-50 degrees celsius water-based extinguishing agent with low-temperature synergistic regulation mechanism

Abstract

The present application relates to a kind of-50 ℃ water-based fire extinguishing agent with low temperature synergistic regulation mechanism.The fire extinguishing agent includes water-based carrier W, low temperature phase stabilizer A, low temperature rheological regulator B, film-forming oxygen barrier synergist C and performance compatibility constraint D by mass fraction.By constructing the formulation feasible region including mass fraction upper and lower limit and total amount conservation constraint, combined with the prediction evaluation based on historical test data and risk weighted correction mechanism, candidate formula is screened and optimized, so that the resulting fire extinguishing agent still maintains liquid stability, low temperature sprayable rheological performance and good film-forming oxygen barrier ability under-50 ℃ and below environment, while meeting the corrosion rate and material compatibility requirements.The fire extinguishing agent significantly reduces the number of tests while ensuring safety margin, improves optimization convergence efficiency, and enhances the adaptability under different working conditions, thereby improving the engineering reliability and stability of-50 ℃ water-based fire extinguishing agent.

Description

Technical Field

[0001] This invention belongs to the field of fire extinguishing agent materials and intelligent optimization technology, specifically relating to a -50℃ water-based fire extinguishing agent with a low-temperature synergistic regulation mechanism and its formulation intelligent optimization method. This fire extinguishing agent can be used in firefighting scenarios in low-temperature environments, and is particularly suitable for applications requiring a balance of low-temperature liquid stability, sprayability, and film-forming oxygen barrier capabilities. Background Technology

[0002] Water-based fire extinguishing agents have advantages such as high cooling efficiency, controllable cost, and environmental friendliness. However, in extremely low temperature environments (such as -50℃), conventional water-based fire extinguishing agents typically exhibit the following problems: 1. Freezing / crystallization and phase separation: At low temperatures, solute precipitates to form crystals or phase separation occurs, leading to storage failure and jet blockage; 2. Increased viscosity at low temperatures leads to poor sprayability: A sharp increase in viscosity makes pumping difficult, reduces atomization, and can even cause blockages; 3. Decreased film formation and coverage: Wetting and spreading at low temperatures and decreased film stability result in insufficient oxygen barrier coverage and increased risk of reignition. 4. Material compatibility issues: Some components may cause corrosion or residue problems in metallic materials, affecting engineering applications.

[0003] Existing technologies largely rely on empirical formulation or local single-index parameter tuning, lacking an engineering method that unifies multi-objective modeling of phase state, rheology, film formation, and material compatibility, and efficiently seeks optimization with limited experimental budget. Therefore, a cryogenic fire extinguishing agent technology solution that combines physical property constraints with intelligent optimization is needed, ensuring that the formulation remains stable and usable at -50℃. Summary of the Invention

[0004] The present invention aims to provide a water-based fire extinguishing agent with a low-temperature synergistic regulation mechanism, which significantly reduces the number of tests, improves optimization convergence efficiency, and enhances adaptability under different working conditions while ensuring safety margin, thereby improving the engineering reliability and stability of the -50℃ water-based fire extinguishing agent.

[0005] To achieve the aforementioned objective, this invention provides a water-based fire extinguishing agent with a low-temperature synergistic regulation mechanism, characterized in that: the fire extinguishing agent comprises, by weight, parts of a water-based carrier. Low-temperature phase-stable molecular weight Low-temperature rheology regulation of molecular mass Film-forming oxygen barrier synergistic molecular weight and performance compatibility constraints submass The formulation parameters of the extinguishing agent are determined by an intelligent optimization method executed by the processor, the method including: constructing a formulation parameter vector. It also defines a feasible formulation domain that includes upper and lower limits for mass parts, non-negativity, and total quantity conservation constraints. The Bayesian surrogate model is trained on the entire trial dataset up to the current iteration in round t to obtain the posterior distribution of the objective function, and then applied to any candidate formulation. Output predicted mean With prediction uncertainty A data acquisition strategy is constructed based on the predicted mean and the predicted uncertainty. And select the sorting order in each iteration. A candidate formulation; the candidate formulation is subjected to parameter-controlled processing. The modulated risk-weighted feasible region projection yields a feasible recipe and outputs a constraint function; opportunity-constrained probabilistic safety screening is performed on the feasible recipe, and a probabilistic violation risk feedback quantity is constructed; the risk feedback quantity is incorporated into the total loss function of the new task to update the meta-parameters. This allows the updated meta-parameters to modulate the projection risk weights and data collection strategies in the next round, thus forming a closed loop of "data collection—projection—screening—feedback—update," iteratively outputting the optimal formula. The water-based fire extinguishing agent was prepared.

[0006] Compared with the prior art, the present invention has the following beneficial effects:

[0007] This invention constructs a closed-loop optimization mechanism based on Bayesian uncertainty prediction, probabilistic chance constraint screening, risk feedback-driven meta-parameter updates, and feasible region projection correction. This mechanism achieves synergistic control of low-temperature phase stability, low-temperature rheological sprayability, and material compatibility. Compared to existing methods that rely on empirical proportions or deterministic screening, this invention significantly reduces the number of experiments, improves optimization convergence efficiency, and enhances adaptability under different operating conditions while ensuring safety margins. This, in turn, improves the engineering reliability and stability of water-based fire extinguishing agents operating at -50℃. Attached Figure Description

[0008] Figure 1 This is a flowchart of the intelligent optimization method provided by the present invention. Detailed Implementation

[0009] To enable those skilled in the art to better understand the technical solutions of the present invention, the clock loop forming processing method and apparatus provided by the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0010] The invention will be described more fully below with reference to the accompanying drawings; however, the embodiments shown may be embodied in different forms, and the invention should not be construed as limited to the embodiments set forth below. Rather, these embodiments are provided so that the invention will be thorough and complete, and that those skilled in the art will fully understand the scope of the invention.

[0011] The accompanying drawings of the embodiments of the present invention are provided to further illustrate the embodiments of the present invention and form part of the specification. They are used together with the detailed embodiments to explain the present invention and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the description of the detailed embodiments with reference to the accompanying drawings.

[0012] The present invention is not limited to the embodiments shown in the accompanying drawings, which illustrate illustrative properties but are not intended to be limiting.

[0013] The -50℃ water-based fire extinguishing agent with a low-temperature synergistic regulation mechanism provided by this invention comprises: a water-based carrier in parts by mass. Low-temperature phase stabilizers Low-temperature rheological regulator Film-forming oxygen barrier synergist and performance compatibility constraints Its formula parameters are determined by an intelligent optimization method executed by the processor.

[0014] Figure 1 This is a flowchart of the intelligent optimization method provided by the present invention, such as... Figure 1 As shown, the intelligent optimization method includes: constructing a formula parameter vector. It also defines a feasible formulation domain that includes upper and lower limits for mass parts, non-negativity, and total quantity conservation constraints. The Bayesian surrogate model is trained on the entire trial dataset up to the current iteration in round t to obtain the posterior distribution of the objective function, and then applied to any candidate formulation. Output predicted mean With prediction uncertainty A data acquisition strategy is constructed based on the predicted mean and the predicted uncertainty. And select the sorting order in each iteration. A candidate formulation; the candidate formulation is subjected to parameter-controlled processing. The modulated risk-weighted feasible region projection yields a feasible recipe and outputs a constraint function; opportunity-constrained probabilistic safety screening is performed on the feasible recipe, and a probabilistic violation risk feedback quantity is constructed; the risk feedback quantity is incorporated into the total loss function of the new task to update the meta-parameters. This allows the updated meta-parameters to modulate the projection risk weights and data collection strategies in the next round, thus forming a closed loop of "data collection—projection—screening—feedback—update," iteratively outputting the optimal formula. The water-based fire extinguishing agent was prepared.

[0015] In this invention, a formula parameter vector is constructed. It also defines a feasible formulation domain that includes upper and lower limits for mass parts, non-negativity, and total quantity conservation constraints. Specifically, it includes the following sub-steps: S1-1 Component Mapping and Parametric Modeling: The water-based fire extinguishing agent to be optimized is divided into five functional categories, and each component is parameterized by mass fraction to obtain the formulation parameter vector. , in, : Recipe parameter vector; : Mass fraction of water-based carrier; Low-temperature phase-stable molecular weight; Low-temperature rheological regulation of molecular mass fraction; : Film-forming oxygen barrier synergistic molecular weight; The performance compatibility constraint sub-mass components: The above five types of components respectively play the functional roles of "cooling carrier, low temperature phase stability, low temperature jettable rheology, film formation oxygen barrier, material compatibility / corrosion control", etc. Using mass components to uniformly parameterize facilitates the continuous optimization of the formula by subsequent algorithms.

[0016] To achieve a synergistic regulation mechanism for low temperatures of -50℃ and below, the fire extinguishing agent includes the following functional components, and each component may be selected from one or more of the following materials in combination.

[0017] The water-based carrier W is used to provide the cooling medium and the main phase change endothermic capacity, and is preferably a low conductivity water system. Specifically, it includes: deionized water, purified water, softened water, or ultrapure water with a conductivity not higher than 10 μS / cm.

[0018] The water-based carrier W is preferably deionized water or purified water with a conductivity of less than 5 μS / cm, in order to reduce the overall conductivity of the system and improve material compatibility.

[0019] The low-temperature phase stabilizer A is used to lower the freezing point and inhibit crystallization and phase separation, and preferably includes a combination of freezing point lowering component A1 and crystallization inhibiting component A2. The freezing point lowering component A1 may include: ethylene glycol, propylene glycol, glycerol, ethylene glycol monomethyl ether, low molecular weight polyether, methanol, or ethanol.

[0020] The anti-crystallization component A2 may include: polyvinyl alcohol (PVA), sodium polyacrylate (PAA-Na), sodium carboxymethyl cellulose (CMC-Na), polyaspartic acid, or organophosphonate crystal form modifiers.

[0021] The low-temperature phase stabilizer A lowers the freezing point and improves the low-temperature phase stability through the synergistic effect of solution colligative freezing point reduction and crystal nucleation inhibition.

[0022] The low-temperature rheology regulator B is used to construct a "static weak network-shear reversible disintegration" rheological structure at low temperatures, thereby balancing storage stability and sprayability. Specifically, it includes: xanthan gum, hydroxyethyl cellulose, carboxymethyl cellulose, polyacrylamide, organobentonite, polyurethane rheology modifiers, or temperature-sensitive polymer materials.

[0023] The low-temperature rheology regulator B is preferably a shear-thinning polymeric rheology modifier.

[0024] The film-forming oxygen barrier synergist C enhances wetting and spreading capabilities and forms a stable covering film, thereby improving the coverage retention capability after fire extinguishing. Specifically, it includes wetting components and film-stabilizing components. The wetting components may include: sodium dodecyl sulfate, alkyl glycosides, amphoteric surfactants (such as betaines), fluorocarbon surfactants, or nonionic surfactants.

[0025] The membrane stabilizing component may include: polyvinyl alcohol, acrylic copolymers, aqueous emulsion film-forming aids, or protein foam stabilizers.

[0026] The film-forming oxygen barrier synergist C improves the coverage time integral by reducing surface tension and enhancing film stability.

[0027] The performance-compatible constraint D is used to reduce corrosion risk and adjust the ionic strength of the system, and preferably includes corrosion inhibitors, pH adjusters, or complexing agents. Specifically, it includes: sodium benzoate, molybdate corrosion inhibitors, borate corrosion inhibitors, phosphate corrosion inhibitor systems, organic amine corrosion inhibitors, triethanolamine, EDTA, or other complexing agents.

[0028] The performance compatibility constraint D is used to inhibit corrosion of copper, aluminum and stainless steel materials and improve material compatibility.

[0029] S1-2 Construction of Mass Conservation and Non-negativity Constraints: Applying total mass conservation and non-negativity constraints to the formulation parameter vector forms the basic feasible conditions: , ,in: : The Middle Each component corresponds to ; The mass fraction conservation constraint means that the total quantity is normalized to 100 parts; Non-negative constraint: This means that the mass fraction of each component cannot be negative.

[0030] S1-3 Upper and Lower Limit Boundaries (Engineering Applicable Range): Based on the solubility of raw materials, low-temperature phase stability, feasibility of jet rheology, and film formation and compatibility requirements, upper and lower limit boundaries for the mass fractions are set for each component: .in: : No. Lower limit of component mass parts; : No. Upper limit of component mass parts.

[0031] Optional implementation: The above upper and lower limits can be set as fixed intervals or scenario-specific intervals; for example, increasing them in lower temperature environments. The upper limit should be appropriately reduced in scenarios with higher requirements for atomization. The upper limit is set to expand the space for feasible exploration while remaining feasible.

[0032] S1-4 Feasible Recipe Domain The set-based definition specifically includes: merging conservation, nonnegativity, and upper and lower bound constraints, and defining the feasible region of the complete solution as: , in: This is the recipe vector; For the first Upper and lower limits of component mass parts; For feasible regions: : Five-dimensional real space.

[0033] S1-5 Unified Encoding and Data Structure Output: and The output is a shared data structure for subsequent algorithm modules, used for candidate recipe generation, projection correction, and trial scheduling; where: As an "input code" for a single recipe sample; This serves as the "configurable constraint space" for subsequent projection operators and candidate selection; for each Associated unique sample identifier With test round identifier This is used to construct the dataset for subsequent experiments.

[0034] This invention uses formula parameter vectorization This transforms the complex formulation design problem into a computable continuous variable optimization problem, enabling the subsequent Bayesian surrogate model and meta-learning strategy to directly learn and update the formulation space, thereby reducing the number of manual trial formulations.

[0035] The Bayesian surrogate model is trained on the entire trial dataset up to the current iteration in round t to obtain the posterior distribution of the objective function, and then applied to any candidate formulation. Output predicted mean With prediction uncertainty Specifically, it includes:

[0036] S2-1 Experimental Data Construction and Consistency Preprocessing: For each set of historical experimental formulations Preparation and testing were performed to obtain performance observation vectors. , forming the first Experimental dataset for round iterations:

[0037] in: For the first The test dataset during round iteration; For the first There are a vector of formula parameters, and they satisfy... ; For the first A vector of real performance observations for each formulation; This represents the number of iteration rounds or the cumulative number of samples.

[0038] Furthermore, specifically including: [the following] Dimensional consistency and normalization are performed to obtain standardized observations for model training. : , in: Normalization operator, used to map indices with different dimensions to a uniform scale; : Normalized performance observation vector.

[0039] Optionally, Minimum-maximum normalization or mean-variance-based standardization can be used to give different metrics comparable weights during training.

[0040] S2-2 Multi-Indicator Target Mapping and Proxy Output Definition: The key performance characteristics of the extinguishing agent are defined as multi-indicator outputs, including at least three of the following: freezing point, low-temperature viscosity, and corrosion rate, forming a target vector: , in: : No. Freezing point of each sample; : No. Apparent viscosity of a sample at -50℃; : No. The corrosion rate of each sample; It may also include: optional extended metrics (such as crystallization increment, phase separation volume fraction increment, coverage integral, etc.).

[0041] And define a Bayesian surrogate model for any candidate formulation The output is the predicted mean. With prediction uncertainty : , in: In the first Formula after round of training The predicted mean (which can be a scalar or a vector); For the formulation The prediction uncertainty (which can be the standard deviation or variance).

[0042] S2-3 Bayesian surrogate model construction: Construct a Gaussian process regression (GP) surrogate model for each indicator. Constructing a Gaussian process regression model: , in: For the first The implicit response function of each index; Gaussian process; Kernel functions (such as RBF kernel or Matérn kernel).

[0043] Specifically, this includes: utilizing training data For any candidate formulation Calculate the posterior predicted distribution: , in: For the first The posterior predictive mean of each indicator; For the first The posterior prediction variance of each indicator; :normal distribution.

[0044] Optionally, for multiple metrics, training can be performed separately. indivual Or use multiple outputs Correlation between modeling indicators.

[0045] Bayesian Neural Network (BNN) surrogate model: building neural networks and weights Treating it as a random variable, define its posterior distribution: , in: These are network weight parameters; It is the prior distribution; Let be the likelihood function.

[0046] Specifically, this includes: obtaining approximate posteriors using variational inference. And trained by minimizing the KL divergence: , in: It is a variational distribution; For variational parameters; Let KL divergence be denoted as KL divergence.

[0047] When making predictions using the following formula, the mean and prediction variance (i.e., the square of the uncertainty) are obtained through Monte Carlo sampling: , , in: Number of samples; For the first Next from The weights obtained from sampling; , Predict the mean and the variance respectively.

[0048] S2-4 Uncertainty Calibration and Update Output: This section calibrates the predicted uncertainty to ensure... It can reflect the characteristic that "sparse data regions are more uncertain, while dense data regions are more certain"; and will output and As input for constructing the S3 acquisition strategy.

[0049] Optionally: Apply a noise term to the GP Alternatively, temperature scaling can be used to calibrate the uncertainty of BNN.

[0050] This invention constructs a Bayesian surrogate model and outputs the predicted mean. With prediction uncertainty By probabilistically modeling the formulation-performance relationship under limited experimental samples, the optimization process can explicitly identify candidate regions that are "high-performance but uncertain" and "fully validated" candidate regions, thereby reducing blind trial formulations and improving optimization efficiency.

[0051] In this invention, a data acquisition strategy is constructed from the predicted mean and the predicted uncertainty. And select the sorting order in each iteration. The candidate formulations specifically include: S3-1 Candidate Recipe Set Generation: From the feasible recipe domain defined in S1... Internally generated candidate recipe set Candidates are used for evaluation and ranking in this iteration: , in: : A set of candidate recipes; : Feasible recipe domain.

[0052] Optional: It can be generated through grid sampling, Latin hypercube sampling, Sobol sequence sampling, or local perturbation based on the historical best neighborhood, to cover both global exploration and local fine search needs.

[0053] S3-2 Obtain Bayesian proxy output and construct a collection strategy: for each candidate recipe The Bayesian surrogate model of S2 is called to output the predicted mean. With prediction uncertainty Based on this, a data collection strategy is constructed: , in: For the first Round-based iterative acquisition strategy; This recipe The predicted mean; For the formula The uncertainty of prediction; This exploration intensity parameter is used to balance "utilization (mean)" and "exploration (uncertainty)"; For iteration rounds.

[0054] when When the size is larger, it tends to explore uncertain areas; when When the size is smaller, it tends to select regions with better prediction performance.

[0055] S3-3 Scheduling of Exploration Intensity Parameters: The exploration intensity parameter is decayed by round, resulting in stronger exploration in the early stages and stronger convergence in the later stages. , in: This initial exploration intensity; The attenuation coefficient; For iteration rounds.

[0056] Optional: It is also possible to Let it be a piecewise constant (as before) (Using larger values ​​in turn, and smaller values ​​in subsequent iterations) to adapt to the requirements of experimental cost and convergence speed.

[0057] S3-4 TopK ranking selection and trial budget allocation: For all candidates... calculate Sort by collection strategy from largest to smallest, and select the top... One candidate formulation was used for real-world preparation and testing: , in: For the first One selected candidate formula; Budget for test samples in a single iteration; To select the first value from largest to smallest The operator.

[0058] Optionally: To avoid excessive concentration of candidates, a minimum distance constraint can be added to the TopK selection. To enhance diversity.

[0059] S3-5 Output Interface and Data Structure Encapsulation: Encapsulating the candidate recipes output by TopK. As input to the S4 projection operator, and with each candidate formulation appended its acquisition strategy score and uncertainty label: , in: This round outputs a data packet, which is used to drive subsequent projections and test scheduling; Candidate formulation The collected scores; Candidate formulation Uncertainty label.

[0060] This invention constructs a data acquisition strategy. By unifying "predictive performance" and "predictive uncertainty" into a sortable index, the algorithm can simultaneously find high-performance formulations and explore sparse data regions, thereby reducing blind trial and error and increasing experimental information gain.

[0061] By selecting only the TopK candidate formulations for real-world testing in each iteration, the testing budget is explicitly incorporated into the optimization process. This allows for stable model updates and performance improvements even under cost-constrained testing conditions, achieving the beneficial effects of reducing the number of experiments and shortening the R&D cycle.

[0062] In this invention, the candidate formulation is subjected to parameter-dependent processing. The modulated risk-weighted feasible region projection yields a feasible recipe, and the output constraint function specifically includes: S4-1 Receive candidate recipes and construct projective input: Receive candidate recipe set: ,in: For the first A vector of candidate formulation parameters; This is the budgeted sample size for this round of trials; each .

[0063] S4-2 Constructing a projection operator modulated by elementary parameters: for each candidate formulation Constructing a feasible region projection problem with risk-weighted terms: , in: The feasible formula after projection; For the subject parameter Modulated projection operator; A feasible recipe domain is constructed for S1; : L2 norm; : Soft penalty weight; : No. One performance constraint function; : By the current meta-parameter Output constraint weights.

[0064] S4-3 Performance constraint function definition: Transform physical and engineering constraints into constraint functions that include at least the freeze point. Low-temperature viscosity constraint function Corrosion rate constraint function : , , , in: The freezing point; The apparent viscosity at -50℃; For corrosion rate; This is the viscosity threshold; Corrosion threshold: If This indicates a violation of Article [number missing]. Item constraint.

[0065] S4-4 Modulation of risk weights by meta-parameters: meta-parameters updated from S6 Generate risk weights: , in: For the first Risk weights for each constraint; : The intermediate representation obtained by mapping meta-parameters; : Normalization function, making .

[0066] S4-5 Projection Result Output and Packaging: Output the set of feasible recipes after projection. , Includes information on the degree of violation: , in: This is the set of formulas after this round of revisions; This is the corresponding violation matrix.

[0067] This invention constructs a risk-weighted feasible region projection operator to modify the original candidate formulation into a feasible formulation that meets the engineering compatibility range, thereby ensuring that the algorithm output is always prepareable and implementable, achieving the beneficial effects of avoiding unrealizable formulations and reducing experimental waste.

[0068] By introducing risk weights modulated by metaparameters This allows the projection direction to change dynamically with risk feedback, thereby enabling high-risk constraints to automatically receive higher correction weights in the projection, achieving the beneficial effect of improving low-temperature stability and safety.

[0069] In this invention, the process of performing opportunity-constrained probabilistic safety screening on the feasible formulation and constructing a probabilistic violation risk feedback quantity specifically includes the following sub-steps: S5-1 Real Performance Testing and Observation Vector Construction: Each sample was fabricated and its performance tested in real-world applications to obtain observation vectors. ,in: For the first The true performance observation vector of each sample; The freezing point; The apparent viscosity at -50℃; For corrosion rate; This represents the increase in the mass fraction of crystallization. This represents the volume fraction increment of phase separation.

[0070] S5-2 Probabilistic Model Construction (Combined with Surrogate Uncertainty): Using the predicted mean and uncertainty output by S2, construct probabilistic models for each performance index: , To predict the mean; To predict variance; :normal distribution.

[0071] S5-3 Chance Constraint Calculation: Calculate the probability that each indicator satisfies the constraints. ,in: : Standard normal cumulative distribution function; numerator represents the difference between the target threshold and the predicted mean; denominator is the predicted standard deviation.

[0072] Similarly: ,in: This refers to the low-temperature viscosity threshold. This represents the corrosion rate threshold.

[0073] S5-4 Opportunity Constraint Screening and Judgment: Setting a Risk Confidence Threshold:

[0074] If the following conditions are met: , , , The sample is then determined to have passed the security screening, where: Allowable failure probability.

[0075] S5-5 Constructing the probabilistic violation risk feedback quantity: Constructing a risk feedback quantity for samples that do not meet the constraints: , in: : No. The probability of violation for each sample; : No. One performance constraint function; : Corresponding risk confidence threshold; This indicates taking the non-negative part; In the formula The probability that the j-th constraint is satisfied. Let j represent the constraint function for the j-th term, where j = 1, 2, 3.

[0076] When all constraints are satisfied: .

[0077] S5-6 Data Output and Update: Add the selected samples to the training dataset: , Will Output to S6 for meta-learning updates.

[0078] This invention transforms performance indicators from deterministic judgments to probabilistic confidence judgments by constructing a chance-constrained probability model, thereby enabling robust decision-making even when model uncertainty exists, and achieving the beneficial effect of improving low-temperature stability and safety.

[0079] This invention constructs a probabilistic violation risk feedback quantity. This allows the degree of constraint violation to be quantified and fed back to the meta-learning module, achieving the beneficial effect of forming a risk-sensitive adaptive optimization closed loop.

[0080] By adding the filtered samples to the training dataset This allows the surrogate model to gradually reduce the uncertainty in the critical boundary region, thereby achieving the beneficial effects of accelerating convergence and reducing the number of trials.

[0081] In this invention, the risk feedback quantity is incorporated into the total loss function of the new task to update the meta-parameters. The updated meta-parameters are used to inversely modulate the next round of projection risk weights and the acquisition strategy, including: S6-1 Task Set Construction and Meta-parameter Initialization: Construct task sets based on different ambient temperature ranges, test conditions, or fire extinguishing scenarios: , in: : Task set; : No. One task scenario; Number of tasks. Defines a shared meta-parameter for all tasks. Used to modulate the risk weight function in S4 .

[0082] S6-2 Constructing a Risk-Driven Comprehensive Loss Function: Constructing a comprehensive loss function based on the risk feedback quantity output from S5 and the actual performance data: , in: Let be the total loss function of the new task; The performance index loss function; This refers to the risk weighting coefficient. This refers to the probability of violation feedback generated by S5. This is the sample size for this round of testing.

[0083] The performance loss function is: , in: Performance weights; The freezing point; Low temperature viscosity; The corrosion rate is represented by the value of .

[0084] S6-3 Gradient Update of Meta-parameters: Perform gradient update on the meta-parameters: in: This is the current meta-parameter; For the updated meta parameters In-task learning rate; To The gradient operator; For iteration rounds.

[0085] S6-4 Cross-Task Quick Adaptation: In new tasks Below, a fast update is performed using only a small number of samples: , in: These are the basic meta-parameters shared across tasks; Parameters adapted for quick adaptation to new tasks; New scenario task.

[0086] S6-5 Inverse Modulation of Projection and Risk Weights by Metaparameters: Updated Metaparameters Recalculated for risk weights in S4: , in: This is the updated constraint weight; This is an intermediate representation generated by the meta-parameter mapping; This is the normalization function.

[0087] In this invention, the updated meta-parameters On the one hand, it is used to modulate the risk-weighted projection and candidate sampling in the next round, thereby changing the newly added test samples and updating the dataset. On the other hand, as inputs to the kernel function / prior / noise or sample weight hyperparameters of the Bayesian surrogate model, the Bayesian surrogate output... and Follow Adaptive updates form a closed-loop coupling across steps.

[0088] This invention constructs a comprehensive loss function that includes risk feedback, thereby enabling constraint violation information to directly participate in the updating of meta-parameters, achieving the beneficial effect of making the optimization process risk-sensitive.

[0089] By inversely modulating the projection operator and risk weights using meta-parameters, high-risk constraints automatically receive higher correction priority in subsequent iterations, thereby improving low-temperature stability and safety margin.

[0090] Through a cross-task rapid adaptation mechanism, parameter adjustments can be completed with only a small number of samples under different ambient temperatures or scenarios, thereby reducing R&D costs and improving engineering promotion efficiency.

[0091] Comparison Implementation

[0092] Table 1. Formulation Grouping Settings

[0093] In Table 1, “√” indicates inclusion, and “√” indicates exclusion.

[0094] Table 2 Comparison of Low-Temperature Phase and Rheological Properties

[0095] Test aperture example: -50 ℃ ± 2 ℃, ts = 168 after standing. Viscosity measured by rheometer at -50℃

[0096] in:

[0097]

[0098] The freezing point; This represents the increase in the mass fraction of crystallization. This represents the volume fraction increment of phase separation; Apparent viscosity; Shear-thinning ratio; This represents the jet flow rate.

[0099] Table 3 Comparison of film-forming coverage and fire extinguishing performance

[0100] Example of test aperture: hot plate Fixed spray dose and distance; statistical window .

[0101] in:

[0102] C(t) represents the instantaneous coverage. The covered area; For the target area; For coverage integral; This is a statistics window.

[0103] Table 4. Example data on material compatibility (corrosion rate)

[0104] Example of test caliber: Immersion at 25℃ for 72 hours, mass loss converted to mm / a (or you can also use mg / (cm²·d), but it must be consistent).

[0105] The results showed that, compared with the traditional water-based control group G2, the freezing point of the present invention group G1 at −50℃ was −55.2℃, and the crystallization increment after standing for 168h was greater. With hierarchical increment The viscosity was significantly reduced, while the low-temperature viscosity and shear thinning ratio met the sprayability requirements, and the spray flow rate was stable and unobstructed. Compared with the control group G3 (which only lowered the freezing point), G1 significantly reduced crystallization / stratification and improved spray stability. Compared with the rheology-free group G4, G1 showed a significant reduction in low-temperature viscosity, increased spray flow rate, and eliminated obstruction, indicating that the rheology regulator plays a key role in low-temperature sprayability. Compared with the film-forming synergy-free group G5, G1 showed a significant increase in coverage integral and the re-ignition phenomenon disappeared, indicating that the film-forming oxygen barrier synergy effectively improved coverage retention. Compared with the compatibility-constraint-free group G6, G1 showed a significant reduction in the corrosion rate of copper and aluminum materials, indicating that the compatibility-constraint can effectively suppress corrosion risk. In summary, this invention achieves a synergistic improvement in phase stability, sprayable rheology, film coverage, and material compatibility at −50℃.

[0106] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An optimization method for a water-based fire extinguishing agent with a low-temperature synergistic regulation mechanism, characterized in that: The extinguishing agent includes, by weight, parts of water-based carrier. Low-temperature phase-stable molecular weight Low-temperature rheology regulation of molecular mass Film-forming oxygen barrier synergistic molecular weight and performance compatibility constraints submass ; The formulation parameters of the extinguishing agent are determined by an intelligent optimization method executed by the processor, the method including: constructing a formulation parameter vector. It also defines a feasible formulation domain that includes upper and lower limits for mass parts, non-negativity, and total quantity conservation constraints. The Bayesian surrogate model is trained on the entire trial dataset up to the current iteration in round t to obtain the posterior distribution of the objective function, and then applied to any candidate formulation. Output predicted mean With prediction uncertainty A data acquisition strategy is constructed based on the predicted mean and the predicted uncertainty. And select the sorting order in each iteration. A candidate formulation; the candidate formulation is subjected to parameter-controlled processing. The modulated risk-weighted feasible region projection yields a feasible recipe and outputs a constraint function; opportunity-constrained probabilistic safety screening is performed on the feasible recipe, and a probabilistic violation risk feedback quantity is constructed; the risk feedback quantity is incorporated into the total loss function of the new task to update the meta-parameters. This causes the updated meta-parameters to modulate the projection risk weights and collection strategies in the next round, thus forming a closed loop of "collection—projection—screening—feedback—update," iteratively outputting the optimal formula. The water-based fire extinguishing agent was prepared.

2. The method according to claim 1, characterized in that, The formula parameter vector With feasible recipe domain They are respectively: ， ， in; For the first Upper and lower limits of component mass parts, i=1,…,5.

3. The method according to claim 2, characterized in that, In the t-th iteration, the dataset based on all test data up to the current time step is: , in: For the first One recipe; For the first Real-world performance observation of each formulation.

4. The method according to claim 3, characterized in that, The data collection strategy Before sorting in each iteration The candidate formulations are as follows: , , in; To explore strength parameters; For the first One candidate formulation; This refers to the sample size in a single round of testing. Select operators for sorting.

5. The method according to claim 4, characterized in that, For candidate formulations Execution subject to meta-parameters Modulated risk-weighted projection yields feasible formulations : , in: For the subject parameter Modulated projection operator; It is a norm 2; Soft penalty weight; For the first One constraint function; These are the constraint weights output by the meta-parameters.

6. The method according to claim 5, characterized in that, The constraint functions include at least the freeze point constraint function. Low-temperature viscosity constraint function Corrosion rate constraint function : ， ， ， in: The freezing point; The apparent viscosity at -50℃; For corrosion rate; This is the viscosity threshold; This represents the corrosion threshold.

7. The method according to claim 6, characterized in that, The probability-constrained safety screening of the feasible formulation includes at least the following: ， ， ， in: Risk confidence threshold; This represents a probability function.

8. The method according to claim 7, characterized in that, The probability of violation risk feedback volume: ， in; For the first Risk confidence threshold for the constraint; Let j be the constraint function for the j-th term; This indicates taking the non-negative part; In the formula Next, the The probability that the constraint is satisfied; Let j represent the constraint function for the j-th term, where j = 1, 2, 3.

9. The method according to claim 8, characterized in that, The comprehensive loss function for the new task is: , in: For the loss function of the new task; This is the performance loss function; Risk weights; The meta-parameters are updated as follows: , The projection risk weights are then inversely modulated using the updated meta-parameters: , in: These are the meta-parameters before and after the update, respectively. The learning rate within the task; For gradient operators; For the first Each constraint weight; For mapping functions; This is the normalization function.

10. A water-based fire extinguishing agent, characterized in that, The optimal formulation output by the method according to any one of claims 1-9 preparation.

Images