Agricultural and animal husbandry information demand response method based on intelligent analysis model

Abstract

The invention discloses an agriculture and animal husbandry information demand response method based on an intelligent analysis model. The method comprises the following steps: acquiring agriculture and animal husbandry multi-source data, denoising and complementing, performing unified time alignment, and generating a time sequence data set; production event data are extracted, an event enhancement sequence is constructed, and three types of sequences of marking, segmentation and weight are generated; inputting the enhanced sequence into an improved DLinear model, performing trend and season decomposition, and predicting a future state; judging whether the predicted value crosses the border or not, and generating trend border-crossing and implicit demand indication information; according to the state and behavior characteristics, causal chain fragments are matched, and an information demand type, an object and an emergency degree are determined; and based on the demand characteristics and the information bearing capacity, generating a multi-stage load response strategy and outputting information. Through event-enhanced time sequence trend analysis, behavior entropy recognition and multi-level response strategy generation, precise recognition, intelligent classification and efficient response of agriculture and animal husbandry information requirements are realized.

Description

Technical Field

[0001] This invention relates to the field of agricultural and animal husbandry informatization and intelligent analysis technology, and in particular to an agricultural and animal husbandry information demand response method based on an intelligent analysis model. Background Technology

[0002] With the continuous improvement of agricultural modernization and digitalization, environmental monitoring equipment, production management systems, and information service platforms are gradually being introduced into agricultural and livestock production processes. These systems collect environmental data such as temperature, humidity, soil, and water quality, as well as production operation data such as irrigation, fertilization, feeding, and pesticide application. They also provide technical guidance, risk warnings, and management suggestions to agricultural and livestock operators. Existing agricultural and livestock information services are typically based on fixed rules, experience thresholds, or simple data statistics methods. After analyzing the collected data, they generate information prompts, which to some extent improves the informatization level of agricultural and livestock production.

[0003] However, existing technologies still have significant shortcomings in practical applications. On the one hand, existing information service systems tend to focus on judging single environmental indicators or historical statistical results, making it difficult to distinguish whether changes in agricultural and pastoral conditions are caused by changes in the natural environment or by human production operations such as irrigation, feeding, and ventilation. This can easily lead to misjudgments or delayed judgments. On the other hand, existing systems typically rely on explicit indicators to trigger information pushes, lacking the ability to analyze changes in the behavior patterns of agricultural and pastoral entities. They are unable to identify the implicit information needs of agricultural and pastoral entities during the production process due to uncertainty, lack of experience, or potential risks, resulting in insufficient targeting and effectiveness of information services.

[0004] Existing agricultural and pastoral information services generally adopt a unified push or simple hierarchical push mechanism in terms of information output, which does not fully consider the information carrying capacity of different entities at different production stages and under different workload conditions. This can easily lead to situations where the information push frequency is too high, the content is redundant, or key prompts are ignored, thus affecting the information response effect.

[0005] Therefore, how to provide a method for responding to agricultural and pastoral information needs based on intelligent analysis models is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method for responding to agricultural and pastoral information demands based on an intelligent analysis model. This invention achieves precise triggering, accurate classification, and reasonable response to agricultural and pastoral information demands by uniformly analyzing multi-source data generated during agricultural and pastoral production, introducing a time-series trend prediction mechanism enhanced by production events, a method for identifying implicit demands based on subject behavioral characteristics, and a multi-level response load balancing information control approach. This invention can effectively distinguish between the impact of natural factors and human production operations in changes in agricultural and pastoral status, promptly identify the explicit and implicit information demands of agricultural and pastoral subjects in production management, and dynamically generate matching information response strategies based on the information carrying capacity of subjects under different production stages and workload conditions. This avoids problems such as information misjudgment, response lag, and information overload, and possesses the advantages of high accuracy in information demand identification, strong response targeting, and good rationality in information output.

[0007] An agricultural and livestock information demand response method based on an intelligent analysis model according to an embodiment of the present invention includes:

[0008] Collect multi-source agricultural and animal husbandry data during the agricultural and animal husbandry production process, preprocess the multi-source agricultural and animal husbandry data, and obtain agricultural and animal husbandry time series datasets;

[0009] Based on the production event data in the agricultural and animal husbandry time series dataset, an event-enhanced sequence is constructed, and the production events corresponding to irrigation, fertilization, feeding, pesticide application, ventilation, and equipment start-up and shutdown are converted into event label sequence, segmented factor sequence, and weight adjustment factor sequence, respectively.

[0010] The event augmentation sequence is input into the improved DLinear model. The event augmentation sequence is decomposed into trend and seasonal terms. The decomposed trend and seasonal terms are then linearly mapped and predicted to generate agricultural and pastoral status prediction results and trend change results within the target time window.

[0011] The main behavioral data in the agricultural and pastoral time series dataset is constructed into a behavioral feature sequence. Based on the trend change results, trend boundary indication information is generated. The behavioral entropy change of the behavioral feature sequence within a continuous time window is calculated to generate implicit demand indication information.

[0012] Based on trend boundary indication information and implicit demand indication information, combined with the constructed agricultural and animal husbandry causal chain fragments, agricultural and animal husbandry information demand is classified to generate information demand type, information demand object and demand urgency.

[0013] Construct a multi-level response load balancing tree, select the target response path from the multi-level response load balancing tree according to the information demand type, the information demand object, the urgency of the demand and the information carrying capacity of the subject, generate the information response strategy, and output the information response content to the target subject according to the information response strategy.

[0014] Optionally, the multi-source agricultural and livestock data includes environmental time-series data, production event data, subject behavior data, and production result data during the agricultural and livestock production process.

[0015] Optionally, the preprocessing of agricultural and animal husbandry multi-source data includes outlier removal, missing value completion, time alignment, format unification, and structured encoding to obtain a preprocessed dataset.

[0016] Optionally, constructing event-enhanced sequences based on production event data in agricultural and livestock time-series datasets includes:

[0017] The agricultural and animal husbandry time series dataset is divided into multiple continuous time windows according to a uniform time step, and production event records within each time window are extracted from the agricultural and animal husbandry time series dataset.

[0018] For each type of production event within each time window, determine the occurrence status and event quantity of the production event within that time window;

[0019] Based on the occurrence status and event quantity of various production events within each time window, an event marker vector corresponding to each time window is generated. The event marker vectors corresponding to each time window are arranged in chronological order to obtain the event marker sequence.

[0020] A segmented factor sequence is generated based on the change relationship between the event marker vectors corresponding to adjacent time windows. When the event marker vectors of adjacent time windows change, the current time window is marked as a new event segment. When the event marker vectors of adjacent time windows do not change, the current time window is kept in the current event segment. The event segment identifiers to which each time window belongs are arranged in chronological order to obtain the segmented factor sequence.

[0021] A weight adjustment factor sequence is generated based on the event label sequence and the segmented factor sequence. The weight adjustment factor is calculated for each time window. The weight adjustment factor is obtained by accumulating the event quantity of various production events within the historical time window according to the decay rule. The cumulative contribution of various production events is weighted and synthesized according to the event type weight. The event label sequence, the segmented factor sequence and the weight adjustment factor sequence are aligned and combined according to the time window to obtain the event enhancement sequence.

[0022] Optionally, the generation of agricultural and pastoral status prediction results and trend change results within the target time window includes:

[0023] The event augmentation sequence is organized into multiple consecutive time windows in chronological order. The length of the input window is determined to be L and the length of the prediction window is determined to be H. For the current time window, the event augmentation data corresponding to the previous L consecutive time windows are selected to form the input subsequence.

[0024] The input subsequence is fed into the improved DLinear model, which adds an event-gated decomposition unit to the decomposition structure of DLinear. The event-gated decomposition unit determines the decomposition parameters of the corresponding time window based on the event tag sequence, segmentation factor sequence and weight adjustment factor sequence in the input subsequence, and decomposes the input subsequence into trend component subsequence and seasonal component subsequence according to the decomposition parameters.

[0025] In the improved DLinear model, the trend component subsequence is predicted. The improved DLinear model adds a piecewise linear mapping unit to the linear mapping structure. The piecewise linear mapping unit selects the corresponding linear mapping parameters for different events based on the piecewise factor sequence, and performs linear mapping on the trend component subsequence to obtain the trend prediction component sequence for the next H time windows.

[0026] In the improved DLinear model, seasonal component sequences are predicted. The piecewise linear mapping unit selects the corresponding linear mapping parameters for different events based on the piecewise factor sequence, and performs linear mapping on the seasonal component sequences to obtain the seasonal predicted component sequences for the next H time windows.

[0027] In the improved DLinear model, the trend prediction component sequence and the seasonal prediction component sequence are fused. The improved DLinear model adds a component fusion unit to the output structure. The component fusion unit aligns the trend prediction component sequence and the seasonal prediction component sequence according to the time window and synthesizes them into an event-enhanced prediction sequence for the next H time windows. Data items corresponding to the agricultural and pastoral status are extracted to obtain the agricultural and pastoral status prediction results. Based on the agricultural and pastoral status prediction results, the trend change results are generated.

[0028] Optionally, the step of generating trend boundary indication information based on trend change results, calculating the behavioral entropy change of the behavioral feature sequence within a continuous time window, and generating implicit demand indication information includes:

[0029] The agricultural and pastoral status prediction results are divided into multiple prediction time windows according to the prediction time sequence. For each prediction time window, the corresponding prediction status value is compared with the management threshold corresponding to the prediction time window to generate trend boundary indication information that indicates whether the prediction status exceeds the management threshold.

[0030] The subject's behavioral data is divided into continuous behavioral time windows according to a uniform time granularity. Within each behavioral time window, the subject's operation type, operation sequence, operation interval time, and operation spatial location are extracted to construct the behavioral feature sequence of the corresponding time window.

[0031] For each behavior time window, the behavior feature sequence is discretized to obtain the behavior category distribution within the time window, and the behavior complexity index of the time window is calculated based on the behavior category distribution.

[0032] Between adjacent consecutive action time windows, the action complexity index is compared, the magnitude and direction of change of the action complexity index are calculated, and the magnitude of change is weighted and corrected by combining the degree of fluctuation of operation interval and the degree of change of operation sequence within the action time window to obtain the change of action entropy.

[0033] Implicit demand indication information is generated based on the change in behavioral entropy. When the change in behavioral entropy shows a continuous upward trend within multiple consecutive behavioral time windows, it is determined that the subject has implicit information needs and corresponding implicit demand indication information is generated. When the change in behavioral entropy remains stable or decreases within consecutive behavioral time windows, implicit demand indication information is not generated.

[0034] Optionally, the information request type, information request object, and urgency level include:

[0035] The trend out-of-bounds indication information and implicit demand indication information are used as input conditions for demand determination. At the same time, the predicted state type corresponding to the trend out-of-bounds indication information and the subject identifier and behavior time window identifier corresponding to the implicit demand indication information are obtained.

[0036] Based on long-term operational data of agricultural and animal husbandry production processes, a set of agricultural and animal husbandry causal chain segments is constructed. The agricultural and animal husbandry causal chain segments take state change - behavior change - management needs as the basic structural unit. Each causal chain segment includes state triggering conditions, behavior triggering conditions, demand type identifier, demand object mapping rules, and demand urgency generation rules.

[0037] When constructing a set of agricultural and pastoral causal chain segments, corresponding causal chain subsets are established for different production stages, different crop types, or different livestock objects. Different combinations of behavioral triggering conditions are configured for the same state change in the causal chain subsets to form an agricultural and pastoral causal chain segment structure.

[0038] The trend out-of-bounds indication information, implicit demand indication information, predicted state type and subject identifier are matched with the state triggering conditions and behavior triggering conditions in the agricultural and animal husbandry causal chain segment set in turn, and the causal chain segments that simultaneously meet the state triggering conditions and behavior triggering conditions are selected as candidate causal chain segments.

[0039] For each candidate causal chain segment, the information demand type is determined based on the demand type identifier corresponding to the causal chain segment, the information demand object is determined based on the demand object mapping rule of the causal chain segment, and the demand urgency is determined based on the demand urgency generation rule of the causal chain segment, combined with trend cross-boundary indication information and implicit demand indication information. The information demand type, information demand object and demand urgency are then output.

[0040] Optionally, the information response generation strategy outputs information response content to the target subject according to the information response strategy, including...

[0041] Construct a multi-level response load balancing tree, which is divided into multiple response levels according to the information output load from low to high. Each response level corresponds to a type of information output complexity. Under each response level, establish a set of candidate response information corresponding to the level.

[0042] When constructing a multi-level response load balancing tree, entry conditions are configured for each response level. The entry conditions are jointly determined by the information demand type, the urgency of the demand, the trend out-of-bounds indication information, and the implicit demand indication information.

[0043] In the multi-level response load balancing tree, information load attributes and demand matching attributes are configured for each candidate response information. The information load attribute represents the degree to which the response information occupies the subject's attention and terminal resources under the target output mode, while the demand matching attribute represents the degree of matching between the response information and the information demand type, information demand object, and demand urgency.

[0044] Based on the information carrying capacity of the subject, the response levels are traversed from top to bottom in the multi-level response load balancing tree. When the cumulative information load of the candidate response information set under the current response level does not exceed the information carrying capacity of the subject, the response level is selected as the target response level. Within the response level, the candidate response information is filtered and sorted according to the requirement matching attributes to generate the target response information set.

[0045] When generating the target response information set, a response stability constraint rule is introduced to constrain the consistency of the information output level within multiple consecutive response cycles. When the information demand type and urgency level do not change within consecutive response cycles, the selected response level is kept from jumping, and only the target response information set is updated to generate the final information response strategy. The information response content is then output to the target subject according to the information response strategy.

[0046] The beneficial effects of this invention are:

[0047] This invention introduces a time-series trend prediction mechanism enhanced by production events, enabling unified modeling and analysis of multi-source data generated during agricultural and livestock production. This allows changes in agricultural and livestock status to simultaneously reflect the impact of natural environmental changes and human production operations, effectively avoiding the misjudgment problems caused by relying solely on single environmental indicators or static thresholds in existing technologies. By structuring production events and incorporating them into trend analysis, this invention can identify potential state boundary violations in advance, improving the timeliness and accuracy of information demand triggering and enhancing the foresight and reliability of agricultural and livestock production management.

[0048] This invention constructs a mechanism for analyzing behavioral complexity and entropy changes based on the characteristics of agricultural and pastoral subjects. It can identify potential implicit information needs from changes in the frequency, sequence, and behavioral patterns of agricultural and pastoral subjects' operations, overcoming the limitations of existing technologies that can only identify explicit risks or needs. By incorporating behavioral changes into the information need determination process, this invention can provide targeted management suggestions and technical guidance to agricultural and pastoral subjects before risks fully manifest or indicators exceed limits, thereby enhancing the practical value and responsiveness of information services.

[0049] This invention constructs a multi-level response load balancing tree, dynamically selecting the information output level and content combination based on different information demand types, urgency levels, and the information carrying capacity of the stakeholders. This effectively solves the problems of excessive information push, mismatched response levels, or ignored key prompts in existing agricultural and pastoral information services. This approach makes information responses more aligned with the actual usage scenarios and production rhythms of agricultural and pastoral stakeholders, ensuring information sufficiency while reducing unnecessary interference, thus improving the availability, stability, and overall service effectiveness of information responses. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a flowchart of an agricultural and pastoral information demand response method based on an intelligent analysis model proposed in this invention;

[0052] Figure 2 This is a schematic diagram of the agricultural and livestock status trend prediction process based on the improved DLinear model, which is a method for responding to agricultural and livestock information demand based on an intelligent analysis model proposed in this invention. Detailed Implementation

[0053] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0054] refer to Figure 1 and Figure 2 A method for responding to agricultural and pastoral information demands based on an intelligent analysis model, comprising:

[0055] Collect multi-source agricultural and animal husbandry data during the agricultural and animal husbandry production process, preprocess the multi-source agricultural and animal husbandry data, and obtain agricultural and animal husbandry time series datasets;

[0056] Based on the production event data in the agricultural and animal husbandry time series dataset, an event-enhanced sequence is constructed, and the production events corresponding to irrigation, fertilization, feeding, pesticide application, ventilation, and equipment start-up and shutdown are converted into event label sequence, segmented factor sequence, and weight adjustment factor sequence, respectively.

[0057] The event augmentation sequence is input into the improved DLinear model. The event augmentation sequence is decomposed into trend and seasonal terms. The decomposed trend and seasonal terms are then linearly mapped and predicted to generate agricultural and pastoral status prediction results and trend change results within the target time window.

[0058] The main behavioral data in the agricultural and pastoral time series dataset is constructed into a behavioral feature sequence. Based on the trend change results, trend boundary indication information is generated. The behavioral entropy change of the behavioral feature sequence within a continuous time window is calculated to generate implicit demand indication information.

[0059] Based on trend boundary indication information and implicit demand indication information, combined with the constructed agricultural and animal husbandry causal chain fragments, agricultural and animal husbandry information demand is classified to generate information demand type, information demand object and demand urgency.

[0060] Construct a multi-level response load balancing tree, select the target response path from the multi-level response load balancing tree according to the information demand type, the information demand object, the urgency of the demand and the information carrying capacity of the subject, generate the information response strategy, and output the information response content to the target subject according to the information response strategy.

[0061] In this embodiment, the multi-source agricultural and animal husbandry data includes environmental time-series data, production event data, subject behavior data, and production result data during the agricultural and animal husbandry production process.

[0062] In this embodiment, the preprocessing of agricultural and animal husbandry multi-source data includes outlier removal, missing value completion, time alignment, format unification, and structured encoding to obtain a preprocessed dataset.

[0063] In this embodiment, the construction of event-enhanced sequences based on production event data in the agricultural and livestock time-series dataset includes:

[0064] The agricultural and animal husbandry time series dataset is divided into multiple continuous time windows according to a uniform time step. Production event records within each time window are extracted from the agricultural and animal husbandry time series dataset. The production event records include irrigation events, fertilization events, feeding events, pesticide application events, ventilation events, and equipment start-up and shutdown events.

[0065] For each type of production event within each time window, determine the occurrence status and event quantity of the production event within the time window, where the occurrence status indicates whether the production event has occurred within the time window, and the event quantity indicates the execution intensity or number of executions of the production event within the time window.

[0066] Based on the occurrence status and event quantity of various production events within each time window, an event marker vector corresponding to each time window is generated. The event marker vectors corresponding to each time window are arranged in chronological order to obtain the event marker sequence.

[0067] A segmented factor sequence is generated based on the change relationship between the event marker vectors corresponding to adjacent time windows. When the event marker vectors of adjacent time windows change, the current time window is marked as a new event segment. When the event marker vectors of adjacent time windows do not change, the current time window is kept in the current event segment. The event segment identifiers to which each time window belongs are arranged in chronological order to obtain the segmented factor sequence.

[0068] A weight adjustment factor sequence is generated based on the event label sequence and the segmented factor sequence. The weight adjustment factor is calculated for each time window. The weight adjustment factor is obtained by accumulating the event quantity of various production events within the historical time window according to the decay rule. The cumulative contribution of various production events is weighted and synthesized according to the event type weight. The event label sequence, the segmented factor sequence and the weight adjustment factor sequence are aligned and combined according to the time window to obtain the event enhancement sequence.

[0069] In this embodiment, the generation of agricultural and pastoral status prediction results and trend change results within the target time window includes:

[0070] The event augmentation sequence is organized into multiple consecutive time windows in chronological order. The length of the input window is determined to be L and the length of the prediction window is determined to be H. For the current time window, the event augmentation data corresponding to the previous L consecutive time windows are selected to form the input subsequence.

[0071] The input subsequence is fed into the improved DLinear model. This improved DLinear model adds an event-gated decomposition unit to the DLinear decomposition structure. The event-gated decomposition unit determines the decomposition parameters for the corresponding time window based on the event tag sequence, segmentation factor sequence, and weight adjustment factor sequence in the input subsequence. According to these parameters, the input subsequence is decomposed into trend component subsequences and seasonal component subsequences. Specifically, the decomposition of the input subsequence into trend component subsequences and seasonal component subsequences according to the decomposition parameters is as follows:

[0072] Based on the event marker sequence, determine whether each time window is in the state of production event action. When the corresponding time window has a production event marker, adjust the participation weight of the time window in the decomposition process so that the contribution of the data of the time window to the trend component and the seasonal component changes.

[0073] The input subsequence is divided into multiple continuous event segments based on the segmented factor sequence. The decomposition window length and decomposition smoothing coefficient are set for different event segments, so that the time windows of different event segments are decomposed at different scales to extract the trend component and the seasonal component.

[0074] The data amplitude of each time window in the input subsequence is weighted according to the weight adjustment factor sequence, so that the time window with higher weight occupies a higher proportion in the trend component and the time window with lower weight occupies a higher proportion in the seasonal component, thus obtaining the trend component subsequence and the seasonal component subsequence that match the intensity of the production event.

[0075] In the improved DLinear model, the trend component subsequence is predicted by adding a piecewise linear mapping unit to the linear mapping structure. This piecewise linear mapping unit selects corresponding linear mapping parameters for different event segments based on the segmented factor sequence, and performs linear mapping on the trend component subsequence to obtain the trend prediction component sequence for the next H time windows. Specifically, the process of performing linear mapping on the trend component subsequence to obtain the trend prediction component sequence for the next H time windows is as follows:

[0076] The trend component subsequence is divided into multiple continuous event segments based on the segmented factor sequence, and a set of linear mapping parameters is associated with each event segment so that different event segments correspond to different mapping relationships.

[0077] Within each event segment, the trend component subsequence within the event segment is linearly mapped using the linear mapping parameter corresponding to that event segment to generate intermediate trend prediction results corresponding to the event segment.

[0078] The intermediate trend forecast results of each event segment are spliced ​​together and aligned with time according to the time window order to form a complete trend forecast component sequence covering H future time windows.

[0079] In the improved DLinear model, seasonal component sequences are predicted. The piecewise linear mapping unit selects corresponding linear mapping parameters for different events based on the piecewise factor sequence, and performs linear mapping on the seasonal component sequences to obtain seasonal prediction component sequences for the next H time windows. Specifically, the process of performing linear mapping on the seasonal component sequences to obtain seasonal prediction component sequences for the next H time windows is as follows:

[0080] The seasonal component subsequence is divided into multiple continuous event segments based on the segmented factor sequence, and a linear mapping parameter corresponding to each event segment is determined for each event segment.

[0081] Within each event segment, the seasonal component subsequences within the event segment are linearly mapped using the linear mapping parameters corresponding to the event segment to generate intermediate seasonal prediction results corresponding to the event segment.

[0082] The intermediate seasonal forecast results of each event segment are spliced ​​together and time-aligned according to the time window order to form a complete seasonal forecast component sequence covering H future time windows.

[0083] In the improved DLinear model, the trend prediction component sequence and the seasonal prediction component sequence are fused. The improved DLinear model adds a component fusion unit to the output structure. The component fusion unit aligns the trend prediction component sequence and the seasonal prediction component sequence according to the time window and synthesizes them into an event-enhanced prediction sequence for the next H time windows. Data items corresponding to the agricultural and pastoral status are extracted to obtain the agricultural and pastoral status prediction results. Based on the agricultural and pastoral status prediction results, the trend change results are generated.

[0084] In this embodiment, the step of generating trend boundary indication information based on trend change results, calculating the behavioral entropy change of the behavioral feature sequence within a continuous time window, and generating implicit demand indication information includes:

[0085] The agricultural and pastoral status prediction results are divided into multiple prediction time windows according to the prediction time sequence. For each prediction time window, the corresponding prediction status value is compared with the management threshold corresponding to the prediction time window to generate trend boundary indication information that indicates whether the prediction status exceeds the management threshold.

[0086] The subject's behavioral data is divided into continuous behavioral time windows according to a uniform time granularity. Within each behavioral time window, the subject's operation type, operation sequence, operation interval time, and operation spatial location are extracted to construct the behavioral feature sequence of the corresponding time window.

[0087] For each behavior time window, the behavior feature sequence is discretized to obtain the behavior category distribution within the time window, and the behavior complexity index of the time window is calculated based on the behavior category distribution.

[0088] Between adjacent consecutive action time windows, the behavioral complexity index is compared, and the magnitude and direction of change of the behavioral complexity index are calculated. The magnitude of change is then weighted and corrected by considering the fluctuations in operation intervals and the changes in operation sequence within the action time window, resulting in the behavioral entropy change. Specifically, the calculation of the magnitude and direction of change of the behavioral complexity index is as follows:

[0089] Obtain the behavior complexity index value corresponding to the current behavior time window and the behavior complexity index value corresponding to the previous adjacent behavior time window. Calculate the difference between the two to obtain the change range of the behavior complexity index.

[0090] The direction of change of the behavior complexity index is determined based on the positive and negative relationship of the difference. When the behavior complexity index value of the current behavior time window is higher than the behavior complexity index value corresponding to the previous behavior time window, the direction of change is determined to be upward. When the behavior complexity index value of the current behavior time window is lower than the behavior complexity index value corresponding to the previous behavior time window, the direction of change is determined to be downward.

[0091] The magnitude of the change is associated with the direction of the change, so that the result of the change in the behavior complexity index includes both the magnitude of the change and the direction of the change.

[0092] Implicit demand indication information is generated based on the change in behavioral entropy. When the change in behavioral entropy shows a continuous upward trend within multiple consecutive behavioral time windows, it is determined that the subject has implicit information needs and corresponding implicit demand indication information is generated. When the change in behavioral entropy remains stable or decreases within consecutive behavioral time windows, implicit demand indication information is not generated.

[0093] In this embodiment, the generation of information demand type, information demand object, and demand urgency includes:

[0094] The trend out-of-bounds indication information and implicit demand indication information are used as input conditions for demand determination. At the same time, the predicted state type corresponding to the trend out-of-bounds indication information and the subject identifier and behavior time window identifier corresponding to the implicit demand indication information are obtained.

[0095] Based on long-term operational data of agricultural and animal husbandry production processes, a set of agricultural and animal husbandry causal chain segments is constructed. The agricultural and animal husbandry causal chain segments take state change - behavior change - management needs as the basic structural unit. Each causal chain segment includes state triggering conditions, behavior triggering conditions, demand type identifier, demand object mapping rules, and demand urgency generation rules.

[0096] When constructing a set of agricultural and pastoral causal chain segments, corresponding causal chain subsets are established for different production stages, different crop types, or different livestock objects. Different combinations of behavioral triggering conditions are configured for the same state change in the causal chain subsets to form an agricultural and pastoral causal chain segment structure.

[0097] The trend out-of-bounds indication information, implicit demand indication information, predicted state type and subject identifier are matched with the state triggering conditions and behavior triggering conditions in the agricultural and animal husbandry causal chain segment set in turn, and the causal chain segments that simultaneously meet the state triggering conditions and behavior triggering conditions are selected as candidate causal chain segments.

[0098] For each candidate causal chain segment, the information demand type is determined based on the demand type identifier corresponding to the causal chain segment, the information demand object is determined based on the demand object mapping rule of the causal chain segment, and the demand urgency is determined based on the demand urgency generation rule of the causal chain segment, combined with trend cross-boundary indication information and implicit demand indication information. The information demand type, information demand object and demand urgency are then output.

[0099] In this embodiment, the information response generation strategy outputs information response content to the target subject according to the information response strategy, including...

[0100] Construct a multi-level response load balancing tree, which is divided into multiple response levels according to the information output load from low to high. Each response level corresponds to a type of information output complexity. Under each response level, establish a set of candidate response information corresponding to the level.

[0101] When constructing a multi-level response load balancing tree, entry conditions are configured for each response level. The entry conditions are jointly determined by the information demand type, the urgency of the demand, the trend out-of-bounds indication information, and the implicit demand indication information.

[0102] In the multi-level response load balancing tree, information load attributes and demand matching attributes are configured for each candidate response information. The information load attribute represents the degree to which the response information occupies the subject's attention and terminal resources under the target output mode, while the demand matching attribute represents the degree of matching between the response information and the information demand type, information demand object, and demand urgency.

[0103] Based on the information carrying capacity of the subject, the response levels are traversed from top to bottom in the multi-level response load balancing tree. When the cumulative information load of the candidate response information set under the current response level does not exceed the information carrying capacity of the subject, the response level is selected as the target response level. Within the response level, the candidate response information is filtered and sorted according to the requirement matching attributes to generate the target response information set.

[0104] When generating the target response information set, a response stability constraint rule is introduced to constrain the consistency of information output levels across multiple consecutive response cycles. When the information demand type and urgency remain unchanged within consecutive response cycles, the selected response level remains constant, and only the target response information set is updated to generate the final information response strategy. The information response content is then output to the target entity according to this strategy. The response stability constraint rule refers to:

[0105] Within multiple consecutive response cycles, the consistency of the information demand type and urgency level generated in each response cycle is judged. When the information demand type is the same and the urgency level is at the same level in adjacent response cycles, the response cycle is determined to meet the stability condition.

[0106] When the response cycle meets the stability condition, the currently selected response level is locked so that the response level is not switched again in subsequent response cycles, and the target response information set is replaced, supplemented or deleted only within the scope of the response level.

[0107] When a change in the type of information need or a change in the urgency of the need across levels is detected, the locked state of the response level is released, allowing the response level to be reselected according to the new information need type and urgency, and an information response strategy is generated based on the new response level.

[0108] Example 1:

[0109] To verify the feasibility of this invention in practice, it was applied to an integrated agro-livestock management entity, which includes a corn-wheat rotation planting area of ​​approximately 360 mu (about 24 hectares) and a standardized breeding area with approximately 210 head of beef cattle. The area experiences high temperatures and humidity in summer, with uneven rainfall distribution, making the production process highly dependent on irrigation scheduling, ventilation control, and feeding management.

[0110] Prior to applying this invention, the business entity had deployed environmental monitoring equipment and a basic production record system to collect information on soil moisture, air temperature and humidity, cattle shed ventilation status, and production operations such as irrigation and feeding. Existing information service systems primarily rely on single environmental indicator thresholds for judgment and alerts; for example, irrigation alerts are triggered when soil moisture falls below a fixed threshold, and increased ventilation is prompted when cattle shed temperature exceeds a threshold. However, in actual operation, when continuous rainfall alternates with artificial irrigation, or during the initial stages of feed structure adjustments, the system frequently experiences delayed status judgments, misjudgments, and mismatches between information pushes and actual management needs. Managers often need to repeatedly review data and rely on experience for judgment.

[0111] After deploying the agricultural and livestock information demand response method based on the intelligent analysis model described in this invention in this scenario, the system first performs unified preprocessing on the collected multi-source agricultural and livestock data to form an agricultural and livestock time-series dataset, ensuring consistency between environmental data, production event data, and subject behavior data in the time dimension. During production operation, the system continuously extracts production events such as irrigation, fertilization, feeding, ventilation, and equipment start-up and shutdown, constructing event enhancement sequences so that trend analysis can clearly distinguish the production operation background before and after state changes occur.

[0112] Based on event-enhanced sequences, the system performs trend analysis on agricultural and pastoral conditions using an improved DLinear structure. During the analysis, event gating decomposition and piecewise linear mapping mechanisms are introduced to effectively reduce the interference of short-term fluctuations caused by human intervention on future judgments. For example, in the case of rainfall immediately following irrigation, the system can identify the portion of water changes caused by human intervention, preventing the triggering of water anomaly alerts after irrigation is completed.

[0113] The system analyzes the behavioral data of livestock farmers, constructing a behavioral feature sequence from the daily number of patrols, patrol time distribution, operation sequence, and data query behavior, and calculating the changes in behavioral complexity within a continuous time window. When the patrol frequency and operation sequence of livestock farmers fluctuate significantly within two to three days after feed adjustments, the system can generate implicit demand indication information, determining that managers have information needs regarding uncertainty about the current production status.

[0114] The system combines trend out-of-bounds indicators with implicit demand indicators and matches them with constructed agricultural and livestock causal chain segments to determine the corresponding information demand type, target audience, and urgency level. During the information response phase, the system uses a multi-level response load balancing tree to select an appropriate information output level based on the current production load and the main information carrying capacity, pushing only necessary and critical information content to avoid information redundancy.

[0115] Table 1. Comparison of Agricultural and Livestock Information Demand Response Effects Before and After Application of the Invention

[0116] Indicator Categories Average value before application Average value after application Improvement range Statistical period Application Area Data source Number of times a status misjudgment is triggered (times / month) 13.8 7.1 ↓48.6% 3 months Planting area + breeding area System Log Information request response delay (minutes) 29 17 ↓41.4% 3 months breeding area Response Log Average number of messages pushed per day (items) 16.2 10.8 ↓33.3% 3 months All regions Push Records Key information reading confirmation rate 54% 72% ↑18 percentage points 3 months breeding area User Feedback Effectiveness of management intervention 63% 75% ↑12 percentage points 3 months Planting area Production Records Managerial subjective satisfaction 3.4 / 5 4.2 / 5 ↑0.8 3 months All regions Questionnaire statistics

[0117] As can be seen from the data in Table 1, the application of this invention has significantly improved the accuracy of agricultural and livestock information services in terms of status judgment. The average number of false status judgments triggered by the system per month decreased from 13.8 times to 7.1 times, a reduction of nearly half. This indicates that by introducing a trend analysis mechanism enhanced by production events, the impact of natural environmental changes and human production operations on status indicators can be effectively distinguished, reducing false triggers caused by simple threshold judgments, and making the triggering of information demands more consistent with the actual production process.

[0118] In terms of information response efficiency and information load control, this invention also demonstrates excellent results. The average information demand response delay was reduced from 29 minutes to 17 minutes, indicating that the system can more promptly identify and respond to genuine information needs. The average daily number of information pushes decreased from 16.2 to 10.8. This reduction in the number of pushes did not weaken the information service capability; on the contrary, the multi-level response load balancing mechanism improved the targeting of information pushes, avoiding information redundancy and information overload for management personnel.

[0119] In terms of information usage effectiveness and subjective experience, the key information reading confirmation rate increased from 54% to 72%, the management intervention effectiveness rate increased from 63% to 75%, and the subjective satisfaction of managers also increased from 3.4 points to 4.2 points. This indicates that while improving the accuracy of information identification, the present invention makes the output information content more aligned with actual management needs, increases the probability of information being adopted and transformed into effective management behavior, and improves the overall practicality and usability of agricultural and livestock information services.

[0120] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for responding to agricultural and pastoral information demands based on an intelligent analysis model, characterized in that, include: Collect multi-source agricultural and animal husbandry data during the agricultural and animal husbandry production process, preprocess the multi-source agricultural and animal husbandry data, and obtain agricultural and animal husbandry time series datasets; Based on the production event data in the agricultural and animal husbandry time series dataset, an event-enhanced sequence is constructed, and the production events corresponding to irrigation, fertilization, feeding, pesticide application, ventilation, and equipment start-up and shutdown are converted into event label sequence, segmented factor sequence, and weight adjustment factor sequence, respectively. The event augmentation sequence is input into the improved DLinear model. The event augmentation sequence is decomposed into trend and seasonal terms. The decomposed trend and seasonal terms are then linearly mapped and predicted to generate agricultural and pastoral status prediction results and trend change results within the target time window. The main behavioral data in the agricultural and pastoral time series dataset is constructed into a behavioral feature sequence. Based on the trend change results, trend boundary indication information is generated. The behavioral entropy change of the behavioral feature sequence within a continuous time window is calculated to generate implicit demand indication information. Based on trend boundary indication information and implicit demand indication information, combined with the constructed agricultural and animal husbandry causal chain fragments, agricultural and animal husbandry information demand is classified to generate information demand type, information demand object and demand urgency. Construct a multi-level response load balancing tree, select the target response path from the multi-level response load balancing tree according to the information demand type, the information demand object, the urgency of the demand and the information carrying capacity of the subject, generate the information response strategy, and output the information response content to the target subject according to the information response strategy.

2. The agricultural and pastoral information demand response method based on an intelligent analysis model according to claim 1, characterized in that, The multi-source agricultural and animal husbandry data includes environmental time-series data, production event data, subject behavior data, and production result data in the agricultural and animal husbandry production process.

3. The agricultural and pastoral information demand response method based on an intelligent analysis model according to claim 1, characterized in that, The preprocessing of multi-source agricultural and pastoral data includes outlier removal, missing value completion, time alignment, format unification, and structured encoding to obtain a preprocessed dataset.

4. The agricultural and pastoral information demand response method based on an intelligent analysis model according to claim 1, characterized in that, The construction of event-enhanced sequences based on production event data in agricultural and pastoral time-series datasets includes: The agricultural and animal husbandry time series dataset is divided into multiple continuous time windows according to a uniform time step, and production event records within each time window are extracted from the agricultural and animal husbandry time series dataset. For each type of production event within each time window, determine the occurrence status and event quantity of the production event within that time window; Based on the occurrence status and event quantity of various production events within each time window, an event marker vector corresponding to each time window is generated. The event marker vectors corresponding to each time window are arranged in chronological order to obtain the event marker sequence. A segmented factor sequence is generated based on the change relationship between the event marker vectors corresponding to adjacent time windows. When the event marker vectors of adjacent time windows change, the current time window is marked as a new event segment. When the event marker vectors of adjacent time windows do not change, the current time window is kept in the current event segment. The event segment identifiers to which each time window belongs are arranged in chronological order to obtain the segmented factor sequence. A weight adjustment factor sequence is generated based on the event label sequence and the segmented factor sequence. The weight adjustment factor is calculated for each time window. The weight adjustment factor is obtained by accumulating the event quantity of various production events within the historical time window according to the decay rule. The cumulative contribution of various production events is weighted and synthesized according to the event type weight. The event label sequence, the segmented factor sequence and the weight adjustment factor sequence are aligned and combined according to the time window to obtain the event enhancement sequence.

5. The agricultural and pastoral information demand response method based on an intelligent analysis model according to claim 1, characterized in that, The predicted results and trend changes of agricultural and pastoral status within the target time window include: The event augmentation sequence is organized into multiple consecutive time windows in chronological order. The length of the input window is determined to be L and the length of the prediction window is determined to be H. For the current time window, the event augmentation data corresponding to the previous L consecutive time windows are selected to form the input subsequence. The input subsequence is fed into the improved DLinear model, which adds an event-gated decomposition unit to the decomposition structure of DLinear. The event-gated decomposition unit determines the decomposition parameters of the corresponding time window based on the event tag sequence, segmentation factor sequence and weight adjustment factor sequence in the input subsequence, and decomposes the input subsequence into trend component subsequence and seasonal component subsequence according to the decomposition parameters. In the improved DLinear model, the trend component subsequence is predicted. The improved DLinear model adds a piecewise linear mapping unit to the linear mapping structure. The piecewise linear mapping unit selects the corresponding linear mapping parameters for different events based on the piecewise factor sequence, and performs linear mapping on the trend component subsequence to obtain the trend prediction component sequence for the next H time windows. In the improved DLinear model, seasonal component sequences are predicted. The piecewise linear mapping unit selects the corresponding linear mapping parameters for different events based on the piecewise factor sequence, and performs linear mapping on the seasonal component sequences to obtain the seasonal predicted component sequences for the next H time windows. In the improved DLinear model, the trend prediction component sequence and the seasonal prediction component sequence are fused. The improved DLinear model adds a component fusion unit to the output structure. The component fusion unit aligns the trend prediction component sequence and the seasonal prediction component sequence according to the time window and synthesizes them into an event-enhanced prediction sequence for the next H time windows. Data items corresponding to the agricultural and pastoral status are extracted to obtain the agricultural and pastoral status prediction results. Based on the agricultural and pastoral status prediction results, the trend change results are generated.

6. The agricultural and pastoral information demand response method based on an intelligent analysis model according to claim 1, characterized in that, The process of generating trend boundary indication information based on trend change results, calculating the behavioral entropy change of the behavioral feature sequence within a continuous time window, and generating implicit demand indication information includes: The agricultural and pastoral status prediction results are divided into multiple prediction time windows according to the prediction time sequence. For each prediction time window, the corresponding prediction status value is compared with the management threshold corresponding to the prediction time window to generate trend boundary indication information that indicates whether the prediction status exceeds the management threshold. The subject's behavioral data is divided into continuous behavioral time windows according to a uniform time granularity. Within each behavioral time window, the subject's operation type, operation sequence, operation interval time, and operation spatial location are extracted to construct the behavioral feature sequence of the corresponding time window. For each behavior time window, the behavior feature sequence is discretized to obtain the behavior category distribution within the time window, and the behavior complexity index of the time window is calculated based on the behavior category distribution. Between adjacent consecutive action time windows, the action complexity index is compared, the magnitude and direction of change of the action complexity index are calculated, and the magnitude of change is weighted and corrected by combining the degree of fluctuation of operation interval and the degree of change of operation sequence within the action time window to obtain the change of action entropy. Implicit demand indication information is generated based on the change in behavioral entropy. When the change in behavioral entropy shows a continuous upward trend within multiple consecutive behavioral time windows, it is determined that the subject has implicit information needs and corresponding implicit demand indication information is generated. When the change in behavioral entropy remains stable or decreases within consecutive behavioral time windows, implicit demand indication information is not generated.

7. The agricultural and pastoral information demand response method based on an intelligent analysis model according to claim 1, characterized in that, The information request type, information request object, and urgency level include: The trend out-of-bounds indication information and implicit demand indication information are used as input conditions for demand determination. At the same time, the predicted state type corresponding to the trend out-of-bounds indication information and the subject identifier and behavior time window identifier corresponding to the implicit demand indication information are obtained. Based on long-term operational data of agricultural and animal husbandry production processes, a set of agricultural and animal husbandry causal chain segments is constructed. The agricultural and animal husbandry causal chain segments take state change - behavior change - management needs as the basic structural unit. Each causal chain segment includes state triggering conditions, behavior triggering conditions, demand type identifier, demand object mapping rules, and demand urgency generation rules. When constructing a set of agricultural and pastoral causal chain segments, corresponding causal chain subsets are established for different production stages, different crop types, or different livestock objects. Different combinations of behavioral triggering conditions are configured for the same state change in the causal chain subsets to form an agricultural and pastoral causal chain segment structure. The trend out-of-bounds indication information, implicit demand indication information, predicted state type and subject identifier are matched with the state triggering conditions and behavior triggering conditions in the agricultural and animal husbandry causal chain segment set in turn, and the causal chain segments that simultaneously meet the state triggering conditions and behavior triggering conditions are selected as candidate causal chain segments. For each candidate causal chain segment, the information demand type is determined based on the demand type identifier corresponding to the causal chain segment, the information demand object is determined based on the demand object mapping rule of the causal chain segment, and the demand urgency is determined based on the demand urgency generation rule of the causal chain segment, combined with trend cross-boundary indication information and implicit demand indication information. The information demand type, information demand object and demand urgency are then output.

8. The agricultural and pastoral information demand response method based on an intelligent analysis model according to claim 1, characterized in that, The information response generation strategy outputs information response content to the target subject according to the information response strategy, including... Construct a multi-level response load balancing tree, which is divided into multiple response levels according to the information output load from low to high. Each response level corresponds to a type of information output complexity. Under each response level, establish a candidate response information set corresponding to the level. When constructing a multi-level response load balancing tree, entry conditions are configured for each response level. The entry conditions are jointly determined by the information demand type, the urgency of the demand, the trend out-of-bounds indication information, and the implicit demand indication information. In the multi-level response load balancing tree, information load attributes and demand matching attributes are configured for each candidate response information. The information load attribute represents the degree to which the response information occupies the subject's attention and terminal resources under the target output mode, while the demand matching attribute represents the degree of matching between the response information and the information demand type, information demand object, and demand urgency. Based on the information carrying capacity of the subject, the response levels are traversed from top to bottom in the multi-level response load balancing tree. When the cumulative information load of the candidate response information set under the current response level does not exceed the information carrying capacity of the subject, the response level is selected as the target response level. Within the response level, the candidate response information is filtered and sorted according to the requirement matching attributes to generate the target response information set. When generating the target response information set, a response stability constraint rule is introduced to constrain the consistency of the information output level within multiple consecutive response cycles. When the information demand type and urgency level do not change within consecutive response cycles, the selected response level is kept from jumping, and only the target response information set is updated to generate the final information response strategy. The information response content is then output to the target subject according to the information response strategy.

Images