Fertilization management decision method and system based on multi-modal knowledge graph

Abstract

The application discloses a kind of based on multimodal knowledge graph's fertilization management decision method and system, it is related to data processing technical field, method includes: obtaining natural language label and fertilization dataset;In first time, obtain first soil external environment information, obtain first matching word frequency, obtain the first similarity of first soil external environment information and sub data group, according to the highest first similarity of sub fertilization data, carry out fertilization management;Obtain environmental data variable threshold, obtain second time, in second time, obtain second soil external environment information, obtain real-time environmental data variable, judge whether it is more than environmental data variable threshold;If more than, obtain second similarity, according to the highest second similarity of sub fertilization data, carry out fertilization management;If not more than, record current soil internal environment information and maintain current fertilization management strategy.The application has the advantages of improving fertilizer utilization, reducing environmental stress risk, reducing the advantages of algorithm power redundancy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a fertilization management decision method and system based on a multi-modal knowledge graph. BACKGROUND

[0002] In existing agricultural production, especially in the fertilization management process of facility agriculture (such as greenhouse), there are problems such as poor dynamic environment adaptability, insufficient decision accuracy, and low resource utilization rate. In order to solve these problems, the existing method usually formulates fertilization strategies based on the static crop growth model in the knowledge graph or the environmental monitoring rules of fixed threshold. Such methods not only cannot cope with frequent fluctuations in environmental parameters such as light, temperature, humidity, and carbon dioxide concentration, but also have the following defects.

[0003] Firstly, the existing method only selects the initial fertilization decision according to a single environmental parameter threshold, which is poor in environmental adaptability. For example, the initial fertilization data is roughly mapped through the crop name in the knowledge graph without quantifying the semantic similarity of the crop label, and the synergistic effect of multiple environmental factors (such as humidity, light, and carbon dioxide concentration) is ignored, resulting in a disconnection between the initial fertilization strategy and the real environment.

[0004] Secondly, the existing technology generally uses fixed thresholds to determine whether to trigger fertilization adjustment (such as re-decision when the humidity change exceeds ±5%). However, when the actual environmental data at the time of making the fertilization decision is highly consistent with the scenario recorded in the knowledge graph, the fertilization strategy should not be frequently changed. At the same time, a certain degree of fluctuation may be noise interference, but the existing technology changes the fertilization strategy as long as the fixed threshold is exceeded, leading to resource abuse. When the similarity between the actual environmental data and the scenario recorded in the knowledge graph is low, the fertilization strategy should be iterated faster, but the existing technology cannot sensitively capture key environmental mutations, leading to decision lag. In this way, the fixed threshold cannot be dynamically adjusted according to the current scenario, causing the dual risks of false triggering or missed triggering.

[0005] Finally, the influence of environmental changes on crops varies by type (such as leafy vegetables being sensitive to short-term temperature changes, while root vegetables being more resistant), but the existing method uses a unified monitoring time interval (such as sampling once every 24 hours). This leads to damage to sensitive crops due to response delay, or increased computational burden for robust crops due to redundant decision-making. SUMMARY

[0006] Due to the above-mentioned defects, the fertilization decision often deviates from the real needs of crops, leading to fertilizer waste, crop stress, or yield reduction risks, and restricting the development of fine agriculture. In view of the above technical problems, the present application provides a fertilization management decision method and system based on a multi-modal knowledge graph.

[0007] A fertilization management decision method based on a multi-modal knowledge graph, comprising: obtaining a plurality of natural language labels representing different crops and a fertilization data set corresponding to each natural language label based on a multi-modal knowledge graph, the fertilization data set comprising a plurality of sub-data groups representing different environmental data and sub-fertilization data corresponding to each sub-data group; obtaining crop information to be fertilized and first soil external environment information at a first time, obtaining a first matching word frequency of the crop information to be fertilized and each natural language label, taking the natural language label corresponding to the maximum first matching word frequency as a label to be processed, and obtaining a first similarity of the first soil external environment information and each sub-data group, and performing fertilization management according to the sub-fertilization data corresponding to the sub-data group with the highest first similarity; obtaining an environmental data variable threshold according to the highest first similarity, and obtaining a second time intervaling a preset time period from the first time according to the crop information to be fertilized, obtaining second soil external environment information at the second time, and obtaining real-time environmental data variables according to the first soil external environment information and the second soil external environment information, and determining whether the real-time environmental data variables exceed the environmental data variable threshold; if the real-time environmental data variables exceed the environmental data variable threshold, obtaining a second similarity of the second soil external environment information and each sub-data group, and performing fertilization management according to the sub-fertilization data corresponding to the sub-data group with the highest second similarity; if the real-time environmental data variables do not exceed the environmental data variable threshold, recording the current soil internal environment information and maintaining the current fertilization management strategy.

[0008] Optionally, obtaining the first matching word frequency of the crop information to be fertilized and each natural language label comprises: obtaining the number of same words between each natural language label and the crop information to be fertilized; and dividing the number of same words corresponding to each natural language label by the total number of words of the crop information to be fertilized to obtain the first matching word frequency of each natural language label.

[0009] Optionally, obtaining the first similarity of the first soil external environment information and each sub-data group comprises: obtaining a corresponding relationship between each type of environmental data in the first soil external environment information and the same type of environmental data in the sub-data group; obtaining a difference between the environmental data under each corresponding relationship based on the corresponding relationship; and obtaining the first similarity according to each difference.

[0010] Optionally, obtaining the environmental data variable threshold according to the highest first similarity comprises: obtaining a standard threshold and a standard similarity; obtaining a change coefficient according to the standard similarity and the first similarity, and obtaining the environmental data variable threshold according to the change coefficient and the standard threshold.

[0011] Optionally, obtaining the second time intervaling a preset time period from the first time according to the crop information to be fertilized comprises: determining a sensitive time period as the preset time period based on the growth characteristics of the crop information to be fertilized; and determining the second time according to the first time and the preset time period.

[0012] Optionally, the obtaining the real-time environment data variable according to the first soil external environment information and the second soil external environment information comprises: comparing the same category of environment data in the first soil external environment information and the second soil external environment information, and obtaining the change range of the same category of environment data between the first time and the second time; and selecting the maximum change range in the multiple change ranges as the real-time environment data variable.

[0013] Also provided is a fertilization management decision system based on a multi-modal knowledge graph, comprising: a data acquisition module, configured to acquire, based on the multi-modal knowledge graph, multiple natural language labels representing different crops and a fertilization data set corresponding to each natural language label, the fertilization data set comprising multiple sub-data groups representing different environment data and sub-fertilization data corresponding to each sub-data group; a first data processing module, configured to acquire, at a first time, to-be-fertilized crop information and first soil external environment information, to acquire a first matching word frequency of the to-be-fertilized crop information and each natural language label, to take the natural language label corresponding to the maximum first matching word frequency as a to-be-processed label, and to acquire a first similarity of the first soil external environment information and each sub-data group, and to perform fertilization management according to the sub-fertilization data corresponding to the sub-data group with the highest first similarity; a second data processing module, configured to acquire an environment data variable threshold according to the highest first similarity, to acquire, according to the to-be-fertilized crop information, a second time interval a preset time period from the first time, to acquire second soil external environment information at the second time, to acquire a real-time environment data variable according to the first soil external environment information and the second soil external environment information, and to determine whether the real-time environment data variable exceeds the environment data variable threshold; a first management module, configured to, when the real-time environment data variable exceeds the environment data variable threshold, acquire a second similarity of the second soil external environment information and each sub-data group, and to perform fertilization management according to the sub-fertilization data corresponding to the sub-data group with the highest second similarity; and a second management module, configured to, when the real-time environment data variable does not exceed the environment data variable threshold, record the current soil internal environment information and maintain the current fertilization management strategy.

[0014] Optionally, the first data processing module is further configured to: acquire the number of same words between each natural language label and the to-be-fertilized crop information; and divide the number of same words corresponding to each natural language label by the total number of words of the to-be-fertilized crop information to obtain the first matching word frequency of each natural language label.

[0015] Optionally, the first data processing module is further configured to: acquire a corresponding relationship between each category of environment data in the first soil external environment information and the same category of environment data in the sub-data group; acquire a difference value between the environment data under each corresponding relationship based on the corresponding relationship; and acquire the first similarity according to each difference value.

[0016] Optionally, the second data processing module is further configured to: obtain a standard threshold and a standard similarity; obtain a change coefficient according to the standard similarity and the first similarity, and obtain an environmental data variable threshold according to the change coefficient and the standard threshold.

[0017] The beneficial effects of the present application are embodied in:

[0018] In the entire fertilization management decision-making method based on the multi-modal knowledge graph, the semantic natural language label system constructed by the multi-modal knowledge graph, combined with the first matching word frequency algorithm to quantify the variety difference of crops, ensures that the initial fertilization strategy accurately matches the real crop demand, and based on the first similarity of the synergistic correlation of multiple environmental factors, the optimal fertilization scheme is selected from multiple sub-data groups to avoid the disconnection of the strategy caused by a single parameter threshold. Further, the similarity-driven environmental variable threshold is introduced to dynamically adjust the threshold strictness, relax the tolerance to avoid false triggering at high similarity, and tighten the threshold to accelerate iteration at low similarity; combined with real-time environmental variable monitoring of the maximum change amplitude, not only prevents the misuse of resources caused by fixed thresholds, but also sensitively triggers strategy updates at key changes, double-avoiding the risk of missed and false judgments. Further, according to the crop growth characteristics, the monitoring interval is customized, the sensitive crops are detected at high frequency to ensure rapid response to environmental mutations, and the robust crops are monitored for a long period to reduce the redundant calculation burden. This mechanism not only ensures key protection while optimizing resource allocation, but also balances computing efficiency and protection performance. Further, when the environmental fluctuation is in the safe interval, the strategy is maintained to avoid disturbance, and the environmental data in the soil is recorded to construct an environment-strategy-state time series database, which not only evaluates the fertilization effect in real time, but also provides incremental training samples for the multi-modal knowledge graph, promoting the upgrade of the multi-modal knowledge graph. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual proportions.

[0020] Figure 1 The flowchart of the fertilization management decision-making method based on the multi-modal knowledge graph of the present application;

[0021] Figure 2 The flowchart of the fertilization management decision-making method based on the multi-modal knowledge graph of the present application;

[0022] Figure 3 The flowchart of part of the step S2 in the fertilization management decision-making method based on the multi-modal knowledge graph of the present application;

[0023] Figure 4Another part of step schematic diagram for S2 in the fertilization management decision method based on the multi-modal knowledge graph of the present application;

[0024] Figure 5 Part of step schematic diagram for S3 in the fertilization management decision method based on the multi-modal knowledge graph of the present application;

[0025] Figure 6 Another part of step schematic diagram for S3 in the fertilization management decision method based on the multi-modal knowledge graph of the present application;

[0026] Figure 7 Another part of step schematic diagram for S3 in the fertilization management decision method based on the multi-modal knowledge graph of the present application. DETAILED DESCRIPTION

[0027] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0028] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0029] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0030] As shown in Figure 1 and Figure 2 A fertilization management decision method based on a multi-modal knowledge graph is provided, in one embodiment, the method comprises:

[0031] S1, obtaining a plurality of natural language labels representing different crops and a fertilization data set corresponding to each natural language label based on a multi-modal knowledge graph, the fertilization data set comprising a plurality of sub-data groups representing different environmental data and sub-fertilization data corresponding to each sub-data group;

[0032] S2, acquiring the to-be-fertilized crop information and the first soil external environment information at the first time, acquiring the first matching word frequency of the to-be-fertilized crop information and each natural language label, taking the natural language label corresponding to the maximum first matching word frequency as a to-be-processed label, acquiring the first similarity of the first soil external environment information and each sub-data group, and performing fertilization management according to the sub-fertilization data corresponding to the sub-data group with the highest first similarity;

[0033] S3, acquiring an environmental data variable threshold according to the highest first similarity, acquiring a second time interval a preset time period from the first time according to the to-be-fertilized crop information, acquiring second soil external environment information at the second time, acquiring real-time environmental data variables according to the first soil external environment information and the second soil external environment information, and judging whether the real-time environmental data variables exceed the environmental data variable threshold;

[0034] S4, if the real-time environmental data variables exceed the environmental data variable threshold, acquiring the second similarity of the second soil external environment information and each sub-data group, and performing fertilization management according to the sub-fertilization data corresponding to the sub-data group with the highest second similarity;

[0035] S5, if the real-time environmental data variables do not exceed the environmental data variable threshold, recording the current soil internal environment information and maintaining the current fertilization management strategy.

[0036] In the embodiment, it is necessary to explain that in S1, the key crop and fertilization association information is acquired from the multi-modal knowledge graph. Specifically, natural language labels representing a plurality of different crops (for example, "tomato", "lettuce", "potato", etc.) in the knowledge graph are extracted, and these labels not only are simple crop names, but also contain semantic entities defined in the graph that can distinguish different crops (including variety differences). At the same time, S1 will retrieve and associate the corresponding "fertilization data set" for each natural language label. This fertilization data set is a structured data collection: it contains a plurality of "sub-data groups", each of which represents a specific environmental condition combination state (for example, a specific combination of indoor temperature range, humidity range, light intensity range, and carbon dioxide concentration range), and for each such environmental condition combination state (i.e., each sub-data group), a specific "sub-fertilization data" (representing the best fertilization recommendation for the crop type or variety under this specific environmental combination, such as the ratio and amount of nitrogen, phosphorus, and potassium) is accurately associated.

[0037] Further, for example, in the knowledge graph, the fertilization data set associated with the natural language label of "tomato" may contain dozens or even hundreds of different sub-data groups. One of the sub-data groups may describe an environment with high light, medium to high temperature, medium humidity, and sufficient carbon dioxide (the specific values are actually obtained through the knowledge graph), and match a sub-fertilization data that promotes fruit enlargement (such as a higher proportion of potassium elements) for this specific combination of environment. The fertilization data set under the "cherry tomato" label has its own uniqueness, which may contain a sub-data group describing a narrower temperature range, a specific light variation pattern, and extremely high humidity, and match a sub-fertilization strategy that requires higher trace element guarantee. At the same time, a sub-data group of the "lettuce" label in the graph may focus on the combination of cool temperature, stable humidity, and avoidance of strong light direct radiation, and its corresponding sub-fertilization data will tend to provide balanced quick-acting nitrogen fertilizer to meet the leaf blade demand of its rapid growth.

[0038] In S2, the crop information and the first soil external environment information (such as uncontrolled environmental parameters such as indoor humidity, light intensity, carbon dioxide concentration) collected at the first time are matched using the fertilization data set corresponding to each natural language label constructed based on the multi-modal knowledge graph in S1. First, the first matching word frequency between the to-be-fertilized crop information and all natural language labels (such as "AB", "ABC", "ABD") in the knowledge graph is calculated, which is essentially to quantify the similarity between the two at the character level. The natural language label with the maximum first matching word frequency is selected as the to-be-processed label, thereby locking the fertilization data set corresponding to it in the knowledge graph, which is exclusive to this variety.

[0039] Further, the current measured first soil external environment information (such as: the current actual humidity in the shed is 70%, the light intensity is medium-high, and the carbon dioxide concentration is appropriate) is evaluated for the fit degree with all environment sub-data groups associated with the to-be-processed label. A plurality of different environment combination state descriptions (sub-data groups) preset in the knowledge graph for the crop variety are traversed, and the first similarity of the current environment and each description combination is calculated one by one. This similarity is a comprehensive index, which reflects the overall closeness of the current environment (containing the instantaneous value of a plurality of specific parameters) and a certain ideal environment combination state (also containing the target value or range of a plurality of parameters) in each parameter category. For example, the graph may store a description "moderate to high humidity, strong light, high carbon dioxide demand" state (represented by a sub-data group), and the similarity of the current actual environment and this description is calculated. Finally, the sub-fertilization data corresponding to the sub-data group with the highest similarity (such as the recommended formula and dosage of high potassium and trace elements under this environment) is selected as the first actual execution of the fertilization management strategy. This ensures that the initial fertilization strategy is no longer a rigid single threshold judgment result, but an optimal choice that comprehensively considers the synergistic effect of multiple environmental factors and best fits the current crop variety and actual environment conditions, laying a solid foundation for the entire dynamic decision-making process.

[0040] In S3, first, the highest first similarity value calculated is used to set an environment data variable threshold. This threshold is no longer fixed, but is dynamically adjusted according to the fit degree of the initial environment matching. Specifically, if the initial measured environment is highly consistent with the ideal state recorded in the knowledge graph (the highest first similarity is high), it is inclined to define the current fertilization strategy as stable and reliable, so a relatively loose (numerical value is larger) threshold is set, allowing larger environmental fluctuations without easily triggering strategy changes, effectively avoiding unnecessary strategy iteration and resource waste caused by small environmental changes or noise interference; on the contrary, if the initial matching degree is low (the highest first similarity is small), it indicates that the current environment deviates far from the ideal state, so a relatively strict (numerical value is smaller) threshold is set, which is more sensitive to environmental changes. As long as a certain degree of environmental variables is monitored, reevaluation will be triggered, thereby accelerating strategy iteration to approach the optimal fertilization scheme more quickly, preventing the risks caused by decision lag. This solves the dual problems of false triggering and missed triggering caused by traditional fixed thresholds.

[0041] Further, S3 will automatically determine a customized preset time interval based on the growth characteristics (such as variety, type) of the crop to be fertilized, which essentially represents the waiting interval for the next environmental monitoring. For example, short-term leafy vegetables (such as lettuce) that are exceptionally sensitive to environmental changes will be identified as "sensitive crops" and set to a very short time interval (such as a few hours) to ensure high-frequency detection of environmental changes to respond to their fragile needs in a timely manner and avoid stress damage; while crops with strong environmental tolerance (such as root vegetables such as carrots) are identified as "robust crops" and set to a longer time interval (such as tens or hundreds of hours). Based on the first time and the calculated preset time interval, the second time is determined and set. When the second time arrives, the second soil external environment information is collected, and by comparing the change amplitudes of all categories of environmental parameters between the first and second times (for example, how much percentage of temperature change, how much percentage of humidity change, how much percentage of light change), and selecting the largest one as the key real-time environmental data variable. Finally, the actual monitored maximum change is compared with the previously dynamically set environmental data variable threshold to scientifically determine whether the strategy needs to be re-evaluated (S4) or the current strategy needs to be maintained (S5) at the second time. This adaptive monitoring interval ensures that sensitive crops are protected in a timely manner, while reducing unnecessary frequent computation burden on robust crops.

[0042] In S4, it is only activated when S3 determines that the real-time environmental data variable exceeds the dynamically set environmental data variable threshold. This means that since the last decision (from the first time to the second time), the actual change amplitude of the environment has exceeded the tolerance range allowed according to the initial matching degree of fitness. For example, in the case of high matching (high first similarity leading to a loose threshold) between the initial environment (first soil external environment information) and the knowledge graph, small ordinary fluctuations will not trigger S4; but if there is a dramatic mutation of key environmental parameters (such as a cold wave leading to a sudden drop in temperature far exceeding the loose threshold), or the initial matching degree is low (low first similarity leading to a strict threshold), even a relatively moderate continuous change exceeds the strict threshold, S4 will be activated. This ensures that strategy updating only occurs when environmental changes have a substantial impact on the adaptability of the current fertilization scheme or the original decision basis has become significantly outdated, effectively avoiding the problems of false triggering (noise interference) and missed triggering (key changes not responded) caused by fixed thresholds, and achieving precise intervention.

[0043] Further, when S4 is triggered, a re-matching process of the same depth as the initial decision (S2) is performed based on the currently newly collected second soil external environment information. The second similarity between the newly collected second soil external environment information and all the environment sub-data groups associated with the to-be-processed tag is calculated again. This process also traverses all the pre-stored environment combination states, comprehensively evaluates the global matching degree between the current latest environment (containing the latest measured values of all category parameters) and each ideal state in the knowledge graph. Finally, the sub-fertilization data corresponding to the sub-data group with the highest second similarity is selected, and a new set of fertilization management strategy is immediately switched to execute. Through this accurate re-matching based on the complete information of the latest environment, S4 can make the fertilization strategy track and adapt to the significant changes in the environment in real time and dynamically, ensure that the nutrient supply of crops always approaches the optimal demand, maximize resource utilization efficiency, and avoid stress risks.

[0044] In S5, the operation is performed when it is determined in S3 that the real-time environment data variable does not exceed the dynamically set environment data variable threshold. This result means that the amplitude of environmental change from the last decision (first time) to the current check point (second time) is still within the safe interval determined according to the initial environment matching degree. It is inferred that the currently executed fertilization management strategy is still adaptive and effective for crops, and the change in the environment has not reached the degree that requires triggering a strategy change. At this time, the core operation of S5 is to maintain the existing fertilization management strategy unchanged. This avoids unnecessary disturbance, additional resource consumption, and interference with the stability of crop growth caused by frequent adjustment of the strategy due to minor, non-critical fluctuations or temporary noise interference, fully embodying the stability of the decision and the intelligent tolerance of the system.

[0045] Further, while maintaining the strategy, the current soil internal environment information (usually referring to factors that can be directly affected by fertilization, such as the real-time nitrogen, phosphorus, and potassium content in the soil, pH value, water state, etc.) is recorded. This recording action establishes a correlation data point between the environment state and the effect of the fertilization strategy, for example, if the maintained fertilization strategy for the environment is a medium nitrogen and high potassium fertilizer, then the recorded soil internal nutrient level at this moment can provide a basis for subsequent evaluation of the nutrient absorption efficiency and remaining condition of the strategy in this stable environment; secondly, the recorded soil internal environment information (including parameters that may change slightly over time, such as nitrogen consumption) and the maintained strategy together constitute a complete environment-action-state time series data. These data will become the basis for optimizing the multi-modal knowledge graph (such as updating the best fertilization recommendation under a specific environment combination), evaluating the long-term strategy effect (such as verifying the sustained effect of a certain fertilizer ratio), and even warning potential stress risks (such as finding that a certain element is abnormally accumulated in the soil).

[0046] In summary, in the entire fertilization management decision-making method based on the multi-modal knowledge graph, the semantic natural language label system constructed by the multi-modal knowledge graph, combined with the first matching word frequency algorithm to quantify the differences between crops, ensures that the initial fertilization strategy accurately matches the real crop demand. At the same time, based on the first similarity of the synergistic correlation of multiple environmental factors, the optimal fertilization scheme is selected from multiple sub-data groups to avoid the disconnection of the strategy caused by a single parameter threshold. Further, the similarity-driven environmental variable threshold is introduced to dynamically adjust the threshold strictness. When the similarity is high, the tolerance is relaxed to avoid false triggering, and when the similarity is low, the threshold is tightened to speed up iteration. Combined with real-time monitoring of the maximum change amplitude of the environmental variable, it not only prevents resource abuse caused by fixed thresholds, but also sensitively triggers strategy updates when key changes occur, avoiding both missed and false judgments. Further, according to the growth characteristics of crops, the monitoring interval is customized. High-frequency detection is used for sensitive crops to ensure rapid response to environmental changes, and the monitoring period is extended for robust crops to reduce redundant computational burden. This mechanism not only ensures key protection, but also optimizes resource allocation, achieving a balance between computational efficiency and protection performance. Further, when the environmental fluctuation is within the safe interval, the strategy is maintained to avoid disturbance, and the soil internal environmental data is recorded to construct an environment-strategy-state time series database. This not only evaluates the fertilization effect in real time, but also provides incremental training samples for the multi-modal knowledge graph, promoting the upgrade of the multi-modal knowledge graph. In summary, the entire technical solution improves fertilizer utilization, reduces environmental stress risk, and reduces algorithm redundancy, providing technical support for fine fertilization in protected agriculture.

[0047] As shown in Figure 3 In one embodiment, the first matching word frequency of each natural language label and the to-be-fertilized crop information in S2 includes:

[0048] S21, the number of identical characters between each natural language label and the to-be-fertilized crop information is obtained;

[0049] S22, the number of identical characters corresponding to each natural language label is divided by the total number of characters of the to-be-fertilized crop information to obtain the first matching word frequency of each natural language label.

[0050] In this embodiment, it should be noted that in S21, semantic analysis at the character level is first performed, and the natural language labels in the knowledge graph (such as "ABC", "ABD", "ABCD") are compared one by one with the to-be-fertilized crop information input by the user (such as "ABCDE"), and the number of identical characters (such as "ABCD" label and input information have 4 identical characters) is accurately counted.

[0051] In S22, the number of identical characters of each label is divided by the total number of characters of the input information (such as "ABCDE" has 5 characters), and the first matching word frequency between 0 and 1 is generated (such as 4 / 5=0.8).

[0052] As Figure 4 shown, in one embodiment, the first soil external environment information in S2 is acquired and the first similarity of each sub-data set includes:

[0053] S23, acquiring the corresponding relationship between each type of environment data in the first soil external environment information and the same type of environment data in the sub-data set;

[0054] S24, based on the corresponding relationship, acquiring the difference between the environment data under each corresponding relationship;

[0055] S25, acquiring the first similarity according to each difference value.

[0056] In this embodiment, it should be noted that in S23, a parameter category mapping relationship is established: the alignment relationship of the same parameters (such as the pairing of the shed temperature and the growth temperature) between the first soil external environment information (such as the measured shed temperature, humidity, and carbon dioxide concentration) and the sub-data set environment description (such as the growth temperature, humidity, and carbon dioxide concentration recorded in the atlas) is identified.

[0057] In S24, for each set of established parameter pairs, the absolute difference between the measured value and the target value is calculated (such as the measured shed temperature 28 degrees Celsius vs. the target growth temperature 25 degrees Celsius → difference 3). This step generates a set of difference vectors, quantifying the degree to which each environment parameter deviates from the ideal state, laying the foundation for comprehensive similarity evaluation.

[0058] In S25, based on the difference value set of S24, the first similarity is calculated, which reflects the core index of overall environment adaptability. The core lies in the differentiated weight fusion. First, each parameter difference value is converted to a similarity contribution value between 0 and 1; then, according to the parameter importance, the weight is allocated, and in general cases, the weight of each parameter is equal; then, the contribution values of all parameters are weighted and summed to generate the total similarity. For example, even if the temperature deviates greatly, if the humidity and light are highly consistent and the weight is high, the overall similarity can still maintain a high level. This model overcomes the single parameter short board and ensures that the decision takes into account the synergistic effect of multiple factors.

[0059] It should also be noted that the first similarity can be represented as:

[0060] ; wherein, is the first similarity, is the number of corresponding relationships, is the summation weight under the ith corresponding relationship, is the difference between the environment data under the ith corresponding relationship, is the environment data corresponding to the sub-data set in the ith corresponding relationship.

[0061] It also needs to be explained that the whole expression of obtaining the first similarity is essentially the adaptability quantization under the synergistic influence of multiple environmental factors. Through the asymmetric attenuation mechanism, differential weight integration and normalization processing, the problem of ignoring the synergistic influence of multiple factors and environmental fluctuation tolerance in traditional methods is solved. Specifically, In the formula, the nonlinear attenuation characteristics of the sensitivity of crops to environmental parameters are included (for example, the temperature has little effect within ±2°C, and the effect increases sharply when it exceeds ±5°C); when is very small (such as a slight change in the environment), tends to 1, indicating that slight fluctuations hardly reduce the similarity; when is significantly increased (such as a sudden cold wave), the function value will decrease sharply, indicating that a large deviation will significantly reduce the adaptability. In this way, the one-size-fits-all determination of the traditional fixed threshold (such as ±5%) is avoided, and the nonlinear response of real crops to the environment is simulated through the function attenuation curve.

[0062] Further, is the relative deviation rate, which eliminates the dimensional differences of parameters. Different types of corresponding relationships (such as temperature, humidity, and carbon dioxide concentration) have different dimensions and numerical scales. Direct comparison of differences will amplify the weight of high-value parameters, such as 500 ppm carbon dioxide concentration fluctuations that can easily mask the effects of 1°C temperature changes. By calculating the relative deviation rate, if the target humidity is 70% and the measured humidity is 65% in the sub-data set, the deviation rate |65-70| / 70=7.1%; if the target carbon dioxide concentration is 800 ppm and the measured carbon dioxide concentration is 750 ppm, the deviation rate |750-800| / 800=6.25%; converting the absolute difference into a dimensionless relative deviation ensures that different corresponding relationships can participate in the similarity evaluation fairly and avoids the dominance of a single high-amplitude parameter in the decision-making result.

[0063] Further, is the weighted sum, which quantifies the synergistic influence of multiple factors. The sum weights of different corresponding relationships are different (for example, the influence weight of temperature on solanaceous fruits is higher than that of carbon dioxide concentration on solanaceous fruits). The weight can be pre-set by those skilled in the art, and in general, the sum weights of each corresponding relationship are the same, as long as they satisfy That is, this is to ensure that the final first similarity is in [0, 1], so that the similarities of different scenarios are comparable; wherein, when all the corresponding relationships are perfectly matched, the first similarity is 1 (ideal adaptation); when all the corresponding relationships are seriously deviated, the first similarity is 0 (completely inadaptation). The weighted sum result strengthens the decision-making power of the key factors and avoids the misjudgment of the overall negative adaptation due to the over-standard single parameter (such as the slight deficiency of carbon dioxide concentration but the high consistency of temperature and humidity, the overall similarity is still high), which is in line with the physiological characteristics of crops.

[0064] As shown in Figure 5 in one embodiment, the environment data variable threshold value obtained according to the highest first similarity in S3 includes:

[0065] S31, obtaining a standard threshold value and a standard similarity;

[0066] S32, obtaining a change coefficient according to the standard similarity and the first similarity, and obtaining an environment data variable threshold value according to the change coefficient and the standard threshold value.

[0067] In this embodiment, it should be noted that in S31, the standard threshold value can be obtained by the following method. First, collect the facility agriculture environment monitoring data spanning more than 1 year, and statistically analyze the fluctuation range that crops can tolerate without damage, and quantify the maximum change amplitude of environmental parameters under the influence of non-stress; for example, leaf vegetables tolerate humidity ± 10% and temperature ± 3℃; fruit and vegetable crops tolerate light ± 15% and temperature ± 5℃; assuming that from the 1-year data of a tomato greenhouse, it is found that when the temperature mutation is ≤±4℃, the fruit setting rate does not decrease significantly, then the temperature threshold value is defined as 4℃. Then, apply a gradient humidity environment mutation to the target crop, record the critical point of yield reduction (such as a 20% decrease in photosynthetic rate), establish the causal relationship between crop physiological damage and parameter mutation, and exclude noise interference; assuming that the tomato experimental group: when the humidity drops by 40%, the leaves wilt, so set the temperature to 40%. Finally, through the same limited number of experiments as described above, the influence of each different type of environmental data on plants is obtained, the critical threshold value is obtained, and the existing random forest algorithm is used to analyze the influence weight of joint mutation of environmental data on yield loss, to determine the dominant parameter threshold value, solve the limitation of single parameter threshold value, for example, humidity mutation is more likely to induce tomato collar rot, and the threshold weight is increased, and the threshold values of all types of environmental data are weighted and summed to obtain the final standard threshold value.

[0068] Further, when obtaining the standard similarity, the environmental combination of all sub-data groups in the atlas can be extracted, and the similarity mean of the intersection point of the historical yield change is calculated, that is, the general similarity benchmark of high yield and sharp decrease is determined, which is used as the standard similarity.

[0069] In S32, the difference between the first similarity and the standard similarity is converted into a change coefficient, and then the environmental data variable threshold is obtained according to the linear response of the change coefficient and the standard threshold. When the first similarity is significantly higher than the standard similarity, a larger threshold (loose tolerance fluctuation) is generated; when the first similarity is significantly lower than the standard similarity, a smaller threshold (strict sensitive monitoring) is generated.

[0070] It should be further pointed out that the environmental data variable threshold obtained in S3 can be expressed as:

[0071] ; wherein, is the environmental data variable threshold, is the standard threshold, is the first similarity, is the standard similarity.

[0072] It should be further pointed out that in the entire expression, is a dynamic response mechanism of the change coefficient, and the dimensionless difference between and is used to ensure that the compensation factor is unbiased. If it is positive compensation ( ), when the measured environment highly matches the atlas ( tends to 1), the function output value exceeds 1, and the environmental data variable threshold is expanded to form a loose buffer zone to prevent the strategy adjustment from being triggered by a slight change in the environment. If it is negative compensation ( ), when the environment is not well adapted, the function output value is less than 1, is contracted to form a strict sensitive zone to accelerate the response to key mutations.

[0073] As shown in Figure 6 , in one embodiment, in S3, the second time interval of a preset time period from the first time is obtained according to the information of the crop to be fertilized, including:

[0074] S33, determining a sensitive time period based on the growth characteristics of the crop to be fertilized and taking it as the preset time period;

[0075] S34, determining the second time according to the first time and the preset time period.

[0076] In this embodiment, it should be pointed out that in S33, the monitoring interval is dynamically preset according to the growth characteristics of the crop to be fertilized, that is, the biological characteristics of the crop. For example, sensitive crops (such as lettuce): a short preset time period (such as 2-4 hours) is used to ensure high-frequency capture of temperature and humidity mutations; robust crops (such as potatoes): a long preset time period (such as 24-48 hours) is used to reduce redundant monitoring load.

[0077] In S34, the first time is added to the preset time period to calculate the accurate second time, forming a timing trigger mechanism.

[0078] As Figure 7 shown, in one embodiment, the real-time environmental data variable is obtained according to the first soil external environment information and the second soil external environment information in S3, including:

[0079] S35, comparing the same category of environmental data in the first soil external environment information and the second soil external environment information, and obtaining the change amplitude of the same category of environmental data between the first time and the second time;

[0080] S36, selecting the maximum change amplitude in the plurality of change amplitudes as the real-time environmental data variable.

[0081] In this embodiment, it should be noted that in S35, the full parameter change amplitude is calculated. Specifically, the change amplitudes of the same environmental parameters (temperature / humidity / light, etc.) between the first time and the second time are compared one by one, that is, the relative change rate.

[0082] In S36, an extreme value screening strategy is adopted: the maximum value is extracted from all parameter change amplitudes as the real-time environmental data variable. This reflects that the influence of environmental mutation on crops follows the short board effect, and the maximum change parameter determines the risk level. Overcome the one-sidedness of the single parameter trigger mechanism. Accurately capture the key environmental mutation point (even if other parameters are stable), and ensure the sensitivity of decision-making.

[0083] Also provided is a fertilization management decision system based on a multi-modal knowledge graph, the system comprising:

[0084] A data acquisition module is configured to acquire a plurality of natural language labels representing different crops and a fertilization data set corresponding to each natural language label based on a multi-modal knowledge graph, wherein the fertilization data set comprises a plurality of sub-data groups representing different environmental data and sub-fertilization data corresponding to each sub-data group.

[0085] A first data processing module is configured to acquire crop information to be fertilized and first soil external environment information at a first time, and to acquire a first matching word frequency of the crop information to be fertilized and each natural language label, and to select a natural language label corresponding to the maximum first matching word frequency as a to-be-processed label, and to acquire a first similarity of the first soil external environment information and each sub-data group, and to perform fertilization management according to the sub-fertilization data corresponding to the sub-data group with the highest first similarity.

[0086] The second data processing module is configured to obtain an environmental data variable threshold according to the highest first similarity, obtain a second time point which is apart from the first time point by a preset time period according to the crop information to be fertilized, obtain second soil external environment information at the second time point, obtain a real-time environmental data variable according to the first soil external environment information and the second soil external environment information, and determine whether the real-time environmental data variable exceeds the environmental data variable threshold.

[0087] The first management module is configured to, when the real-time environmental data variable exceeds the environmental data variable threshold, obtain a second similarity between the second soil external environment information and each sub-data group, and perform fertilization management according to sub-fertilization data corresponding to a sub-data group with the highest second similarity.

[0088] The second management module is configured to, when the real-time environmental data variable does not exceed the environmental data variable threshold, record current soil internal environment information and maintain a current fertilization management strategy.

[0089] In one embodiment, the first data processing module is further configured to: obtain a number of same words between each natural language label and the crop information to be fertilized; and divide the number of same words corresponding to each natural language label by a total number of words of the crop information to be fertilized to obtain a first matching word frequency of each natural language label.

[0090] In one embodiment, the first data processing module is further configured to: obtain a corresponding relationship between each type of environmental data in the first soil external environment information and the same type of environmental data in the sub-data group; obtain a difference between the environmental data under each corresponding relationship based on the corresponding relationship; and obtain the first similarity according to each difference.

[0091] In one embodiment, the second data processing module is further configured to: obtain a standard threshold and a standard similarity; obtain a change coefficient according to the standard similarity and the first similarity, and obtain the environmental data variable threshold according to the change coefficient and the standard threshold.

[0092] In the present embodiment, it should be noted that, as to the above-mentioned fertilization management decision system based on the multi-modal knowledge graph, the specific manner of performing operations has been described in detail in the embodiments of the fertilization management decision method based on the multi-modal knowledge graph, and will not be described in detail here.

[0093] The preferred embodiments of the present disclosure are described in detail above with reference to the accompanying drawings, but the present disclosure is not limited to the specific details in the above-described embodiments. Within the technical concept of the present disclosure, various simple modifications can be made to the technical solutions of the present disclosure, and these simple modifications all belong to the protection scope of the present disclosure.

[0094] It should be further noted that the combination of one or several letters of "A, B, C, D, E" described in the above embodiments can represent different plant names or varieties, and the same combination of letters in different embodiments can represent different plant names or varieties.

[0095] It should be further noted that each specific technical feature described in the above embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, the disclosure will not further describe various possible combinations.

[0096] In addition, various different embodiments of the disclosure can also be combined in any manner, as long as they do not deviate from the idea of the disclosure, and should also be considered as disclosed by the disclosure.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some or all of the technical features thereof; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. A method for fertilization management decision based on a multi-modal knowledge graph, characterized in that, The method comprises the following steps: obtaining a plurality of natural language labels representing different crops and a plurality of fertilizer data sets corresponding to the natural language labels based on a multi-modal knowledge graph, wherein each of the fertilizer data sets comprises a plurality of sub-data groups representing different environmental data and sub-fertilizer data corresponding to each of the sub-data groups; obtaining crop information to be fertilized and first external soil environment information at a first time, obtaining a first matching word frequency of the crop information to be fertilized and each of the natural language labels, taking the natural language label corresponding to the maximum first matching word frequency as a label to be processed, obtaining a first similarity of the first external soil environment information and each of the sub-data groups, and performing fertilization management according to the sub-fertilizer data corresponding to the sub-data group with the highest first similarity; obtaining a standard threshold and a standard similarity, obtaining a change coefficient based on the standard similarity and the first similarity, and obtaining an environmental data variable threshold based on the change coefficient and the standard threshold; Wherein, the environment data variable threshold is represented as: ; wherein, is the environment data variable threshold, is the standard threshold, is the first similarity, is the standard similarity; obtaining second external soil environment information at a second time interval of a preset time period from the first time based on the crop information to be fertilized, obtaining a real-time environmental data variable based on the first external soil environment information and the second external soil environment information, and determining whether the real-time environmental data variable exceeds the environmental data variable threshold; if the real-time environmental data variable exceeds the environmental data variable threshold, obtaining a second similarity of the second external soil environment information and each of the sub-data groups, and performing fertilization management according to the sub-fertilizer data corresponding to the sub-data group with the highest second similarity; if the real-time environmental data variable does not exceed the environmental data variable threshold, recording the current internal soil environment information and maintaining the current fertilization management strategy.

2. The fertilization management decision method based on a multi-modal knowledge graph according to claim 1, characterized in that, The method comprises the following steps: obtaining the number of identical words between each of the natural language labels and the crop information to be fertilized; dividing the number of identical words corresponding to each of the natural language labels by the total number of words in the crop information to be fertilized to obtain the first matching word frequency of each of the natural language labels. 3.The multi-modal knowledge graph based fertilization management decision method according to claim 1, characterized in that, The method comprises the following steps: obtaining the corresponding relationship between each type of environmental data in the first external soil environment information and the same type of environmental data in the sub-data group; obtaining the difference between the environmental data under each corresponding relationship based on the corresponding relationship; obtaining the first similarity based on each of the differences. 4.The multi-modal knowledge graph based fertilization management decision method according to claim 1, characterized in that, The method comprises the following steps: determining a sensitive time period as the preset time period based on the growth characteristics of the crop information to be fertilized; determining the second time based on the first time and the preset time period. 5.The multi-modal knowledge graph based fertilization management decision method according to claim 1, characterized in that, The method comprises the following steps: comparing the same type of environmental data in the first external soil environment information and the second external soil environment information, and obtaining the change amplitude of the same type of environmental data between the first time and the second time; selecting the maximum change amplitude from the plurality of change amplitudes as the real-time environmental data variable. 6.A fertilization management decision system based on a multi-modal knowledge graph, characterized in that, The system is used to implement the fertilization management decision method based on the multi-modal knowledge graph as claimed in any one of claims 1 to 5, and the system comprises: The data acquisition module is configured to acquire, based on the multi-modal knowledge graph, a plurality of natural language labels representing different crops and a fertilizer data set corresponding to each natural language label, the fertilizer data set including a plurality of sub-data groups representing different environmental data and sub-fertilizer data corresponding to each sub-data group. The first data processing module is configured to acquire, at a first time, crop information to be fertilized and first external soil environment information, acquire a first matching word frequency of the crop information to be fertilized and each natural language label, take a natural language label corresponding to a maximum first matching word frequency as a label to be processed, acquire a first similarity of the first external soil environment information and each sub-data group, and perform fertilization management according to sub-fertilizer data corresponding to a sub-data group with a highest first similarity. The second data processing module is configured to acquire an environmental data variable threshold according to the highest first similarity, acquire a second time interval a preset time period from the first time according to the crop information to be fertilized, acquire second external soil environment information at the second time, acquire real-time environmental data variables according to the first external soil environment information and the second external soil environment information, and determine whether the real-time environmental data variables exceed the environmental data variable threshold. The first management module is configured to acquire a second similarity of the second external soil environment information and each sub-data group when the real-time environmental data variables exceed the environmental data variable threshold, and perform fertilization management according to sub-fertilizer data corresponding to a sub-data group with a highest second similarity. The second management module is configured to record current internal soil environment information and maintain a current fertilization management strategy when the real-time environmental data variables do not exceed the environmental data variable threshold.

7. The multi-modal knowledge graph based fertilization management decision system according to claim 6, wherein, The first data processing module is further configured to: acquire a number of same words between each natural language label and the crop information to be fertilized; divide the number of same words corresponding to each natural language label by a total number of words of the crop information to be fertilized to obtain a first matching word frequency of each natural language label. 8.The multi-modal knowledge graph based fertilization management decision system of claim 6, wherein, The first data processing module is further configured to: acquire a corresponding relationship between each type of environmental data in the first external soil environment information and the same type of environmental data in the sub-data group; acquire a difference between environmental data under each corresponding relationship based on the corresponding relationship; acquire the first similarity according to each difference. 9.The multi-modal knowledge graph based fertilization management decision system of claim 6, wherein, The second data processing module is further configured to: acquire a standard threshold and a standard similarity; acquire a change coefficient according to the standard similarity and the first similarity, and acquire the environmental data variable threshold according to the change coefficient and the standard threshold.

Images