Multi-agent collaborative reasoning eutectic hydrate salt phase change material design method and system

Abstract

The application provides a method and system for designing eutectic hydrate salt phase change materials through multi-agent collaborative reasoning, and belongs to the field of artificial intelligence and material design. The method first constructs a structured eutectic hydrate salt database; analyzes agents to generate target windows according to target melting points and latent heat of phase change; retrieves agents to generate candidate salt pairs from the database according to the target windows, and compatibility agents perform compatibility evaluation on the candidate salt pairs; the matching enumeration agent executes fixed grid matching enumeration on the candidate salt pairs that pass the compatibility screening, and the prediction agent performs unimodal gating discrimination on the candidate formula; for the candidate formula that passes the unimodal gating, the predicted melting point and latent heat of phase change are output, and a comprehensive scoring function is constructed; the decision coordination agent grades and outputs the candidate formula according to the compliance threshold; otherwise, trigger the ternary adaptive expansion to introduce a third component to generate a ternary salt group, and then enter the compatibility evaluation. The application shortens the cycle of phase change material design and improves the design success rate.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and materials design, specifically relating to a design method and system for eutectic hydrated salt phase change materials based on multi-agent collaborative reasoning. Background Technology

[0002] Hydrated salt phase change materials are a class of inorganic energy storage materials capable of storing and releasing latent heat of phase change during the phase change process. They typically possess characteristics such as high volumetric heat storage density, non-flammability, good safety, relatively low cost, and wide availability of raw materials. Therefore, they have broad application prospects in low- and medium-temperature energy storage fields such as building energy-saving temperature control, solar energy and waste heat utilization, cold and heat energy storage, and thermal management. To meet the matching requirements of phase change temperature and energy density for different application scenarios, engineering often uses eutectic strategies to control the composition and ratio of hydrated salt systems, thereby achieving a downward shift or precise positioning of the phase change temperature range, while maintaining a relatively high latent heat of phase change ΔH output to a certain extent.

[0003] However, the design and optimization of eutectic hydrated salt systems is a typical high-complexity formulation problem: the candidate space is determined by the selection of salt species / hydration states, the combination of binary or ternary components, and the continuous ratio, naturally exhibiting significant combinatorial explosion; at the same time, hydrated salt systems are significantly affected by factors such as hydration structure, ion interactions, and crystallization kinetics, and even small perturbations in the ratio can affect the melting point (T). m The coupled changes in key indicators such as latent heat of phase transition (ΔH) and phase transition peak shape (single-peak / multi-peak) can induce risks such as multiphase coexistence, phase separation, and decreased cycle stability. Therefore, the traditional research and development route that relies on empirical salt selection and point-by-point ratio scanning often requires a large number of trial-and-error experiments, which is not only inefficient, time-consuming, and resource-intensive, but also makes it difficult to achieve the target T while ensuring a stable single-peak phase transition. m Reliable alignment of / ΔH and reproducible recipe window output. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a design method and system for eutectic hydrated salt phase change materials based on multi-agent collaborative reasoning. This method utilizes knowledge-guided retrieval from a large language model, structural compatibility rule constraints, unimodal phase change gating prediction, and melting point (T0) estimation. m Multi-objective regression prediction of phase change latent heat (ΔH) enables end-to-end automated screening from thermal performance targets to experimental formulation windows.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows:

[0006] In a first aspect, embodiments of the present invention provide a method for designing eutectic hydrated salt phase change materials based on multi-agent collaborative reasoning, the method comprising the following steps:

[0007] Step S1: Collect the structural and thermophysical property information of pure hydrated salts and eutectic systems to form a unified structured eutectic hydrated salt database; and set up a collaborative multi-agent system with process orchestration and state management by a central controller; the multi-agent system includes an analysis agent, a retrieval agent, a compatibility agent, a proportioning traversal agent, a prediction agent, and a decision coordination agent;

[0008] Step S2: Analyze the target melting point and latent heat of phase transition received by the intelligent agent from the user, perform standardized analysis, and generate a structured target window;

[0009] Step S3: The retrieval agent generates a set of candidate salt pairs from the structured eutectic hydrated salt database according to the target window. Each set of candidate salt pairs includes salt A and salt B. Salt A is set as the temperature-regulating component, and salt B is set as the enthalpy-regulating component.

[0010] Step S4: The compatibility agent evaluates the compatibility of candidate salt pairs or ternary salt groups; eliminates candidate salt pairs or ternary salt groups that do not meet the hard constraints, and generates compatibility weight information for the candidate salt pairs or ternary salt groups that pass the screening.

[0011] Step S5: The proportioning traversal agent performs fixed grid proportioning enumeration on the candidate salt pairs or ternary salt groups that have passed the compatibility screening, generates a set of binary or ternary candidate formulations, and implements uniqueness control and duplicate item removal.

[0012] Step S6: The predictive agent performs single-peak gating discrimination on binary or ternary candidate formulations; for candidate formulations that pass the single-peak gating, the predicted melting point Tm_pred and the predicted latent heat of phase change ΔH_pred are output; a target proximity term is constructed based on the deviation of Tm_pred from the target melting point and the deviation of ΔH_pred from the target latent heat of phase change, and a comprehensive scoring function is constructed by combining compatibility weight information, the comprehensive score is calculated and multi-target ranking is achieved, and the process proceeds to step S7; for candidate formulations that do not pass the single-peak gating, the process proceeds to step S8.

[0013] Step S7: The decision coordination agent determines whether the binary or ternary candidate formulation meets the minimum compliance threshold based on the comprehensive score; if it does, it outputs binary and / or ternary candidate formulations in stages and exports the result report; if it does not meet the threshold, it proceeds to step S8.

[0014] Step S8: Determine whether the current formulation is binary or ternary; if it is binary, trigger the ternary adaptive expansion strategy with the current binary candidate formulation as the seed to automatically introduce the third component salt C to generate a ternary salt group, and proceed to step S4; if it is ternary, output a failure diagnosis report.

[0015] As a preferred embodiment of the present invention, the structured eutectic hydrated salt database includes pure hydrated salt entries and eutectic system entries; wherein, the pure hydrated salt entries at least include chemical composition information, hydration state information and thermophysical property fields of the salt; the eutectic system entries at least include system type, component information, proportion information and corresponding phase transition thermophysical property data, as well as peak shape labels and data source identifiers.

[0016] In a preferred embodiment of the present invention, the pure hydrated salt is an inorganic salt hydrate containing water of crystallization and a portion of inorganic salts. The cation system includes, but is not limited to, alkali metal, alkaline earth metal, aluminum and transition metal ions. The anion system includes, but is not limited to, sulfate, nitrate, halides, carbonate, phosphate / pyrophosphate / silicate, hydroxide and carboxylate. The hydration number n is 1 or an integer greater than 1.

[0017] In a preferred embodiment of the present invention, when the analytical agent performs standardized parsing, it adopts a parsing strategy of prioritizing large language models and hierarchical tolerance backoff; it prioritizes extracting numerical values ​​and units from natural language input and generating structured target windows through large language models; when parsing fails or the output does not conform to the format, it automatically switches to a parsing method based on regular expressions / rules to ensure that the target parameters can be stably standardized.

[0018] In a preferred embodiment of the present invention, when generating a candidate salt set, the retrieval agent preferentially uses knowledge-guided retrieval from a large language model, and falls back to rule-based retrieval when necessary; wherein, the large language model is integrated in a task-oriented agent manner and satisfies a three-layer architecture, including at least: an interface encapsulation layer for request encapsulation and exception handling, an agent abstraction layer for defining the single responsibility and standard input and output of each agent, and a hint engineering layer for constraining model behavior and output format.

[0019] In a preferred embodiment of the present invention, hard constraint screening is used when conducting compatibility assessment; the hard constraint screening includes at least thermal compatibility and chemical compatibility.

[0020] In a preferred embodiment of the present invention, the ratio enumeration is performed by generating candidate ratio points using a discrete grid enumeration method, and satisfies the constraint that the sum of the ratios of each group is 1 and a preset tolerance ε is set.

[0021] In a preferred embodiment of the present invention, the comprehensive scoring function is a multi-objective scoring function, and the formula is as follows:

[0022] Score={|Tm_pred-Tm_target|+0.02×|ΔH_pred-ΔH_target|}×W_compat(1)

[0023] In equation (1), W_compat represents the compatibility weighting coefficient; Tm_pred represents the predicted melting point; Tm_target represents the target melting point; ΔH_pred represents the predicted latent heat of phase change; and ΔH_target represents the target latent heat of phase change.

[0024] As a preferred embodiment of the present invention, the graded output adopts a threshold system based on |ΔT| and |ΔΔH|, which includes at least three levels: "fully satisfied", "well satisfied" and "acceptably satisfied", and only candidates that meet or exceed the minimum compliance threshold are retained in the report result table.

[0025] Secondly, embodiments of the present invention also provide a multi-agent collaborative reasoning design system for eutectic hydrated salt phase change materials. The system includes: a database generation module, a central controller, an analysis agent, a retrieval agent, a compatibility agent, a proportioning traversal agent, a prediction agent, and a decision coordination agent; wherein...

[0026] The database generation module is used to collect the structural and thermophysical property information of pure hydrated salts and eutectic systems to form a unified structured eutectic hydrated salt database.

[0027] The central controller is used to orchestrate the process and manage the state of the collaborative multi-agent system, record the data of the entire process and export the result report; it is also used to trigger a ternary adaptive expansion strategy to automatically introduce a third component salt C to generate a ternary salt group for binary candidate formulations that fail to pass the single-peak gating or fail to meet the minimum compliance threshold, and forward it to the compatibility agent; if it is a ternary formulation, a failure diagnosis report is output.

[0028] The analytical agent is used to receive the target melting point and target latent heat of phase transition input by the user, perform standardized analysis, and generate a structured target window;

[0029] The retrieval agent is used to generate a set of candidate salt pairs from the structured eutectic hydrated salt database according to the target window. Each set of candidate salt pairs includes salt A and salt B, where salt A is the temperature-regulating component and salt B is the enthalpy-regulating component.

[0030] The compatibility agent is used to evaluate the compatibility of candidate salt pairs or ternary salt groups; eliminate candidate salt pairs or ternary salt groups that do not meet the hard constraints; and generate compatibility weight information for the candidate salt pairs or ternary salt groups that pass the screening.

[0031] The ratio traversal agent is used to perform fixed grid ratio enumeration on candidate salt pairs or ternary salt groups that have passed compatibility screening, generate a set of binary or ternary candidate formulations, and implement uniqueness control and duplicate item removal.

[0032] The predictive agent is used to perform single-peak gating discrimination on binary or ternary candidate formulations; for candidate formulations that pass the single-peak gating, the predicted melting point Tm_pred and the predicted latent heat of phase change ΔH_pred are output; a target proximity term is constructed based on the deviation of Tm_pred from the target melting point and the deviation of ΔH_pred from the target latent heat of phase change, and a comprehensive scoring function is constructed by combining compatibility weight information, the comprehensive score is calculated and multi-objective ranking is achieved, and the decision coordination agent is activated; for candidate formulations that fail the single-peak gating, they are forwarded to the central controller;

[0033] The decision-coordination agent is used to determine whether the binary or ternary candidate formulation meets the minimum compliance threshold based on the comprehensive score; if it does, it outputs binary and / or ternary candidate formulations in stages, exports the result report, and sends the result to the central controller; if it does not meet the threshold, it forwards the candidate formulation to the central controller.

[0034] The solutions of the embodiments of the present invention have the following beneficial effects:

[0035] The multi-agent collaborative reasoning method and system for designing eutectic hydrated salt phase change materials provided in this invention overcomes the problems in eutectic hydrated salt formulation design, such as a large candidate space, high trial-and-error costs due to ratio sensitivity, and the relationship between phase behavior (single-peak / multi-peak) and thermal properties (T). m The limitations of simultaneous alignment ( / ΔH), the ineffective utilization of unstructured knowledge such as literature experience, and the lack of traceability in the screening process mean that researchers only need to input the target T. m / ΔH, which can automatically complete candidate retrieval, compatibility filtering, ratio traversal, single-peak gating and T through natural language interaction. m / ΔH prediction decision, and output formulation windows and basis that can be directly used for experimental verification, greatly simplify the complex process from target demand to formulation recommendation, reduce blind ratio scanning and repeated experiments, and thus accurately and efficiently accelerate the targeted discovery and development process of low temperature high latent heat eutectic hydrated salt phase change materials.

[0036] Of course, implementing any product or method of the present invention does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram illustrating the design principle of the eutectic hydrated salt phase change material based on multi-agent collaborative reasoning as described in an embodiment of the present invention.

[0039] Figure 2 This is a flowchart of the design method for eutectic hydrated salt phase change materials using multi-agent collaborative reasoning, as described in an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of the area under the receiver operating characteristic curve (AUC-ROC) of the single-peak gated classification model in this embodiment of the invention.

[0041] Figure 4 This is a schematic diagram comparing the measured and predicted values ​​of the ΔH prediction regression model in an embodiment of the present invention.

[0042] Figure 5 T is an embodiment of the present invention. m A diagram showing the comparison between measured and predicted values ​​of a predictive regression model;

[0043] Figure 6 A schematic diagram of the DSC test curve of a sample prepared for applying the design results of the embodiments of the present invention. Detailed Implementation

[0044] After discovering the aforementioned problems, the inventors of this application conducted a detailed study on the design problems of existing eutectic hydrated salt phase change materials. The study found that with the development of machine learning technology, using data-driven models to predict and screen material properties provides a new approach for eutectic hydrated salt formulation design. Machine learning can establish a mapping relationship between thermophysical properties and components / ratios in a multi-dimensional feature space, thereby reducing experimental trial-and-error costs to some extent. However, in practical applications, existing methods still face several limitations: on the one hand, constructing and using machine learning models usually requires strong data processing and programming skills, objectively raising the barrier to entry for researchers without an algorithmic background; on the other hand, the model reasoning process often lacks transparency and interpretability, making it difficult for researchers to understand the basis of predictions and form executable formulation decisions accordingly; more importantly, many methods mainly focus on numerical regression prediction, making it difficult to simultaneously handle nonlinear constraints such as "phase behavior reliability (single-peak / multi-peak)" and "structural / chemical feasibility (compatibility constraints)" commonly found in eutectic systems, and also making it difficult to effectively integrate unstructured knowledge such as literature experience and mechanistic priors, thus easily leading to situations where the candidate scale is too large and unexperimentable, or the candidates are too scarce, resulting in false negatives.

[0045] In recent years, large language models have demonstrated strong capabilities in natural language processing, document information extraction, and knowledge reasoning, offering the possibility of integrating unstructured material knowledge and assisting in goal-driven screening. However, general-purpose large language models typically lack customized adaptations for knowledge in areas such as eutectic hydrated salt phase transition mechanisms, salt hydration structures, and compatibility rules. They also fall short in consistently transforming empirical rules into computable screening strategies and forming a unified decision chain with structured data-driven predictions. For these reasons, there is an urgent need for a method that can achieve efficient convergence in ultra-large binary / ternary formulation spaces while simultaneously considering structural / chemical feasibility, unimodal phase transition reliability, and the target T. m / ΔH multi-index aligned intelligent eutectic hydrate salt formulation screening and decision-making technology, to output formulation windows that can be directly used for experimental verification and traceable screening basis.

[0046] It should be noted that the defects in the above-mentioned prior art solutions are all the result of the inventors' practice and careful research. Therefore, the discovery process of the above problems and the solutions proposed by the embodiments of the present invention in the following text should be the inventors' contributions to the present invention.

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. It should be noted that, without conflict, the embodiments and features in the embodiments of the present invention can also be combined with each other.

[0048] It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In the description of the embodiments of the present invention, the terms "first," "second," "third," "fourth," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. In addition, sometimes a subscript such as W1 may be written in a non-subscript form such as W1, and their meanings are consistent unless the distinction is emphasized.

[0049] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0050] Following the above in-depth analysis, this invention provides a multi-agent collaborative reasoning method and system for designing eutectic hydrated salt phase change materials. The method resolves the target melting point (Tm_target) and target latent heat of phase change (ΔH_target) given by the user in natural language into structured targets. Candidate salts are retrieved and assigned roles using a structured database of pure hydrated salts and eutectic structures. Candidate salt pairs are generated and their proportions are enumerated under thermodynamic and chemical compatibility rules. A single-peak gating classification model is used to determine the reliability of the phase change peak shape, and a regression prediction model is used to jointly predict Tm and ΔH. Furthermore, a multi-target scoring and hierarchical decision-making system is constructed based on target proximity and compatibility weighting, outputting candidate salts that can be directly used in experiments, their corresponding proportioning windows, and traceable screening metadata. In this invention, eutectic hydrated salt products with low-temperature, high latent heat target-guided formulations are prepared based on the output results. The prepared samples are verified by differential scanning calorimetry (DSC), and the prepared products generally achieve the corresponding latent heat values. This invention can increase the density of compliant candidate hits while compressing the combination space, and reduce the cost of trial and error.

[0051] like Figure 1 and Figure 2 As shown, the multi-agent collaborative reasoning method for designing eutectic hydrated salt phase change materials includes the following steps:

[0052] Step S1 involves collecting structural and thermophysical property information of pure hydrated salts and eutectic systems to form a unified structured eutectic hydrated salt database, which supports candidate retrieval, compatibility rule solidification, feature engineering construction, and machine learning model training and evaluation. Simultaneously, a collaborative multi-agent system is set up, with a central controller managing the process orchestration and state management. This multi-agent system includes an analysis agent, a retrieval agent, a compatibility agent, a proportioning traversal agent, a prediction agent, and a decision coordination agent.

[0053] In this step, the central controller is used for process orchestration and state management of all intelligent agents, and adopts a sequential search strategy of "binary priority and ternary compensation". When the prediction and decision-making stage returns diagnostic information of failure to meet the target or single peak failure, the central controller triggers strategy adjustment and automatically expands the tolerance window or starts the ternary compensation process.

[0054] The structured eutectic hydrated salt database includes entries for pure hydrated salts and eutectic systems. Each pure hydrated salt entry contains at least information on the salt's chemical composition, hydration state, and thermophysical properties. Each eutectic system entry contains at least the system type (binary or ternary), component information, proportioning information, and corresponding phase transition thermophysical properties (Tm, crystallization temperature, ΔH, supercooling), as well as peak shape labels (single-peak / multi-peak) and data source identifiers (Document DOI and experimental batch number). This forms a unified structured record for compatibility rule solidification, feature engineering construction, and predictive model training and evaluation. For example, the database may contain 65 pure hydrated salts and 200 eutectic system formulation records. These 200 records are counted according to "component-proportion-source," and the same salt pair or salt ternary may correspond to different proportions, forming multiple records to characterize the eutectic system's sensitivity to proportion disturbances and its thermophysical property distribution. Furthermore, the pure hydrated salt is an inorganic salt hydrate containing water of crystallization and some inorganic salts. The cation system includes, but is not limited to, alkali metal, alkaline earth metal, aluminum and transition metal ions, etc., and the anion system includes, but is not limited to, sulfate, nitrate, halides (chlorine / bromine / iodine), carbonate, phosphate / pyrophosphate / silicate, hydroxide and carboxyl ions, etc., and the hydration number n is 1 or an integer greater than 1; for example, but not limited to: Na2SO4·10H2O, Na2CO3·10H2O, Na2HPO4 4·12H2O, Na2S2O3·5H2O, Na2SiO3·9H2O, NaOH·H2O, CaCl2·6H2O, Ca(NO3)2·4H2O, CaBr2·6H2O, MgCl2· 6H2O, MgSO4·7H2O, LiNO3·3H2O, CH3COONa·3H2O, Al(NO3)3·9H2O, Al2(SO2)3·18H2O, Zn(NO3)2·6H2O, etc.

[0055] In step S2, the user determines the target melting point and target latent heat of phase transition based on the design objectives of the eutectic hydrated salt phase change material; the analysis agent receives the target melting point and target latent heat of phase transition input by the user, performs standardized analysis, generates a structured target window, and initializes a graded tolerance backoff mechanism to ensure the accessibility and robustness of the screening process.

[0056] In this step, when analyzing the agent's standardization parsing, a parsing strategy of prioritizing large language models and hierarchical tolerance fallback is adopted. Firstly, the large language model is used to extract numerical values ​​and units from the natural language input and generate a structured target window. When parsing fails or the output does not conform to the format, it automatically switches to a regular expression / rule-based parsing method to ensure that the target parameters can be stably standardized.

[0057] The graded tolerance backoff mechanism includes a three-level backoff strategy. In the initial level, the screening window for salt A is limited to a candidate range above the target temperature zone, and the latent heat threshold for salt B is set to ΔH_min. When no candidate formulation meeting the compliance threshold is obtained at this level, the screening constraints are relaxed step by step according to preset rules: In the first level, the salt A window is kept at [Tm_max+1, Tm_max+50] and the latent heat threshold for salt B is lowered to 0.9×ΔH_min; In the second level, while maintaining the latent heat threshold for salt B at 0.9×ΔH_min, the lower limit of the salt A window is relaxed to [Tm_max−5, Tm_max+50]; In the third level, while maintaining the salt A window at [Tm_max-5, Tm_max+50], the latent heat threshold for salt B is further relaxed to 0.8×ΔH_min; until a candidate meeting the compliance threshold is obtained or the preset maximum backoff level is reached. During this process, Tm_max is the upper limit of the target temperature range, and ΔH_min is the lower limit of the target latent heat. The above window boundaries and coefficients can be set as configurable parameters.

[0058] Step S3: The retrieval agent retrieves and generates a candidate salt set from the structured eutectic hydrated salt database according to the target window, and completes the role assignment: salt A is set as the temperature regulating component, salt B is set as the enthalpy regulating component, and salt A and salt B constitute a candidate salt pair.

[0059] In this step, when generating the candidate salt set, the retrieval agent prioritizes using the knowledge-guided retrieval of the large language model, and falls back to rule-based retrieval when necessary, to obtain a set of candidate salts that meet the target orientation. The large language model is integrated in a task-oriented agent manner and meets a three-layer architecture, including at least: an interface encapsulation layer for request encapsulation and exception handling, an agent abstraction layer for defining the single responsibility and standard input and output of each agent, and a hint engineering layer for constraining model behavior and output format, thereby ensuring the stability and reusability of the multi-round inference process.

[0060] When the retrieval agent performs candidate salt retrieval, it adopts a dual-path approach of knowledge-guided retrieval and rule-based backoff. For salt A, temperature-adjusting candidates are selected based on the empirical law of eutectic cooling; for salt B, enthalpy-adjusting candidates are selected based on the contribution tendency of ΔH and considering the uncertainty law of eutectic ΔH enhancement / weakening; when the large language model is unavailable or fails, rule-based retrieval is used to complete candidate screening based on the target window and threshold, and the same hierarchical backoff strategy is followed.

[0061] Step S4: The compatibility agent performs compatibility evaluation on candidate salt pairs or ternary salt groups; eliminates candidate salt pairs or ternary salt groups that do not meet the hard constraints, and generates compatibility weight information for the selected candidate salt pairs or ternary salt groups for subsequent comprehensive ranking.

[0062] In this step, a hard constraint screening method is used for compatibility assessment. The hard constraint screening includes at least thermal compatibility and chemical compatibility. When the phase transition temperature difference of a candidate salt pair exceeds a preset threshold, it is determined to be thermally incompatible and eliminated; when a candidate salt pair has the risk of chemical instability reactions such as precipitation, hydrolysis, or gas release, it is determined to be chemically incompatible and eliminated; only candidates that simultaneously meet both thermal and chemical compatibility requirements are retained for subsequent proportioning and prediction stages.

[0063] Step S5: Proportional grid traversal and uniqueness control. The proportional traversal agent performs fixed grid proportional enumeration on the candidate salt pairs or ternary salt groups that have passed the compatibility screening, generates a set of binary or ternary candidate formulations, and implements uniqueness control and duplicate removal.

[0064] In this step, the ratio enumeration uses a discrete grid enumeration method to generate candidate ratio points, and satisfies the constraint that the sum of the ratios of each group is 1 and sets a preset tolerance ε. At the same time, uniqueness constraints and duplicate removal are performed on the salt combination and ratio points. The ratio is preferably represented by mass fraction.

[0065] The discrete grid enumeration includes binary and ternary systems: For the binary system, the mass fraction of salt A is discretely evaluated within a preset ratio range at a preset step size to generate multiple candidate points, and the mass fraction of salt B is determined by the remaining share; for the ternary system, preferably, the mass fraction of salt C is first discretely evaluated within a preset range at a preset step size, and then the mass fraction of salt A is discretely evaluated within the remaining share at a preset step size, and the mass fraction of salt B is determined by the remaining share, thereby forming multiple ternary candidate ratio points. The salt species identities satisfy the distinct constraint (salt A, salt B, and salt C are all different), and combinations of the same salt species regardless of order are normalized and deduplicated; the tolerance ε is a preset positive number used to determine whether the sum of the ratios satisfies the constraint and to control numerical errors. Further, the ratio range, step size, and number of candidate points are configurable parameters that can be adjusted according to the target window, candidate size, or computational budget. Preferably, the feature vector is in a unified input format. The binary system zero-fills the ternary slots to maintain consistent input dimensions; the feature vector is composed of component physicochemical descriptors and proportioning information to achieve a unified binary / ternary prediction interface.

[0066] In one executable embodiment, during normalization, to improve model interpretability and enhance the characterization of physicochemical trends, normalized proxy features are constructed based on the original descriptor, including but not limited to: hydration density = hydration number / molecular weight; hydrogen bond density = (number of hydrogen bond donors + number of hydrogen bond acceptors) / molecular weight; unit atomic polarity = topological polar surface area / number of heavy atoms. The above proxy features are used to characterize the effects of hydration enrichment, hydrogen bonding ability, and polarity distribution on crystal packing, phase stability, and phase transition thermal properties.

[0067] Step S6: The predictive agent performs single-peak gating discrimination on binary or ternary candidate formulations, outputs the single-peak probability, and marks the availability of candidates accordingly, providing gating results for subsequent prediction and ranking. For candidate formulations that pass the single-peak gating, the predicted melting point Tm_pred and the predicted latent heat of phase change ΔH_pred are output. A target proximity term is constructed based on the deviation of Tm_pred from the target melting point and the deviation of ΔH_pred from the target latent heat of phase change. A comprehensive scoring function is constructed by combining compatibility weight information, the comprehensive score is calculated, and multi-target ranking is achieved. Proceed to step S7. For candidate formulations that do not pass the single-peak gating, proceed to step S8.

[0068] In this step, the single-peak gating outputs a single-peak probability. When the single-peak probability is below a threshold, the candidate is marked as not participating in performance ranking, but its gating and prediction information is retained for diagnostic analysis and subsequent ternary compensation triggering. This candidate is placed at the end of the comprehensive ranking or does not enter the report result table, but only enters the diagnostic table. Preferably, the single-peak gating uses a random forest classifier with a threshold of 0.5.

[0069] The predictive agent trains, evaluates, and selects a single-peak gated classification model and a thermal performance regression model to obtain an optimal model combination for online inference, thereby improving the stability and prediction accuracy of end-to-end screening.

[0070] A dataset for model training and evaluation is constructed. This dataset consists of eutectic system entries and their corresponding labels / values, where: the sample labels for the unimodal gating classification task characterize the peak shape type (unimodal / multimodal) of the candidate formulation; and the sample labels for the thermal performance regression task are the measured values ​​of Tm and ΔH. The data is cleaned and standardized, including but not limited to unit unification, outlier handling, missing value handling, and duplicate sample handling. A unified input feature vector is constructed based on the feature engineering module to simultaneously support the prediction interfaces for binary and ternary systems. To maintain input dimension consistency, the binary system can have zero-filling or equivalent placeholder processing in the third component slot, allowing both binary and ternary candidates to reuse the same prediction interface and inference process.

[0071] A multi-model candidate set is established, and training is performed separately for classification and regression tasks. Optionally, the classification model candidate set includes linear models, support vector machines, decision trees, random forests, gradient boosting trees, and XGBoost; the regression model candidate set includes linear regression, support vector regression, decision tree regression, random forest regression, gradient boosting regression, and XGBoost regression. Each candidate model is trained using a consistent data partitioning method, a unified feature input format, and the same training process, and hyperparameter optimization is performed to form a comparable candidate model group.

[0072] Candidate models are evaluated against benchmarks to determine the optimal model. For single-peak gating classification tasks, a comprehensive evaluation can be conducted using metrics such as accuracy, precision, recall, F1 score, and area under the receiver operating characteristic curve (AUC-ROC) to balance the risks of false admission and false rejection during the gating process. For Tm and ΔH regression tasks, metrics such as root mean square error (RMSE) and coefficient of determination (R²) can be used to characterize the prediction error level and the fit's explanatory power. The system can select classification and regression models independently, meaning that the single-peak gating model and the thermal performance regression model do not need to come from the same algorithm family, thus achieving better overall performance in the two sub-tasks of peak shape discrimination and thermal performance prediction.

[0073] The comprehensive scoring function is a multi-objective scoring function, and its formula is as follows:

[0074] Score={|Tm_pred-Tm_target|+0.02×|ΔH_pred-ΔH_target|}×W_compat(1)

[0075] In equation (1), W_compat represents the compatibility weighting coefficient; Tm_pred represents the predicted melting point; Tm_target represents the target melting point; ΔH_pred represents the predicted latent heat of phase change; and ΔH_target represents the target latent heat of phase change.

[0076] By introducing W_compat, a comprehensive scoring result is obtained, thereby achieving a unified coupled ranking of structural robustness and target proximity.

[0077] The compatibility weighting coefficient is determined by calculating the compatibility weight information obtained from the compatibility agent. In one executable embodiment, the compatibility weighting is based on structural preference counting. For each candidate, the number of preferences satisfied, N_match, is calculated, and W_compat = 1.0 - 0.2 × N_match is defined; where preferences include at least: cation sharing or correlation, anion charge consistency, and single-atom / multi-atom anion type matching. Preferably, W_compat is set with a lower bound W_min (W_min > 0), and W_min is used when the calculated value is less than W_min to ensure the numerical stability of the scoring function. The compatibility of the ternary system A–B, A–C, and B–C is evaluated respectively, and the lowest compatibility among the three is used as the compatibility level of the ternary system to ensure the overall structural consistency of the system.

[0078] Step S7: The decision coordination agent determines whether the binary or ternary candidate formulation meets the minimum compliance threshold based on the comprehensive score; if it does, it outputs binary and / or ternary candidate formulations in a tiered manner and exports the result report; if it does not meet the threshold, it proceeds to step S8.

[0079] In this step, the graded output adopts a threshold system based on |ΔT| and |ΔΔH|. The threshold system includes multiple compliance thresholds, with at least three levels: "fully satisfied," "well satisfied," and "acceptably satisfied." Only candidate formulations that meet or exceed the lowest compliance threshold are included in the report results table. Where ΔT = Tm_pred − Tm_target, ΔΔH = ΔH_pred − ΔH_target. Preferably, fully satisfied means: |ΔT| ≤ 2 ℃ and |ΔΔH| ≤ 10 J·g. - ¹; Goodly satisfies: |ΔT|≤4 ℃ and |ΔΔH|≤20 J·g - ¹; Acceptable conditions are: |ΔT|≤6 ℃ and |ΔΔH|≤30 J·g - ¹.

[0080] The results report should include at least the salt type identifier, ratio, unimodal probability, Tm_pred, ΔH_pred, score, compliance rating, and metadata records for reproducible experiments and traceability procedures. It may also include audit fields to ensure reproducibility and traceability.

[0081] In step S8, the central controller determines whether the current formulation is binary or ternary. If it is binary, the current binary candidate formulation (corresponding to the salt A-salt B combination and its ratio point) is used as a seed to trigger the ternary adaptive expansion strategy to automatically introduce the third component salt C to generate a ternary salt group, and then proceed to step S4. If it is ternary, a failure diagnosis report is output.

[0082] In this step, the ternary adaptive expansion strategy is binary priority followed by ternary compensation. When a binary combination fails to achieve a single peak, salt C that shares at least one cation or anion with salt A or salt B is selected, with priority given to salt C with a low Tm value to promote single peak formation. When a binary combination fails to meet the standard due to excessively high Tm_pred, salt C with shared ions and low Tm value is selected to lower the eutectic temperature. When a binary combination fails to meet the standard due to insufficient ΔH_pred value, salt C with shared ions and high ΔH value is selected to enhance ΔH. The compatibility screening, sizing traversal, single peak gating, and thermal performance prediction steps are repeatedly performed on the generated ternary candidates.

[0083] Based on the same idea, this invention also provides a multi-agent collaborative reasoning system for designing eutectic hydrated salt phase change materials.

[0084] The system includes: a database generation module, a central controller, an analysis agent, a retrieval agent, a compatibility agent, a matching and traversal agent, a prediction agent, and a decision coordination agent; wherein...

[0085] The database generation module is used to collect the structural and thermophysical property information of pure hydrated salts and eutectic systems to form a unified structured eutectic hydrated salt database.

[0086] The central controller is used to orchestrate the process and manage the state of the collaborative multi-agent system, record the data of the entire process and export the result report; it is also used to automatically introduce a third component salt C to generate a ternary salt group for binary candidate recipes that fail to pass the single-peak gating or fail to meet the minimum compliance threshold, using the current binary candidate recipe as a seed, and forwarding it to the compatibility agent; if it is a ternary, it outputs a failure diagnosis report.

[0087] The analytical agent is used to receive the target melting point and target latent heat of phase transition input by the user, perform standardized analysis, and generate a structured target window;

[0088] The retrieval agent is used to generate a set of candidate salt pairs from the structured eutectic hydrated salt database according to the target window. Each set of candidate salt pairs includes salt A and salt B, where salt A is the temperature-regulating component and salt B is the enthalpy-regulating component.

[0089] The compatibility agent is used to evaluate the compatibility of candidate salt pairs or ternary salt groups; eliminate candidate salt pairs or ternary salt groups that do not meet the hard constraints; and generate compatibility weight information for the candidate salt pairs or ternary salt groups that pass the screening.

[0090] The ratio traversal agent is used to perform fixed grid ratio enumeration on candidate salt pairs or ternary salt groups that have passed compatibility screening, generate a set of binary or ternary candidate formulations, and implement uniqueness control and duplicate item removal.

[0091] The predictive agent is used to perform single-peak gating discrimination on binary or ternary candidate formulations; for candidate formulations that pass the single-peak gating, the predicted melting point Tm_pred and the predicted latent heat of phase change ΔH_pred are output; a target proximity term is constructed based on the deviation of Tm_pred from the target melting point and the deviation of ΔH_pred from the target latent heat of phase change, and a comprehensive scoring function is constructed by combining compatibility weight information, the comprehensive score is calculated and multi-objective ranking is achieved, and the decision coordination agent is activated; for candidate formulations that fail the single-peak gating, they are forwarded to the central controller;

[0092] The decision-coordination agent is used to determine whether the binary or ternary candidate formulation meets the minimum compliance threshold based on the comprehensive score; if it does, it outputs binary and / or ternary candidate formulations in stages, exports the result report, and sends the result to the central controller; if it does not meet the threshold, it forwards the candidate formulation to the central controller.

[0093] The system or device for executing the method in this embodiment of the invention can be a terminal or a server. The system includes a processor, a memory, and / or a transceiver, etc., and is connected via a communication bus. Each module can be implemented by a processor, a memory, and / or a transceiver, etc. The processor can be, but is not limited to, one or more microprocessors (MPUs), central processing units (CPUs), network processors (NPs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), and other programmable logic devices, discrete gates, transistor logic devices, discrete hardware components, etc., or can be configured to implement one or more integrated circuits of this invention. The processor can perform various functions by running or executing software programs in the memory and calling data in the memory. The memory includes Random Access Memory (RAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read-Only Memory (CD-ROM), and / or Non-Volatile Memory (NVM), etc. The transceiver is used to communicate with network devices or terminal devices, and includes a receiver and a transmitter. The memory and transceiver can be integrated with the processor or exist independently.

[0094] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0095] It should also be noted that the multi-agent collaborative reasoning eutectic hydrated salt phase change material design system described in this embodiment corresponds to the multi-agent collaborative reasoning eutectic hydrated salt phase change material design method. The description and limitations of the method also apply to the system, and will not be repeated here.

[0096] The multi-agent collaborative reasoning method and system for designing eutectic hydrated salt phase change materials provided in this embodiment are applied to the development of eutectic hydrated salts. Users set the target melting point Tm_target = 14 °C and the target latent heat of phase change ΔH_target = 180 J·g based on the target phase change material. - ¹. The analysis agent screens salt A based on the phase transition temperature window Tm∈[Tm_target+1,Tm_target+50] ℃ to utilize the eutectic cooling effect, and salt B based on the latent heat threshold ΔH≥ΔH_target to ensure heat storage capacity. This yields 45 candidate salts for A and 23 candidate salts for B, which are then cross-combined to generate 1,035 candidate salt pairs. After screening and deduplication, 910 different salt pairs are obtained, of which 538 are feasible due to compatibility hard constraints. Nine binary mass fraction ratio points are enumerated for each salt pair, resulting in 4,842 binary candidate formulations. The rule-based fallback mode outputs 1,284 candidate formulations that meet the minimum threshold, of which 16 are "fully satisfied," 314 are "well satisfied," and 954 are "acceptably satisfied." This demonstrates that the system can still complete a full closed-loop screening from target input to candidate output without relying on a large language model.

[0097] For the single-peak gated classification task, several classification models were compared and evaluated. The results show that the Random Forest classification model performs more balancedly in terms of overall performance, achieving a high F1 score (0.7501) and high accuracy (78.31%), while maintaining a high AUC-ROC (0.8109). The XGBoost classification model performs similarly (F1 score 0.7374, accuracy 76.93%, AUC-ROC 0.8036). The Gradient Boosting Tree classification model achieves a high AUC-ROC (0.8422), but its F1 score and accuracy are relatively low (F1 score 0.6341, accuracy 70.53%). Linear models and Support Vector Machines are relatively weak in recall and generalization, while the Decision Tree model performs moderately but has poor consistency. The AUC-ROC of the optimal model in the single-peak gated classification task is illustrated below. Figure 3 As shown.

[0098] For the thermal performance regression task, various regression models were compared and evaluated. The results show that in ΔH regression, the XGBoost regression model achieves a better error level and goodness of fit (RMSE of 26.6987 J·g). - ¹, R² = 0.7475), followed by the random forest regression model (RMSE = 27.0070 J·g). - ¹, R² is 0.7416); in Tm regression, the XGBoost regression model also performs better (RMSE 2.8556, R² 0.9636), followed by the Random Forest regression model (RMSE 3.1702, R² 0.9552). The comparison between the measured and predicted values ​​of the preferred model in the ΔH regression task is illustrated below. Figure 4 As shown; in the Tm regression task, the comparison between the measured and predicted values ​​of the optimal model is illustrated as follows. Figure 5 As shown.

[0099] A set of ternary candidate formulations was selected from the system's output of compliant candidates for experimental verification. The salts were Na₂SO₄·10H₂O, Na₂CO₃·10H₂O, and Na₂HPO₄·12H₂O, with a mass fraction ratio of 0.292:0.358:0.350. Samples were prepared according to the formulations and subjected to DSC testing. The DSC test curves are shown below. Figure 6 As shown, a single endothermic peak was obtained, with test results of Tm = 15.74 ℃ and ΔH = 188 J·g. - ¹, thus demonstrating that the candidate formulations output by the system of the present invention under low temperature and high latent heat target conditions are experimentally verifiable and feasible.

[0100] Therefore, the multi-agent collaborative reasoning method and system for eutectic hydrated salt phase change material design provided in this invention constructs a multi-agent collaborative reverse design process oriented towards the target phase change temperature and target latent heat of phase change; proposes an adaptive search mechanism of "binary priority, ternary compensation"; establishes a chain-like screening process coupling compatibility hard constraints, single-peak gating, and thermal performance regression prediction; and achieves automated closed-loop output of candidate formulations by driving the introduction of salt C through failure mode and joint scoring of target proximity and compatibility; the beneficial effects achieved include:

[0101] Goal-driven efficient convergence: Based on the user-given Tm / ΔH target, it realizes automated candidate retrieval, rule-based hard screening, ratio traversal and predictive ranking in a super-large binary / ternary formulation space, significantly reducing the number of blind ratio scanning and trial-and-error experiments;

[0102] Unified Coupling of Multiple Constraints: Structural / chemical feasibility (compatibility rules), phase behavior reliability (single-peak gating), and thermal performance proximity (Tm / ΔH multi-objective scoring) are uniformly constrained in the same chain process, reducing the risk of multi-peak / multi-phase and phase separation, and improving the experimental feasibility of candidates;

[0103] Adaptive strategy of binary priority and ternary compensation: By adopting the "binary priority and ternary compensation" strategy and the adaptive introduction of salt C based on failure mode, automatic compensation is achieved for unsolvable or low-hit targets, thereby improving the robustness of the system and the reachability of the target.

[0104] Low barrier to entry and reusability: Users can trigger end-to-end filtering by inputting target parameters in natural language, reducing the barrier to programming and data processing; since each intelligent agent adopts standardized input and output and a unified data base, it can be migrated to other eutectic material systems by replacing the salt pool and the target.

[0105] Traceability and auditability: While outputting the candidate salt type-ratio window, it also provides single-peak probability, predicted value, score and metadata records to ensure that the results are reproducible, the process is traceable and the basis is interpretable.

[0106] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0107] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed, and is not intended to limit the scope of the claimed invention, but merely to illustrate preferred embodiments of the invention. Those skilled in the art should understand that the scope of the invention is not limited to the specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A design method for eutectic hydrated salt phase change materials based on multi-agent collaborative reasoning, characterized in that, The method includes the following steps: Step S1: Collect the structural and thermophysical property information of pure hydrated salts and eutectic systems to form a unified structured eutectic hydrated salt database; and set up a collaborative multi-agent system with process orchestration and state management by a central controller; the multi-agent system includes an analysis agent, a retrieval agent, a compatibility agent, a proportioning traversal agent, a prediction agent, and a decision coordination agent; Step S2: Analyze the target melting point and latent heat of phase transition received by the intelligent agent from the user, perform standardized analysis, and generate a structured target window; Step S3: The retrieval agent generates a set of candidate salt pairs from the structured eutectic hydrated salt database according to the target window. Each set of candidate salt pairs includes salt A and salt B. Salt A is set as the temperature-regulating component, and salt B is set as the enthalpy-regulating component. Step S4: The compatibility agent evaluates the compatibility of candidate salt pairs or ternary salt groups; eliminates candidate salt pairs or ternary salt groups that do not meet the hard constraints, and generates compatibility weight information for the candidate salt pairs or ternary salt groups that pass the screening. Step S5: The proportioning traversal agent performs fixed grid proportioning enumeration on the candidate salt pairs or ternary salt groups that have passed the compatibility screening, generates a set of binary or ternary candidate formulations, and implements uniqueness control and duplicate item removal. Step S6: The predictive agent performs single-peak gating discrimination on binary or ternary candidate formulations; for candidate formulations that pass the single-peak gating, the predicted melting point Tm_pred and the predicted latent heat of phase change ΔH_pred are output; a target proximity term is constructed based on the deviation of Tm_pred from the target melting point and the deviation of ΔH_pred from the target latent heat of phase change, and a comprehensive scoring function is constructed by combining compatibility weight information, the comprehensive score is calculated and multi-target ranking is achieved, and the process proceeds to step S7; for candidate formulations that do not pass the single-peak gating, the process proceeds to step S8. Step S7: The decision-coordination agent determines whether the binary or ternary candidate formulation meets the minimum compliance threshold based on the comprehensive score. If the conditions are met, binary and / or ternary candidate formulations will be output in stages, and a results report will be exported. If not satisfied, proceed to step S8; Step S8: Determine whether the current formula is binary or ternary; If it is a binary salt, the current binary candidate formulation is used as the seed to trigger the ternary adaptive expansion strategy to automatically introduce the third salt component C to generate a ternary salt group, and then proceed to step S4. If it is a ternary variable, then a failure diagnosis report will be output.

2. The method according to claim 1, characterized in that, The structured eutectic hydrated salt database includes pure hydrated salt entries and eutectic system entries; wherein, the pure hydrated salt entries include at least the chemical composition information, hydration state information, and thermophysical property fields of the salt; the eutectic system entries include at least the system type, component information, proportion information, and corresponding phase transition thermophysical property data, peak shape labels, and data source identifiers.

3. The method according to claim 2, characterized in that, The pure hydrated salt is an inorganic salt hydrate containing water of crystallization and some inorganic salts. The cation system includes, but is not limited to, alkali metal, alkaline earth metal, aluminum and transition metal ions. The anion system includes, but is not limited to, sulfate, nitrate, halides, carbonate, phosphate / pyrophosphate / silicate, hydroxide and carboxylate. The hydration number n is 1 or an integer greater than 1.

4. The method according to claim 1, characterized in that, When the analytical agent performs standardized parsing, it adopts a parsing strategy that prioritizes large language models and uses hierarchical tolerance and backoff. It prioritizes extracting numerical values ​​and units from natural language input and generating structured target windows through large language models. When parsing fails or the output does not conform to the format, it automatically switches to a parsing method based on regular expressions / rules to ensure that the target parameters can be stably standardized.

5. The method according to claim 1, characterized in that, When generating the candidate salt set, the retrieval agent prioritizes using knowledge-guided retrieval from the large language model, and falls back to rule-based retrieval when necessary. The large language model is integrated in a task-oriented agent manner and satisfies a three-layer architecture, including at least: an interface encapsulation layer for request encapsulation and exception handling, an agent abstraction layer for defining the single responsibility and standard input and output of each agent, and a hint engineering layer for constraining model behavior and output format.

6. The method according to claim 1, characterized in that, When conducting compatibility assessments, hard constraint screening is employed; the hard constraint screening includes at least thermal compatibility and chemical compatibility.

7. The method according to claim 1, characterized in that, The ratio enumeration uses a discrete grid enumeration method to generate candidate ratio points, and satisfies the constraint that the sum of the ratios of each group is 1 and sets a preset tolerance ε.

8. The method according to claim 1, characterized in that, The comprehensive scoring function is a multi-objective scoring function, and its formula is as follows: Score={|Tm_pred-Tm_target|+0.02×|ΔH_pred-ΔH_target|}×W_compat(1) In equation (1), W_compat represents the compatibility weighting coefficient; Tm_pred represents the predicted melting point; Tm_target represents the target melting point; ΔH_pred represents the predicted latent heat of phase change; and ΔH_target represents the target latent heat of phase change.

9. The method according to claim 1, characterized in that, The tiered output adopts a threshold system based on |ΔT| and |ΔΔH|, which includes at least three levels: "fully satisfied", "well satisfied", and "acceptably satisfied", and only candidates that meet or exceed the minimum compliance threshold are retained in the report results table.

10. A multi-agent collaborative reasoning design system for eutectic hydrated salt phase change materials, characterized in that, The system includes: a database generation module, a central controller, an analysis agent, a retrieval agent, a compatibility agent, a matching and traversal agent, a prediction agent, and a decision coordination agent; wherein... The database generation module is used to collect the structural and thermophysical property information of pure hydrated salts and eutectic systems to form a unified structured eutectic hydrated salt database. The central controller is used to orchestrate the process and manage the state of the collaborative multi-agent system, record the data of the entire process and export the result report; it is also used to trigger a ternary adaptive expansion strategy to automatically introduce a third component salt C to generate a ternary salt group for binary candidate formulations that fail to pass the single-peak gating or fail to meet the minimum compliance threshold, and forward it to the compatibility agent; if it is a ternary formulation, a failure diagnosis report is output. The analytical agent is used to receive the target melting point and target latent heat of phase transition input by the user, perform standardized analysis, and generate a structured target window; The retrieval agent is used to generate a set of candidate salt pairs from the structured eutectic hydrated salt database according to the target window. Each set of candidate salt pairs includes salt A and salt B, where salt A is the temperature-regulating component and salt B is the enthalpy-regulating component. The compatibility agent is used to evaluate the compatibility of candidate salt pairs or ternary salt groups; eliminate candidate salt pairs or ternary salt groups that do not meet the hard constraints; and generate compatibility weight information for the candidate salt pairs or ternary salt groups that pass the screening. The ratio traversal agent is used to perform fixed grid ratio enumeration on candidate salt pairs or ternary salt groups that have passed compatibility screening, generate a set of binary or ternary candidate formulations, and implement uniqueness control and duplicate item removal. The predictive agent is used to perform single-peak gating discrimination on binary or ternary candidate formulations; for candidate formulations that pass the single-peak gating, the predicted melting point Tm_pred and the predicted latent heat of phase change ΔH_pred are output; a target proximity term is constructed based on the deviation of Tm_pred from the target melting point and the deviation of ΔH_pred from the target latent heat of phase change, and a comprehensive scoring function is constructed by combining compatibility weight information, the comprehensive score is calculated and multi-objective ranking is achieved, and the decision coordination agent is activated; for candidate formulations that fail the single-peak gating, they are forwarded to the central controller; The decision-coordination agent is used to determine whether the binary or ternary candidate formulation meets the minimum compliance threshold based on the comprehensive score; if it does, it outputs binary and / or ternary candidate formulations in stages, exports the result report, and sends the result to the central controller; if it does not meet the threshold, it forwards the candidate formulation to the central controller.

Images