A non-toxic, multi-purpose water-based fire extinguishing agent

Abstract

This invention discloses a non-toxic, multi-purpose water-based fire extinguishing agent. The water-based fire extinguishing agent formulation is represented as a mass fraction vector, and the preparation process and spraying conditions are represented as vectors, with a joint feasible region constructed. A surrogate model is trained based on a sample dataset, outputting the performance prediction mean and variance. A set of constraint functions, including non-toxicity constraints and cross-scenario multi-purpose constraints, is constructed, and the probability of satisfaction is calculated. A cross-scenario comprehensive performance function is constructed, and the next round of test points is selected and iteratively updated using a risk constraint acquisition function. The optimal prepared fire extinguishing agent is output when the termination condition is met. This method reduces the number of tests while improving the non-toxic safety margin and adaptability to A / B class scenarios.

Description

Technical Field

[0001] This invention relates to a non-toxic, multi-purpose water-based fire extinguishing agent, and more particularly to a non-toxic, multi-purpose water-based fire extinguishing agent formulation and process optimization method, as well as the non-toxic, multi-purpose water-based fire extinguishing agent itself, belonging to the field of fire protection materials and artificial intelligence optimization design technology. Background Technology

[0002] Current research and development of water-based fire extinguishing agents mainly relies on human experience and single-index optimization methods, which has the following problems:

[0003] 1. Difficulty in coordinating multiple objectives: There are coupling conflicts among objectives such as non-toxicity, low corrosivity, low electrical conductivity, freeze-thaw stability, and Class A / B fire extinguishing efficiency;

[0004] 2. High experimental costs: Requires numerous orthogonal or full-factor experiments;

[0005] 3. Insufficient stability: The optimal laboratory formulation cannot guarantee batch consistency;

[0006] 4. Insufficient scenario adaptability: It is difficult to simultaneously handle both solid and liquid fires;

[0007] 5. The terms "non-toxic and multi-purpose" are often used qualitatively and lack a safety margin in a probabilistic sense.

[0008] Therefore, there is a need for a computable, iterative, verifiable, and batch-stable method for the formulation and process optimization of a non-toxic, multi-purpose water-based fire extinguishing agent. Summary of the Invention

[0009] To achieve the aforementioned objectives, this invention provides a non-toxic, multi-purpose water-based fire extinguishing agent formulation and process optimization method, as well as a non-toxic, multi-purpose water-based fire extinguishing agent. Under the premise of ensuring non-toxicity and material compatibility safety margins, it achieves synergistic optimization design and stable output of water-based fire extinguishing agents in multiple scenarios such as Class A and Class B, while reducing testing costs and the risk of fluctuations in mass production.

[0010] To achieve the aforementioned objective, this invention provides a formulation and process optimization method for a non-toxic, multi-purpose water-based fire extinguishing agent, which is executed by a processor and includes:

[0011] S1) Represent the extinguishing agent formulation as a formulation vector x; represent the process parameters as a parameter vector u, and construct the joint feasible region. ;

[0012] S2) Based on sample dataset Train the agent model, in order to The average value of the input-output performance prediction for the trained surrogate model variance of prediction ,in, These represent the j-th formula vector, process parameter vector, and performance vector, respectively.

[0013] S3) Construct a set of constraint functions This includes at least: non-toxic constraints and cross-scenario multi-purpose constraints; and calculates the probability of constraint satisfaction based on the predicted variance:

[0014]

[0015] in For the first The failure probability allowed by each constraint. Represents probability;

[0016] S4) Construct a cross-scenario comprehensive performance function based on the performance prediction:

[0017] ,

[0018] in These are performance sub-functions for at least two typical fire scenarios. The risk term is determined by the predicted mean and the predicted variance; These are the weighting coefficients;

[0019] S5) Construct the risk constraint acquisition function:

[0020] and solve

[0021] ,

[0022] According to the above Preparation and testing to obtain new samples To update the agent model; wherein Based on the comprehensive performance function Expected improvements;

[0023] S6) When the termination condition is met, output the optimized formula and process parameters. It satisfies the constraint, satisfies the probability condition, and makes To achieve the preset optimal criteria.

[0024] To achieve the aforementioned objective, this invention also provides a non-toxic, multi-purpose water-based fire extinguishing agent, which is formulated and processed using the optimized parameters determined by the above method. Prepare and satisfy the minimum performance constraints corresponding to both scenario A and scenario B. , And restrictions on prohibited substances and conductivity constraints Established at the same time.

[0025] Compared with existing technologies, this invention represents the fire extinguishing agent formulation as a mass fraction vector x, the preparation process as a process vector u, and constructs a joint feasible region. This transforms the traditional formula optimization problem, which relies on experience-based adjustments, into a structured and computable multivariate optimization problem. Consequently, the coupling relationship between formula variables and preparation variables can be modeled in a unified manner, avoiding local optima caused by single-factor experiments and achieving system-level synergistic optimization between formula performance, process stability, and spraying conditions.

[0026] By constructing a multi-task neural network, the predicted performance values ​​can be simultaneously calculated. variance of prediction This allows the system to obtain not only performance predictions but also prediction confidence intervals, enabling subsequent optimization decisions to consider the risks arising from model uncertainty. This avoids blindly selecting high-risk formulations in sparse data regions, significantly improving the reliability and safety of the R&D phase. Compared to traditional deterministic regression, this invention can statistically control the failure probability, improving the safety margin for industrial applications.

[0027] By constructing constraint functions such as prohibited substance functions, pH window functions, corrosion threshold functions, and conductivity functions, and introducing a probabilistic constraint mechanism, the "non-toxic" designation is transformed from a qualitative statement into a calculable, verifiable, and quantifiable statistical safety constraint. This avoids ignoring batch fluctuation risks based solely on a single passing test, and improves the safety and stability of fire extinguishing agents under long-term storage and large-scale production conditions. This mechanism is particularly suitable for applications involving electrical equipment, confined spaces, and densely populated environments.

[0028] By constructing a cross-scenario mismatch risk function and incorporating it into the comprehensive performance function for robust optimization, the optimization results can simultaneously meet the minimum performance requirements for Class A solid fires and Class B liquid fires, thus avoiding the performance mismatch problems of "effective for Class A but ineffective for Class B" or "good smoke suppression but high re-ignition rate" in traditional technologies.

[0029] By constructing a data acquisition function that includes expected improvement terms and constraint satisfaction probability terms, each round of experimentation is conducted in a region of "high efficiency + high safety probability," significantly reducing the number of invalid experiments. This typically reduces experimental costs by more than 30%, while ensuring a balance between exploration and utilization, thus improving R&D efficiency.

[0030] By introducing CVaR or quantile robust optimization, the lower limit of performance can be controlled under extremely adverse conditions, avoiding significant performance degradation of extinguishing agents when temperature fluctuates, batch micro-deviations occur, or spraying conditions change, thereby improving reliability in extreme environments and enhancing product engineering usability.

[0031] By incorporating injection pressure, nozzle orifice diameter, or atomized particle size into the process vector u, the performance optimization of fire extinguishing agents is no longer limited to static physicochemical indicators, but is coupled with actual injection conditions, improving cooling efficiency and coverage uniformity, reducing the risk of secondary reignition, and enhancing on-site adaptability.

[0032] By collecting online detection vector z, batch deviations can be detected and automatically corrected in real time, avoiding performance degradation due to minor process drifts, improving batch production consistency, reducing returns or quality risks, and enhancing industrialization capabilities. Attached Figure Description

[0033] Figure 1 This is an engineering drawing of the optimized method for the formulation and process of the non-toxic multi-purpose water-based fire extinguishing agent provided by the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings.

[0035] 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.

[0036] 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.

[0037] 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.

[0038] Figure 1 These are engineering drawings illustrating the optimized formulation and process of the non-toxic, multi-purpose water-based fire extinguishing agent provided by this invention. Figure 1 As shown, this invention provides an optimized method for the formulation and process of a non-toxic, multi-purpose water-based fire extinguishing agent, which is executed by a processor and includes:

[0039] S1) Joint parametric modeling of formulation and process, specifically including:

[0040] S1-1) Establish raw material and process fields: Obtain the set of candidate raw materials for extinguishing agents M and the set of process / injection controllable parameters U, and configure upper and lower limits for the mass percentage of each raw material. Configure feasible ranges for each process parameter ,in, Indicates the first The lower limit value of each process parameter; Indicates the first The upper limit of each process parameter.

[0041] S1-2) Vectorization of formulation variables: Representing the extinguishing agent formulation as a mass fraction vector:

[0042]

[0043] And apply mass conservation and boundary constraints: ,in, : No. Mass fraction of the raw materials; : Number of raw materials; : No. The lower / upper limit of the mass fraction of each raw material. In this invention, the mass fraction can also be a percentage.

[0044] For example, the formulation variables include: water-based carrier, film-forming / wetting synergistic component, cooling and phase stabilizing component, rheological adhesion regulating component, smoke suppression and free radical inhibition component, and anti-corrosion and microbial control component, wherein the water-based carrier component includes one or more of the following: deionized water, softened water, and a small amount of buffer salt (such as potassium dihydrogen phosphate).

[0045] Film-forming / wetting synergistic components, used to improve spreading and covering ability, mainly include one or more of alkyl polysaccharides, cocamidopropyl betaine, fatty alcohol polyoxyethylene ether, and siloxane-modified surfactants (such as polyether-modified siloxanes).

[0046] Cooling and phase-stabilizing components are used to lower the freezing point and enhance low-temperature stability. They mainly include one or more of the following: propylene glycol, glycerol, ethylene glycol (in small amounts), and potassium acetate / formate.

[0047] Rheology and adhesion modulators are used to enhance adhesion and reduce sagging. They mainly include one or more of xanthan gum, hydroxypropyl methylcellulose, sodium carboxymethyl cellulose, sodium polyacrylate, and modified starch.

[0048] Smoke suppressant and free radical inhibitor components are used to inhibit the generation of combustion free radicals and smoke. These mainly include one or more of the following: ammonium dihydrogen phosphate, ammonium polyphosphate, urea, melamine, and sodium bicarbonate. Preservative and microbial control components are used to improve storage stability and equipment compatibility. These include one or more of the following: sodium benzoate, potassium sorbate, isothiazolinone preservatives (low dosage), sodium molybdate (corrosion inhibitor), and benzotriazole (copper corrosion inhibitor).

[0049] In one embodiment, the formula variable It includes: 60–85% deionized water, 2–10% alkyl polysaccharide and fatty alcohol polyoxyethylene ether compound system, 3–20% propylene glycol or glycerol as phase stabilizing component, 0.1–1.5% xanthan gum or cellulose ether as rheology regulating component, 0.5–5% phosphate or nitrogen-containing compound as smoke-suppressing free radical inhibiting component, and 0.05–1% anti-corrosion and corrosion-inhibiting component.

[0050] S1-3) Vectorization of process and spraying conditions: The preparation process and spraying conditions are expressed as follows:

[0051]

[0052] in The set of feasible processes consists of the range of each process parameter and feasible rules; : No. Process / operating condition parameters (such as temperature, stirring speed, shear rate, feeding time, injection pressure, atomized particle size, etc.); Number of parameters.

[0053] S1-4) Constructing the joint feasible region:

[0054]

[0055] This invention unifies the parameterization of formulation and process / injection conditions and constructs a joint feasible domain. This transforms the empirical formulation problem into a computable joint optimization problem, allowing subsequent agent model learning, probability constraint evaluation, and data acquisition function selection to all occur within the same feasible space, thus avoiding the mismatch risk caused by "optimizing only the formulation without considering the process".

[0056] S2) Proxy model training and uncertainty output: Specifically including:

[0057] S2-1) Sample Dataset Construction: Forming a sample dataset based on experimental / production records:

[0058]

[0059] in, Let j be the quality score vector in the sample dataset; Let j be the j-th manufacturing process vector in the sample dataset; Let y be the j-th real performance index vector in the sample dataset (covering at least non-toxic related indicators and cross-scenario performance indicators), where the performance index vector y includes prohibited substances, target pH, conductivity and corrosion rate, which satisfy the following constraints respectively.

[0060] Restricted substances Let the set of prohibited and restricted substances be: For each prohibited or restricted substance Define its mass percentage content in the formula as The upper limit threshold is ,but:

[0061]

[0062] in, : Collection of prohibited and restricted substances (such as organohalogenated flame retardants, heavy metal salts, and preservatives restricted by regulations). Restricted and prohibited substances The content (mass percentage or mass concentration) in the formula must be consistent with... (same unit) Restricted and prohibited substances The maximum allowable threshold; This indicates that all prohibited or restricted substances are within acceptable limits.

[0063] This invention uses After normalization, it is easier to compare threshold scales of different substances in a unified manner.

[0064] pH window constraint Given a target pH range Let the predicted or measured pH be ,but:

[0065]

[0066] in, In the formula With process / operating conditions The pH value can be output by a proxy model or detected online. Allowed pH range (lower / upper limits, e.g., 6.5–8.5). : This indicates that the pH value falls within the window.

[0067] Conductivity constraint Assume the predicted or measured conductivity is The upper limit of live adapter is ,but:

[0068] in, Electrical conductivity (can be output by a proxy model or detected online); Upper limit threshold for electrical conductivity (related to standards / application scenarios) This indicates that the conductivity does not exceed the limit, reducing the risk of secondary electrical problems.

[0069] Corrosion constraint Let the set of materials be... The corrosion rate was obtained from testing (containing at least carbon steel, aluminum alloy, and copper). The threshold is ,but:

[0070] ,

[0071] in, Materials (carbon steel / aluminum / copper, etc.); In materials Corrosion rate or corrosion evaluation index (mm / a or equivalent index) on the surface; :Material The allowable corrosion threshold; This indicates that all critical materials meet low corrosion requirements.

[0072] S2-2) Establishing a multi-task agent model: Training the agent model to... The mean of the input and output performance predictions. variance of prediction :

[0073] ,

[0074] ;

[0075] S2-3) Heteroscedasticity Likelihood Training: Training is performed using a heteroscedastic regression loss function.

[0076] ,

[0077] in , ,

[0078] Indicates the first True values ​​of each performance metric; Indicates the first Variance of each indicator.

[0079] By training a surrogate model that simultaneously outputs the mean and variance of performance predictions, the effect of quantifying the prediction risk is achieved while predicting performance. This provides a basis for subsequent probability constraint calculations and risk constraint collection functions, reducing the probability of failure caused by blind experimentation in sparse data regions.

[0080] S3) The set of probability constraints and the calculation of the probability of satisfying them, specifically including:

[0081] S3-1) Construction of the constraint function set: Constructing the constraint set:

[0082]

[0083] These include at least: non-toxic constraints (such as prohibited substances, pH window, irritation or migration indicators, etc.); multi-purpose constraints across scenarios (Class A and Class B minimum performance threshold constraints); and optional constraints such as conductivity, corrosion, freeze-thaw stability, etc., where Q is a positive integer greater than or equal to 2.

[0084] S3-2) Probability Satisfaction Calculation: The mean of the performance prediction based on the output of S2 and the variance of the prediction For each constraint function Uncertainty propagation is performed to obtain the predicted mean of the constraint function. With variance :

[0085] ,

[0086] ,

[0087] in, The mean vector of performance predictions output by the surrogate model; This corresponds to the performance prediction variance vector; constraint function The gradient with respect to the performance vector; This is the diagonal covariance matrix composed of the prediction variances.

[0088] Under the first-order linear approximation, the constraint function can be regarded as an approximately normal random variable:

[0089]

[0090] The probability that the constraint is satisfied is: ,

[0091] in: This is the cumulative distribution function of the standard normal distribution.

[0092] The final constraint satisfaction condition is written as:

[0093] , among which, among which For the first The failure probability allowed by each constraint. It represents probability.

[0094] This invention achieves the effect of controlling failure risk in a statistical sense by transforming "non-toxic" and "multi-purpose" into probabilistic constraints and calculating the probability of satisfying the constraints. This avoids batch fluctuation failures caused by relying solely on a single qualified test, thereby improving the safety margin and engineering reliability.

[0095] S4) Construction of cross-scenario comprehensive performance function, specifically including:

[0096] S4-1) Constructing the scene performance sub-function: Prediction performance based on the output of S2 Construct performance functions for at least two typical fire scenarios:

[0097] ,

[0098] ,

[0099] in, Corresponding to Class A solid fire scenario, This corresponds to a Class B liquid fire scenario.

[0100] In this invention, indicators that are considered "better smaller than better" (such as fire extinguishing time, reignition probability, smoke density, corrosion, and conductivity) are normalized using the following formula:

[0101]

[0102] For indicators that are considered "better the higher" (such as coverage retention, film formation score, cooling score, etc.), the following formula is used for normalization:

[0103] ,

[0104] in: ; : Corresponding performance index value (can be a predicted value) ); The lower / upper bound of this indicator (which can be taken from standards, historical data quantiles, or engineering experience ranges).

[0105] The normalization result falls on It facilitates weighted summation.

[0106] In this invention, the performance of Class A scenarios is... (Solid fire):

[0107] ,

[0108] The predicted mean and predicted standard deviation of four indicators are taken from the output of the surrogate model:

[0109] ,

[0110] ,

[0111] ,

[0112] ,

[0113] And define the correspondence of the normalization functions:

[0114] ,

[0115] ,

[0116] ,

[0117] ,

[0118] in, : The predicted mean of extinguishing time for Class A fires; : Variance of prediction for Class A fire extinguishing time; , : Predicted mean / variance of the probability of re-ignition in Class A; , : Predicted mean / variance of attachment / retention score; , : Predicted mean / variance of cooling rating; : No. The standard deviation of the indicator forecast; Performance weighting coefficient; Uncertainty penalty weighting coefficient; Uncertainty penalty intensity coefficient.

[0119] Performance in Class B scenarios (Liquid Fire):

[0120]

[0121] in, ,

[0122] ,

[0123] ,

[0124] ,

[0125] ,

[0126] Normalization function correspondence:

[0127] ,

[0128] ,

[0129] ,

[0130] ,

[0131] ,

[0132] in, : Average predicted extinguishing time for Class B fires; : Mean predicted probability of re-ignition in Class B; : Predicted mean of the smoke suppression index; : Predicted mean of film formation score; : Predicted mean surface tension; : The predicted variance of the corresponding indicator; : Predictive standard deviation; Category B performance weights; Uncertainty penalty weight; Uncertainty penalty intensity coefficient.

[0133] Defined as "the better the smoke suppression, the higher the value", then... Change to .

[0134] S4-2) Constructing Risk Terms: Introducing uncertainty and cross-scenario mismatch into risk terms:

[0135] ,

[0136] Optionally, the following definitions can be made:

[0137] ,

[0138] ,

[0139] in, Performance sub-functions for two types of scenarios; Minimum performance threshold

[0140] These are the weighting coefficients.

[0141] S4-3) Construct the comprehensive performance function:

[0142]

[0143] in, Importance weights of two types of scenarios

[0144] This invention constructs a cross-scenario performance function and introduces uncertainty and cross-scenario mismatch into the risk term, thereby achieving the effect of simultaneously optimizing the lower limit of performance in multiple scenarios and suppressing high uncertainty solutions. This avoids multi-purpose mismatch of "effective in one scenario but ineffective in another scenario" and improves stable adaptability.

[0145] S5: Risk constraint acquisition function and point selection iteration, specifically including:

[0146] S5-1) Constructing the risk constraint acquisition function It is used to weigh "performance improvement" against "risk control":

[0147] ,

[0148] in, : Expected improvement term, representing the expected improvement in overall performance relative to the current best performance; : No. The probability of satisfying each constraint;

[0149] S5-2) Solve for the next round of test points within the joint feasible region: ,

[0150] in, Iteration rounds; The next round of selected formulas and process points.

[0151] S5-3) Perform preparation and testing and update the dataset: New samples were obtained through preparation and testing. And updated:

[0152] ,

[0153] S5-4) Based on the updated Retrain or incrementally update the S2 proxy model to proceed to the next iteration.

[0154] This invention constructs a collection function by multiplying the expected improvement term with the constraint satisfaction probability and selects points within the feasible region. This achieves the effect of maximizing overall performance while ensuring the probability of non-toxicity and multi-purpose safety, thereby reducing the number of invalid experiments, lowering R&D costs, and improving convergence efficiency and success rate.

[0155] S6: Termination conditions and optimal solution output: Specifically includes:

[0156] S6-1) Set termination conditions: The iteration will terminate when at least one of the following conditions is met: The number of iterations reaches the upper limit. The maximum value of the acquisition function increases less than the threshold. :

[0157] ,

[0158] in, Convergence threshold.

[0159] The optimal solution satisfies all probability constraints and achieves the overall performance target:

[0160]

[0161] in, Target performance threshold; Represents probability; Indicates the first The probability of failure allowed by a constraint (risk tolerance).

[0162] S6-2) Output the optimal formula and process parameters :

[0163] and output the corresponding prediction performance. With risk assessment results.

[0164] This invention achieves the effect of obtaining the optimal formula and process that meets the probabilistic safety margin of non-toxicity and multi-purpose under a limited experimental budget by setting a termination condition based on the convergence of the acquisition function and the satisfaction of probabilistic constraints and outputting the optimal solution, thereby improving R&D efficiency and the reproducibility of results.

[0165] According to one embodiment of the present invention, a non-toxic multi-purpose water-based fire extinguishing agent is also provided, wherein the water-based fire extinguishing agent is formulated and processed using the optimized method described above. Prepare and satisfy the minimum performance constraints corresponding to both scenario A and scenario B. , And restrictions on prohibited substances and conductivity constraints Established at the same time.

[0166] According to one embodiment, the present invention also provides an online detection method, specifically comprising:

[0167] S7-1) Batch finished product online testing data collection (formation) ):

[0168] Specifically, this includes: 1) using optimized formulas and process parameters. After preparation, an online detection vector z is collected for each batch of finished product before filling or on the filling line. The online detection vector includes at least two or more of the following: surface tension, viscosity, pH, and conductivity; preferably:

[0169] ,

[0170] in, Surface tension (mN / m); Apparent viscosity (mPa·s); pH (dimensionless); Electrical conductivity (µS / cm or mS / cm)

[0171] 2) To ensure comparability, the online detection is performed under temperature conditions consistent with production or after temperature compensation, and the detection temperature is recorded. With batch identifier .

[0172] S7-2) Online Indicator Prediction (Forming) ):

[0173] Specifically, this includes: 1) Calling the trained proxy model. For the current batch of formula and process parameters Calculate the mean predictive performance:

[0174]

[0175] in, Proxy model; Model parameters

[0176] 2) From the mean of prediction performance Extract online observable indicators, or through mapping functions. The predicted online detection vector is obtained:

[0177] in For the "Indicator Selection / Mapping" operator, such as direct selection In .

[0178] S7-3) Deviation Calculation and Trigger Judgment (Forming) (and triggering conditions): Specifically, this includes: 1) Calculating the online detection deviation vector: ,

[0179] 2) Calculate the deviation norm: ,

[0180] in For online detection metrics; : No. Individual indicator deviations (such as) deviation)

[0181] 3) Set quality control thresholds ,when:

[0182]

[0183] This triggers calibration; preferably, a component threshold is further set. When any component satisfies Calibration is also triggered at the same time; Comprehensive quality control threshold; Component threshold.

[0184] S7-4) Calibration Calculation and Process Parameter Update (Output) )

[0185] Specifically, this includes: 1) Building or calling the calibration model According to the deviation Output process parameter calibration values:

[0186] ,

[0187] in, Model parameters.

[0188] 2) Update process parameters: ,

[0189] in, : Process calibration vector.

[0190] 3) To ensure that the updated process remains within the feasible set For the updated Perform projection or truncation:

[0191]

[0192] in Will Project to feasible region .

[0193] S7-5) Probability Constraint Review and Release

[0194] Specifically, this includes: 1) After the process is updated, based on the updated... With the established formula Call the proxy model to get the updated prediction With uncertainty .

[0195] 2) For the set of key constraints Calculate the probability of satisfying:

[0196] : No. The probability of failure allowed by each constraint; : The probability of constraint satisfaction.

[0197] 3) When any constraint fails to meet the probability threshold, perform secondary calibration or trigger manual review / line stop strategy, and record the abnormal batch label for subsequent model updates.

[0198] This invention collects online detection vectors for batches of finished products. and the predicted values ​​of the proxy model Deviation calculations are performed to transform quality fluctuations during the production process into quantifiable deviations. This makes quality control judgments calculable and automated.

[0199] By setting quality control thresholds and with Triggering calibration allows for timely intervention in batches that exceed permissible fluctuations, thereby reducing the risk of non-toxic constraint or multi-purpose performance failure caused by batch drift.

[0200] Output by constructing a calibration model And update process parameters This allows for the rapid correction of production deviations without altering the formula, thereby improving the consistency and repeatability of industrialized batches.

[0201] By updating the process By performing probability verification, subsequent batches can meet the non-toxic and multi-purpose constraints in a statistical sense, thereby improving safety margin and reliability.

[0202] Comparative experiment

[0203] Table 1. Explanation of experimental groups and controls (E / C1 / C2)

[0204] Table 2. Standardized Test Conditions (Example)

[0205] Table 3 Comparison of Multi-purpose Effects

[0206] As shown in Table 3, the present invention belongs to group E. , Simultaneous achievement rate in both scenarios Significantly higher than C1 And superior to C2 .

[0207] This invention achieves the effect of enabling the same formula to meet the minimum performance requirements in both Class A and Class B scenarios by using a cross-scenario performance function as the optimization objective and the minimum threshold of both scenarios as the criterion, thereby demonstrating stronger multi-purpose adaptability.

[0208] As shown in Table 3, Group E had a shorter fire extinguishing time. s、 s), with a lower probability of reignition ( , Both are superior to C1 and C2.

[0209] This invention improves operational reliability by jointly optimizing fire extinguishing time and reignition probability in the performance function and suppressing cross-scenario mismatch, thereby shortening fire extinguishing time and reducing the risk of reignition.

[0210] Table 4 Comparison of Non-Toxic / Compatibility Indicators (Compliance Rate + Fluctuation)

[0211] Note: Pure water may naturally have lower conductivity and corrosion resistance, but it is usually significantly weaker than specialized water-based agents in key performance aspects such as "multi-purpose use / smoke suppression / reignition control", as shown in Tables 3 and 5.

[0212] As shown in Table 4, Group E achieved a 100% compliance rate in key indicators such as pH, conductivity, and corrosion, and also outperformed C1 / C2 in terms of multi-purpose performance as shown in Table 3.

[0213] This invention achieves the effect of improving multi-purpose performance while maintaining stable compliance with non-toxicity and compatibility standards by incorporating non-toxicity and material compatibility into the constraint system and considering them simultaneously during the optimization process.

[0214] Table 5 Comparison of smoke suppression effects ( (and improvement rate)

[0215] As shown in Table 5, the smoking suppression index of group E is... The smoking suppression improvement rate was lower than 0.60 for C1. This invention improves safety in confined spaces or liquid fire scenarios by incorporating and optimizing optical smoke suppression indicators into the performance of Class B scenarios, thereby reducing smoke density and improving visibility.

[0216] Table 6. Comparison of Batch Consistency (Industrial Stability)

[0217] illustrate: Adopting Perform according to their respective Normalized deviation calculations facilitate cross-dimensional synthesis.

[0218] As shown in Table 6, the online deviation of group E is... The inter-batch CV of the key indicator is less than C1, indicating lower batch volatility.

[0219] This invention achieves the effect of reducing batch fluctuations and improving quality consistency by evaluating the deviation between the online detection vector and the prediction benchmark, combined with process stability control, thereby enhancing the stability and repeatability of large-scale production.

[0220] 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. A formulation and process optimization method for a non-toxic, multi-purpose water-based fire extinguishing agent, characterized in that, Executed by the processor, including: S1) Represent the extinguishing agent formulation as a formulation vector x; represent the process parameters as a parameter vector u, and construct the joint feasible region. ; S2) Based on sample dataset Train the agent model, in order to The mean of the input-output performance predictions for the trained agent model. With prediction variance ,in, These represent the j-th formula vector, process parameter vector, and performance vector, respectively. S3) Construct a set of constraint functions This includes at least: non-toxic constraints and cross-scenario multi-purpose constraints; and calculates the probability of constraint satisfaction based on the predicted variance. , in For the first The failure probability allowed by each constraint. Represents probability; S4) Construct a cross-scenario comprehensive performance function based on the performance prediction: ， in These are performance sub-functions for at least two typical fire scenarios. The risk term is determined by the predicted mean and predicted variance; These are the weighting coefficients; S5) Construct the risk constraint acquisition function: and solve ， According to the above Preparation and testing to obtain new samples To update the agent model; wherein Based on the comprehensive performance function Expected improvements; S6) When the termination condition is met, output the optimized formula and process parameters. It satisfies the constraint, satisfies the probability condition, and makes To achieve the preset optimal criteria.

2. The method according to claim 1, characterized in that, The performance vector y includes at least a non-toxicity-related index, and the non-toxicity-related index includes at least a constraint on the content of prohibited or restricted substances: , in, : Collection of prohibited and restricted substances; Restricted and prohibited substances Content in the formula; Restricted and prohibited substances The maximum allowable threshold; This indicates that all prohibited or restricted substances are within acceptable limits.

3. The method according to claim 1, characterized in that, The performance vector y includes at least a multi-purpose related index; the multi-purpose related index is used to construct cross-scenario multi-purpose constraints, and the cross-scenario multi-purpose constraints include at least minimum performance constraints for two typical fire scenarios, namely: ， ， in: This is the preset minimum performance threshold; If and only if both conditions are met and At that time, the fire extinguishing agent reaches the minimum effectiveness requirement in both types of scenarios, thereby meeting the constraint of multi-purpose use across scenarios.

4. The method according to claim 1, characterized in that, The prediction variance Obtained through at least one of heteroscedastic regression or deep ensemble.

5. The method according to claim 1, characterized in that, The risk items At least includes cross-scenario mismatch risk and / or uncertainty risk : , ,in These are the weighting coefficients; This is the preset minimum performance threshold.

6. The method according to claim 1, characterized in that, The cross-scenario comprehensive performance function Using the Conditional Value at Risk (CVaR) robustness criterion: , and maximize Substitution Maximization As the optimization objective, among which This refers to the quantile level parameter.

7. The method according to claim 1, characterized in that, This also includes the optimization of the formula and process parameters. The obtained batch of finished product online detection vector z is collected and based on... Trigger calibration, where For the surrogate model's predicted values ​​of online metrics; when Time output calibration value Update the process parameters to ensure that subsequent batches meet the probability condition of the constraint satisfaction; wherein, This is the quality control threshold.

8. A non-toxic, multi-purpose water-based fire extinguishing agent, characterized in that, The water-based fire extinguishing agent is formulated and processed according to the method of any one of claims 1–7. Prepare and satisfy the minimum performance constraints corresponding to both scenario A and scenario B. , And restrictions on prohibited substances and conductivity constraints Established at the same time.

Images