Knowledge graph-based compound fertilizer proportioning optimization method and system

Abstract

The application discloses a knowledge graph-based mixed fertilizer proportioning optimization method and system, and belongs to the technical field of mixed fertilizer proportioning optimization. Under the condition that the relative humidity of the working environment is relatively high and the mixed material needs to maintain the stability of the particle size structure during transportation and spreading, an executable proportioning result is output for a given candidate raw material list and target nutrient index. The method uses a fertilizer knowledge graph to depict the multi-hop coupling and closed-loop influence of the hygroscopic properties of the candidate raw materials and the incompatible relationship and eutectic interaction relationship between the candidate raw materials, generates the upper limit and lower limit of the proportion of each candidate raw material and forms a proportion window constraint, and solves the mixed fertilizer proportioning that meets the target nutrient index under the constraint. When there is no feasible solution, the upper limit of the proportion is kept unchanged, the segregation constraint is relaxed to update the lower limit of the proportion, and then the solution is recalculated, so as to improve the executability and reproducibility of the proportioning result under the common constraint of controlled hygroscopic caking risk and stable particle size structure.

Description

Technical Field

[0001] This invention relates to the field of fertilizer blending ratio optimization technology, and more specifically, to a method and system for fertilizer blending ratio optimization based on knowledge graphs. Background Technology

[0002] Blended fertilizer formulation optimization technology is commonly used in on-site bulk blending operations at grassroots fertilizer blending stations or distribution centers. This technology aims to quickly generate a calculable blended fertilizer formulation scheme given a list of candidate raw materials and target nutrient indicators in a blending order. In this scenario, existing technologies typically impose proportional constraints on each candidate raw material based on preset process limits or raw material usage specifications, and solve for the formulation result by satisfying the target nutrient indicators as the objective or constraint.

[0003] The existing technology has the following shortcomings:

[0004] When the relative humidity of the working environment is high and the mixture needs to maintain a stable particle size structure during transportation and application, in addition to the target nutrient index, it is also necessary to consider both the controllable risk of hygroscopic agglomeration and the constraint of stable particle size structure. This makes the generation of proportion constraints and the solution of proportions more prone to conflict. Specifically, when the combination of candidate raw materials is complex and there are incompatible relationships and eutectic interactions between them, the coupling of hygroscopic properties and multi-hop relationships between candidate raw materials may introduce risk transmission and amplification effects, making it difficult to stably suppress the risk of hygroscopic agglomeration by relying solely on single raw material properties or simple rules. At the same time, the lower limit of proportion needs to be determined under the condition that the degree of segregation risk does not exceed the preset segregation threshold. The degree of segregation risk is a global quantity at the candidate raw material list level, which can easily lead to inconsistencies between the lower limit of proportion of each raw material and the global segregation constraint and the feasibility of proportion, thereby affecting the executability of the final proportion result.

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for optimizing the ratio of compound fertilizers based on knowledge graphs, in order to solve the problems mentioned in the background art. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Knowledge graph-based methods for optimizing the formulation of blended fertilizers include:

[0008] S101, obtain the candidate raw material list, target nutrient index and relative humidity of the working environment, and obtain the hygroscopic properties and particle size index of each candidate raw material, as well as the eutectic interaction and incompatibility relationship between the candidate raw materials in the fertilizer knowledge graph;

[0009] S102, for each candidate raw material, starting from the candidate raw material, a multi-hop transmission path is retrieved along the incompatibility relationship and a closed-loop interaction configuration is detected to form an interaction subgraph of the candidate raw material; within the interaction subgraph, based on the hygroscopic properties of the candidate raw material and other candidate raw materials in the interaction subgraph and combined with the eutectic interaction relationship in the interaction subgraph, the combined hygroscopic coupling strength of the candidate raw material is determined; and based on the particle size index of the candidate raw material and the particle size index of other candidate raw materials in the interaction subgraph, the particle size overlap of the candidate raw material is determined.

[0010] S103, for each candidate raw material, the upper limit of the proportion of the candidate raw material is determined based on the combined hygroscopic coupling strength, particle size overlap, and relative humidity of the working environment. The greater the particle size overlap, the smaller the upper limit of the proportion. The degree of segregation risk is determined by the dispersion of the particle size index of each candidate raw material in the candidate raw material list. The lower limit of the proportion of the candidate raw material is determined based on the comparison between the degree of segregation risk and the preset segregation threshold. When the degree of segregation risk does not exceed the preset segregation threshold, the lower limit of the proportion is determined according to the adjustment rule. When the degree of segregation risk exceeds the preset segregation threshold, the lower limit of the proportion is determined according to the conservative lower limit. The upper and lower limits of the proportions corresponding to each candidate raw material are summarized into a proportion window table.

[0011] S104 uses the proportion window table as a constraint and aims to ensure that the obtained ratio meets the target nutrient index to determine the proportion of each candidate raw material. When there is no feasible solution that meets the target nutrient index, the upper limit of the proportion of each candidate raw material remains unchanged. The separation threshold caliber is updated according to the relaxation level or the lower limit of the proportion is updated according to the adjustment rule caliber. Based on this, the lower limit of the proportion of each candidate raw material is recalculated and the solution is re-solved to output the mixed fertilizer ratio result.

[0012] In a preferred embodiment, the fertilizer knowledge graph includes at least a set of raw material nodes with candidate raw materials as entity nodes, attribute fields that record the hygroscopic properties and particle size index of the candidate raw materials, and relationship edges used to characterize the eutectic interaction relationship and incompatibility relationship between candidate raw materials. The incompatibility relationship is a relationship edge that characterizes the stability conflict or adverse reaction of two candidate raw materials under mixing conditions, and the eutectic interaction relationship is a relationship edge that characterizes the formation of eutectic or the induction of hygroscopic related interactions between two candidate raw materials.

[0013] In a preferred embodiment, the nutritional contribution coefficient of each candidate raw material is also obtained. The nutritional contribution coefficient is used to characterize the content of target nutrients per unit mass of the candidate raw material. In S104, the target nutrient content corresponding to the obtained ratio is calculated based on the nutritional contribution coefficient and compared with the target nutrient index to establish the constraint conditions that meet the target nutrient index.

[0014] In a preferred embodiment, the hygroscopic property includes at least one of the critical relative humidity indicator and the hygroscopic rate indicator, and the critical relative humidity indicator and the hygroscopic rate indicator are experimental calibration results obtained according to preset test specifications or historical operating condition statistical results obtained according to preset statistical calibrators, and are written into the fertilizer knowledge graph.

[0015] In a preferred embodiment, the particle size index is generated from the particle size distribution detection data of the candidate raw materials through a preset mapping rule and written into the fertilizer knowledge graph. The preset mapping rule at least specifies the particle size distribution interval division method and maps the statistical characteristics of the particle size distribution to scalar values ​​by a lookup table mapping method.

[0016] In a preferred embodiment, in S102, the multi-hop transfer path is searched along the incompatibility relation, the search depth does not exceed the preset maximum number of hops, and the path with duplicate nodes is truncated; the closed-loop interaction configuration is a closed-loop path formed by the starting candidate raw material returning to itself through the incompatibility relation, the length of the closed-loop path does not exceed the preset upper limit of the closed-loop length, and the consistency of the node sequence of the closed-loop path is used as the criterion for determining duplicate closed loops and deduplicating the count; the interaction subgraph includes the starting candidate raw material and the candidate raw material nodes covered by the above path and the incompatibility relation edges, and retains the eutectic interaction relation edges existing between the candidate raw material nodes.

[0017] In a preferred embodiment, the combined hygroscopic coupling strength is obtained by fusing the cumulative risk term, the amplified risk term, and the hygroscopic attribute term. The cumulative risk term is the sum of the risk weights of incompatible relation edges on the multi-hop transmission path within the interaction subgraph. The amplified risk term is the amplification amount calculated by the number of closed-loop paths and the closed-loop length within the interaction subgraph using a preset amplification factor. The hygroscopic attribute term is determined by the hygroscopic properties of the starting candidate raw material. The eutectic interaction relation is used to apply a gain correction with a preset gain coefficient to the hygroscopic attribute term. The fusion method is normalized weighted summation, and the weights are given by a preset weight table.

[0018] In a preferred embodiment, the degree of particle size overlap is formed by the difference set between the particle size guide number of the starting candidate raw material and the particle size guide number of other candidate raw materials in the interaction subgraph. The difference set is obtained by looking up a table through a preset mapping function. The overall difference of the difference set is represented by the mean of the absolute difference. The smaller the overall difference, the greater the degree of particle size overlap. The difference set is aggregated by weighted mean. The weight is determined by the shortest path length from the starting point to the corresponding candidate raw material node in the interaction subgraph.

[0019] In a preferred embodiment, the segregation risk level is determined by the dispersion of the particle size index of each candidate raw material in the candidate raw material list. The dispersion is the ratio of the standard deviation to the mean of the particle size index. The segregation threshold is determined by a preset threshold table corresponding to the model of the mixing equipment. The lower limit of the proportion is determined according to a preset minimum proportion rule. The preset minimum proportion rule is obtained by looking up the minimum proportion value according to the candidate raw material type, and then adjusting the minimum proportion value downward according to the adjustment rule when the segregation risk level does not exceed the segregation threshold. The lower limit of the proportion is adjusted more weakly as the segregation risk level is closer to the segregation threshold. When determining the separation threshold, the lower limit of the proportion is determined according to a conservative lower limit caliber. The conservative lower limit caliber includes at least making the downward adjustment correction degenerate to zero, so that the lower limit of the proportion is equal to the minimum proportion value. When there is no feasible solution that satisfies the target nutrient index, the upper limit of the proportion remains unchanged, the relaxation level is a non-negative integer, the separation threshold caliber or the downward adjustment correction rule caliber of the lower limit of the proportion is updated according to the relaxation level, and the lower limit of the proportion is recalculated accordingly. The upper and lower limits of the proportion of each candidate raw material are summarized into a proportion window table. The proportion window table includes at least the candidate raw material identifier, the upper limit of the proportion, the lower limit of the proportion, and the relative humidity field of the working environment.

[0020] The knowledge graph-based fertilizer blending ratio optimization system includes a candidate raw material graph data acquisition and input set construction unit, an interaction subgraph construction and coupling index calculation unit, a proportion window constraint determination and window table generation unit, a ratio optimization solution and result output unit, and a preset parameter storage area. The candidate raw material graph data acquisition and input set construction unit, the interaction subgraph construction and coupling index calculation unit, the proportion window constraint determination and window table generation unit, and the ratio optimization solution and result output unit are respectively used to execute S101, S102, S103, and S104. The preset parameter storage area is used to store and provide the maximum number of hops, the upper limit of closed loop length, the risk weight mapping rule, the eutectic gain correction rule, the fusion weight, the proportion upper limit shrinkage mapping rule, the segregation threshold, the minimum proportion rule, the downward adjustment correction rule, the relaxation level, the iteration upper limit, and the tolerance threshold for use by the above units.

[0021] The effects and advantages of the knowledge graph-based method and system for optimizing the ratio of compound fertilizers:

[0022] This invention provides a knowledge graph-based method and system for optimizing the proportion of blended fertilizers. By utilizing a fertilizer knowledge graph, it structurally characterizes the multi-hop transmission and closed-loop interaction configurations of the hygroscopic properties and incompatibilities of candidate raw materials. Combined with gain correction of eutectic interaction relationships, the risk constraint of hygroscopic agglomeration under high humidity and complex combinations of candidate raw materials no longer relies on single raw materials or simple rules, thus better suppressing the distortion of proportion constraints caused by amplified risks. Simultaneously, this invention determines a lower limit of proportions under the condition that the segregation risk does not exceed a preset segregation threshold, and together with the upper limit of proportions, forms a proportion window constraint as the solution entry point. This ensures that the generation of proportion constraints and the proportion solution are connected under the same caliber, reducing the risk of infeasibility or unexecutability. When a feasible solution does not exist, only the segregation constraint is relaxed and the lower limit of proportions is updated before recalculation, maintaining the consistency and reproducibility of the solution logic without changing the existing input conditions. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0024] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0025] Figure 3 This is a diagram illustrating the risks of traditional strategies.

[0026] Figure 4 This is a diagram illustrating the risk comparison of the strategy of this invention. Detailed Implementation

[0027] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] This invention provides a knowledge graph-based method for optimizing the proportion of blended fertilizers. It is applicable to on-site bulk blending scenarios at grassroots fertilizer blending stations or distribution centers. Under conditions where a blending order provides a list of candidate raw materials and target nutrient indicators, and the relative humidity of the operating environment is relatively high, the method requires that the blended material maintain a stable particle size structure during transportation and application, and outputs an executable fertilizer blending ratio. In this scenario, the ratio calculation typically uses a preset process upper limit or raw material usage specification upper limit to form a basic proportion upper limit, combined with a preset minimum proportion rule obtained by looking up a table according to the candidate raw material type to form a lower proportion limit. Then, the proportion of each candidate raw material is determined with the target nutrient indicators as the objective. The default premise of this approach is that the risk of hygroscopic agglomeration can be directly explained by the hygroscopic properties of the candidate raw materials, and that stable particle size structure can be maintained without explicitly characterizing multi-hop transmission paths and closed-loop interaction configurations of incompatible relationships.

[0029] However, when candidate raw material combinations are complex and incompatible relationships and eutectic interactions coexist, the influence between candidate raw materials may form multi-hop transmission paths along incompatible relationships and result in closed-loop interaction configurations. This couples hygroscopic properties with multi-hop relationships and amplifies the effect under the gain correction of eutectic interactions. At the same time, when the difference in particle size index is small, leading to increased particle size overlap, the risk of hygroscopic agglomeration is more easily amplified. If only the basic upper limit of proportion and the preset minimum proportion rule are relied upon, it is easy to get the result that "the upper limit and lower limit of proportion seem calculable but are actually more wrong": the upper limit of proportion may not be able to effectively suppress the risk of hygroscopic agglomeration, or the proportion constraint may be too conservative, making the proportion solution infeasible. Furthermore, the lower limit of proportion also needs to be determined under the condition that the degree of segregation risk does not exceed the preset segregation threshold. The degree of segregation risk is a global quantity at the candidate raw material list level, which further couples and conflicts between the lower limit of proportion of each raw material and the global segregation constraint and the existence of feasible solutions.

[0030] Therefore, the technical problem to be solved by the present invention is, under the premise of a given list of candidate raw materials and target nutrient indicators in a compounding order, how to use fertilizer knowledge graph to simultaneously characterize the hygroscopic properties and multi-hop relationship coupling of candidate raw materials, and under the joint constraints of stable particle size structure and controlled risk of hygroscopic agglomeration, generate upper and lower limits of proportion for each candidate raw material, thereby obtaining an executable compounding fertilizer ratio result.

[0031] Based on the above design, this invention constructs a complete process for optimizing the ratio of compound fertilizers based on knowledge graphs, consisting of steps S101 to S104 sequentially. (Refer to...) Figure 1 , Figure 1 This is a schematic diagram of the method flow of the present invention, which includes:

[0032] Step S101 involves acquiring candidate raw material graph data and constructing an input set. This step is used during the initialization phase of the blending operation to perform graph mapping and standardized encapsulation on external inputs, completing the basic input construction for the data preparation phase of this method. This step also reads the preset ratio parameter set X201 to obtain order parsing rules, missing data handling rules, particle size mapping rules, and relationship filtering rules. The processing standards for candidate raw material identifier parsing, node attribute reading, and relationship edge filtering are unified according to these rules. This step reads the blending order and environmental input set X101 and the fertilizer knowledge graph X102, parses the order requirements, and queries the corresponding raw material nodes and relationship edges in X102 according to the candidate raw material identifier list. It obtains the hygroscopic properties and particle size index of the candidate raw materials, and acquires the eutectic interaction relationships and incompatibility relationships between candidate raw materials. After data alignment, this is encapsulated into a candidate raw material graph input set R101, which is used in the subsequent step S102 to construct an interaction subgraph along the incompatibility relationships. X101 provides the candidate raw material list, target nutrient indicators, and relative humidity of the operating environment for this operation. X102 serves as a knowledge base, providing hygroscopic properties and particle size index fields for raw material nodes, as well as relation edges to characterize eutectic interactions and incompatibilities. R101, as a standardized spectral data carrier circulating within this method, registers key attributes and relationships using candidate raw materials as index objects, shielding the impact of external spectral structure differences on subsequent calculation methods.

[0033] Step S102, the construction of the interaction subgraph and the calculation of coupling indices, is used to perform multi-hop retrieval of incompatibility relationships among candidate raw materials and detect closed-loop interaction configurations after data preparation. This constructs an interaction subgraph for each candidate raw material and calculates the combined hygroscopic coupling strength and particle size overlap within the subgraph. This step reads the candidate raw material spectrum input set R101 and the preset ratio parameter set X201, and outputs the interaction subgraph result set R102 and the risk and particle size coupling result set R103. X201 provides parameters for multi-hop retrieval depth, path truncation, closed-loop deduplication, and risk fusion and mapping. R102 registers the candidate raw material nodes and incompatibility relationship edges covered by the multi-hop transmission path, using the starting candidate raw material as an index, and retains the eutectic interaction relationship edges between covered nodes and registers the closed-loop interaction configuration. R103 registers the combined hygroscopic coupling strength and particle size overlap corresponding to each candidate raw material for use in step S103 to generate the candidate raw material proportion window table R104.

[0034] Step S103 determines the candidate raw material proportion window. Based on the coupled risk quantification, this step performs upper limit shrinkage of the proportion based on moisture absorption risk and lower limit correction of the proportion based on segregation risk for each candidate raw material, thus completing the construction of the feasible region constraint for the proportion solution. This step reads the risk and particle size coupling result set R103, the candidate raw material spectrum input set R101, and the mixing order and environment input set X101. It also reads the proportion preset parameter set X201 to obtain the basic proportion upper limit caliber, shrinkage mapping caliber, segregation threshold caliber, minimum proportion rule, and downward adjustment correction rule. It calculates the proportion boundary values ​​and generates a candidate raw material proportion window table R104 for subsequent step S104 to read as the constraint solution entry point. The candidate raw material proportion window table R104 summarizes the proportion upper limit, proportion lower limit, and relative humidity of the working environment using candidate raw materials as indexes, serving as a unified window structure for this method to bear segregation and moisture absorption constraints. In this step, the particle size guide number used to calculate the degree of segregation risk is provided by the candidate raw material spectrum input set R101. The particle size guide number is a scalar attribute obtained from the fertilizer knowledge graph in step S101 and packaged according to a unified standard.

[0035] Step S104, the ratio optimization solution and result output, is used to perform constraint solving and infeasibility relaxation iteration with the goal of meeting the target nutrient index, based on the determined ratio window constraints, to generate the final ratio scheme of the blended fertilizer. This step reads the candidate raw material ratio window table R104, the blending order and environmental input set X101, and the ratio preset parameter set X201. Based on the ratio window constraints and normalization constraints, it calculates the target nutrient content based on the nutrient contribution coefficient and determines whether the target nutrient index is met. The nutrient contribution coefficient is obtained from the fertilizer knowledge graph according to the given X201 or the graph call caliber specified by X201. When the constraints are infeasible, the upper limit of the ratio remains unchanged, and the separation threshold caliber is updated according to the relaxation level given by X201 or the lower limit of the ratio adjustment correction rule caliber is updated. The lower limit of the ratio is recalculated and iteratively solved. The final output set of the mixed fertilizer ratio, R105, serves as the final output of this method. R105 records the final determined raw material proportions and the verification results of the target nutrient indicators, indexed by candidate raw materials. It can optionally record infeasibility markers, relaxation levels, and iteration counts for verification and traceability. The preset parameter set X201 provides update methods for relaxation levels, iteration limits, segregation threshold update rules, or downward adjustment correction rules, and provides the tolerance criteria required for determining target nutrient satisfaction.

[0036] For ease of understanding, the following embodiments are described under a unified system architecture, which can be modified accordingly to meet actual needs. The system consists of a candidate raw material chromatogram data acquisition and input set construction unit, an interaction subgraph construction and coupling index calculation unit, a proportion window constraint determination and window table generation unit, a proportion optimization solution and result output unit, and a preset parameter storage area. (Refer to...) Figure 2, Figure 2 This is a schematic diagram of the system structure of the present invention.

[0037] Specifically, the system inputs include at least the blending order and environmental input set X101, the fertilizer knowledge graph X102, and the set of preset ratio parameters X201.

[0038] The preset parameter storage area stores and maintains the parameter groups and rule definitions in the preset parameter set X201, and provides a unified reading interface to each functional unit to avoid duplicate definitions or definition drift of the same parameter in different units. The parameter groups include at least the order parsing and alignment parameter group, the missing data handling and granular mapping parameter group, the multi-hop retrieval parameter group, the loop closure detection parameter group, the risk fusion parameter group, the upper limit contraction parameter group, the separation constraint parameter group, the relaxation level and iteration upper limit definition, and the tolerance definition for target nutrient satisfaction. When executing the corresponding method steps, each functional unit reads the required parameters from the preset parameter storage area according to the parameter groups to determine the retrieval depth, loop closure deduplication, risk fusion weight, contraction mapping rule, separation threshold, minimum proportion rule, downward adjustment rule, relaxation level, iteration upper limit, and tolerance threshold, thereby ensuring the reproducibility of system operation and the consistency of parameter definitions.

[0039] The candidate raw material map data acquisition and input set construction unit is used to execute method step S101. It receives the blending order and environmental input set X101 and the fertilizer knowledge graph X102, and reads the order parsing, missing data processing, particle size mapping and relationship filtering caliber from the preset parameter storage area, and outputs the candidate raw material map input set R101.

[0040] The interactive subgraph construction and coupling index calculation unit is used to execute method step S102. It receives the candidate raw material spectrum input set R101 and reads the multi-hop retrieval parameter group, closed-loop detection parameter group and risk fusion parameter group from the preset parameter storage area, and outputs the interactive subgraph result set R102 and the risk and particle size coupling result set R103.

[0041] The proportion window constraint determination and window table generation unit is used to execute method step S103. It receives the risk and particle size coupling result set R103, the candidate raw material spectrum input set R101 and the mixing order and environment input set X101, and reads the upper limit shrinkage parameter group and the segregation constraint parameter group from the preset parameter storage area. The segregation constraint parameter group includes at least the segregation threshold minimum proportion rule and the downward adjustment correction rule. It outputs the candidate raw material proportion window table R104.

[0042] The ratio optimization solution and result output unit is used to execute method step S104. It receives the candidate raw material ratio window table R104 and the mixing order and environment input set X101, and reads the relaxation level, iteration upper limit and tolerance caliber from the preset parameter storage area. When it is not feasible, it keeps the ratio upper limit unchanged and relaxes the segregation constraint to update the ratio lower limit before iterating and solving, and outputs the mixed fertilizer ratio result set R105.

[0043] The implementation process and operational effects of the method of the present invention will be described in detail below with reference to specific embodiments. It should be understood that the embodiments are only used to illustrate the technical solution of the present invention, and not to limit it. The relevant steps, parameters, and module divisions can be appropriately adjusted without changing the essence of the invention.

[0044] The implementation process of step S101 includes parsing the candidate raw material list, retrieving and mapping node attributes, retrieving and extracting relation edges, and encapsulating and writing back data.

[0045] Before execution, the relevant rule items for this step are read from the preset parameter set X201 and executed accordingly. These rule items include at least order parsing rule items, missing item handling rule items, granularity mapping rule items, and relationship filtering rule items. The order parsing rule items are used to locate the candidate raw material entry field, candidate raw material identifier field, target nutrient index field, and relative humidity field in X101, and provide deduplication and sorting rules, as well as a marking strategy for unidentified identifiers. It also provides the alignment criteria between the candidate raw material identifier and the raw material node identifier field in the fertilizer knowledge graph X102. The missing item handling rule items are used to determine missing hygroscopic properties and missing granularity guide numbers, and provide default value completion criteria and missing mark criteria. The granularity mapping rule items are used to uniformly convert multiple granularity representation methods in X102 into granularity guide numbers to ensure consistency in subsequent calculations. The particle size guide number is generated from the particle size distribution detection data of candidate raw materials using preset mapping rules and written into the fertilizer knowledge graph X102. The preset mapping rules specify at least the method for dividing the particle size distribution intervals and map the statistical characteristics of the particle size distribution to scalar values ​​using a lookup table mapping method. The relationship filtering rule item is used to limit the processing to only two types of relationship edges: eutectic interaction relationships and incompatible relationships, and to limit the retention to only relationship edges whose endpoints all fall within the candidate raw material identifier list. It also provides the processing criteria for direction processing, merging of duplicate relationships, self-loop processing, and empty relationship lists. This step is only performed according to the criteria given in X201, and no new coefficients or threshold values ​​are introduced in this step.

[0046] In the candidate raw material list parsing, starting from the mixing order and environmental input set X101, a list of candidate raw material items is extracted according to the field positioning criteria given by X201. A candidate raw material identifier list is then obtained according to the identifier extraction rules given by X201. Finally, a standard candidate raw material identifier list is generated according to the deduplication and sorting rules given by X201. Simultaneously, the relative humidity and target nutrient index of the working environment are extracted from X101 according to the field positioning criteria given by X201 and temporarily stored as execution context parameters for this step, to be read by subsequent steps. If there are missing fields or unidentified candidate raw material identifiers, a parsing anomaly flag is registered according to the marking strategy given by X201. Optionally, the parsing anomaly flag can be written to R101. The parsing anomaly flag is not a required input field for subsequent steps.

[0047] In node attribute retrieval and mapping, the standard candidate raw material identifier list is first normalized according to the identifier alignment standard given by the preset parameter set X201, resulting in a graph query identifier list for graph lookup. The identifier alignment standard specifies at least the character normalization method, alias equivalence rules, and mapping method from candidate raw material identifiers to raw material node identifiers. Then, the corresponding raw material node is queried in the fertilizer knowledge graph X102 using the graph query identifier list as an index. For candidate raw materials that do not match the query, a mismatch marker is registered according to the mismatch handling rules given by X201, and the entity boundary of the candidate raw material in the candidate raw material list is retained. The mismatch marker is only used to indicate graph missingness and does not trigger candidate raw material removal. For candidate raw material nodes that match the query, their hygroscopic attribute field and particle size guidance number field are read. The hygroscopic attribute includes at least one of the critical relative humidity indicator and the hygroscopic rate indicator. The critical relative humidity indicator and the hygroscopic rate indicator are experimental calibration results obtained according to preset test specifications or historical operating condition statistics obtained according to preset statistical standards, and are written into the fertilizer knowledge graph X102. The particle size guide number corresponds to a scalar value generated from the particle size distribution detection data of candidate raw materials using a preset mapping rule. This preset mapping rule includes at least the particle size distribution interval division method and the mapping method that looks up the statistical characteristics of the particle size distribution and converts them into scalar values. If X102 has multiple particle size characterization methods, they are uniformly converted into particle size guide numbers according to the particle size mapping rule given in X201 and written to the node attribute record. For cases where hygroscopic properties or particle size guide numbers are missing, a default value is written according to the missing value handling rule given in X201, and a missing flag can be optionally registered. The missing flag and default value are only used to indicate data reliability and do not change the entity boundaries of the candidate raw material list.

[0048] In relation edge retrieval and extraction, based on the relation edge set of the fertilizer knowledge graph X102, incompatible relations are those that characterize stability conflicts or adverse reactions between two candidate raw materials under mixing conditions, while eutectic interaction relations are those that characterize the formation of eutectic or induced hygroscopic interactions between two candidate raw materials. First, only eutectic interaction relations and incompatible relations are retained according to the relation type whitelist given by X201. Then, according to the internal relation determination conditions given by X201, only relations whose endpoints both fall within the candidate raw material identifier list are retained. A standard relation edge set is generated according to the direction processing rules and duplicate relation merging rules given by X201. Relationship type identifiers and endpoint information are registered for incompatible relations, and for eutectic interaction relations. If the eutectic interaction relation list or incompatible relation list corresponding to a candidate raw material is empty, it is written into an empty list according to the empty relation list definition given by X201 and optionally registered as an empty marker. The coefficients required for the risk weight of the relation edge and the correction of the eutectic gain are given by X201 and used in step S102. This step only registers the endpoints of the relation edge and the relation type identifier, and does not extract the above coefficients on the spectrum side.

[0049] In the data encapsulation and write-back process, node attribute records and relation edge sets are aligned and assembled according to the preset structure of the candidate raw material graph input set R101. This ensures that each candidate raw material identifier is associated with its corresponding hygroscopic properties, particle size index, eutectic interaction list, and incompatible relationship list, forming a structured record with the candidate raw material as the key. The candidate raw material identifier, hygroscopic properties, particle size index, eutectic interaction list, and incompatible relationship list are required fields in R101. Missing data flags, parsing anomaly flags, and X201 rule version identifiers are optional fields in R101. Missing data flags and parsing anomaly flags are only used to indicate data credibility and order parsing anomalies, while the X201 rule version identifier is only used for reproduction and auditing. None of these optional fields are required input fields for subsequent steps S102 and S103. Finally, the assembled data is written into the candidate raw material graph input set R101. The R101 produced in this step provides a cleaned and standardized spectral data foundation for the subsequent step S102 to construct the interactive subgraph, so that subsequent steps do not need to repeatedly query the original spectral data and avoid caliber drift.

[0050] The implementation process of step S102 includes parameter reading and index initialization, multi-hop retrieval and loop closure detection, interactive subgraph construction and write-back, combined hygroscopic coupling strength calculation, particle size overlap degree calculation and result set write-back.

[0051] In parameter reading and index initialization, the maximum number of hops, path truncation rules, closed-loop length upper limit, closed-loop deduplication rules, incompatibility risk weight mapping rules, eutectic gain correction rules, fusion weights and normalization methods, the mapping rules from shortest path length to weights, and the mapping rules from overall difference to particle size overlap are read from the preset parameter set X201. These rules in the preset parameter set X201 are stored in the form of parameter keys and parameter values. Parameter values ​​include at least one of enumeration rules, thresholds, and mapping rules. During step execution, intermediate quantities are read by parameter keys and calculated according to the corresponding rules; they are not required to be expanded in tabular form in the specification. Subsequently, the standard candidate raw material identifier list and the corresponding hygroscopic properties and particle size index numbers of the candidate raw materials are read from the candidate raw material spectrum input set R101, and the incompatibility relationship list and eutectic interaction relationship list between candidate raw materials are read to form the input index structure for subsequent retrieval and calculation.

[0052] In multi-hop retrieval and loop closure detection, for each candidate raw material identifier in the standard candidate raw material identifier list, this candidate raw material is taken as the starting candidate raw material. Multi-hop retrieval is then performed along the incompatible relation list within R101 to obtain a set of multi-hop transmission paths originating from the starting candidate raw material. The number of hops in the multi-hop retrieval does not exceed the maximum number of hops given by X201, and paths with duplicate nodes are truncated according to the path truncation rules given by X201. Based on the set of multi-hop transmission paths, it is detected whether there is a closed loop path formed by the starting candidate raw material returning to itself via incompatible relations. Closed loop paths that satisfy the condition that the closed loop length does not exceed the upper limit of the closed loop length given by X201 are registered as a closed loop interaction configuration candidate set. The closed loop interaction configuration candidate set is then deduplicated according to the closed loop deduplication rules given by X201. Deduplication is determined based at least on the consistency of the node sequence of the closed loop path, thus obtaining the closed loop interaction configuration set of the starting candidate raw material.

[0053] In the construction and write-back of the interaction subgraph, the set of candidate raw material nodes and the set of incompatible edges covered by the multi-hop propagation path set are extracted. The starting candidate raw material, together with the above-mentioned node set and incompatible edge set, are used to construct the skeleton of the interaction subgraph for that starting candidate raw material. Subsequently, based on the eutectic interaction relationship list in R101, eutectic interaction relationship edges whose endpoints all fall within the above-mentioned node set are retrieved and retained, and these are incorporated into the interaction subgraph to form a complete interaction subgraph. The list of nodes, the list of incompatible edges, the list of eutectic interaction edges, and the set of closed-loop interaction configurations in the interaction subgraph are written into the interaction subgraph result set R102, completing the structured registration indexed by the starting candidate raw material.

[0054] In the calculation of the combined hygroscopic coupling strength, for each starting candidate raw material, the calculation scope is based on its corresponding interaction subgraph. The cumulative risk term is calculated according to the caliber given by X201. The risk term and hygroscopic attribute term are amplified, and a normalized weighted sum is performed according to the normalization method and weighting weight given by X201 to obtain the combined hygroscopic coupling strength. The cumulative risk term is determined at least as follows: for each incompatible relation edge on the multi-hop transmission path within the interaction subgraph, the risk weight is obtained according to the relation type identifier of the incompatible relation edge and the incompatible relation risk weight mapping rule given by X201. The risk weights are then aggregated according to the path aggregation method and multi-path aggregation method given by X201 to obtain the cumulative risk term. In an optional implementation, the cumulative risk term is the sum of the risk weights of incompatible relation edges on each multi-hop transmission path within the interaction subgraph, and the summation is performed on all multi-hop transmission paths of the same starting candidate raw material to obtain the cumulative risk term. The amplification risk term should be determined at least as follows: based on the number and length of closed loops in the closed-loop interaction configuration set, and the amplification amount should be calculated according to the amplification factor given by X201. The hygroscopic property term should be determined at least as follows: the hygroscopic property of the starting candidate raw material is read from R101 as the basic hygroscopic amount, and based on the eutectic interaction relationship edges in the interaction subgraph, the gain correction amount is determined according to the eutectic gain correction rule given by X201, and the gain correction is applied to the basic hygroscopic amount to obtain the corrected hygroscopic property term.

[0055] In the particle size overlap calculation and result set write-back, for each starting candidate material, the calculation scope is based on its corresponding interaction subgraph. The particle size guide number of the starting candidate material is read from R101, and the particle size guide number of other candidate materials in the interaction subgraph is read one by one to form a particle size guide number difference set. For each difference in the difference set, the weight is determined according to the shortest path length from the starting candidate material to the corresponding candidate material in the interaction subgraph according to the mapping rule given by X201, and the weighted absolute difference mean of the difference set is calculated as the overall difference. The shortest path length is the minimum number of hops calculated along the incompatible relationship edge in the interaction subgraph, and the direction processing caliber of the shortest path length is given by X201. Then, the overall difference is mapped to the particle size overlap degree according to the mapping function given by X201. The smaller the overall difference, the greater the particle size overlap degree. Finally, the combined hygroscopic coupling strength and particle size overlap degree corresponding to each candidate material are written into the risk and particle size coupling result set R103 and associated with the candidate material identifier for registration. Optionally, if X201 requires intermediate items to be retained for review, the cumulative risk item can be registered simultaneously in R103, and the intermediate value markers of the risk item and the corrected moisture-wicking property item can be amplified. However, the intermediate item is not a required input field for subsequent steps.

[0056] The implementation process of step S103 includes upper limit shrinkage calculation, segregation risk assessment, lower limit correction calculation, and window table summary and write-back.

[0057] In the upper limit shrinkage calculation, for each candidate raw material, the combined hygroscopic coupling strength and particle size overlap are read from the risk and particle size coupling result set R103, and the relative humidity of the working environment is obtained from the mixing order and environmental input set X101. The basic proportion upper limit is obtained from the ratio preset parameter set X201 according to the candidate raw material identifier, or from X201 according to the raw material type identifier corresponding to the candidate raw material, wherein the selection rules by identifier or by type are given by X201. The basic proportion upper limit is the preset process upper limit or the raw material usage specification upper limit. The hygroscopic shrinkage coefficient is a coefficient used to perform shrinkage on the basic proportion upper limit according to the combined hygroscopic coupling strength, particle size overlap, and relative humidity of the working environment. The hygroscopic shrinkage coefficient is obtained according to the mapping rules given by X201. Subsequently, the proportion upper limit is obtained by performing shrinkage calculation on the basic proportion upper limit based on the hygroscopic shrinkage coefficient. As the particle size overlap increases, the hygroscopic shrinkage coefficient increases, thereby reducing the proportion upper limit.

[0058] In the segregation risk assessment, the particle size guide number of all candidate raw materials in the candidate raw material list is extracted from the candidate raw material spectrum input set R101 to form a particle size guide number set of the candidate raw material list. The segregation risk level is calculated according to the ratio of the standard deviation to the mean of the particle size guide number of each candidate raw material in the candidate raw material list. The segregation risk level is a global quantity at the candidate raw material list level and is used to characterize the particle size structure stability risk of the current raw material combination.

[0059] In the lower limit correction calculation, the calculated segregation risk level is compared with the segregation threshold given by the pre-set parameter set X201. The model of the blending equipment is obtained from the blending order and environmental input set X101. If X101 does not provide it, it is pre-given by X201. The corresponding segregation threshold is obtained from X201 based on the model of the blending equipment for comparison with the segregation risk level. Under the condition that the segregation risk level does not exceed the segregation threshold, the raw material type identifier corresponding to the candidate raw material is preferentially obtained by mapping the candidate raw material identifier by the pre-set parameter set X201, or provided by the raw material node attribute field of the fertilizer knowledge graph and carried in the candidate raw material graph input set R101. Then, according to the minimum proportion rule given by the pre-set parameter set X201, the minimum proportion value is obtained according to the raw material type identifier, and the minimum proportion value is adjusted downward according to the downward adjustment rule given by the pre-set parameter set X201 to obtain the lower limit of the proportion. The downward adjustment rule is the rule used to adjust the minimum proportion value downward to obtain the lower limit of the proportion. The conservative lower limit is the lower limit determination method used when the segregation risk level exceeds the segregation threshold. When the segregation risk level exceeds the segregation threshold, the lower limit of the proportion is output according to the conservative lower limit method given by X201 to form the candidate raw material proportion window table R104. The conservative lower limit method includes at least setting the downward adjustment coefficient to zero so that the lower limit of the proportion degenerates to the minimum proportion value.

[0060] In the summary and write-back of the window table, the upper and lower limits of the proportion calculated for each candidate raw material, along with the relative humidity of the working environment, are uniformly written into the candidate raw material proportion window table R104 and associated with the candidate raw material identifier. This allows the subsequent step S104 to solve the mixed fertilizer ratio that meets the target nutrient index within the proportion window constraint without having to repeatedly calculate the upper and lower limits of the proportion.

[0061] To ensure the determinism of boundary calculations, this embodiment provides the calculation methods for the upper limit of proportion, the degree of segregation risk, and the lower limit of proportion, and clarifies the parameter form and calling method of the preset parameter set X201 in the calculation. Let... This is the upper limit of the percentage. The upper limit of the basic proportion, The coefficient of moisture absorption and shrinkage is... To combine the hygroscopic coupling strength, For the degree of overlap in particle size, The relative humidity of the working environment, To determine the degree of risk of separation, This is the set of granularity guide numbers for the candidate raw material list. This represents the quantity of candidate raw materials.

[0062] Basic percentage upper limit Given by X201, and can be obtained by candidate raw material identifier or raw material type identifier. If X201 gives both a process upper limit and a usage specification upper limit, the lower of the two values ​​shall be taken as the basic proportion upper limit. Moisture absorption shrinkage coefficient. Obtained by the mapping rules given by X201, the mapping rules include at least the combined hygroscopic coupling strength segmented point set, the particle size overlap degree segmented point set, the relative humidity of the working environment segmented point set, and the corresponding coefficient value set, and specify the interpolation method or the nearest value method and the upper limit of the coefficients. To facilitate the use of standardized terminology, , , Normalize according to the upper and lower bounds given by X201. , , Normalization is performed linearly and truncated to the range of zero to one: , , .

[0063] The moisture absorption shrinkage coefficient is obtained from the mapping rule of X201. ;in, Let X201 be the piecewise mapping function given by X201. To meet the aperture requirement where increased particle size overlap leads to a decrease in the upper limit of the proportion, X201 should ensure that... When unchanged Follow and Monotonic and unreduced. The upper limit of the percentage is calculated based on the contraction coefficient and the range is truncated. .

[0064] The degree of segregation risk is calculated as the ratio of the standard deviation to the mean of the granularity guide number set. The mean and standard deviation can be calculated using either the population or sample caliber, the choice of which is given by X201; if not given, the population caliber is used. , .

[0065] To avoid the ratio becoming unstable due to an excessively small mean, the mean is set to the lower limit given by X201. Enforcement protection, specifically And based on this, the degree of segregation risk is calculated. .

[0066] The lower limit of the percentage is determined based on the minimum percentage rule and the downward adjustment rule given by X201. Let... This is the minimum percentage value, determined by the raw material type identifier according to the minimum percentage rule given in X201. Let... The segregation threshold is obtained from X201 according to the mixing equipment model. When At that time, the minimum percentage value is adjusted downward according to the adjustment rule given in X201 to obtain the lower limit of the percentage. The adjustment rule adopts a downward adjustment coefficient that monotonically does not increase with the degree of separation risk. And limit it to zero to Within the range, and when The closer The current adjustment is weaker. Specifically, , ;in Let X201 be the piecewise mapping function. When At that time, the lower limit of the proportion is output according to the conservative lower limit caliber given by X201 to ensure that the candidate raw material proportion window table R104 can be generated. The conservative lower limit caliber includes at least the following: This causes the lower limit of the proportion to degenerate into the minimum proportion value, i.e. .

[0067] The implementation process of step S104 includes the construction of constraints and decision criteria, solving under window constraints and feasibility branching, and infeasibility relaxation iteration and result encapsulation output.

[0068] In the construction of constraints and decision criteria, the upper and lower limits of the proportion of each candidate raw material are extracted from the candidate raw material proportion window table R104 as hard constraints of the feasible region for solving, and a normalization constraint is established that the sum of the proportions of all candidate raw materials equals one. Simultaneously, target nutrient indicators are read from the blending order and environmental input set X101. Target nutrient indicators include at least the set of target nutrient types and the target value or target range corresponding to each target nutrient. The nutrient contribution coefficient is preferentially given by the pre-set ratio parameter set X201 and mapped according to the candidate raw material identifier and the target nutrient type. If X201 does not provide a nutrient contribution coefficient, it is obtained from the fertilizer knowledge graph and called according to the same mapping criteria. Let the first... The proportion of each candidate raw material is , No. The nutrient contribution coefficient of the target nutrient is Then the first The calculated value of the target nutrient is .

[0069] When the target nutrient index is a target range, let the lower bound of that target range be... The upper boundary is Then the condition for satisfying the target range is: .

[0070] When the target nutrient index is the target value, let the target value be... The tolerance threshold is configured in X201 according to the target nutrient type. When X201 provides only a uniform tolerance threshold, all target nutrients share the same tolerance threshold. The criteria for satisfying the target value are as follows: .

[0071] The above criteria are used as the standard for determining a feasible solution. When multiple feasible solutions exist, the final ratio solution is selected according to the secondary objective given by X201. The secondary objective takes the form of minimizing nutrient deviation. Let the reference target value be... When the target nutrient index is the target value, take When the target nutrient index is within the target range, take Let the deviation weights be... Given X201, the secondary objective is .

[0072] In step S104, the target nutrient index constraint is set as follows: The calculation formula and corresponding judgment formula shall be used as the standard. The solver will only determine the ratio solution as a feasible solution when the ratio window constraint and the target nutrient index constraint are satisfied.

[0073] In the solution process under window constraints and the feasibility branch, the proportion window constraint, normalization constraint, and target nutrient index constraint serve as boundaries, and the secondary target caliber given by X201 is used as the selection criterion for constraint optimization. A linear programming solver is used when both constraints and objectives are linear; a nonlinear programming solver is used when nonlinear mappings or nonlinear objectives exist. The stopping condition for the solution process includes at least an upper limit on the number of iterations and a convergence tolerance, given by the pre-set parameter set X201. The candidate raw material proportion combinations obtained from the solution are used as the ratio solution, and the feasibility solution is checked against the aforementioned feasible solution criteria to see if they satisfy the target nutrient index constraint and proportion window constraint. If satisfied, the result is encapsulated and output; otherwise, the infeasibility relaxation strategy is triggered, and the infeasibility relaxation iteration process begins.

[0074] In the infeasible relaxation iteration and result encapsulation output, iterative relaxation is performed according to the relaxation level upper limit and iteration upper limit given by the preset parameter set X201. The relaxation level is a non-negative integer, initially set to zero, and incremented in each iteration until the relaxation level upper limit is reached. In each iteration, the upper limit of the proportion of each candidate raw material in the candidate raw material proportion window table R104 remains unchanged.

[0075] Separation risk level The granularity guide number set from the candidate raw material list is always calculated according to the caliber given in step S103, and the segregation risk level is not considered as a relaxation target. Each iteration only updates the segregation threshold caliber by default. Specifically, X201 provides an updated threshold table or update rule from the relaxation level to the segregation threshold, and the updated threshold is obtained by reading it according to the current relaxation level. Then, the calculation method of the lower limit of the proportion in step S103 is reused to recalculate the lower limit of the proportion of each candidate raw material. The updated lower limit of the proportion is written back to R104 and the solution and feasibility determination are re-executed.

[0076] In a further optional implementation, if no feasible solution satisfying the target nutrient index is found after reaching a preset relaxation level, the adjustment rule for the lower limit of the proportion is allowed to be further updated. Specifically, X201 provides an update rule or rule parameter table from the relaxation level to the adjustment rule, and the updated rule is obtained by reading it according to the current relaxation level. Or its parameters, or read the updated The lower limit of the proportion is recalculated and written back to R104 before resolving. Once a feasible solution is obtained during the iteration process, the final determined proportions of each candidate raw material, the calculated values ​​of each target nutrient, and the verification results of the target nutrient indicators are written into the compound fertilizer ratio result set R105, and the final relaxation level and the actual number of iterations are recorded for traceability. If no feasible solution is found after reaching the iteration limit, an infeasibility flag is recorded in R105 according to the failure handling criteria given in X201, and step S104 ends.

[0077] To illustrate the constraint effect of this invention on the solution of proportions when a closed-loop interaction configuration exists within the interaction subgraph, based on the same candidate raw material list, the same relative humidity of the working environment, and the same preset parameter set for proportions, the proportions of a certain candidate raw material were solved using both the traditional strategy and the strategy of this invention, and the results are as follows: Figure 3 and Figure 4 As shown in the figure. The horizontal axis represents the proportion of candidate raw materials, and the vertical axis represents the combined hygroscopic coupling strength and the degree of segregation risk, respectively. The shaded diagonal lines represent the feasible proportion window determined by the lower and upper limits of the proportion, the cross-shaded areas represent the high region of combined hygroscopic coupling strength triggered by the closed-loop interaction configuration, the dashed lines represent the preset segregation threshold, and the narrow boxes represent the finally determined proportion of candidate raw materials. Figure 3 It is evident that traditional strategies, while meeting target nutrient indicators, do not impose explicit exclusion constraints on the high-hygroscopic coupling intensity region. The determined candidate raw material proportions are prone to falling into or approaching this high region, thus amplifying the hygroscopic-related risks as the proportion changes under the same operating environment relative humidity. Figure 4 As can be seen, the present invention determines the upper limit of the proportion based on the combined hygroscopic coupling strength and particle size overlap obtained by knowledge graph calculation, and determines the lower limit of the proportion under the condition that the segregation risk does not exceed the preset segregation threshold, forming a candidate raw material proportion window table and using it as a constraint to participate in the proportion solution, so that the finally determined candidate raw material proportion is within the feasible proportion window and avoids the high region of combined hygroscopic coupling strength, thereby reducing the probability of risk amplification triggered by closed-loop interaction configuration and reducing the occurrence of re-solution and relaxation iteration.

[0078] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0079] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0080] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0081] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0082] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing the ratio of compound fertilizers based on knowledge graphs, characterized in that, include: S101, obtain the candidate raw material list, target nutrient index and relative humidity of the working environment, and obtain the hygroscopic properties and particle size index of each candidate raw material, as well as the eutectic interaction and incompatibility relationship between the candidate raw materials in the fertilizer knowledge graph; S102, for each candidate raw material, starting from the candidate raw material, a multi-hop transmission path is retrieved along the incompatibility relationship and a closed-loop interaction configuration is detected to form an interaction subgraph of the candidate raw material; within the interaction subgraph, based on the hygroscopic properties of the candidate raw material and other candidate raw materials in the interaction subgraph and combined with the eutectic interaction relationship in the interaction subgraph, the combined hygroscopic coupling strength of the candidate raw material is determined; and based on the particle size index of the candidate raw material and the particle size index of other candidate raw materials in the interaction subgraph, the particle size overlap of the candidate raw material is determined. S103, for each candidate raw material, the upper limit of the proportion of the candidate raw material is determined based on the combined hygroscopic coupling strength, particle size overlap and relative humidity of the working environment. The greater the particle size overlap, the smaller the upper limit of the proportion. The degree of segregation risk is determined by the dispersion of the particle size index of each candidate raw material in the candidate raw material list. The lower limit of the proportion of the candidate raw material is determined based on the comparison between the degree of segregation risk and the preset segregation threshold. When the degree of segregation risk does not exceed the preset segregation threshold, the lower limit of the proportion is determined according to the adjustment rule. When the degree of segregation risk exceeds the preset segregation threshold, the lower limit of the proportion is determined according to the conservative lower limit. The upper limit and lower limit of the proportion corresponding to each candidate raw material are summarized into a proportion window table. S104 uses the proportion window table as a constraint and aims to ensure that the obtained ratio meets the target nutrient index to determine the proportion of each candidate raw material. When there is no feasible solution that meets the target nutrient index, the upper limit of the proportion of each candidate raw material remains unchanged. The separation threshold caliber is updated according to the relaxation level or the lower limit of the proportion is updated according to the adjustment rule caliber. Based on this, the lower limit of the proportion of each candidate raw material is recalculated and the solution is re-solved to output the mixed fertilizer ratio result.

2. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 1, characterized in that, The fertilizer knowledge graph includes at least a set of raw material nodes with candidate raw materials as entity nodes, attribute fields that record the hygroscopic properties and particle size index of candidate raw materials, and relationship edges used to characterize the eutectic interaction relationship and incompatibility relationship between candidate raw materials. The incompatibility relationship is a relationship edge that characterizes the stability conflict or adverse reaction of two candidate raw materials under mixing conditions, and the eutectic interaction relationship is a relationship edge that characterizes the formation of eutectic or the induction of hygroscopic related interactions between two candidate raw materials.

3. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 1, characterized in that, The nutritional contribution coefficient of each candidate raw material is also obtained. The nutritional contribution coefficient is used to characterize the content of target nutrients per unit mass of candidate raw materials. In S104, the target nutrient content corresponding to the obtained ratio is calculated based on the nutritional contribution coefficient and compared with the target nutrient index to establish the constraint conditions that meet the target nutrient index.

4. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 2, characterized in that, The hygroscopic properties include at least one of the critical relative humidity indicator and the hygroscopic rate indicator, and the critical relative humidity indicator and the hygroscopic rate indicator are experimental calibration results obtained according to the preset test specifications or historical operating condition statistical results obtained according to the preset statistical caliber, and are written into the fertilizer knowledge graph.

5. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 2, characterized in that, The particle size guide number is generated from the particle size distribution detection data of candidate raw materials through preset mapping rules and written into the fertilizer knowledge graph. The preset mapping rules at least specify the particle size distribution interval division method and map the statistical characteristics of particle size distribution into scalar values ​​by table lookup mapping.

6. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 2, characterized in that, In S102, the multi-hop transmission path is searched along the incompatible relation, the search depth does not exceed the preset maximum number of hops, and the path with duplicate nodes is truncated; the closed-loop interaction configuration is the closed-loop path formed by the starting candidate raw material returning to itself through the incompatible relation, the length of the closed-loop path does not exceed the preset upper limit of the closed-loop length, and the consistency of the node sequence of the closed-loop path is used as the basis for determining duplicate closed loops and deduplicating the count. The interaction subgraph includes the starting candidate raw material and the candidate raw material nodes covered by the above path, as well as the incompatible relationship edges, and retains the eutectic interaction relationship edges between the candidate raw material nodes.

7. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 6, characterized in that, The combined hygroscopic coupling strength is obtained by fusing the cumulative risk term, the amplified risk term, and the hygroscopic attribute term. The cumulative risk term is the sum of the risk weights of incompatible relation edges on the multi-hop transmission path within the interaction subgraph. The amplified risk term is the amplification amount calculated by the number and length of closed loop paths within the interaction subgraph using a preset amplification factor. The hygroscopic attribute term is determined by the hygroscopic properties of the starting candidate raw material. The eutectic interaction relation is used to apply a preset gain coefficient to the gain correction of the hygroscopic attribute term. The fusion method is normalized weighted summation, and the weights are given by a preset weight table.

8. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 6, characterized in that, The degree of particle size overlap is formed by the difference set between the particle size guide number of the starting candidate raw material and the particle size guide number of other candidate raw materials in the interaction subgraph. The difference set is obtained by looking up a table through a preset mapping function. The overall difference of the difference set is represented by the mean of the absolute difference. The smaller the overall difference, the greater the degree of particle size overlap. The difference set is aggregated by weighted mean. The weight is determined by the shortest path length from the starting point to the corresponding candidate raw material node in the interaction subgraph.

9. The method for optimizing the ratio of compound fertilizers based on knowledge graphs according to claim 1, characterized in that, The segregation risk level is determined by the dispersion of the particle size index of each candidate raw material in the candidate raw material list. The dispersion is the ratio of the standard deviation to the mean of the particle size index. The segregation threshold is determined by a preset threshold table corresponding to the model of the mixing equipment. The lower limit of the proportion is determined according to a preset minimum proportion rule. The preset minimum proportion rule is obtained by looking up the minimum proportion value according to the candidate raw material type, and then adjusting the minimum proportion value downward according to the adjustment rule when the segregation risk level does not exceed the segregation threshold. The lower limit of the proportion is adjusted more weakly as the segregation risk level is closer to the segregation threshold. When the segregation risk level exceeds the segregation threshold, the lower limit is adjusted downward according to the following rules. The conservative lower limit is used to determine the lower limit of the proportion. The conservative lower limit includes at least making the downward adjustment degenerate to zero so that the lower limit of the proportion is equal to the minimum proportion value. When there is no feasible solution that satisfies the target nutrient index, the upper limit of the proportion remains unchanged, the relaxation level is a non-negative integer, the separation threshold is updated according to the relaxation level or the downward adjustment rule of the lower limit of the proportion is updated, and the lower limit of the proportion is recalculated accordingly. The upper and lower limits of the proportion of each candidate raw material are summarized into a proportion window table. The proportion window table includes at least the candidate raw material identifier, the upper limit of the proportion, the lower limit of the proportion, and the relative humidity of the working environment.

10. A knowledge graph-based system for optimizing the proportion of compound fertilizers, used to implement the knowledge graph-based method for optimizing the proportion of compound fertilizers as described in any one of claims 1-9, characterized in that, It includes a candidate raw material chromatogram data acquisition and input set construction unit, an interactive subgraph construction and coupling index calculation unit, a proportion window constraint determination and window table generation unit, a proportion optimization solution and result output unit, and a preset parameter storage area; The candidate raw material spectrum data acquisition and input set construction unit, the interaction subgraph construction and coupling index calculation unit, the proportion window constraint determination and window table generation unit, and the proportion optimization solution and result output unit are respectively used to execute S101, S102, S103 and S104; the preset parameter storage area is used to store and provide the above units with the maximum number of hops, the upper limit of closed loop length, the risk weight mapping rule, the eutectic gain correction rule, the fusion weight, the proportion upper limit shrinkage mapping rule, the segregation threshold, the minimum proportion rule, the down-adjustment correction rule, the relaxation level, the iteration upper limit and the tolerance threshold.

Images