In this embodiment, the reagent consumption deviation of the current specific action is compared with the reagent consumption deviation threshold.

[0221] Specifically, when the reagent consumption deviation exceeds the reagent consumption deviation threshold, the specific action is identified as the target operation action that triggers abnormal resource transmission; when the reagent consumption deviation does not exceed the reagent consumption deviation threshold, the same judgment is performed on the next specific action in the inversion result; the reaction temperature fluctuation is not involved in the direct judgment of the target operation action, but is saved as temperature status data when matching subsequent abnormal causes.

[0222] Furthermore, when extracting the abnormal transmission path of the target operation action, the corresponding associated link identifier is retrieved from the abnormal transmission identifier set based on the action node number, resource node number, spatial location number, and time period number corresponding to the target operation action. After retrieving the associated link identifier, the starting link node, ending link node, transmission path number, matrix row number, and matrix column number are read to form the abnormal transmission path corresponding to the target operation action. This abnormal transmission path is used to limit the extraction range of waste liquid discharge and instrument occupancy time.

[0223] Furthermore, when extracting the waste liquid discharge volume and instrument occupancy time based on the abnormal transmission path, first read the waste liquid weighing record, waste liquid level record or waste liquid registration record within the time period corresponding to the abnormal transmission path, and add up the waste liquid records belonging to the same resource node within the same time period to obtain the waste liquid discharge volume.

[0224] Then, the activation time and shutdown time in the instrument usage record are read, and the activation time is subtracted from the shutdown time to obtain the instrument usage duration. When the activation time and shutdown time are the same, the minimum timing unit of 1 minute preset in the laboratory management platform is used as the instrument usage duration to avoid the divisor being 0 in subsequent ratio calculations.

[0225] In this embodiment, the consumable wear rate is calculated by the waste liquid discharge volume and the instrument occupancy time.

[0226] Specifically, the waste liquid discharge volume is divided by the instrument occupancy time, and the result is used as the consumable wear rate. This consumable wear rate is linked to the target operation action, abnormal transmission path, experimental batch number and time cycle number and stored for subsequent process evaluation index calculation.

[0227] As a preferred embodiment, the preset evaluation weights are determined by the experimental item parameter table.

[0228] Specifically, in the parameter table, the weight of consumable consumption rate is set to 0.5, the weight of reagent consumption deviation is set to 0.3, and the weight of reaction temperature fluctuation is set to 0.2, with the sum of the three weights being 1. Before calculating the process evaluation index, the consumable consumption rate, reagent consumption deviation, and reaction temperature fluctuation are normalized respectively. During normalization, the maximum and minimum values ​​of the corresponding index are first read in the current experimental batch. The minimum value is subtracted from the current index value, and then divided by the difference between the maximum and minimum values ​​to obtain the normalized value of the index. When the maximum and minimum values ​​are equal, the normalized value of the index is written as 0.

[0229] Subsequently, the three normalized values ​​are multiplied by their corresponding evaluation weights, and the three products are added together to obtain the process evaluation index.

[0230] Furthermore, when locking the location of a process in the preset laboratory topology map based on the process evaluation indicators, the operation process number, action node number, resource node number and spatial location number corresponding to the target operation action are read first, and then the corresponding node is retrieved in the laboratory topology map.

[0231] Specifically, the laboratory topology includes operation link nodes, action nodes, resource nodes, equipment nodes, and spatial location nodes. Nodes are connected by operation sequence edges, action resource edges, equipment occupancy edges, and spatial mapping edges. After retrieving the node corresponding to the target operation action, the process evaluation index is written into the operation link node directly connected to that node. If the same target operation action corresponds to more than two operation link nodes, the operation link node with the same number as the starting or ending link node in the exception propagation path is selected first. If there are still more than two matching results, the operation link node that appears earlier in the experimental process configuration table is selected. The selected operation link node is identified as a wasted link location.

[0232] Furthermore, when matching the location of the waste step in the preset abnormal cause database, the system first reads the operation step number, resource node number, target operation action, reagent consumption deviation, reaction temperature fluctuation, and consumable consumption rate corresponding to the waste step location.

[0233] Specifically, the abnormal cause database pre-stores the correspondence between operation step number, resource type, action type, indicator conditions and abnormal causes; during matching, candidate cause records are first filtered according to operation step number; then cause records are filtered according to resource node number and target operation action; finally, reagent consumption deviation, reaction temperature fluctuation and consumable consumption rate are compared with the indicator conditions in the candidate cause records; cause records that meet the indicator conditions are extracted as abnormal causes of resource consumption.

[0234] In this embodiment, the index conditions in the anomaly cause database are determined by benchmark experimental data.

[0235] Specifically, for 30 consecutive batches of data from the same experimental project that passed the experiment without any equipment alarm records, the baseline average and baseline standard deviation of reagent consumption deviation, reaction temperature fluctuation, and consumable consumption rate are calculated respectively; the baseline average plus 3 times the baseline standard deviation is used as the cause matching limit for the corresponding indicator; when the current indicator value exceeds the corresponding cause matching limit, it is determined that the indicator condition corresponding to the abnormal cause is met; the cause record is bound to the experimental project number, operation link number, resource type, and action type, and the corresponding cause matching limit is regenerated when the experimental process or equipment configuration is changed.

[0236] Furthermore, when generating optimization management strategies based on the causes of abnormal resource consumption and process evaluation indicators, the strategy records corresponding to the causes of abnormalities are read from the strategy rule table.

[0237] Specifically, the strategy rule table pre-stores the correspondence between anomaly causes, process evaluation index ranges, and management strategies. The process evaluation index range is determined by the benchmark process evaluation index: the process evaluation indexes of 30 batches of benchmark experimental data are added together and divided by the quantity to obtain the average value of the benchmark process evaluation index; then the standard deviation of the benchmark process evaluation index is calculated; the benchmark average value plus one standard deviation is used as the upper limit of level 1, the benchmark average value plus two standard deviations is used as the upper limit of level 2, and the range exceeding the upper limit of level 2 is used as the range of level 3; when the process evaluation index falls into level 1, a prompting strategy is generated; when it falls into level 2, a review strategy is generated; when it falls into level 3, a mandatory handling strategy is generated.

[0238] In this embodiment, the specific strategy is generated according to the cause of abnormal resource consumption.

[0239] Specifically, when the abnormal resource consumption is caused by a metering pump calibration deviation, a metering pump calibration task, a pre-dropping no-load verification task, and a dropping volume verification rule are generated; when the abnormal resource consumption is caused by repeated dropping, a dropping action interlock rule and an operation confirmation rule are generated; when the abnormal resource consumption is caused by a heating module response delay, a heating module maintenance task and a temperature control curve verification task are generated; when the abnormal resource consumption is caused by an abnormal waste liquid discharge, a waste liquid diversion inspection task and a waste liquid container replacement rule are generated. Each optimization management strategy is bound to the target operation action, the location of the waste link, the responsible equipment number, the generation time, and the strategy level.

[0240] Ultimately, the causes of abnormal resource consumption, the location of wasteful processes, process evaluation indicators, target operational actions, abnormal transmission paths, consumable consumption rates, and optimization management strategies are combined to form the overall process management results for the chemical laboratory.

[0241] Example 9: Figure 2 This is a structural block diagram of the local terminal of an exemplary electronic device (machine) of the present invention; as shown... Figure 2 As shown, the electronic device of the present invention includes a processor 11, a memory 12, a storage space 13 for storing program code, and program code 14 for executing the method steps according to the present invention. The program code 14 for executing the method steps according to the present invention is used to execute the above-described control logic.

[0242] Figure 3 This is a structural block diagram of the network end of an exemplary electronic device of the present invention; as shown below. Figure 3 As shown, the present invention also provides an electronic device (machine), which may include at least one processor 210, at least one memory 230 communicatively connected to the processor, and a communication bus 240 and a communication interface 220 connecting different system components (including the memory 230 and the processor 210). The processor 210, the memory 230 and the communication interface 220 are connected through the communication bus 240 and communicate with each other. The communication interface 220 is used for data interaction with external devices. The memory 230 stores a machine-executable program that can be executed by the processor, and the processor 210 can execute the above-mentioned control logic by calling the machine-executable program.

[0243] Communication bus 240 represents one or more of several bus architectures, including a memory bus or memory controller, peripheral bus, graphics acceleration port, processor, or local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnection (PCI) bus.

[0244] Electronic devices typically include a variety of computer system readable media, which can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0245] Memory 230 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The electronic device may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 230 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the control logic described above.

[0246] A program / utility having a set (at least one) of program modules can be stored in memory 230. Such program modules include, but are not limited to, an operating system, one or more applications, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment.

[0247] Machine-executable programs for performing this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, C++, and Python, and may also include specialized engineering languages ​​such as R. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0248] The present invention also discloses a storage medium on which a machine-executable program as described above is stored.

[0249] The aforementioned storage medium may be any combination of one or more computer-readable media. Computer-readable media may be, for example, computer-readable signal media or computer-readable storage media. Computer-readable storage media include, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium may be, for example, any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0250] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0251] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0252] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for full-process management of a chemical laboratory based on data analysis and processing, characterized in that, include: S1. Obtain the first operation sequence and the corresponding first resource consumption data of the entire process in the chemical laboratory. Use a preset filtering algorithm to denoise the first operation sequence and the first resource consumption data to obtain the second operation sequence and the second resource consumption data. S2. Based on the second operation sequence and the second resource consumption data, construct the action resource coupling tensor and perform time-delay attention decoupling to obtain the consumption fluctuation characteristics that include time and space dimensions. S3. By consuming fluctuation characteristics, a multi-level directed graph structure is constructed to map nodes and determine the link transmission relationship matrix between nodes in different operation links. S4. Based on the link transmission relationship matrix, calculate the consumption change between adjacent link nodes and perform normalization to obtain the quantified transmission effect numerical sequence. S5. Obtain the numerical sequence of the transmission effect. If a specific value in the numerical sequence of the transmission effect is greater than a preset threshold, it is determined that there is abnormal transmission behavior and an abnormal transmission identifier set is obtained. S6. Using the set of abnormal propagation identifiers, the isolated forest algorithm is used to perform abnormal distribution detection and cluster analysis to determine the coordinates of the wasteful location nodes that cause abnormal propagation behavior; S7. Based on the coordinates of the waste location node, trace back to the specific actions in the corresponding second operation action sequence, determine the target operation action that caused the abnormal transmission of resources, and generate the full process management results of the chemical laboratory based on the target operation action, including the cause of abnormal resource consumption, the location of the waste link, the process evaluation results, and the optimization management strategy.

2. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 1, characterized in that: S1 includes: Acquire a first operation sequence and first resource consumption data. The first operation sequence includes reagent addition and stirring actions. The first resource consumption data includes reagent usage and power consumption. Align the first operation sequence and the first resource consumption data using timestamps to obtain an initial sequence. Based on the initial sequence, the mean filtering algorithm is used to calculate the average values ​​of reagent usage and power consumption within a preset time window. The deviation between the actual value and the average value in the initial sequence is calculated. If the deviation exceeds the preset threshold, it is determined that the sequence node corresponding to the actual value has isolated data. The isolated data in the initial sequence are removed by the mean filtering algorithm, and the isolated data are then smoothly replaced by the mean value to obtain the denoised sequence. Extract continuous action features and real consumption features from the denoised sequence, and determine the second operation action sequence and the second resource consumption data based on the action features and consumption features.

3. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 1, characterized in that: S2 includes: Obtain the second operation action sequence and the second resource consumption data, extract the action frequency of the second operation action sequence and the consumable consumption of the second resource consumption data, and construct the action resource coupling tensor based on the action frequency and consumable consumption. For the execution space mapping of the action-resource coupling tensor, the node association relationship between action nodes and resource nodes inside the action-resource coupling tensor is extracted, and the spatial distribution matrix is ​​obtained based on the node association relationship. The trajectory offset data is obtained from the second operation sequence, and the spatial distribution matrix is ​​weighted according to the trajectory offset data to obtain the weighted spatial matrix. A self-attention mechanism is used to decouple the weighted space matrix by time delay in the time dimension, extract the periodic comparison difference of the weighted space matrix in different time periods, and determine the dimension splitting vector based on the periodic comparison difference; Peak values ​​are extracted from the weighted space matrix based on the dimensional splitting vector to obtain the consumption peak. If the consumption peak is greater than a preset threshold, consumption fluctuation features including time and space dimensions are generated based on the consumption peak.

4. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 1, characterized in that: S3 includes: Obtain the first consumption fluctuation feature, extract the fluctuation amplitude based on the first consumption fluctuation feature, map the fluctuation amplitude to a preset fluctuation amplitude range and divide it into multiple fluctuation levels, use the fluctuation levels as nodes, extract the pointing relationship between nodes based on the time dimension in the first consumption fluctuation feature, and construct a directed graph based on the nodes and the pointing relationship. For nodes in the directed graph extraction operation, the transmission path is determined based on the pointing relationship between the nodes in the operation. Node mapping is performed on the transmission path to obtain the node mapping relationship. The path weight is calculated based on the node mapping relationship and the transmission frequency on the transmission path. Obtain the path weight. If the path weight is greater than the preset threshold, extract the hierarchical association strength between the nodes of the operation link based on the path weight. Input the hierarchical association strength into the preset efficiency evaluation function for calculation to determine the transmission efficiency between the nodes of the operation link. By analyzing the transmission efficiency of each node in the operation process, a transmission efficiency vector is obtained. A relation matrix is ​​constructed based on the transmission efficiency vector, and the relation matrix is ​​concatenated to obtain the transmission relation matrix.

5. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 1, characterized in that: S4 includes: Obtain the link transmission relationship matrix, extract the adjacent link nodes corresponding to the non-zero elements in the link transmission relationship matrix, obtain the node load of the adjacent link nodes in the service transmission process, calculate the node load difference between adjacent link nodes, and obtain the consumption change. Obtain the consumption change, extract the extreme values ​​of all consumption changes to determine the maximum and minimum values, calculate the difference between the maximum and minimum values, and obtain the range value; Determine if the range value is greater than zero. If the range value is greater than zero, subtract the minimum value from the change in consumption to obtain the change difference value. Divide the change difference value by the range value to obtain the normalized value. Obtain the normalized values, and combine the normalized values ​​according to the order of adjacent nodes in the link transmission relationship matrix to obtain the numerical sequence of transmission effects.

6. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 1, characterized in that: S5 includes: Obtain the numerical sequence of the transmission effect, extract the transmission effect value corresponding to each node in the numerical sequence of the transmission effect, obtain a preset transmission effect threshold, calculate the absolute value of the difference between the transmission effect value and the preset transmission effect threshold, and obtain the deviation magnitude. Obtain the deviation magnitude. If the deviation magnitude is greater than the preset deviation tolerance threshold, it is determined that the node has abnormal transmission behavior. Extract the node identifier with abnormal transmission behavior and the corresponding associated link identifier, and combine them to generate an abnormal transmission identifier set.

7. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 1, characterized in that: S6 includes: Obtain the set of anomaly propagation identifiers, extract the resource consumption and state deviation from the set of anomaly propagation identifiers, combine the resource consumption and state deviation to construct an anomaly feature matrix; An isolated forest algorithm is used to detect abnormal distributions in the abnormal feature matrix. Multiple isolated trees are constructed, and the average path length of each data point in the abnormal feature matrix in the multiple isolated trees is calculated. An abnormal score is obtained based on the average path length. If the abnormal score is greater than a preset threshold, the corresponding data point is determined to be an abnormal feature point, and a set of abnormal feature points is obtained.

8. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 7, characterized in that: S6 further includes: Cluster analysis is performed on the set of anomalous feature points using the mean clustering algorithm. The Euclidean distance between each anomalous feature point in the set is calculated. Based on the Euclidean distance, the set of anomalous feature points is divided into at least one anomalous cluster. The mean of all anomalous feature points in at least one anomalous cluster is calculated to obtain the coordinates of the center point of the anomalous cluster. Obtain the network topology diagram corresponding to the coordinates of the center point of the abnormal cluster. Find the adjacent nodes connected to the coordinates of the center point of the abnormal cluster in the network topology diagram. Compare the resource consumption of the adjacent nodes and identify the adjacent node with the largest resource consumption as the waste location node coordinate that caused the abnormal transmission behavior.

9. The method for full-process management of a chemical laboratory based on data analysis and processing according to claim 1, characterized in that: S7 includes: The coordinate mapping matrix is ​​obtained based on the coordinates of the waste positioning node. The container transfer trajectory is extracted through the coordinate mapping matrix. The transfer probability of the container transfer trajectory is calculated using a hidden Markov model. Based on the transfer probability, the action sequence is inverted to obtain the specific actions in the operation action sequence. For specific actions, extract reagent consumption deviation and reaction temperature fluctuation. If the reagent consumption deviation is greater than the preset threshold, then the specific action is identified as the target operation that triggers abnormal resource transfer. Extract the abnormal transmission path of the target operation action, extract the waste liquid discharge volume and instrument occupancy time based on the abnormal transmission path, calculate the ratio of waste liquid discharge volume to instrument occupancy time, and obtain the consumable wear rate.

10. A method for full-process management of a chemical laboratory based on data analysis and processing according to claim 9, characterized in that: The S7 also includes: By combining the consumable consumption rate with the preset evaluation weights, process evaluation indicators are obtained. Based on the process evaluation indicators, the location of the waste process is locked in the preset laboratory topology map. By matching the location of wasteful processes with a pre-set database of abnormal causes, the causes of abnormal resource consumption are extracted. Based on the causes of abnormal resource consumption and process evaluation indicators, optimization management strategies are generated to obtain the full-process management results of the chemical laboratory.

Images