Nfpp-based composite material and method for directional assembly of nfpp functional units

Abstract

The application provides an NFPP-based composite material and a directional assembly method of an NFPP functional unit, and relates to the technical field of sodium ion batteries. The assembly method comprises the following steps: defining a design variable space and a multi-objective function set; obtaining initial sample points; constructing an initial training data set; constructing a model and defining a multi-objective acquisition function; preparing an NFPP composite material and testing the compaction density and the rate performance index; obtaining a target combination scheme and mixing to obtain a target NFPP composite material. Through the closed-loop iteration of the proxy model and the acquisition function, the optimal proportion is quickly locked in a huge parameter space through a small amount of physical verification, and the research and development cost and period are significantly reduced. The output of the Pareto frontier non-dominated solution resolves the physical conflict between the compaction density and the rate performance, and ensures the engineering reliability and comprehensive performance of the composite material.

Description

Technical Field

[0001] This application relates to the field of sodium-ion battery technology, and more specifically, to an NFPP-based composite material and a method for directional assembly of NFPP functional units. Background Technology

[0002] Sodium-ion batteries have attracted widespread attention in energy storage and related fields due to their inherent working mechanism. Among them, sodium iron pyrophosphate (NFPP) is considered one of the cathode materials with high application potential in sodium-ion batteries due to its relatively stable three-dimensional framework structure and suitable operating voltage. The physical morphology and electrochemical properties of the cathode material itself directly determine the overall performance of the entire battery system; therefore, continuous optimization of this type of material is a core focus in this field.

[0003] In existing materials preparation and research processes, to overcome the inherent limitations of single-structure materials, it is common practice to comprehensively control the physical and chemical parameters of the materials. Common techniques include mixing and proportioning material particles with different characteristics, or introducing specific elements, in order to balance different dimensions of indicators such as electrode compaction density and electrochemical performance. The parameter design space of materials typically encompasses multiple dimensions, including composition space, particle size space, and doping space.

[0004] However, the aforementioned multi-dimensional parameter control approach faces significant bottlenecks in practical implementation. On the one hand, multiple variables such as ratio, particle size, and doping are intertwined, resulting in a vast design variable space and numerous potential parameter combinations. On the other hand, core target indicators such as electrode compaction density, rate performance, and cycle life often exhibit contradictory relationships, and pursuing improvements in a single performance metric can easily lead to the degradation of other performance indicators. Faced with this multi-objective performance conflict and high-dimensional design space, traditional parameter verification schemes (such as orthogonal experiments and grid search) have extremely low screening efficiency.

[0005] In summary, existing methods for preparing and optimizing sodium iron pyrophosphate cathode materials mainly face the challenges of an excessively large multi-parameter space and conflicting performance objectives. Traditional experimental verification methods based on conventional trial and error often require hundreds or even thousands of experiments, making it difficult to efficiently and systematically search for the optimal combination of various performance objectives within a reasonable timeframe. This results in inefficiency and high resource consumption.

[0006] In view of the above, this application is hereby submitted. Summary of the Invention

[0007] The purpose of this application is to provide an NFPP-based composite material and an oriented assembly method for NFPP functional units. The assembly method uses a Gaussian process proxy model and a data-driven closed-loop iteration with a multi-objective acquisition function to resolve the inherent conflict between compaction density and rate performance in a large design space with extremely low physical verification cost, and efficiently obtains a material combination ratio with good comprehensive electrochemical performance.

[0008] In order to achieve the above-mentioned objectives of this application, the following technical solution is adopted: In a first aspect, this application provides a method for oriented assembly of NFPP functional units, comprising: Define the design variable space and multi-objective function set for NFPP functional units; the NFPP functional units include power units, bridge units, and energy units; the design variable space includes the mix proportions of the power units, bridge units, and energy units; the multi-objective function set includes compaction density and ratio performance indicators; Multiple initial sample points are obtained in the design variable space, and each initial sample point contains a set of initial proportions; NFPP composite materials are prepared according to each initial proportion, and the compaction density and rate performance index of each NFPP composite material are tested to construct an initial training dataset; Based on the initial training dataset, Gaussian process surrogate models are constructed for each objective function in the multi-objective function set, and multi-objective acquisition functions are defined based on each Gaussian process surrogate model; The sample points to be verified are determined by maximizing the multi-objective acquisition function. The sample points to be verified include a set of ratios to be verified. The NFPP composite material to be verified is prepared according to the ratios to be verified, and the compaction density and ratio performance index of the NFPP composite material to be verified are tested to update the initial training dataset and the Gaussian process surrogate model. This step is iterated until the preset termination condition is met. After the preset termination condition is met, the non-dominated solution on the current Pareto front is output to obtain the target combination scheme of the NFPP functional units. The corresponding functional units are then mixed according to the proportions in the target combination scheme to obtain the target NFPP composite material.

[0009] In an optional implementation, the step of outputting the non-dominated solution on the current Pareto front to obtain the target combination scheme of the NFPP functional unit includes: Output all non-dominated solutions on the current Pareto front; Based on the preset performance index priority, select one non-dominated solution from the non-dominated solutions as the target combination scheme; Wherein, if the performance index priority is ratio performance priority, then the non-dominated solution with the largest ratio performance index is selected; if the performance index priority is compaction density priority, then the non-dominated solution with the largest compaction density is selected.

[0010] In an optional implementation, the design variable space further includes the median particle size of the power unit, the bridge unit, and the energy unit; Wherein, the power unit is a nanoscale primary particle with a median particle size D50 of 300nm~800nm; and / or, the bridging unit is a porous secondary particle with a median particle size D50 of 0.8μm~8μm; and / or, the energy unit is a dense secondary particle with a median particle size D50 of 8μm~25μm.

[0011] In an optional implementation, the energy unit is doped with element M; the design variable space also includes the type and amount of element M. The chemical formula of the doped energy unit is Na₄Fe. 3-x M x (PO4)2P2O7, wherein the element M is selected from at least one of Mg, Al, Ti, Mn, Zr, and Cr, and the doping amount x satisfies 0. <x≤0.2。

[0012] In an optional implementation, the ratio is a mass fraction ratio; The mass fraction ratio of the power unit's mass fraction WA, the bridge unit's mass fraction WB, and the energy unit's mass fraction WC satisfies: WA + WB + WC = 1, and the values ​​of WA, WB, and WC are all in the range of 0 to 1. In the design variable space, WA and WB are used as independent variables for optimization; WC is calculated based on the values ​​of WA and WB and the constraint condition WA+WB+WC=1.

[0013] In an optional implementation, the rate performance metric is the 10C / 0.2C capacity retention rate; and / or, The multi-objective function set also includes a cycle lifetime index; the cycle lifetime index is the capacity retention rate after 1000 cycles at 1C; and / or, The Gaussian process surrogate model is a multi-output Gaussian process, and the kernel function used in the multi-output Gaussian process is either the Matrn kernel or the quadratic exponential kernel; and / or, The multi-objective acquisition function includes at least one of the following: a desired hypervolume boosting function, a parallel desired hypervolume boosting function, or a weighted sum acquisition function; and / or, The step of determining the sample points to be verified by maximizing the multi-objective acquisition function includes: locating the extreme value region of the multi-objective acquisition function in the design variable space using a global search; using points within the extreme value region as initial values, performing local fine-tuning using a local search algorithm to determine the sample points to be verified; and / or, The number of initial sample points obtained in the design variable space is 10 to 20; and / or, The preset termination condition includes at least one of the following conditions: A. The number of iterations reaches the preset maximum number of iterations, and the maximum number of iterations is 30 to 50 times; B. The maximum value of the multi-target acquisition function is lower than a preset threshold; C. In a series of preset number of iterations, the relative improvement of the Pareto front is less than a preset proportion.

[0014] Secondly, this application provides an NFPP-based composite material, which is prepared by the NFPP functional unit orientation assembly method as described in any of the foregoing embodiments.

[0015] Thirdly, this application provides a positive electrode sheet comprising the NFPP-based composite material as described in the foregoing embodiments.

[0016] Fourthly, this application provides a sodium-ion battery, including the positive electrode sheet as described in the foregoing embodiments.

[0017] Fifthly, this application provides an electrical device including a sodium-ion battery as described in the foregoing embodiments.

[0018] Compared with existing technologies, this application achieves efficient data characterization of complex material systems by constructing a design variable space encompassing the proportions of power units, bridge units, and energy units, and by building an initial training dataset incorporating compaction density and rate performance indicators. Based on this, it utilizes a Gaussian process surrogate model and a multi-objective acquisition function for iterative iteration, changing the traditional blind R&D model that relies on extensive physical trial and error and parameter grid traversal. This proactive learning search process can converge quickly and accurately in a vast multidimensional continuous parameter space with very few actual tests, thereby significantly shortening the R&D cycle of new materials and significantly reducing experimental material consumption and time costs.

[0019] To address the inherent physical constraints between compaction density and rate performance, this application directly outputs a nondominated solution on the Pareto front, breaking through the bottleneck of simply pursuing a single performance indicator leading to severe degradation of another performance metric. This process can comprehensively quantify the nonlinear synergistic effects when functional units of different scales are mixed, accurately identifying the optimal balance point between various conflicting indicators, and meeting the stringent performance requirements of different application scenarios.

[0020] Furthermore, from parameter space definition and mathematical model recommendation to physical material preparation and testing feedback, the entire implementation process forms an effective closed loop of virtual and real interaction. Each model iteration incorporates real physical test data to correct prediction biases, ensuring that the final output target combination ratio is not merely a theoretical calculation result, but a real process parameter with high engineering reliability. The sodium iron pyrophosphate pyrophosphate composite material obtained by mixing according to these parameters can effectively achieve the best comprehensive electrochemical performance designed for it. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 A basic flowchart illustrating an NFPP functional unit orientation assembly method provided in this application embodiment; Figure 2 A detailed logical architecture diagram of the algorithm iterative optimization in the assembly method provided in the embodiments of this application; Figure 3 This is a scanning electron microscope (SEM) image of the electrode cross-section prepared by the optimal scheme (Scheme T2) in Example 2 of this application; Figure 4 This is a comparison of X-ray diffraction (XRD) spectra of the energy unit before and after doping with element M in the embodiments of this application. Detailed Implementation

[0023] The embodiments of this application will be described in detail below with reference to examples. However, those skilled in the art will understand that the following examples are for illustrative purposes only and should not be considered as limiting the scope of this application. Unless otherwise specified in the examples, conventional conditions or conditions recommended by the manufacturer shall apply. Reagents or instruments whose manufacturers are not specified are all commercially available conventional products.

[0024] refer to Figure 1 This application provides a method for the directional assembly of NFPP functional units. Its core concept lies in effectively addressing the technical challenges of large parameter spaces and high trial-and-error costs in multi-objective performance optimization of complex material systems through a data-driven strategy. The assembly method specifically includes the following implementation steps: Step S100: Define the design variable space and multi-objective function set for the NFPP functional unit.

[0025] In this embodiment, the NFPP functional units within the system include at least three structural units: power units, bridge units, and energy units. When these three functional units are mixed and assembled, their structural morphology can form pore filling and framework support at both the macroscopic and microscopic levels, thereby affecting the overall physical stacking state of the electrode and the ion / electron transport channels. Therefore, the design variable space at least includes the proportions of the power units, bridge units, and energy units during mixing. Simultaneously, considering the often mutually restrictive physical properties within the material system, a multi-objective function set is defined, which includes at least compaction density and rate performance indicators.

[0026] Step S200: In order to establish the algorithm's preliminary understanding of objective physical laws, multiple initial sample points are obtained in the design variable space, and each initial sample point contains a set of initial proportions; NFPP composite materials are prepared according to each initial proportion, and the compaction density and ratio performance index of each NFPP composite material are tested to construct an initial training dataset.

[0027] In practice, experimental design methods can be used to uniformly or randomly obtain the initial points in space. Then, NFPP composite materials are prepared according to the initial proportions, and the actual physical test values ​​of the compaction density and rate performance of each NFPP composite material are obtained by standard physical characterization and electrochemical testing methods. Based on this, the initial training dataset is constructed.

[0028] Step S300: Based on the initial training dataset, construct a Gaussian process surrogate model for each objective function in the multi-objective function set, and define a multi-objective acquisition function based on each Gaussian process surrogate model.

[0029] It should be noted that the underlying mathematical logic of the Gaussian process surrogate model and the subsequent expected hypervolume enhancement function (EHVI) used in this application is derived from the well-known Bayesian statistical learning theory. The core of this application lies in the specific engineering mapping and combination application of this type of well-known mathematical model with the multidimensional physical design space (ratio, particle size, doping amount) and electrochemical indicators (compaction density, rate of change) of the NFPP functional unit.

[0030] In specific mathematical implementations, for a set of multi-objective functions Each objective function in For each, a corresponding Gaussian process surrogate model is constructed. The Gaussian process is derived from the mean function. Sum of covariance functions (i.e., kernel functions) The complete definition, its mathematical representation is: ; In this embodiment, the kernel function is preferably a squared exponential kernel, used to measure the similarity between design points with different proportions, and its function expression is: ; In the formula, The signal variance determines the amplitude of the function; The length scale determines the smoothness of the prediction function as the design variables change. For any unsampled experimental mix, this surrogate model can not only provide the predicted mean of performance, but also simultaneously output the prediction variance (characterizing uncertainty), providing a data foundation for subsequent active learning.

[0031] Gaussian process surrogate models can fit the nonlinear mapping relationship between proportion parameters and multi-objective performance, and provide the predicted mean of the objective function and the predicted variance characterizing the prediction uncertainty for any unsampled proportion. Based on the established Gaussian process surrogate models, a multi-objective acquisition function is defined to achieve a balance between reducing prediction uncertainty (exploration) and approximating the optimal prediction value (utilization), thereby guiding the selection of subsequent physical experiments.

[0032] Step S400: Enter the iterative optimization process of alternating physical experiments and model predictions: determine the sample points to be verified by maximizing the multi-objective acquisition function, the sample points to be verified include a set of ratios to be verified; prepare the NFPP composite material to be verified according to the ratios to be verified, and test the compaction density and rate performance index of the NFPP composite material to be verified, so as to update the initial training dataset and the Gaussian process surrogate model, and iterate this step until the preset termination condition is met.

[0033] In this embodiment, the expected hypervolume increase (EHVI) is preferably used as the multi-objective acquisition function. This function can accurately calculate the expected hypervolume increment of any candidate sampling point relative to the current Pareto front, and its mathematical expression is defined as: EHVI(x)=E[max(0,HV(P∪{x})-HV(P))]; In the formula, This represents a hypervolume dominated by non-dominated solution sets in a multidimensional target space. This is the set of Pareto frontiers that have been explored so far; This represents the mathematical expectation under the current Gaussian process prediction distribution.

[0034] During the iterative optimization process, which alternates between physical experiments and model predictions, the computer system proactively recommends the sampling point parameters with the greatest physical verification value for the next round by solving for the maximum value of the multi-objective acquisition function. The mathematical expression for this optimization process is: ; In the formula, This represents the multi-target acquisition function (such as the aforementioned EHVI function). This encompasses the overall design variable space, including proportions, particle size, and doping. Through this extreme value calculation, the algorithm can accurately balance "exploring unknown regions" with "utilizing known high-yield regions," significantly reducing the number of ineffective physical material preparations.

[0035] This step involves iterative execution, continuously improving the model's prediction accuracy and bringing it closer to the optimal solution region until a preset termination condition is met. This preset termination condition could be, for example, reaching a set maximum number of experimental iterations or the performance gains converging.

[0036] Step S500: After the preset termination condition is met, output the non-dominated solution on the current Pareto front to obtain the target combination scheme of the NFPP functional units, and mix the corresponding functional units according to the proportion in the target combination scheme to obtain the target NFPP composite material.

[0037] In this step, after satisfying the preset termination condition, the algorithm converges within the design variable space and outputs the non-dominated solution on the current Pareto front, thus obtaining the target combination scheme of the NFPP functional units. The Pareto front represents the optimal trade-off boundary between compaction density and rate performance achievable under the current material system. Subsequently, according to the proportions in the target combination scheme, the corresponding functional units are mixed using conventional physical mixing and other preparation processes to obtain the target NFPP composite material that balances multiple performance objectives. This mechanism achieves an effective closed loop of computational science guiding physical material preparation.

[0038] To further explain in detail the underlying logical architecture of steps S100 to S500 in the actual data processing and optimization process, refer to... Figure 2 , Figure 2 The detailed logical architecture of the algorithm iterative optimization in this assembly method is shown.

[0039] like Figure 2As shown, at the specific algorithm execution level, firstly, in step S1, a design space encompassing multiple dimensions such as sizing, particle size, and doping is established, and the electrochemical objective function to be maximized is determined. Subsequently, in step S2, statistical methods such as Latin hypercube sampling (LHS) are used to generate 10-20 initial data points to complete the cold start. In the core iterative optimization stage (steps S3 to S5), the system uses a multi-output Gaussian process surrogate model with a specific kernel function (such as the Matern kernel or the quadratic exponential kernel) to fit the complex physical black-box function, and explores and utilizes acquisition function balancing algorithms such as expected hypervolume improvement (EHVI) to continuously recommend the next set of sampling points with the most verification value. When the termination condition is met, the algorithm enters step S6, outputs the Pareto front composed of all non-dominated solutions, and finally selects the final target execution scheme based on the priority settings of specific application scenarios such as high power, high energy, or long lifetime.

[0040] In this embodiment, a design variable space encompassing the proportions of power units, bridge units, and energy units was constructed. By combining a Gaussian process surrogate model with a multi-objective acquisition function through iterative iteration, a small amount of physical verification replaced the traditional large-scale blind trial and error. This allowed for the rapid locking of optimal proportion parameters within an extremely large parameter space, significantly reducing R&D experimental costs and shortening the cycle. Simultaneously, the Pareto front non-dominated solution directly output by the algorithm effectively resolved the inherent conflict between compaction density and rate performance, achieving precise quantitative control of the synergistic effect of particles at different physical scales. This closed-loop feedback iterative process, continuously driven by real physical test data, ensures that the final mixing proportions have extremely high reliability for engineering mass production, enabling the final composite material to genuinely exhibit excellent comprehensive electrochemical performance.

[0041] In some implementations, step S500, which outputs the non-dominated solution on the current Pareto front to obtain the target combination scheme of the NFPP functional unit, specifically includes the process of outputting a set of candidate solutions by the algorithm and making a directional selection based on the actual physical application scenario. Specifically, it includes: Step S510: Output all non-dominated solutions on the current Pareto front.

[0042] Because there is often a mutually restrictive physical relationship between electrode compaction density and rate performance, pursuing high compaction often leads to a decrease in rate performance. In the multi-objective optimization process, the algorithm converges to a solution set. Any non-dominated solution within this set cannot improve the performance of another objective without weakening the performance of at least one objective. The boundary formed by these non-dominated solutions in the multi-dimensional objective space is the Pareto front. Outputting all non-dominated solutions essentially provides a complete optimal alternative formula library containing various performance trade-off strategies.

[0043] Step S520: Based on preset performance index priorities, select one non-dominated solution from the non-dominated solutions as the target combination scheme. Since different downstream terminal products have significantly different focuses on battery performance, it is necessary to convert specific application requirements into performance index priorities.

[0044] In the specific execution of the selection logic: (1) If the performance index priority is rate performance priority, then the non-dominated solution with the largest rate performance index is selected.

[0045] If the current preset scenario is a high-power application scenario (e.g., requiring the battery to have extremely strong fast charging and discharging capabilities), the corresponding performance index priority is set to rate performance priority. In this case, the system or operator will directly select the non-dominated solution with the largest rate performance index value from all non-dominated solutions at the Pareto front as the target combination scheme. This scheme can provide an optimal ion / electron conduction network at the microstructure level, thereby maximizing the satisfaction of high-power output requirements.

[0046] (2) If the performance index priority is compaction density priority, then the non-dominated solution with the largest compaction density is selected.

[0047] If the current preset scenario is a high-energy application scenario (e.g., requiring the maximum energy loading within a limited volume), the corresponding performance index priority is set to compaction density priority. In this case, the non-dominated solution with the largest compaction density value will be directly selected as the target combination scheme. This scheme can achieve the tightest integration between functional units at each scale in terms of physical stacking, thereby maximizing the active material content per unit volume of the electrode.

[0048] Through the aforementioned global output based on the Pareto front and the directional selection mechanism based on performance index priority, this assembly method can flexibly and accurately output the optimal material formulation adapted to different extreme or equilibrium conditions under the same set of experimental data and model framework, achieving a high degree of fit between material design and end application.

[0049] In some implementations, in order to more precisely control the microstructure of the electrode and the electrochemical reaction kinetics, the design variable space also includes the median particle size of the power unit, the bridge unit, and the energy unit when defining the design variable space.

[0050] When the median particle size distribution of powder particles is continuously adjusted within a certain range, different particle size combinations will directly determine the contact resistance, porosity, and packing density between particles in the electrode, thereby directly affecting the electrode's packing mode and final electrochemical performance. By incorporating the median particle size into the input features of the Gaussian process surrogate model, the algorithm can simultaneously find the particle size combination that achieves the optimal gradation while optimizing the proportions.

[0051] Specifically, the three functional units within the system have distinctly different physical morphologies and size ranges, resulting in a nonlinear physical synergistic effect during assembly: (1) The power unit is a nanoscale primary particle with a median particle size D50 of 300 nm to 800 nm. For example, it can be 300 nm, 350 nm, 400 nm, 450 nm, 500 nm, 550 nm, 600 nm, 700 nm, 750 nm, 800 nm, etc. Nanoscale primary particles of this size have a huge specific surface area and extremely short ion transport paths, which are mainly used to provide excellent rate performance and to enter the micropores formed by the stacking of other large particles as a tiny filler in the gradation assembly.

[0052] (2) The bridging unit is a porous secondary particle with a median particle size D50 of 0.8 μm to 8 μm; for example, it can be 0.8 μm, 1.5 μm, 2.0 μm, 3.0 μm, 4.0 μm, 5.0 μm, 6.0 μm, 7.0 μm, 7.5 μm, 8.0 μm, etc. The mesoscopic-scale porous particles play a role in structural support and dynamic buffering in the system. Their internal porous structure facilitates rapid wetting of the electrolyte, and their moderate size allows them to perfectly fill the large gaps created by the micron-sized large particles.

[0053] (3) The energy unit is a dense secondary particle with a median particle size D50 of 8μm to 25μm. For example, it can be 8μm, 10μm, 12μm, 14μm, 16μm, 18μm, 20μm, 22μm, 24μm, 25μm, etc. These micron-sized dense secondary particles have extremely high self-density and mainly serve as a macroscopic rigid skeleton in the mixed system to significantly improve the overall compaction density of the electrode.

[0054] By assembling the three types of units with specific morphology and scale gradients in a coordinated manner, and accurately matching their median particle size under the guidance of the algorithm, the inherent physical defects of single-morphology particles can be effectively overcome, and a smooth three-dimensional ion and electron conduction network can be constructed while maintaining high compaction density.

[0055] In some embodiments, to further improve the ion conduction kinetics of large-sized particles from the microscopic crystal lattice level and greatly broaden the optimization search dimension for multi-objective performance optimization, the energy unit is doped with element M; correspondingly, when defining the design variable space, the design variable space further includes the type and doping amount of the element M.

[0056] Since the micron-scale dense energy unit mainly provides a high compaction density of the electrode plate, its intrinsic sodium ion diffusion path is relatively long. By incorporating the microscopic doping variables into the unified surrogate model search space, the model can simultaneously find the most matched chemical modification scheme while searching for the optimal physical mixing ratio, so as to realize the synergy between physical mesoscopic assembly and microscopic crystal lattice regulation.

[0057] Specifically, the chemical formula of the doped energy unit is Na4Fe 3-x M x (PO4)2P2O7, wherein the element M is selected from at least one of Mg, Al, Ti, Mn, Zr and Cr.

[0058] The ionic radii of the above metal elements are different from that of iron ions. By in-situ substituting part of the iron atoms in the crystal lattice, moderate expansion of the crystal lattice or rearrangement of the electronic structure can be effectively induced, thereby broadening the diffusion channels for sodium ions inside the framework, lowering the migration energy barrier, and internally improving the rate response capability of the energy unit and the structural retention rate during long cycles.

[0059] Meanwhile, to strike a balance between improving kinetics and maintaining the intrinsic high specific capacity of the material, the doping amount x must be strictly controlled, which satisfies 0<x≤0.2. For example, the doping amount x may be 0.01, 0.03, 0.05, 0.08, 0.10, 0.12, 0.15, 0.18, 0.19, 0.20 and so on. By limiting the doping amount within this trace range, it can be ensured that the material maintains the stable three-dimensional framework structure of sodium iron pyrophosphate phosphate unchanged, avoiding phase separation caused by excessive doping or excessive loss of electrochemically active iron sites, thus ensuring high capacity output of the composite material during cycles. Through global iterative optimization in this doping space and ratio space by the model algorithm, the excellent parameter combination that meets specific application scenarios can be efficiently identified.

[0060] In specific embodiments, the power unit, the bridge unit and the energy unit can be independently prepared by respectively controlling the reaction time, stirring speed, sintering temperature and the addition amount of the templating agent in the precursor synthesis process (such as a co-precipitation method or a solid-phase method). For the energy unit doped with the element M, a metal salt compound containing the specific element M can be introduced according to the stoichiometric ratio in the precursor mixing stage, and high-temperature solid-phase sintering is performed to realize in-situ crystal lattice doping of the metal element.

[0061] In some implementations, to ensure that the data generated by the algorithm is compatible with real material preparation processes and to maximize the algorithm's optimization efficiency, the ratio is limited to a mass fraction ratio. In the industrial manufacturing process of battery electrodes, the mixing of each active component and the preparation of the slurry both rely on precise mass weighing; using mass fractions can directly guide subsequent physical experimental operations.

[0062] Specifically, the mass fraction ratios of the power unit (WA), the bridge unit (WB), and the energy unit (WC) are defined to satisfy: WA + WB + WC = 1. This constraint establishes the physical conservation law of the mixed system, meaning the total mass of the electrode active material is closed, and any increase in the proportion of one component is necessarily accompanied by a decrease in the proportion of other components. Furthermore, the values ​​of WA, WB, and WC are all within the range of 0 to 1; for example, they can be 0 wt%, 10 wt%, 22 wt%, 35 wt%, 48 wt%, 55 wt%, 68 wt%, 75 wt%, 88 wt%, 100 wt%, etc. (where wt% represents the mass percentage, corresponding to a coefficient value between 0 and 1). This range ensures that the amount of each component added conforms to physical reality, avoiding logical errors such as negative mass or exceeding the total amount.

[0063] In the design variable space, WA and WB are used as independent variables for optimization; WC is calculated based on the values ​​of WA and WB and the constraint condition WA+WB+WC=1.

[0064] In the actual surrogate modeling and optimization process, directly searching for WA, WB, and WC in three-dimensional space would lead to the algorithm encountering a large number of invalid coordinate points that do not satisfy the "sum of 1" constraint, severely slowing down computational efficiency and reducing the prediction accuracy of the Gaussian process surrogate model. Therefore, this assembly method performs dimensionality reduction processing of the simplex dimension in the design variable space: WA and WB are treated as independent variables for active optimization exploration; while WC is no longer used as a dimension actively searched by the algorithm, but as a passively derived variable, directly calculated based on the values ​​of WA and WB and the constraint condition WA+WB+WC=1.

[0065] Through the aforementioned dimensionality reduction mechanism, the originally constrained three-dimensional proportioning space is transformed into a purely two-dimensional space of independent variables. Each time the algorithm recommends the next sampling point, the system only needs to output a specific combination of WA and WB in the two-dimensional plane, and then complete the proportioning scheme through a simple subtraction operation (WC = 1 - WA - WB). This approach not only significantly reduces the dimensionality of the design space, enabling the algorithm to quickly converge the model with a very small number of physical experiments (e.g., 40 to 70 total experiments), but also effectively avoids the generation of invalid parameter combinations, ensuring the executability of each recommended proportion in physical experiments.

[0066] In some implementations, to meet the core requirements of the application scenario, the rate performance index is the capacity retention rate at 10C / 0.2C; and / or, the multi-objective function set also includes a cycle lifetime index; the cycle lifetime index is the capacity retention rate after 1000 cycles at 1C. These indices truly reflect the mutual constraints between the ion transport network within the material system and the macroscopic physical packing framework.

[0067] In some implementations, when establishing the virtual mapping relationship, the Gaussian process proxy model is preferably a multi-output Gaussian process. Unlike independently constructing a single-target model, the multi-output Gaussian process, by constructing a covariance matrix between indicators, can effectively capture and learn the implicit physical linkages between compaction density, rate performance, and cycle life, thereby improving prediction accuracy under small sample conditions. Furthermore, the kernel function used in the multi-output Gaussian process is a Matrn kernel or a squared exponential kernel, to accurately measure the similarity and spatial distance between different combinations of formulation parameters.

[0068] In some implementations, to enable precise navigation within a broad design space, the multi-objective acquisition function includes at least one of the desired hypervolume boosting function, the parallel desired hypervolume boosting function, or the weighted sum acquisition function. Such acquisition functions do not require subjective weighting of performance based on experience; instead, they inherently support multi-objective optimization, achieving an optimal balance between reducing prediction uncertainty and approaching performance limits.

[0069] In some implementations, when performing the acquisition function maximization operation, since its response surface often exhibits multi-peak characteristics, the system adopts a combined global and local optimization strategy. Specifically, step S400, which determines the sample points to be verified by maximizing the multi-objective acquisition function, includes: Step S410: Use a global search to locate the extreme region of the multi-objective acquisition function in the design variable space.

[0070] Step S420: Using the points in the extreme value region as initial values, a local search algorithm is used to perform local fine optimization to determine the sample points to be verified.

[0071] In this embodiment, when performing the maximization operation of the acquisition function, since its response surface often exhibits multi-peak characteristics, the system adopts a global and local optimization strategy: first, a global search is used to evaluate the large-scale range of the design variable space to locate the extreme value region of the multi-objective acquisition function; then, using the points in the extreme value region as initial values, a local search algorithm is introduced to perform local fine-tuning and ramping, thereby accurately extracting the theoretical highest point of the multi-objective acquisition function and transforming it into the physical sample point to be verified.

[0072] In some implementations, strict cost control and convergence management were applied to the scale of the virtual-real interaction physical experiment. During the cold start phase, the number of initial sample points obtained in the design variable space was 10 to 20. For example, it could be 10, 11, 12, 14, 15, 16, 17, 18, 19, 20, etc. This initial scale effectively balances the accuracy of basic modeling with early-stage R&D costs.

[0073] In some implementations, during the iterative optimization phase, the preset termination condition includes at least one of the following conditions: A. The number of iterations reaches a preset maximum number of iterations, which is 30 to 50 times; for example, it can be 30, 32, 35, 38, 40, 42, 45, 48, 49, 50, etc. This rigid constraint avoids the indefinite extension of the R&D cycle.

[0074] B. The maximum value of the multi-target acquisition function is lower than the preset threshold; this phenomenon indicates that all potential high-yield formulation areas within the design space have been fully explored.

[0075] C. In a predetermined number of consecutive iterations, the relative improvement of the Pareto front is less than a predetermined proportion. This phenomenon indicates that the comprehensive physical properties of the material system have reached its intrinsic limit, and the algorithm has fully converged in the multi-objective space. If any of the above conditions are met, the system stops recommending new physical verification experiments and outputs the final combination scheme.

[0076] This application provides an NFPP-based composite material, which is prepared strictly using the aforementioned NFPP functional unit directional assembly method. Due to the adoption of a data-driven optimization strategy based on multi-objective Bayesian optimization, the composite material is physically manifested as a mixture with precise multi-scale gradation characteristics. Internally, it simultaneously contains power units composed of nanoscale primary particles, bridging units composed of mesoscopic porous secondary particles, and energy units composed of micron-scale dense secondary particles. Furthermore, the mass fraction ratio between each functional unit (as well as the potential median particle size and microcrystalline lattice doping amount) precisely falls within the Pareto optimal front range for indicators such as compaction density and rate performance. This composite material, obtained through iterative output and mixing using a specific algorithm, fundamentally overcomes the inherent physical defects of single-morphology sodium iron pyrophosphate phosphate particles. Through the nonlinear synergistic effect of large particles constructing the framework, medium particles providing transition, and small particles filling micropores, it achieves a comprehensive balance and directional customization of high compaction density, high rate performance, and long cycle life at the material bulk level.

[0077] This application provides a positive electrode sheet whose active material coating contains an NFPP-based composite material prepared by the specific assembly method described above. Regarding the macroscopic physical structure and microscopic pore distribution of the electrode sheet, the introduction of multi-scale NFPP functional units with Pareto optimal proportions results in excellent packing density and a three-dimensional ion conduction network within the positive electrode sheet. Specifically, micron-scale dense energy units construct the high-density framework of the electrode sheet, mesoscopic porous bridging units effectively improve the wetting dynamics of the electrolyte within the electrode sheet, and nanoscale power units fully fill the tiny gaps in the electrode sheet framework, providing extremely short solid-phase diffusion paths. This synergistic physical effect of particle size distribution allows the positive electrode sheet to completely overcome the engineering contradiction in traditional electrode manufacturing where "pursuing high compaction often sacrifices rate performance," enabling it to exhibit extremely high electrode compaction density and excellent maximum current discharge capacity retention when meeting specific application scenarios (such as high-power or high-energy scenarios).

[0078] This embodiment provides a sodium-ion battery, including the positive electrode sheet as described above.

[0079] In a specific battery system structure, besides the positive electrode, the sodium-ion battery typically includes a matching negative electrode, a separator to separate the positive and negative electrodes, an electrolyte responsible for ion conduction, and a battery casing for overall encapsulation. Thanks to the deep control over the physical stacking structure and electrochemical kinetics of the positive electrode material, this sodium-ion battery can effectively meet the stringent requirements of various application scenarios, such as high power, high energy, or long lifespan. In actual industrial production and physical form design, the battery can be packaged in various ways, including but not limited to cylindrical, prismatic, pouch, or button cell structures; and in terms of functional application classification, it can include, but is not limited to, high-power batteries, high-capacity energy storage batteries, and balanced batteries that combine power and capacity.

[0080] This embodiment provides an electrical device, including the aforementioned sodium-ion battery, which serves as the core power source or key energy storage unit for providing operating power. Because the sodium-ion battery overcomes inherent defects and achieves comprehensive performance improvements at the microscopic and mesoscopic assembly levels, the device can obtain more stable, efficient, and sustained power supply support during long-term operation or under extreme high-rate discharge conditions. In practical engineering implementation and end-user applications, the scope of this electrical device is extremely broad, specifically including but not limited to: various new energy transportation vehicles (e.g., pure electric vehicles, plug-in hybrid electric vehicles, electric mopeds, electric bicycles, and electric balance scooters), medium and large-scale energy storage system equipment (e.g., grid-scale wind-solar hybrid energy storage power stations, home distributed energy storage microgrids, and backup power supplies for communication base stations), electric industrial or garden tools (e.g., cordless drills, electric lawnmowers, and electric forklifts), and specific portable smart consumer electronics products.

[0081] The present application is further illustrated below with specific embodiments. However, it should be understood that these embodiments are merely for the purpose of more detailed illustration and should not be construed as limiting the present application in any way.

[0082] Example 1 This embodiment provides a Bayesian optimization-based NFPP functional unit oriented assembly method, which mainly focuses on the combination of power unit A and bridge unit B, to examine and verify the dual-objective optimization capability of the method in two-dimensional proportioning space and high-power application scenarios.

[0083] In this embodiment, the power unit A is a nanoscale primary particle Na4Fe3(PO4)2P2O7 with a median particle size D50 = 350 nm; the bridge unit B is a porous secondary particle Na4Fe3(PO4)2P2O7 with a median particle size D50 = 1.5 μm.

[0084] 1. Experimental methods and procedures: Step S1, define the design space and objective function: The design variable space X is defined as the mass fraction WA (range 0.30~0.70, continuous step size) of power unit A and the mass fraction WB (range 0.30~0.70, continuous step size) of bridge unit B, with the constraint WA+WB=1. The objective function is defined as: electrode compaction density ρ (g / cm³, the objective is to maximize) and 10C / 0.2C capacity retention rate R (%, the objective is to maximize).

[0085] Step S2, construct the initial training dataset: 12 initial sample points are selected in the design space X using Latin hypercube sampling (LHS), and the W value of each sample point is... A Uniformly distributed in the range of 0.30 to 0.70, W B =1-W A NFPP composite materials were prepared according to the proportions of each sample point, and electrodes were fabricated using the standard electrode preparation process. The compaction density and 10C rate capacity retention of the electrodes were tested. The specific initial datasets are shown in Table 1. Table 1. Initial training dataset for Example 1

[0086] Step S3, construct Gaussian process surrogate models: construct Gaussian process surrogate models for the objective functions of electrode compaction density and magnification performance respectively, using a squared exponential kernel, and optimize the hyperparameters of the kernel function by maximizing the logarithmic marginal likelihood.

[0087] Step S4, define the multi-objective acquisition function: use the expected hypervolume lift (EHVI) as the acquisition function to calculate the expected hypervolume increment of the candidate point relative to the current Pareto front.

[0088] Step S5, Iterative Optimization: Maximize the EHVI sampling function on the current model (first generate 1000 uniform candidate points globally, then perform a local fine-grained search) to recommend the next sampling point, prepare and test, and then update the model. Set the maximum number of iterations to 30.

[0089] 2. Experimental Results and Analysis: After 30 iterations (42 total experiments), the algorithm converged to the Pareto hypervolume after the 22nd iteration. The non-dominated schemes included in the final Pareto front are shown in Table 2. Table 2. Pareto frontier nondominated schemes output in Example 1

[0090] For high-power applications (requiring 10C retention rate ≥ 80% and electrode compaction density ≥ 2.35 g / cm³), 3From the Pareto front, option P2 (WA=0.52, WB=0.48) is selected as the final optimal objective combination.

[0091] Example 2 This embodiment provides a Bayesian optimization-based NFPP functional unit oriented assembly method, which includes three functional units: power unit A, bridge unit B, and energy unit C. It is used to examine the optimization capability in a three-dimensional ratio space and a scenario of synergistic optimization of three objectives: high power, high energy, and long lifespan.

[0092] In this embodiment, the power unit A is a nanoscale primary particle Na4Fe3(PO4)2P2O7 with a median particle size D50 = 700 nm; the bridging unit B is a porous secondary particle Na4Fe3(PO4)2P2O7 with a median particle size D50 = 6.0 μm; and the energy unit C is a dense secondary particle Na4Fe 2.95 Ti 0.05 (PO4)2P2O7 (Ti doped, doping amount x=0.05), median particle size D50=22μm.

[0093] 1. Experimental methods and procedures: Step S1, the design variable space X is defined as: W A (0.10~0.35), W B (0.20~0.55), W C (0.30~0.70), with constraint W A +W B +W C =1, and transform the three-dimensional proportioning space into a two-dimensional independent variable (W) C =1-W A -W B The objective functions are defined as three: electrode compaction density (maximization), capacity retention at 10C / 0.2C (maximization), and capacity retention after 1000 cycles at 1C (maximization). The energy unit C uses Ti-doped Na₄Fe₂. 2.95 Ti 0.05 (PO4)2P2O7 (doping amount x=0.05).

[0094] The subsequent steps are similar to those in Example 1. Initial sampling is performed using Latin hypercube sampling (15 initial sample points). Three independent Gaussian process surrogate models are constructed, and iterative optimization is performed using the EHVI acquisition function, with a maximum number of iterations set to 40.

[0095] 2. Experimental Results and Analysis: After 40 iterations (55 total experiments), the algorithm converged, and the final output of the non-dominated scheme of the three-dimensional Pareto front is shown in Table 3: Table 3. Three-dimensional Pareto front schemes output in Example 2

[0096] Comparative Example 1 This comparative example uses the traditional grid search method to optimize the proportion of functional units of NFPP material as a comparison and verification of Example 1, in order to examine the efficiency and accuracy limitations of the traditional exhaustive search when dealing with multi-dimensional design space.

[0097] 1. Experimental methods and results: A full factorial search was performed on WA (0.30~0.70, step size 0.05) and WB (0.30~0.70, step size 0.05, and WA+WB=1) with a step size of 5%, requiring a total of 9 sizing schemes to be tested. Furthermore, if particle size (3 levels) and doping (5 levels) are considered simultaneously, the number of particle size combinations involving 3 functional units (3...) becomes even more complex. 3 =27 types) and 5 doping levels, the total number of factor combinations will increase to 9×27×5=1215 types. Experiments of this scale are difficult to implement one by one in engineering practice.

[0098] The test was conducted using the same objective function as in Example 1. The optimal solution found under the same experimental precision was: WA=0.50, WB=0.50, electrode compaction density of 2.30 g / cm³, and 10C retention rate of 82%.

[0099] 2. Results Analysis: The optimal solution found in Comparative Example 1 is completely covered by solution P2 (WA=0.52, WB=0.48, ρ=2.35, R=83%) in the final Pareto front output of Example 1. That is, Example 1 has better rate performance when the compaction density is comparable or even higher.

[0100] Comparative Example 2 This comparative example uses the traditional orthogonal experimental design method to optimize the proportion of functional units of NFPP material as a comparison and verification of Example 1, in order to examine the limitations of conventional statistical experimental methods in resolving multi-objective conflicts.

[0101] 1. Experimental methods and results: Using L9(3) 4 An orthogonal experimental design was used to conduct orthogonal experiments on WA (three levels: 0.35, 0.50, 0.65) and WB (three levels: 0.35, 0.50, 0.65, and WA+WB=1), with a total of 9 experiments. The optimal scheme was found to be: WA=0.50, WB=0.50, and the electrode compaction density was 2.30 g / cm³. 3 The 10C retention rate was 82%.

[0102] 2. Results Analysis: The optimal scheme found in Comparative Example 2 has lower electrode compaction density and ratio performance than scheme P2 in the Pareto front output of Example 1.

[0103] Comparative Example 3 This comparative example provides basic performance data of single-component electrodes for three NFPP functional units (power unit A, bridge unit B, and energy unit C) as a benchmark for comparison and verification of the embodiments of this application, in order to examine the physical limitations of single-morphology materials and the synergistic gain effect brought about by multi-scale assembly.

[0104] 1. Experimental methods and results: Comparative electrodes were prepared using 100 wt% pure power unit A, 100 wt% pure bridge unit B, and 100 wt% pure energy unit C as active materials, respectively, following the same standard electrode preparation process as in the aforementioned embodiments. The electrode compaction density and 10C rate capacity retention were then tested.

[0105] The test results are as follows: (1) 100% pure power unit A (nanoscale primary particles): electrode compaction density is 1.95 g / cm³ 3 The 10C retention rate was 89.1%.

[0106] (2) 100% pure bridge unit B (mesoscopic porous secondary particles): electrode compaction density is 2.06 g / cm³ 3 The 10C retention rate was 85.6%.

[0107] (3) 100% pure energy unit C (micron-sized dense secondary particles): electrode compaction density is 2.21 g / cm³ 3 The 10C retention rate was 70.5%.

[0108] 2. Results Analysis: Particles with a single morphology exhibit severe performance polarization. Although pure power unit A has excellent rate performance, its electrode compaction density is extremely low, resulting in a serious deficiency in volumetric energy density; although pure energy unit C has a high compaction density, its high-rate discharge performance is extremely poor due to the long ion diffusion path.

[0109] Comparing the basic data of this comparative example with the balanced scheme T2 (WA=0.18, WB=0.32, WC=0.50) output in Example 2 of this application, it can be seen that the electrode compaction density of scheme T2 reaches 2.35 g / cm³. 3 (Higher than 2.21 g / cm³ for pure micron-sized particles) 3 Furthermore, the 10C retention rate remained at 84% (far higher than the 70.5% of pure micron-sized particles).

[0110] This directly proves that the multi-scale directional assembly method of this application is not a simple linear summation of performance, but rather a significant synergistic effect of physical structure produced by the perfect gradation filling of large, medium and small particles in the micropores, breaking the inherent bottleneck of mutual exclusion between compaction density and rate performance of a single material.

[0111] Comprehensive experimental results and analysis: To visually compare the optimization efficiency and effectiveness of the embodiments of this application with those of traditional comparative examples, the core data of embodiments 1 and 2 and comparative examples 1-3 are summarized in Table 4: Table 4. Summary of the effects of the embodiments and comparative examples of this application

[0112] in conclusion: (1) In terms of algorithm optimization efficiency: Comparative Example 1 uses the traditional grid search method, which only searches for 9 schemes in the ratio dimension, and cannot take into account the particle size and doping dimensions. The optimal scheme found (WA=0.50, WB=0.50) is not the global optimum. If Comparative Example 1 were to search all dimensions simultaneously (ratio × particle size × doping = 1215 schemes), the experimental scale would not be feasible in industrial practice. However, Example 1 of this application only requires 42 experiments and Example 2 only requires 55 experiments, which means that it converges efficiently in a multi-dimensional design space that includes ratio, particle size and doping, and obtains a material combination scheme with better performance indicators, which fully demonstrates the efficiency advantage of the method of this application in high-dimensional space optimization.

[0113] Meanwhile, compared to the traditional single-solution output (Comparative Examples 1 and 2), the method in this application can simultaneously output the complete Pareto front, providing flexible solution options for different industrial application scenarios.

[0114] (2) Regarding the synergistic effect of materials: Comparing Example 2 (multi-scale unit assembly) of this application with Comparative Example 3 (single-diameter pure phase material), it can be seen that this application, through the synergistic assembly and graded filling of large, medium and small functional units, enables the electrode compaction density to break through the limit of physical packing of micron-sized pure large particles (from 2.21 g / cm³). 3 Increased to 2.35g / cm 3 This approach effectively improves the sluggish dynamics of large particles, significantly increasing the 10C rate retention rate (from 70.5% to 84%). This fully demonstrates that the directional assembly scheme of this application generates significant nonlinear synergistic gains in the intrinsic structure of the material.

[0115] Characterization and analysis of material microstructure and phase structure To further verify the synergistic gain effect brought about by the multi-scale directional assembly and lattice doping from the microscopic mechanism, the micromorphology and phase structure of the materials and electrodes prepared in the embodiments of this application were characterized.

[0116] refer to Figure 3 , Figure 3 This is a scanning electron microscope (SEM) image of the electrode cross-section prepared by the optimal scheme (Scheme T2) in Example 2 of this application.

[0117] like Figure 3 As shown, under the optimal mass ratio output by the Bayesian algorithm, the electrode exhibits a highly dense and multi-level microstructure. It can be clearly observed that micron-sized energy units (large particles) constitute the main rigid framework of the electrode, while mesoscale bridging units (medium particles) and nanoscale power units (small particles) fully and uniformly fill the microscopic voids created by the overlapping of the large particles. This precise gradation and interlocking of multi-scale particles directly explains, from a physical morphology perspective, why this electrode can break through the packing limit of pure-phase large particles and achieve an excellent compaction density of 2.35 g / cm³. Simultaneously, the interspersed medium and small particles effectively maintain abundant capillary channels, shortening the solid-phase ion diffusion distance, thereby ensuring excellent discharge capacity at high rates (10C).

[0118] refer to Figure 4 , Figure 4 This is a comparison of X-ray diffraction (XRD) spectra of the energy unit before and after doping with element M in the embodiments of this application.

[0119] like Figure 4 As shown, comparing the curves before and after doping, it can be seen that after the energy unit is doped with metal element M, the peak intensity of its main diffraction peak is improved, and the full width at half maximum (FWHM) does not increase significantly, indicating that the material still maintains a high degree of crystallinity. More importantly, the diffraction peaks after doping are slightly shifted towards lower angles, and no new impurity peaks are observed to be generated throughout the process. This microscopic characterization phenomenon conclusively proves that the dopant element M has successfully entered the three-dimensional lattice of sodium iron pyrophosphate in situ (causing a moderate expansion of the lattice parameters, resulting in a leftward shift of the diffraction peaks, which is beneficial for widening the sodium ion transport channels). Moreover, this trace doping did not cause lattice distortion, nor did it induce phase separation or lead to a decrease in crystallinity. Thus, while maintaining the intrinsic structural stability of the material, its electrochemical kinetic performance is significantly improved.

[0120] Comparative Experiment and Effect Analysis of Element Doping in Energy Units To further verify the actual contribution of the micro-lattice modification feature of "doping element M within the energy unit" to the electrochemical performance of the electrode, this experimental example is added for benchmark comparison.

[0121] 1. Experimental Method: The optimal balanced scheme T2 output in Example 2 is locked (macroscopic ratios are fixed as follows: power unit WA=0.18, bridge unit WB=0.32, energy unit WC=0.50). During the electrode fabrication process, only the microscopic doping state of the energy unit C is changed, and the following three test groups are set up: Test group 1 (corresponding to scheme T2 in Example 2): using Ti-doped energy unit C, with a doping amount x=0.05, and the chemical formula Na4Fe 2.95 Ti 0.05 (PO4)2P2O7.

[0122] Test group 2 (Extended verification group): Energy unit C doped with Mg element, doping amount x=0.08, chemical formula Na4Fe 2.92 Mg 0.08 (PO4)2P2O7.

[0123] Control group (undoped baseline): Pure phase NFPP with no doping was used as the energy unit C, with the chemical formula Na4Fe3(PO4)2P2O7.

[0124] The electrochemical performance of the three groups of electrodes was tested under the same conditions.

[0125] The test results are shown in Table 5 below: Table 5. Comparison of Electrochemical Performance Before and After Energy Unit Doping

[0126] 2. Results Analysis: According to the test data in Table 5, under the premise that the macroscopic physical proportions (multi-scale gradation) are exactly the same, the compaction density of the three groups of electrodes is basically consistent, all ranging from 2.34 to 2.35 g / cm³. 3 The performance was at a high level. However, there were significant differences in kinetics and long-cycle performance: the undoped control group had a 10C high-rate retention rate of only 76.2% and a 1000-cycle retention rate of only 85.3%; while test groups 1 and 2, which introduced trace elements M (Ti or Mg) for in-situ lattice substitution, achieved a 10C rate retention rate of over 82% and a long-cycle retention rate exceeding 92% without damaging the main framework.

[0127] Experimental results demonstrate that microscopic modification, which induces moderate lattice expansion and widens sodium ion diffusion channels through doping with element M, can successfully compensate for the defects of low intrinsic ionic conductivity and susceptibility to microcracks in long-term charge-discharge cycles in micron-sized large particles (energy units). This characteristic, combined with multi-scale physical gradation, forms an excellent "microscopic-macroscopic" dual synergy, providing essential technical support for improving the overall performance of composite materials.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for oriented assembly of NFPP functional units, characterized in that, include: Define the design variable space and multi-objective function set for NFPP functional units; The NFPP functional unit includes a power unit, a bridge unit, and an energy unit; The design variable space includes the proportion of the power unit, the bridge unit, and the energy unit when mixed; the multi-objective function set includes compaction density and rate performance index. Multiple initial sample points are obtained in the design variable space, and each initial sample point contains a set of initial proportions; NFPP composite materials are prepared according to each initial proportion, and the compaction density and rate performance index of each NFPP composite material are tested to construct an initial training dataset; Based on the initial training dataset, Gaussian process surrogate models are constructed for each objective function in the multi-objective function set, and multi-objective acquisition functions are defined based on each Gaussian process surrogate model; The sample points to be verified are determined by maximizing the multi-objective acquisition function, and the sample points to be verified contain a set of ratios to be verified. Prepare the NFPP composite material to be verified according to the specified ratio to be verified, and test the compaction density and rate performance index of the NFPP composite material to be verified in order to update the initial training dataset and the Gaussian process proxy model, and iterate this step until the preset termination condition is met. After the preset termination condition is met, the non-dominated solution on the current Pareto front is output to obtain the target combination scheme of the NFPP functional units. The corresponding functional units are then mixed according to the proportions in the target combination scheme to obtain the target NFPP composite material.

2. The NFPP functional unit orientation assembly method as described in claim 1, characterized in that, The output of the non-dominated solution on the current Pareto front yields the target combination scheme of the NFPP functional unit, including: Output all non-dominated solutions on the current Pareto front; Based on the preset performance index priority, select one non-dominated solution from the non-dominated solutions as the target combination scheme; Wherein, if the performance index priority is ratio performance priority, then the non-dominated solution with the largest ratio performance index is selected; if the performance index priority is compaction density priority, then the non-dominated solution with the largest compaction density is selected.

3. The NFPP functional unit orientation assembly method as described in claim 1, characterized in that, The design variable space also includes the median particle size of the power unit, the bridge unit, and the energy unit; Wherein, the power unit is a nanoscale primary particle with a median particle size D50 of 300nm~800nm; and / or, the bridging unit is a porous secondary particle with a median particle size D50 of 0.8μm~8μm; and / or, the energy unit is a dense secondary particle with a median particle size D50 of 8μm~25μm.

4. The NFPP functional unit orientation assembly method as described in claim 1, characterized in that, The energy unit is doped with element M; the design variable space also includes the type and doping amount of element M; The chemical formula of the doped energy unit is Na₄Fe. 3-x M x (PO4)2P2O7, wherein the element M is selected from at least one of Mg, Al, Ti, Mn, Zr, and Cr, and the doping amount x satisfies 0. <x≤0.2。 5. The NFPP functional unit orientation assembly method as described in claim 1, characterized in that, The ratio is a mass fraction ratio; The mass fraction ratio of the power unit's mass fraction WA, the bridge unit's mass fraction WB, and the energy unit's mass fraction WC satisfies: WA + WB + WC = 1, and the values ​​of WA, WB, and WC are all in the range of 0 to 1. In the design variable space, WA and WB are used as independent variables for optimization; WC is calculated based on the values ​​of WA and WB and the constraint condition WA+WB+WC=1.

6. The NFPP functional unit orientation assembly method as described in claim 1, characterized in that, The rate performance index is the capacity retention rate at 10C / 0.2C; and / or, The multi-objective function set also includes a cycle lifetime index; the cycle lifetime index is the capacity retention rate after 1000 cycles at 1C; and / or, The Gaussian process surrogate model is a multi-output Gaussian process, and the kernel function used in the multi-output Gaussian process is either the Matrn kernel or the quadratic exponential kernel; and / or, The multi-objective acquisition function includes at least one of the following: a desired hypervolume boosting function, a parallel desired hypervolume boosting function, or a weighted sum acquisition function; and / or, The step of determining the sample points to be verified by maximizing the multi-objective acquisition function includes: locating the extreme value region of the multi-objective acquisition function in the design variable space using a global search; using points within the extreme value region as initial values, performing local fine-tuning using a local search algorithm to determine the sample points to be verified; and / or, The number of initial sample points obtained in the design variable space is 10 to 20; and / or, The preset termination condition includes at least one of the following conditions: A. The number of iterations reaches the preset maximum number of iterations, and the maximum number of iterations is 30 to 50 times; B. The maximum value of the multi-target acquisition function is lower than a preset threshold; C. In a series of preset number of iterations, the relative improvement of the Pareto front is less than a preset proportion.

7. An NFPP-based composite material, characterized in that, It was prepared using the NFPP functional unit orientation assembly method as described in any one of claims 1-6.

8. A positive electrode sheet, characterized in that, Including the NFPP-based composite material as described in claim 7.

9. A sodium-ion battery, characterized in that, Including the positive electrode sheet as described in claim 8.

10. An electrical appliance, characterized in that, Including the sodium-ion battery as described in claim 9.

Images