An integrated management method for laboratory environmental testing data
By introducing a Gaussian process regressor and version vector mechanism into the laboratory environmental monitoring data management system, the problems of global constraints and physical correlations between laboratories were solved, and the integrated management and causal consistency maintenance of distributed laboratory environmental monitoring data were realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PANJIN ZHENGNENG TECH CO LTD
- Filing Date
- 2026-01-29
- Publication Date
- 2026-05-26
AI Technical Summary
In the laboratory management scenarios of large institutions, the environmental monitoring data management systems that operate independently in each laboratory may violate global constraints or physical correlation constraints between parameters, resulting in the disruption of the consistency of integrated management, and clock drift leading to incorrect priority determination.
By obtaining the physical constraint equations and globally shared constraints between environmental parameters, a joint feasible region is defined. Iterative search is performed using a Gaussian process regressor and version vector mechanism to generate optimal threshold candidates. Configuration update messages are broadcast, and configuration merging or rollback is performed through causal order determination and constraint satisfaction checks to ensure that global constraints are satisfied.
It enables integrated management of environmental monitoring data in a distributed multi-laboratory scenario, ensuring the physical consistency of environmental parameter threshold combinations and the coordination and unification of global constraints, and avoiding priority determination errors caused by physical clock asynchrony.
Smart Images
Figure CN121636255B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of laboratory information management and environmental monitoring technology, and more specifically, to an integrated management method for laboratory environmental monitoring data. Background Technology
[0002] In large-scale laboratory management scenarios, multiple laboratories are typically distributed across different buildings or locations. Each laboratory deploys environmental monitoring sensors to continuously collect environmental data such as temperature, humidity, dew point, and air pressure. Each laboratory operates its own environmental data management system but shares certain environmental control equipment (such as zonal air conditioning systems and centralized gas supply systems). To achieve integrated management of environmental monitoring data, it is necessary to uniformly configure the data alert thresholds for each laboratory. However, physical constraints exist between various environmental parameters (e.g., dew point is determined by both temperature and humidity), and the load capacity of shared equipment creates a global constraint on the threshold settings for each laboratory.
[0003] In existing technologies, independent optimization of early warning thresholds by each laboratory may violate global constraints or physical relationships between parameters. Due to network latency and management independence, if one laboratory modifies the management configuration of environmental monitoring data, other laboratories may not be aware of it in time and may make conflicting modifications simultaneously, breaking global constraints and disrupting the consistency of integrated management. Furthermore, in a distributed scenario, the local clocks of each laboratory may deviate. Simply determining the order of configuration updates based on timestamps can lead to incorrect priority determinations due to clock drift, resulting in inconsistent environmental monitoring data management configuration states. Summary of the Invention
[0004] This invention provides an integrated management method for laboratory environmental testing data, which solves the technical problems in related technologies, such as physical clock asynchrony leading to incorrect priority determination and distributed independent optimization potentially breaking global constraints.
[0005] This invention provides an integrated management method for laboratory environmental testing data, comprising:
[0006] Obtain the physical constraint equations between environmental parameters involved in environmental monitoring data from each laboratory and the global shared constraint conditions of shared equipment. Convert the constraint expressions into inequality constraints for decision variables of early warning thresholds for each laboratory. Define the joint feasible region based on the inequality constraints and attach an initial version vector to the management configuration, with each component initialized to zero.
[0007] Within the joint feasible domain, a set of initial threshold candidate points that satisfy the constraints is generated by sampling. For each candidate point, backtesting is performed using historical environmental detection data. The false alarm rate and false negative rate are calculated to generate a local evaluation sample set.
[0008] A proxy distribution is established by fitting a Gaussian process regressor to the local evaluation sample set. The acquisition function is improved by using constrained expectation and iteratively searched within the joint feasible region to output the local optimal threshold candidate.
[0009] The component at the corresponding position in the incremental version vector of this laboratory is packaged with the local optimal threshold candidate and the updated version vector to generate a configuration update message, which is then broadcast to other laboratories.
[0010] The receiver uses the version vector to compare components one by one to determine the causal relationship between the message and the local state, and combines the received threshold with the local threshold to perform a test on the global constraints, generating a constraint satisfaction mark.
[0011] If the constraint satisfaction flag is true, the maximum value of each component of the local version vector and the received version vector is merged, and the configuration update is accepted; if the constraint satisfaction flag is false, the priority is determined based on the causal order of the version vectors, and the non-priority option rolls back the local configuration modification.
[0012] Non-priority parties re-perform sampling and iterative search within the remaining feasible region after deducting the priority parties' usage, generate a corrected local threshold, and broadcast it.
[0013] Output the final coordinated environmental monitoring data management configuration and the current version vector to the early warning system.
[0014] Furthermore, the physical constraint equations include the dew point temperature constraint equation, and converting the constraint expressions into inequality constraints includes:
[0015] Let the upper limit of temperature threshold be the first upper limit threshold, the upper limit of humidity threshold be the second upper limit threshold, and the upper limit of dew point temperature be the upper limit of dew point threshold. The constraint function is defined as follows: calculate the corresponding dew point temperature based on the first upper limit threshold and the second upper limit threshold, and subtract the upper limit of dew point threshold from the dew point temperature to obtain the constraint function value. When the constraint function value is less than or equal to zero, it means that the constraint is satisfied.
[0016] The transformation of the global shared constraint conditions includes: setting the total load capacity limit of the shared equipment as the load limit value, the sum of the expected load demand of each laboratory under the threshold configuration minus the load limit value to obtain the global constraint function value, and when the global constraint function value is less than or equal to zero, it means that the global constraint is satisfied;
[0017] The method for calculating the expected load demand is as follows: based on the laboratory's lower temperature threshold, upper temperature threshold, and environmental reference temperature, calculate the difference between the lower temperature threshold and the environmental reference temperature, and the difference between the environmental reference temperature and the upper temperature threshold. Take the maximum value of the two and the maximum value of zero, and multiply it by the power corresponding to the unit temperature difference to obtain the expected load demand.
[0018] Furthermore, the sampling process within the joint feasible region to generate an initial set of candidate threshold points that satisfy the constraints includes:
[0019] The Latin hypercube sampling method is adopted. The boundary range of the joint feasible region and the number of sampling points are input. The value range of each dimension is divided into an interval equal to the number of sampling points. A point is randomly selected in each interval. A candidate point set is generated by randomly pairing the intervals of each dimension.
[0020] Before sampling, the thresholds of each dimension are preprocessed by normalization to map the value range of each dimension to a unified standard interval; when outputting the local optimal threshold candidate, the optimization results in the normalization space are denormalized to restore the actual physical threshold.
[0021] Furthermore, backtesting using historical environmental monitoring data includes:
[0022] The time window range of the historical environmental monitoring data used for backtesting is a preset time window length shifted backward from the current moment, and the time window length covers the seasonal variation cycle of laboratory environmental parameters;
[0023] The false alarm rate is calculated as follows: In the historical environmental monitoring data, the environmental parameter value at each moment is compared with the candidate threshold. If the environmental parameter value exceeds the threshold range but no abnormal event actually occurs at that moment, it is counted as a false alarm. The number of false alarms is divided by the total number of moments in the historical data to obtain the false alarm rate.
[0024] The false negative rate is calculated as follows: if the environmental parameter value does not exceed the threshold range but an abnormal event actually occurs at that moment, it is counted as one false negative. The number of false negatives is divided by the total number of historical data moments to obtain the false negative rate.
[0025] Furthermore, establishing a surrogate distribution by fitting a Gaussian process regressor to the local evaluation sample set includes:
[0026] The candidate threshold is used as input, and the weighted sum of the false positive rate and the false negative rate is used as the output objective function value. The objective function is defined as: the weight coefficient multiplied by the false positive rate plus one minus the weight coefficient multiplied by the false negative rate, where the weight coefficient takes the value of an open interval between zero and one.
[0027] The Gaussian process regressor uses a radial basis function kernel as the covariance kernel function, which is defined as: the signal variance multiplied by an exponential function, where the exponent of the exponential function is negative, and the square of the Euclidean distance between the two threshold points is divided by twice the square of the length scale parameter. The mean function of the Gaussian process regressor uses a zero mean function, and the hyperparameters are optimized by maximizing the marginal log-likelihood.
[0028] Furthermore, improving the acquisition function using constraint expectations in the iterative search within the joint feasible region includes:
[0029] The constrained expectation improvement acquisition function is defined as the product of the standard expectation improvement function and the probability that the threshold point satisfies the joint feasible region constraint.
[0030] The standard expected improvement function is defined as: the difference between the current optimal objective function value and the predicted mean multiplied by the value of the cumulative distribution function of the standard normal distribution at the standardized deviation, plus the predicted standard deviation multiplied by the value of the probability density function of the standard normal distribution at the standardized deviation; the standardized deviation is the difference between the current optimal objective function value and the predicted mean divided by the predicted standard deviation.
[0031] Each iteration selects a candidate point that maximizes the acquisition function for backtesting evaluation. The evaluation results are added to the sample set and the Gaussian process regressor is updated. The iteration terminates when the number of iterations reaches the preset maximum number of iterations, or when the improvement of the maximum value of the acquisition function in a consecutive preset number of iterations is less than the convergence threshold.
[0032] Furthermore, the methods for calculating the probability that the threshold point satisfies the joint feasible region constraint include:
[0033] An independent Gaussian process regressor is established for each constraint function. Each Gaussian process regressor outputs the mean and standard deviation of the predicted distribution of the constraint function at the threshold point.
[0034] Assuming that the constraints are independent of each other, the probability of satisfying a constraint is calculated as the product of the probabilities of all constraint functions. The probability of each constraint function is the value of the cumulative distribution function of the standard normal distribution at the negative predicted mean divided by the predicted standard deviation.
[0035] Furthermore, using component-by-component comparison of version vectors to determine the causal relationship between messages and local state includes:
[0036] Let the received version vector be the received vector, and the local version vector be the local vector;
[0037] If every component of the received vector is less than or equal to the corresponding component of the local vector, and at least one component is strictly less than, then the received message is determined to be causally prior to the local state.
[0038] If every component of the local vector is less than or equal to the corresponding component of the received vector, and at least one component is strictly less than, then the local state is determined to be causally prior to the received message.
[0039] If two vectors have an intersection size relationship, it is determined to be a concurrent state. In the case of a concurrent state, the priority is determined according to the preset deterministic rules, and the non-priority one rolls back the local configuration modification.
[0040] Furthermore, the non-priority party re-performs sampling and iterative search within the remaining feasible region after deducting the priority party's occupation, including:
[0041] The remaining feasible region is defined as the intersection of the original joint feasible region and the non-priority threshold and the priority threshold that jointly satisfy the global constraints;
[0042] The remaining feasible region is calculated as follows: the thresholds of the priority side are treated as constants and substituted into the original constraint functions to generate a new set of constraint functions that only concern the threshold variables of the non-priority side; for globally shared constraints, after substituting the priority side thresholds, they are converted into constraints between the load demand of the non-priority side being less than or equal to the total load capacity limit minus the sum of the load demands of the priority side.
[0043] The non-priority side re-executes sampling, backtesting, and Bayesian optimization iterative search within the remaining feasible domain to generate a corrected local threshold. The corrected local threshold is then updated and broadcast as a version vector, and this process is repeated until the global constraint is satisfied.
[0044] This invention provides an integrated management system for laboratory environmental monitoring data, comprising:
[0045] The constraint definition module is used to obtain the physical constraint equations of environmental parameters and globally shared constraints, convert constraint expressions into inequality constraints and define the joint feasible region, and add version vectors for management configuration.
[0046] The sampling and evaluation module is used to sample and generate a set of threshold candidate points within the joint feasible domain, and to backtest and generate a local evaluation sample set using historical environmental detection data.
[0047] The search module is optimized to fit a Gaussian process regressor to establish a proxy distribution, and the constraint expectation is used to improve the acquisition function and iteratively search to generate local optimal threshold candidates.
[0048] The message broadcasting module is used to increment the local version vector component, generate configuration update messages, and broadcast them to other laboratories.
[0049] The causality determination module is used to determine causal relationships by comparing each component of the version vector and to perform joint feasible region checks to generate constraint satisfaction labels.
[0050] The configuration merging module is used to perform configuration merging or rollback based on constraint satisfaction flags, determine the priority side and trigger non-priority side re-optimization;
[0051] The output module is used to output the final coordinated environmental monitoring data management configuration and the current version vector to the early warning system.
[0052] The beneficial effects of this invention are as follows:
[0053] This invention integrates a version vector mechanism with a Bayesian optimization framework. It establishes a causal order relationship for cross-laboratory environmental monitoring data management configuration updates by comparing components of the version vectors, overcoming the defect of incorrect priority determination caused by physical clock asynchrony. It defines a joint feasible region using physical constraint equations and performs Gaussian process proxy optimization within the feasible region, ensuring the physical consistency of various environmental parameter threshold combinations. It overcomes the defect of distributed independent optimization potentially breaking global constraints by using a configuration merging process based on constraint satisfaction checks and causal order priority determination. Finally, it uses an iterative process of re-optimizing within the remaining feasible region using non-priority options, enabling the configuration coordination process to converge in constraint conflict scenarios. Therefore, this invention achieves unified causal consistency maintenance and constraint coordination for integrated management configuration of environmental monitoring data in distributed multi-laboratory scenarios. Attached Figure Description
[0054] Figure 1 This is a flowchart of an integrated management method for laboratory environmental testing data according to the present invention;
[0055] Figure 2 This is a bar chart comparing the optimized load requirements of various laboratories according to the present invention;
[0056] Figure 3 This is a mixed-bar graph of the Bayesian optimization iterative convergence process of the present invention;
[0057] Figure 4 This is a scatter plot showing the scatter distribution of false alarm rate and false negative rate of threshold candidate points in this invention;
[0058] Figure 5 This is a dew point temperature thermogram of the temperature-humidity threshold combination of the present invention. Detailed Implementation
[0059] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.
[0060] At least one embodiment of the present invention discloses an integrated management method for laboratory environmental testing data, such as... Figure 1 As shown, it includes the following steps:
[0061] Step 1: Obtain the physical constraint equations for environmental parameters and the globally shared constraints, and generate the joint feasible domain definition and management configuration records with version vectors.
[0062] Obtain the physical constraint equations and globally shared constraints among the environmental parameters involved in the environmental monitoring data of each laboratory. The physical constraint equations express the physical relationships between the environmental parameters, and the globally shared constraints express the overall limitations imposed by shared equipment on the thresholds of each laboratory. Transform the constraint expressions in the physical constraint equations and globally shared constraints into inequality constraints for the decision variables of the early warning thresholds of each laboratory.
[0063] Let the decision variable for the early warning threshold of each laboratory be... ,in The laboratory is numbered. The inequality constraint form after the transformation of the physical constraint equations is as follows:
[0064]
[0065] in, For the first A constraint function, The total number of laboratories. Let represent the total number of constraints. Based on the above inequality constraints, define the joint feasible region. :
[0066]
[0067] At the same time, an initial version vector is added to the integrated management configuration. The vector dimension equals the total number of laboratories, and each component is initialized to zero. Each laboratory maintains a counter for its own location and generates a management configuration record with a version vector.
[0068] In one implementation example, a research institution has three laboratories: a chemistry laboratory, a biological culture laboratory, and a precision instrument laboratory, numbered 1, 2, and 3 respectively. They share a central air conditioning system with a maximum total cooling / heating capacity. kW, ambient reference temperature Power per unit temperature difference kW / °C. The upper limit of the allowable dew point temperature for each laboratory is set to [value missing]. The initial configuration parameters for each laboratory are shown in the table below:
[0069] Table 1 Initial configuration parameters for each laboratory:
[0070]
[0071] Based on the configuration in the table above, the expected load requirement for the chemical laboratory is as follows:
[0072] kW, the expected load requirement of the biological culture laboratory is The expected load demand of the precision instrument laboratory is kW. kW, the total load requirement of the three laboratories is The current configuration exceeds the shared device capacity limit of 30kW, therefore violating the global sharing constraint. The initial version vector was set to... .
[0073] Figure 5 The dew point temperature values calculated under different combinations of upper limits for temperature and humidity thresholds are shown.
[0074] Step 2: Sample within the joint feasible region to generate an initial threshold candidate point set that satisfies the constraints, and use historical environmental detection data to backtest and generate a local evaluation sample set.
[0075] When local laboratories need to adjust the warning thresholds for environmental monitoring data, within the joint feasible domain The initial threshold candidate point set that satisfies the constraints is generated using Latin hypercube sampling. ,in This is the first threshold candidate point. This is the second threshold candidate point. For the first There are threshold candidate points. This represents the number of sampling points.
[0076] For each candidate point Backtesting was conducted using historical environmental monitoring data from our laboratory. During the backtesting process, the number of false alarms and false negatives under the candidate threshold configurations were counted, and the false alarm rate was calculated. and underreporting rate Generate a local evaluation sample set:
[0077]
[0078] Furthermore, the calculation methods for false alarm rate and false negative rate are as follows: In historical environmental monitoring data, the environmental parameter value at each moment is compared with the candidate threshold. If the environmental parameter value exceeds the threshold range but no abnormal event actually occurs at that moment, it is counted as a false alarm; if the environmental parameter value does not exceed the threshold range but an abnormal event actually occurs at that moment, it is counted as a false negative. underreporting rate ,in Number of false alarms The number of times the report was missed. This represents the total number of moments in the historical data.
[0079] Furthermore, the time window range for the historical environmental monitoring data used in the backtesting is:
[0080]
[0081] in For the current moment, The backtesting time window length should be selected to cover the seasonal variation cycle of laboratory environmental parameters to ensure that the backtesting results can reflect the early warning effect of the threshold configuration under different environmental conditions.
[0082] The chemical laboratory needs to adjust its warning threshold due to current total load exceeding limits. Eight candidate points were generated within the joint feasible domain using Latin hypercube sampling. For each candidate point, backtesting was performed using historical environmental monitoring data from the past 180 days (4320 time points in total). The number of false alarms and false negatives were counted, and the false alarm rate and false negative rate were calculated. The sampling and backtesting evaluation results for some candidate points are shown in the table below:
[0083] Table 2. Sampling and backtesting evaluation results of threshold candidate points in the chemical laboratory:
[0084]
[0085] Taking candidate point 1 as an example, the expected load demand corresponding to the lower limit of the temperature threshold of 21.5°C is:
[0086] The kW is lower than the initial configuration of 12kW. A local evaluation sample set was generated based on the above backtest results. .
[0087] Step 3: Fit a Gaussian process regressor to generate a surrogate distribution, and use constrained expectation to improve the acquisition function and iteratively search to generate local optimal threshold candidates.
[0088] With candidate threshold The input is the weighted sum of the false positive rate and the false negative rate, and the output is the objective function value.
[0089]
[0090] in, This is a weighting coefficient used to balance the relative importance of false alarm rate and false negative rate.
[0091] Furthermore, weighting coefficients The value should be determined based on the laboratory's safety level and operational needs: for laboratories with high safety requirements (such as hazardous chemical storage laboratories), a smaller value should be set. To reduce the false alarm rate; for laboratories sensitive to false alarms (such as precision instrument laboratories requiring continuous operation), a larger value should be set. This value is used to reduce the false alarm rate.
[0092] Furthermore, since both false positive rate and false negative rate are indicators that should be minimized, the objective function... To minimize the objective, the goal of Bayesian optimization is to search for... The threshold parameter for obtaining the minimum value.
[0093] Based on local evaluation sample set Fit a Gaussian process regressor. The Gaussian process regressor establishes a surrogate distribution for the objective function, and for any threshold point... The proxy distribution gives the predicted mean. and prediction variance .
[0094] Improving the acquisition function in the joint feasible region using constraint expectation Inner iterative search. The constraint expectation improvement acquisition function is defined as:
[0095]
[0096] in, For the standard expectation improvement function, The threshold point represents the probability that the joint feasible region constraint is satisfied.
[0097] In each iteration, candidate points that maximize the acquisition function are selected for backtesting and evaluation. The evaluation results are added to the sample set, and the Gaussian process regressor is updated. The process continues until the preset maximum number of iterations is reached. or continuous The improvement in the maximum value of the acquisition function in each iteration is less than the convergence threshold. The iteration terminates when the output proxy distribution predicts the optimal threshold parameter. After the iteration terminates, the optimal threshold parameter predicted by the output proxy distribution is used as the local optimal threshold candidate. .
[0098] Weighting coefficient of chemistry laboratory Based on this, the objective function value of each candidate point is calculated. For example, the objective function value of candidate point 1 is... .
[0099] A Gaussian process regressor is fitted based on a locally evaluated sample set, with a maximum number of iterations set. Number of convergence checks Convergence threshold The key iteration records in the Bayesian optimization process are shown in the table below:
[0100] Table 3. Bayesian optimization iterative process in the chemistry laboratory:
[0101]
[0102] The iteration terminates at the 12th iteration, satisfying the convergence condition, and the local optimal threshold candidate is output as follows:
[0103]
[0104] The corresponding expected load demand is kW.
[0105] Figure 3 The convergence trend of the objective function value (weighted sum of false alarm rate and false negative rate) and the changes in the acquisition function value are shown in the chemistry laboratory during 12 maximum iterations.
[0106] Figure 4 The distribution of false alarm and false negative rates for threshold candidate points generated by sampling within the joint feasible domain in the chemical laboratory is shown.
[0107] Step 4: Increment the local version vector component, generate a management configuration update message, and broadcast it to other labs.
[0108] Increment the component corresponding to this laboratory in the local version vector. Let the laboratory number be... The updated version vector is:
[0109]
[0110] Candidate local optimal thresholds With the updated version vector Package the data, generate management configuration update messages, and broadcast them to the environmental data management systems of other laboratories.
[0111] The chemistry lab will transfer the local version vector from Updated to ,
[0112] Threshold candidates With version vector The configuration update message is packaged and broadcast to the biological culture laboratory and the precision instrument laboratory.
[0113] Step 5: The receiver uses the version vector to determine the causal relationship and performs a joint feasible region test to generate constraint satisfaction markers.
[0114] After receiving the management configuration update message, the receiver uses the version vector to compare each component to determine the causal relationship between the message and its local state. Let the received version vector be... The local version vector is .
[0115] The causal order determination rule for version vectors is as follows:
[0116] like Each component is less than or equal to If the corresponding component is less than , and at least one component is strictly less than , then the received message precedes the local state in causal order.
[0117] like Each component is less than or equal to If the corresponding component is less than , and at least one component is strictly less than , then the local state precedes the received message in causal order.
[0118] If two vectors are not comparable (but have an intersection size relationship), then it is a concurrent state.
[0119] Based on the above rules, a causal order determination result is generated. Simultaneously, the received threshold and the local threshold are substituted into the global constraints for a joint feasible region check to determine whether the joint configuration satisfies all constraints, generating a constraint satisfaction flag.
[0120] Furthermore, the criterion for determining constraint satisfaction is: the received threshold value... With local threshold Substitute all constraint functions together If for all All meet If the constraint satisfaction flag is true, then the constraint satisfaction flag is set to true; if there exists any Make If the constraint satisfaction flag is false, then the constraint satisfaction flag is set to false.
[0121] While the chemistry lab broadcast the configuration update message, the precision instrument lab independently executed steps 2 through 4 to obtain the local optimal threshold candidate. The corresponding expected load demand is kW, update the version vector to And broadcast a configuration update message.
[0122] After receiving configuration update messages from the chemistry lab and the precision instrument lab, the biological culture lab compares its version vectors. The local version vector for the biological culture lab is... The version vectors of the two received messages are respectively and .
[0123] The comparison results show: the version vector of the chemical laboratory message Version vector of messages from the Precision Instruments Laboratory There is a cross-size relationship (first component) , third component This indicates a concurrent state.
[0124] Joint feasible domain verification was conducted: the optimized load requirement for the chemistry laboratory is 8kW, the optimized load requirement for the precision instrument laboratory is 12kW, and the current load requirement for the biological culture laboratory is 8kW. The sum of the three is... kW kW, the globally shared constraint is satisfied.
[0125] Simultaneously test dew point temperature constraints: upper threshold of chemical laboratory. The corresponding dew point temperature is approximately 17.8°C. Upper limit of threshold for precision instrument laboratory The corresponding dew point temperature is approximately 16.5°C. The physical constraints are satisfied. Therefore, the constraint satisfaction flag is set to true.
[0126] Step 6: Perform configuration merging or rollback based on constraint satisfaction flags to generate synchronized unified management configuration.
[0127] If the constraint satisfaction flag is true (the joint configuration satisfies all constraints), then perform the configuration merging operation: merge the maximum value of each component of the local version vector and the receiving version vector.
[0128]
[0129] Accept configuration updates and generate synchronized unified management configurations.
[0130] If the constraint satisfaction flag is false (joint configuration violates constraints), the conflict resolution process is triggered: the priority is determined based on the causal order of the version vector. If both parties are in a concurrent state, the priority is determined according to a preset deterministic rule (such as lexicographical order of laboratory numbers). The non-priority party rolls back its local configuration changes.
[0131] Since the constraint satisfaction flag is true, the biological culture laboratory performs a configuration merge operation.
[0132] Local version vector With the two received version vectors and By performing a step-by-step merging process that takes the maximum value from each component, we obtain the merged version vector:
[0133]
[0134] The biological culture laboratory receives two configuration updates, generating a synchronized, unified management configuration.
[0135] In another scenario, the optimization threshold for a precision instrument laboratory is: Corresponding load demand kW, then the total load requirement of the three laboratories is kW kW, the constraint satisfaction flag is set to false, triggering the conflict resolution process. Since the chemistry lab and the precision instrument lab are in a concurrent state, according to the lexicographical order of lab numbers, the chemistry lab (number 1) is the priority lab, and the precision instrument lab (number 3) is the non-priority lab, requiring a rollback of local configuration modifications.
[0136] Step 7: Non-priority parties re-execute Bayesian optimization search within the remaining feasible region to generate a corrected local threshold.
[0137] The non-priority side re-executes the Bayesian optimization search within the remaining feasible region after deducting the space occupied by the priority side. The remaining feasible region is defined as:
[0138]
[0139] Furthermore, the specific calculation method for the remaining feasible region is as follows: The threshold already determined by the priority side... Treating them as constants and substituting them into the original constraint functions generates a new set of constraint functions that only concern the non-priority threshold variables.
[0140] For global shared constraints After substituting the priority threshold, it is converted to ,in The numbers on the right are the laboratory numbers for non-priority laboratories, and the remaining load capacity is shown on the right after deducting the load capacity occupied by priority laboratories.
[0141] In the remaining feasible region Re-execute the Bayesian optimization process from steps 2 to 3 to generate the corrected local threshold. Update, broadcast, and merge the corrected local threshold using steps 4 to 6, repeating this process until the global constraint is satisfied.
[0142] The precision instrumentation laboratory is a non-priority area, so the remaining feasible region is calculated. The chemical laboratory has a determined load requirement of 8kW, and the biological culture laboratory currently has a load requirement of 8kW, for a total of 16kW.
[0143] Remaining load capacity is kW. The lower limit of the temperature threshold for precision instrument laboratories must meet the following requirements. kW, that is .
[0144] The precision instrument laboratory re-executed Bayesian optimization within the remaining feasible region, expanding the search range for the lower limit of the temperature threshold from the original... shrink to .
[0145] After 8 iterations, the process converged, yielding the corrected local threshold. The corresponding expected load demand is kW kW, satisfying the remaining feasible region constraints. The Precision Instruments Laboratory updated the version vector to... The broadcast will update the corrected configuration.
[0146] Step 8: Output the final coordinated environmental monitoring data management configuration and the current version vector to the early warning system.
[0147] The final coordinated environmental monitoring data management configuration and the current version vector are output to the early warning systems of each laboratory, completing a round of configuration synchronization for integrated environmental monitoring data management.
[0148] After all laboratories have completed configuration synchronization, the final coordinated integrated management configuration is shown in the table below:
[0149] Table 4: Final Integrated Management Configuration After Coordination
[0150]
[0151] The total expected load of the three laboratories is kW kW satisfies the globally shared constraints. The version vectors of each laboratory are eventually synchronized to... This indicates that the chemistry lab has performed one configuration update, the precision instrument lab has performed two configuration updates (including one conflict correction), and the biological culture lab has not initiated a configuration update. This version vector will be output to the early warning systems of each lab as the baseline version for the next round of configuration updates.
[0152] It should be noted that the physical constraint equations in step 1 above include, but are not limited to, the dew point temperature constraint equation. Dew point temperature With temperature and relative humidity The relationship can be represented as:
[0153]
[0154] in, , and These are empirical constants. Based on the dew point temperature constraint equation, the temperature and humidity thresholds must be set to ensure that the corresponding dew point temperatures are within a reasonable range.
[0155] Furthermore, the empirical constant in the above dew point temperature equation takes the value of , (Unit: degrees Celsius), the empirical constant values are applicable to a temperature range. to Relative humidity range to A standard laboratory environment.
[0156] Furthermore, the above constraint functions The specific construction method is as follows:
[0157] The physical constraint equations are transformed into inequalities concerning threshold decision variables. Taking dew point temperature constraints as an example, let the upper limit of the temperature threshold be... The upper limit of the humidity threshold is The upper limit of the allowable dew point temperature is The corresponding constraint function is ,when The time indicates that the constraint is satisfied.
[0158] Furthermore, the specific form of the above-mentioned global sharing constraint is as follows: Let the upper limit of the total load capacity of the shared devices be... Each laboratory In threshold configuration The expected load demand is Then the global shared constraint function is ,when This indicates that the threshold configurations of each laboratory collectively meet the load capacity limits of the shared equipment.
[0159] Furthermore, the expected load demand function The calculation method is as follows:
[0160] Estimate the load demand of shared equipment based on the leniency of the threshold configuration. Taking temperature control as an example, suppose a laboratory... The upper limit of the temperature threshold is The lower limit of the temperature threshold is The ambient reference temperature is The cooling or heating power corresponding to a unit temperature difference is The expected load demand is:
[0161] , which represents the shared device power required to maintain the threshold range.
[0162] It should be noted that the Latin hypercube sampling in step 2 above is a hierarchical random sampling method, and the input is the joint feasible region. Boundary range and number of sampling points The output is a set of candidate points that satisfy the feasible region constraints. The sampling process divides the value range of each dimension into equal parts. Sampling is achieved by randomly selecting a point within each interval and randomly pairing intervals of each dimension.
[0163] It should be noted that, due to the decision variables of the early warning threshold... The thresholds include various environmental parameters (temperature, humidity, air pressure, etc.). Since the physical dimensions and numerical ranges of these parameters differ, before sampling in step 2 and Gaussian process modeling in step 3, the thresholds for each dimension need to undergo min-max normalization preprocessing to map the value ranges of each dimension to a unified standard interval. The normalized thresholds are used for Latin hypercube sampling, distance calculation of the Gaussian process covariance kernel, and the Bayesian optimization search process. Step 3 outputs locally optimal threshold candidates. At this time, it is necessary to perform an inverse normalization transformation on the optimization results in the normalization space, restore the standardized values of each dimension to the actual physical thresholds of the corresponding environmental parameters, and generate management configurations that can be directly applied to the early warning system.
[0164] It should be noted that the Gaussian process regressor in step 3 above is a non-parametric Bayesian regression method. The input is the threshold candidate points, and the output is the predicted distribution of the objective function. The Gaussian process is defined by the mean function and the covariance kernel function. The covariance kernel function can be a radial basis function kernel.
[0165]
[0166] in, Represents an exponential function. For signal variance, This is the length scale parameter.
[0167] Furthermore, signal variance The range of values is Used to control the overall amplitude of the Gaussian process prediction distribution; length scale parameter The range of values is This is used to control the rate of correlation decay between adjacent threshold points; a larger one... The value indicates that the objective function changes relatively smoothly.
[0168] The mean function of the aforementioned Gaussian process regressor is a zero-mean function, and the hyperparameters are optimized by maximizing the marginal log-likelihood.
[0169] It should be noted that the standard expected improvement function in step 3 above... The specific definition is:
[0170]
[0171] in, , The optimal objective function value among the currently observed samples. For the Gaussian process regressor at the threshold point The predicted mean at that location, For the Gaussian process regressor at the threshold point The predicted standard deviation at that location The cumulative distribution function of the standard normal distribution. It is the probability density function of the standard normal distribution.
[0172] It should be noted that the probability of satisfying the constraints in step 3 above... The calculation method is as follows: for each constraint function Establish independent Gaussian process regressors, and set the output constraint function of each Gaussian process regressor at the threshold point. The predicted distribution mean at the location and the predicted standard deviation Assuming all constraints are independent, the probability of satisfying a constraint is calculated as follows:
[0173]
[0174] in, The cumulative distribution function of the standard normal distribution. This represents the total number of constraints.
[0175] In this embodiment, to accelerate the convergence speed of Bayesian optimization, a multi-starting point parallel acquisition strategy can be adopted during the iterative search process in step 3. Specifically, in each iteration, multiple candidate points that result in higher acquisition function values are selected simultaneously for parallel backtesting evaluation. After batching the multiple evaluation results into the sample set, the Gaussian process regressor is updated uniformly.
[0176] In this embodiment, to improve the storage and transmission efficiency of version vectors, a sparse version vector representation can be used when the number of laboratories is large. Only the positions and values of non-zero components are stored and transmitted; the corresponding version components are not transmitted for laboratories whose configurations have not been modified.
[0177] In this embodiment of the application, in order to handle a large number of concurrent updates after network partition recovery, a batch merging mechanism can be introduced in the conflict resolution process of step 6. When multiple concurrent configuration update messages are received, the messages are sorted according to the causal order of the version vector, and constraint checks and merging are performed in sequence. The configurations that can coexist are merged in batches, reducing the number of rollbacks and re-optimizations.
[0178] Figure 2 This demonstrates the changes in load requirements for each laboratory from the initial configuration to the final coordinated configuration in steps 1 through 8.
[0179] This implementation integrates a version vector mechanism with a Bayesian optimization framework. It establishes a causal order for cross-laboratory environmental monitoring data management configuration updates by comparing version vectors component by component, overcoming the defect of incorrect priority determination caused by physical clock asynchrony. It defines a joint feasible region using physical constraint equations and performs Gaussian process proxy optimization within the feasible region, ensuring the physical consistency of various environmental parameter threshold combinations. It overcomes the defect of distributed independent optimization potentially breaking global constraints by using a configuration merging process based on constraint satisfaction checks and causal order priority determination. Finally, it uses an iterative process of re-optimizing within the remaining feasible region using non-priority options, enabling the configuration coordination process to converge in constraint conflict scenarios. Therefore, this implementation achieves unified causal consistency maintenance and constraint coordination for integrated management configuration of environmental monitoring data in distributed multi-laboratory scenarios.
[0180] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.
Claims
1. A method for integrated management of laboratory environmental testing data, characterized in that, Includes the following steps: Obtain the physical constraint equations between environmental parameters involved in environmental monitoring data from each laboratory and the global shared constraint conditions of shared equipment. Convert the constraint expressions into inequality constraints for decision variables of early warning thresholds for each laboratory. Define the joint feasible region based on the inequality constraints and attach an initial version vector to the management configuration, with each component initialized to zero. Within the joint feasible domain, a set of initial threshold candidate points that satisfy the constraints is generated by sampling. For each candidate point, backtesting is performed using historical environmental detection data. The false alarm rate and false negative rate are calculated to generate a local evaluation sample set. A proxy distribution is established by fitting a Gaussian process regressor to the local evaluation sample set. The acquisition function is improved by using constrained expectation and iteratively searched within the joint feasible region to output the local optimal threshold candidate. The joint feasible region is defined as the set of all combinations of thresholds that make the values of each constraint function less than or equal to zero, and the dimension of the version vector is equal to the total number of laboratories. The Latin hypercube sampling method is adopted. The boundary range of the joint feasible region and the number of sampling points are input. The value range of each dimension is divided into an interval equal to the number of sampling points. A point is randomly selected in each interval. A candidate point set is generated by randomly pairing the intervals of each dimension. Before sampling, the thresholds of each dimension are normalized and preprocessed to map the value range of each dimension to a unified standard range. When outputting the local optimal threshold candidate, the optimization results in the normalization space are denormalized to restore the actual physical threshold. The time window range of the historical environmental monitoring data used for backtesting is a preset time window length shifted backward from the current moment, and the time window length covers the seasonal variation cycle of laboratory environmental parameters; The false alarm rate is calculated as follows: In the historical environmental monitoring data, the environmental parameter value at each moment is compared with the candidate threshold. If the environmental parameter value exceeds the threshold range but no abnormal event actually occurs at that moment, it is counted as a false alarm. The number of false alarms is divided by the total number of moments in the historical data to obtain the false alarm rate. The false negative rate is calculated as follows: if the environmental parameter value does not exceed the threshold range but an abnormal event actually occurs at that moment, it is counted as one false negative. The number of false negatives is divided by the total number of moments in the historical data to obtain the false negative rate. The component at the corresponding position in the incremental version vector of this laboratory is packaged with the local optimal threshold candidate and the updated version vector to generate a configuration update message, which is then broadcast to other laboratories. The receiver uses the version vector to compare components one by one to determine the causal relationship between the message and the local state, and combines the received threshold with the local threshold to perform a test on the global constraints, generating a constraint satisfaction mark. If the constraint satisfaction flag is true, the maximum value of each component of the local version vector and the received version vector is merged, and the configuration update is accepted; if the constraint satisfaction flag is false, the priority is determined based on the causal order of the version vectors, and the non-priority option rolls back the local configuration modification. The constrained expectation improvement acquisition function is defined as the product of the standard expectation improvement function and the probability that the threshold point satisfies the joint feasible region constraint. The standard expected improvement function is defined as: the difference between the current optimal objective function value and the predicted mean multiplied by the value of the cumulative distribution function of the standard normal distribution at the standardized deviation, plus the predicted standard deviation multiplied by the value of the probability density function of the standard normal distribution at the standardized deviation; the standardized deviation is the difference between the current optimal objective function value and the predicted mean divided by the predicted standard deviation. Each iteration selects the candidate point that maximizes the acquisition function for backtesting evaluation, adds the evaluation results to the sample set and updates the Gaussian process regressor; the iteration terminates when the number of iterations reaches the preset maximum number of iterations, or when the improvement of the maximum value of the acquisition function in a consecutive preset number of iterations is less than the convergence threshold. Let the received version vector be the received vector, and the local version vector be the local vector; If every component of the received vector is less than or equal to the corresponding component of the local vector, and at least one component is strictly less than, then the received message is determined to be causally prior to the local state. If every component of the local vector is less than or equal to the corresponding component of the received vector, and at least one component is strictly less than, then the local state is determined to be causally prior to the received message. If two vectors have an intersection size relationship, it is determined to be a concurrent state; for concurrent states, the priority is determined according to the preset deterministic rules, and the non-priority one rolls back the local configuration modification. Non-priority parties re-perform sampling and iterative search within the remaining feasible region after deducting the priority parties' usage, generate a corrected local threshold, and broadcast it. Output the final coordinated environmental monitoring data management configuration and the current version vector to the early warning system.
2. The integrated management method for laboratory environmental monitoring data according to claim 1, characterized in that, The physical constraint equations include dew point temperature constraint equations. Converting the constraint expressions into inequality constraints includes: Let the upper limit of temperature threshold be the first upper limit threshold, the upper limit of humidity threshold be the second upper limit threshold, and the upper limit of dew point temperature be the upper limit of dew point threshold. The constraint function is defined as follows: calculate the corresponding dew point temperature based on the first upper limit threshold and the second upper limit threshold, and subtract the upper limit of dew point threshold from the dew point temperature to obtain the constraint function value. When the constraint function value is less than or equal to zero, it means that the constraint is satisfied. The transformation of the global shared constraint conditions includes: setting the total load capacity limit of the shared equipment as the load limit value, the sum of the expected load demand of each laboratory under the threshold configuration minus the load limit value to obtain the global constraint function value, and when the global constraint function value is less than or equal to zero, it means that the global constraint is satisfied; The method for calculating the expected load demand is as follows: based on the laboratory's lower temperature threshold, upper temperature threshold, and environmental reference temperature, calculate the difference between the lower temperature threshold and the environmental reference temperature, and the difference between the environmental reference temperature and the upper temperature threshold. Take the maximum value of the two and the maximum value of zero, and multiply it by the power corresponding to the unit temperature difference to obtain the expected load demand.
3. The integrated management method for laboratory environmental monitoring data according to claim 1, characterized in that, The method of establishing a proxy distribution by fitting a Gaussian process regressor to a local evaluation sample set includes: The candidate threshold is used as input, and the weighted sum of the false positive rate and the false negative rate is used as the output objective function value. The objective function is defined as: the weight coefficient multiplied by the false positive rate plus one minus the weight coefficient multiplied by the false negative rate, where the weight coefficient takes the value of an open interval between zero and one. The Gaussian process regressor uses a radial basis function kernel as the covariance kernel function, which is defined as: the signal variance multiplied by an exponential function, where the exponent of the exponential function is negative, and the square of the Euclidean distance between the two threshold points is divided by twice the square of the length scale parameter. The mean function of the Gaussian process regressor uses a zero mean function, and the hyperparameters are optimized by maximizing the marginal log-likelihood.
4. The integrated management method for laboratory environmental monitoring data according to claim 1, characterized in that, Methods for calculating the probability that a threshold point satisfies the joint feasible region constraint include: An independent Gaussian process regressor is established for each constraint function. Each Gaussian process regressor outputs the mean and standard deviation of the predicted distribution of the constraint function at the threshold point. Assuming that the constraints are independent of each other, the probability of satisfying a constraint is calculated as the product of the probabilities of all constraint functions. The probability of each constraint function is the value of the cumulative distribution function of the standard normal distribution at the negative predicted mean divided by the predicted standard deviation.
5. The integrated management method for laboratory environmental monitoring data according to claim 1, characterized in that, The re-sampling and iterative search of the non-priority party within the remaining feasible region after deducting the space occupied by the priority party includes: The remaining feasible region is defined as the intersection of the original joint feasible region and the non-priority threshold and the priority threshold that jointly satisfy the global constraints; The remaining feasible region is calculated as follows: the thresholds of the priority side are treated as constants and substituted into the original constraint functions to generate a new set of constraint functions that only concern the threshold variables of the non-priority side; for globally shared constraints, after substituting the priority side thresholds, they are converted into constraints between the load demand of the non-priority side being less than or equal to the total load capacity limit minus the sum of the load demands of the priority side. The non-priority side re-executes sampling, backtesting, and Bayesian optimization iterative search within the remaining feasible domain to generate a corrected local threshold. The corrected local threshold is then updated and broadcast as a version vector, and this process is repeated until the global constraint is satisfied.
6. An integrated management system for laboratory environmental monitoring data, used to execute the integrated management method for laboratory environmental monitoring data as described in any one of claims 1 to 5, characterized in that, include: The constraint definition module is used to obtain the physical constraint equations of environmental parameters and globally shared constraints, convert constraint expressions into inequality constraints and define the joint feasible region, and add version vectors for management configuration. The sampling and evaluation module is used to sample and generate a set of threshold candidate points within the joint feasible domain, and to backtest and generate a local evaluation sample set using historical environmental detection data. The search module is optimized to fit a Gaussian process regressor to establish a proxy distribution, and the constraint expectation is used to improve the acquisition function and iteratively search to generate local optimal threshold candidates. The message broadcasting module is used to increment the local version vector component, generate configuration update messages, and broadcast them to other laboratories. The causality determination module is used to determine causal relationships by comparing each component of the version vector and to perform joint feasible region checks to generate constraint satisfaction labels. The configuration merging module is used to perform configuration merging or rollback based on constraint satisfaction flags, determine the priority side and trigger non-priority side re-optimization; The output module is used to output the final coordinated environmental monitoring data management configuration and the current version vector to the early warning system.