A smart agriculture remote monitoring method based on the Internet of Things

Abstract

The present application relates to the technical field of data processing, and especially relates to a kind of wisdom agricultural remote monitoring method based on Internet of Things, the method obtains the multidimensional monitoring data sequence of any crop and is divided into at least two monitoring data sequences, each monitoring data sequence is decomposed into at least two time scale subsequence, constructs the causal skeleton diagram under each time scale;According to the data correlation between the corresponding subsequence of any two monitoring data sequences in each causal skeleton diagram, obtain the total direction index of causal relationship;According to the data correlation between any two monitoring data sequences, obtain the intensity index of causal relationship, according to the total direction index of causal relationship and the intensity index of causal relationship, obtain the optimized causal relationship between any two monitoring data sequences;Optimized causal relationship between each two monitoring data sequences is used to build Bayesian network structure model, improve the credibility of causal relationship in Bayesian network structure model, realize crop intelligent monitoring.

Description

A Smart Agriculture Remote Monitoring Method Based on the Internet of Things Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for remote monitoring of smart agriculture based on the Internet of Things. Background Technology

[0002] As a core means of realizing refined management in modern agriculture, the IoT-based smart agriculture remote monitoring system plays a crucial role in ensuring high-yield, high-quality, and efficient crop production through accurate diagnosis of the root causes of environmental anomalies and timely decision-making responses. However, agricultural monitoring data exhibits strong spatiotemporal coupling and multi-scale periodicity. Traditional monitoring methods, which largely rely on isolated threshold alarms and superficial statistical analysis, fail to effectively identify the root causes of multiple alarms from the monitoring platform. Without intervention, the optimal time for intervention may be missed, potentially leading to economic losses due to reduced crop yields.

[0003] Existing technologies typically employ PC algorithms to construct Bayesian network structure models, effectively expressing the probabilistic dependencies between variables, diagnosing and predicting the causal relationships between various monitoring data of crop growth, and tracing the root cause of abnormal events from massive IoT data.

[0004] However, the PC algorithm relies on graph theory rules to infer causal direction, which can easily lead to causal relationships that are directionless or confusing. Furthermore, the PC algorithm uses an inherent temporal data granularity analysis framework, which cannot take into account agricultural data at multiple time scales. Specifically, it ignores the multi-temporal rhythm characteristics of agricultural monitoring data, resulting in low reliability of directional judgment. That is, when the algorithm runs on short-term data, although it can capture instantaneous causes such as equipment failure and irrigation, it will miss causal patterns that require long-term observation to become apparent, such as accumulated temperature → growth cycle. This leads to insufficient credibility and limited practicality of the Bayesian network structure model generated by the current technology in complex agricultural environments.

[0005] Therefore, how to improve the credibility of causal relationships in Bayesian network structure models based on the multi-scale periodicity characteristics of agricultural monitoring data has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a smart agriculture remote monitoring method based on the Internet of Things to solve the problem of how to improve the credibility of causal relationships in Bayesian network structure models based on the multi-scale periodicity characteristics of agricultural monitoring data.

[0007] This invention provides a method for remote monitoring of smart agriculture based on the Internet of Things, which includes the following steps:

[0008] Obtain a multidimensional monitoring data sequence of any crop within a preset historical period, and divide the multidimensional monitoring data sequence into at least two monitoring data sequences according to the dimensions.

[0009] Each monitoring data sequence is decomposed into subsequences at least two time scales. The PC algorithm is used to construct a causal skeleton graph of all subsequences at each time scale. The nodes in the causal skeleton graph are subsequences, and the edges are the causal relationships between the nodes.

[0010] For any two monitoring data sequences, based on the data correlation between the corresponding subsequences in each causal skeleton diagram, the causal relationship direction index of the two monitoring data sequences in each causal skeleton diagram is obtained. Based on the causal relationship direction index of the two monitoring data sequences in each causal skeleton diagram, the total causal relationship direction index between the two monitoring data sequences is obtained.

[0011] Based on the data correlation between any two monitoring data sequences, obtain the causal relationship strength index between any two monitoring data sequences; based on the total causal relationship direction index and the causal relationship strength index, obtain the optimized causal relationship between any two monitoring data sequences.

[0012] By utilizing the optimized causal relationship between every two monitoring data sequences, a Bayesian network structure model of any crop is constructed for intelligent monitoring of the future growth stages of any crop.

[0013] Preferably, the step of obtaining the causal relationship direction index of any two monitoring data sequences in each causal skeleton diagram based on the data correlation between the corresponding subsequences of any two monitoring data sequences in each causal skeleton diagram includes:

[0014] For any causal skeleton diagram, the subsequences corresponding to any two monitoring data sequences in any causal skeleton diagram are denoted as target subsequence groups. The lag time range between the target subsequence groups is obtained. Based on the data correlation between the target subsequences in the target subsequence groups, the standardized cross-correlation coefficient of the target subsequence groups at each lag time within the lag time range is obtained.

[0015] If the target subsequence group has a causal relationship in any causal skeleton graph, then the causal relationship direction index of any two monitoring data sequences in any causal skeleton graph is obtained based on the difference between the standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range.

[0016] Preferably, the step of obtaining the causal relationship direction index of any two monitoring data sequences in any causal skeleton diagram based on the difference between the standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range includes:

[0017] For any lag time within the lag time range, if any lag time is greater than or equal to 0, the standardized cross-correlation coefficient of the target subsequence group at any lag time is recorded as a positive correlation coefficient; if any lag time is less than 0, the standardized cross-correlation coefficient of the target subsequence group at any lag time is recorded as a negative correlation coefficient.

[0018] The standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range are divided into a set of positive correlation coefficients and a set of negative correlation coefficients. The sum of the maximum absolute values ​​of the positive correlation coefficients in the positive correlation coefficient set and the negative correlation coefficients in the negative correlation coefficient set is obtained to obtain the total correlation coefficient. The difference between the maximum absolute values ​​of the positive correlation coefficients in the positive correlation coefficient set and the negative correlation coefficients in the negative correlation coefficient set is obtained to obtain the correlation coefficient difference. The ratio of the correlation coefficient difference to the total correlation coefficient is obtained to obtain the causal relationship direction index of any two monitoring data sequences in any causal skeleton diagram.

[0019] Preferably, obtaining the overall causal relationship direction index between any two monitoring data sequences based on the causal relationship direction index in each causal skeleton diagram includes:

[0020] Bootstrap resampling is performed on the subsequences corresponding to any two monitoring data sequences at each time scale to obtain the Bootstrap confidence of any two monitoring data sequences at each time scale. The formula for calculating the overall causal direction index between any two monitoring data sequences is as follows:

[0021] ;

[0022] in, This is the overall directional index of the causal relationship between any two monitoring data sequences; The j-th causal relationship direction indicator for any two monitoring data sequences; The Bootstrap confidence level at the time scale corresponding to the j-th causal relationship direction indicator between any two monitoring data sequences; The number of causal relationship direction indicators for any two monitoring data sequences; Set a preset consistency sensitivity coefficient; For the causal skeleton diagram corresponding to the j-th causal relationship direction index of any two monitoring data sequences, the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value between the subsequences corresponding to the two monitoring data sequences is defined. For any two monitoring data sequences, in each causal skeleton diagram with causal relationship directional indicators, the median of the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value between the corresponding subsequences is used; For any two monitoring data sequences, in each causal skeleton diagram with causal relationship directional indicators, the maximum absolute value of the cross-correlation coefficient between the corresponding subsequences is the standardized maximum lag time corresponding to the maximum lag time. For any two monitoring data sequences, the minimum lag time corresponding to the standardized cross-correlation coefficient between the corresponding subsequences in each causal skeleton diagram with causal relationship directional indicators is the maximum absolute value between them. It is an exponential function with the natural constant as the base.

[0023] Preferably, obtaining the causal relationship strength index between any two monitoring data sequences based on the data correlation between the two sequences includes:

[0024] The maximum value of the standardized cross-correlation coefficient between the maximum absolute value of any two monitoring data sequences in each causal skeleton diagram is obtained and denoted as the maximum standardized cross-correlation coefficient. The time scale corresponding to the maximum standardized cross-correlation coefficient is denoted as the target time scale, and the lag time corresponding to the maximum standardized cross-correlation coefficient is denoted as the target lag time.

[0025] Based on the target lag time and the causal relationship direction index in the causal skeleton diagram of any two monitoring data sequences at the target time scale, obtain the causal variable sequence and the result variable sequence corresponding to any two monitoring data sequences;

[0026] Linear fitting is performed on the causal variable sequence and the outcome variable sequence to obtain a fitting curve, and the slope of the fitting curve is used as an index of the causal relationship strength between any two monitoring data sequences.

[0027] Preferably, the step of obtaining the causal variable sequence and the outcome variable sequence corresponding to the arbitrary two monitoring data sequences based on the target lag time and the causal relationship direction index in the causal skeleton diagram of the arbitrary two monitoring data sequences at the target time scale includes:

[0028] The causal relationship direction index of any two monitoring data sequences in the causal skeleton diagram at the target time scale is a vector. Based on the causal relationship direction index of any two monitoring data sequences in the causal skeleton diagram at the target time scale, the target subsequences corresponding to any two monitoring data sequences in the target time scale are divided into cause subsequences and result subsequences.

[0029] Remove the last target lag time monitoring data from the causal subsequence to obtain the causal variable sequence, and remove the first target lag time monitoring data from the result subsequence to obtain the result variable sequence.

[0030] Preferably, obtaining the optimized causal relationship between any two monitoring data sequences based on the overall causal direction index and the causal strength index includes:

[0031] The direction in the total causal direction index is recorded as the causal direction between any two monitoring data sequences, the value in the total causal direction index is recorded as the causal confidence between any two monitoring data sequences, the causal strength index is recorded as the effect strength between any two monitoring data sequences, and the causal direction, causal confidence, and effect strength between any two monitoring data sequences are combined to form the optimized causal relationship between any two monitoring data sequences.

[0032] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0033] In this invention, a causal skeleton diagram of all subsequences at each time scale is constructed to reflect the causal relationship between any two monitoring data sequences at different time scales. A causal relationship direction index is obtained to preliminarily determine the causal direction and directional certainty of any two monitoring data sequences in each causal skeleton diagram; the larger the value of the causal relationship direction index, the greater the directional certainty of the causal direction. A total causal relationship direction index is obtained, which, combined with different time scales, comprehensively reflects the final causal direction and the directional confidence of any two monitoring data sequences. A causal relationship strength index is obtained to quantify the actual impact of the causal relationship between any two monitoring data sequences; the larger the causal relationship strength index, the greater the causal effect strength. An optimized causal relationship is obtained to construct a Bayesian network structure model, outputting a complete causal map with direction, confidence, and effect strength, ultimately achieving intelligent monitoring and source tracing diagnosis of crop causal relationships. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 is a flowchart of a smart agriculture remote monitoring method based on the Internet of Things provided in Embodiment 1 of the present invention. Detailed Implementation

[0036] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0037] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0038] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0039] Referring to Figure 1, which is a flowchart of a smart agriculture remote monitoring method based on the Internet of Things provided in Embodiment 1 of the present invention, as shown in Figure 1, the method may include:

[0040] Step S101: Obtain a multidimensional monitoring data sequence of any crop within a preset historical period, and divide the multidimensional monitoring data sequence into at least two monitoring data sequences according to the dimensions.

[0041] As a core means of realizing the refined management of modern agriculture, the IoT-based smart agriculture remote monitoring system aims to automatically and accurately trace the root path of abnormal events from massive IoT data, thereby upgrading the regulatory decision-making from "passive alarm" to "active intervention" and ensuring the stable operation of the agricultural production system and maximizing economic benefits. Existing technologies typically use PC algorithms to construct Bayesian network structure models to diagnose and predict the causal relationships between various monitoring data of crop growth.

[0042] This embodiment utilizes a multi-source sensor network deployed within a smart agriculture remote monitoring system to collect multi-dimensional monitoring data of crops in real time. This data includes, but is not limited to, environmental data (temperature, humidity, light, CO2), equipment status (irrigation, ventilation, supplemental lighting), and crop monitoring data (images, growth indicators). Since the monitoring method is the same for each crop, this embodiment uses any one crop as an example, acquiring multi-dimensional monitoring data for each monitoring moment within a preset historical time period to form a multi-dimensional monitoring data sequence. Because it is necessary to diagnose and predict the causal relationships between various monitoring data related to crop growth, the obtained multi-dimensional monitoring data sequence is divided into at least two monitoring data sequences according to dimensions, with one dimension corresponding to one monitoring data sequence. This is used to analyze the causal relationships of any crop during its growth stages, achieving intelligent crop monitoring. In this embodiment, the preset historical time period is the historical maturity cycle of any crop (from sowing to maturity). The multi-dimensional monitoring data is collected every 10 minutes, but this is not limited and can be set according to the specific implementation scenario.

[0043] Among them, the collected multidimensional monitoring data needs to be preprocessed to transform the original and messy IoT sensor data into high-quality, spatiotemporally aligned multi-scale structured data to obtain multidimensional monitoring data sequences. Data preprocessing is an existing technology, and will be briefly described here: (1) Spatiotemporal alignment and standardization: The collected multidimensional monitoring data is stamped with a unified high-precision timestamp to solve the problem of clock asynchrony between sensor devices, the sensor readings are mapped to specific regulatory areas, the "device-location-crop" association is established, the units are unified and the dimensions are standardized to eliminate the influence of measurement scale differences; (2) Data quality cleaning and repair: Based on statistical methods and domain knowledge, outliers (such as monitoring data that exceed the reasonable range of agronomy) are identified and processed, and the spatiotemporal KNN interpolation algorithm is used to repair missing data. The readings of adjacent sensors and similar times are used to intelligently fill in the detection of sensor health status, and devices with continuous abnormalities are marked and alarmed.

[0044] Because PC algorithms rely on graph theory rules for causal direction inference, they are prone to causal relationships that are directionless or confusing. Furthermore, PC algorithms use an inherent temporal data granularity analysis framework, which cannot take into account agricultural data at multiple time scales. Specifically, they ignore the multi-temporal rhythm characteristics of agricultural monitoring data, resulting in low reliability of directional judgments. That is, when the algorithm runs on short-term data, although it can capture instantaneous causes such as equipment failure and irrigation, it will miss causal patterns that require long-term observation to become apparent, such as accumulated temperature → growth cycle. This leads to insufficient credibility and limited practicality of the Bayesian network structure model generated by existing technologies in complex agricultural environments.

[0045] Therefore, this embodiment constructs causal skeleton diagrams at different time scales. By obtaining the causal relationship direction index of each pair of monitoring data sequences in each causal skeleton diagram, the causal relationship and direction of each pair of monitoring data sequences are comprehensively analyzed to obtain the total causal relationship direction index. Then, the causal relationship strength index between each pair of monitoring data sequences is obtained to quantify the actual impact of the causal relationship. In turn, the optimized causal relationship between each pair of monitoring data sequences is obtained, and a Bayesian network structure model is constructed to intelligently monitor the future growth stage of any crop, ultimately realizing intelligent monitoring and source tracing diagnosis of crop causal relationships.

[0046] Step S102: Decompose each monitoring data sequence into subsequences at least two time scales, and use the PC algorithm to construct a causal skeleton graph of all subsequences at each time scale. The nodes in the causal skeleton graph are subsequences, and the edges are the causal relationships between the nodes.

[0047] Because agricultural monitoring data has multiple time scales, traditional single-scale analysis may miss important causal relationships or introduce spurious associations. Causal relationships on long time scales (such as fertilizer → growth status) cannot be detected on short time scales, and causal relationships on short time scales (such as watering → humidity) will be ignored on long time scales.

[0048] Therefore, in this embodiment, multi-scale resampling is performed on each monitoring data sequence, decomposing each monitoring data sequence into sub-sequences at three time scales: minute-level (preserving short-term fluctuation characteristics), hour-level (capturing daily patterns), and day-level (reflecting long-term trends). A causal skeleton graph of all sub-sequences at each time scale is constructed using the PC algorithm. Nodes in the causal skeleton graph represent sub-sequences, and edges represent causal relationships between nodes (including connections between nodes but not directional information; that is, if nodes are connected by an edge, it indicates a causal relationship between the two nodes, but the direction of the causal relationship is uncertain). The multi-scale resampling decomposition of time scales and the PC algorithm for constructing the causal skeleton graph are existing technologies and will not be elaborated upon here.

[0049] Thus, we have obtained the causal framework diagrams at three time scales.

[0050] Step S103: For any two monitoring data sequences, based on the data correlation between the corresponding subsequences in each causal skeleton diagram, obtain the causal relationship direction index of the two monitoring data sequences in each causal skeleton diagram, and based on the causal relationship direction index of the two monitoring data sequences in each causal skeleton diagram, obtain the total causal relationship direction index between the two monitoring data sequences.

[0051] The causal skeleton diagram illustrates the causal relationships between subsequences at each time scale. For any two monitoring data sequences, if the corresponding subsequences of the two monitoring data sequences at any time scale have a causal relationship in the causal skeleton diagram, it indicates that there is a causal relationship between the two monitoring data sequences. However, the causal relationship in the causal skeleton diagram does not include directional information. Moreover, since the direction judgment of the PC algorithm is based on graph structure rules, it is unreliable in agricultural scenarios with multimodal time series data.

[0052] Therefore, this embodiment introduces cross-correlation analysis as a more reliable directional evidence. Based on the data correlation between the corresponding subsequences of any two monitoring data sequences in each causal skeleton diagram, the causal relationship direction index of any two monitoring data sequences in each causal skeleton diagram is obtained, and the causal direction of any two monitoring data sequences in each causal skeleton diagram and the directional certainty of the causal direction are preliminarily determined.

[0053] For any causal skeleton graph, the method for obtaining the causal relationship direction index of any two monitoring data sequences in any causal skeleton graph based on the data correlation between the corresponding subsequences of any two monitoring data sequences in any causal skeleton graph is as follows:

[0054] (1) The subsequences corresponding to any two monitoring data sequences in any causal skeleton diagram are recorded as target subsequence groups. The lag time range between the target subsequence groups is obtained. Based on the data correlation between the target subsequences in the target subsequence group, the standardized cross-correlation coefficient of the target subsequence group at each lag time in the lag time range is obtained.

[0055] In this embodiment, one-quarter of the length of the target subsequence in the target subsequence group is used as the lag time range between the target subsequence groups. Assuming that the number of monitored data in the target subsequence is 100, the lag time range is [-25, 25]. There is no limitation here, and it can be set according to the specific implementation scenario. The positive and negative signs only represent the leading result of the target subsequence. For example, for target subsequence 1 and target subsequence 2, a positive lag time means that target subsequence 1 leads target subsequence 2, a negative lag time means that target subsequence 2 leads target subsequence 1, and a lag time of 0 means that target subsequence 1 and target subsequence 2 are instantaneously correlated and there is no lag relationship.

[0056] In one implementation, the target subsequence group corresponding to the monitoring data sequence X and monitoring data sequence Y in the i-th causal skeleton graph ( and ), and lag time For example, the target subsequence group ( and In the lag time The formula for calculating the standardized cross-correlation coefficient at point is:

[0057] ;

[0058] in, For the target subsequence group ( and In the lag time Standardized cross-correlation coefficients at the location; For target subsequence The kth monitoring data point; For target subsequence The average value of the monitoring data; For target subsequence The Middle One monitoring data point; For target subsequence The average value of the monitoring data; N is the target subsequence. or The amount of monitoring data in; The method for calculating the standardized cross-correlation coefficient is existing technology and will not be elaborated here.

[0059] Similarly, the target subsequence group is obtained ( and The standardized cross-correlation coefficient at each lag time within the lag time range.

[0060] (2) If the target subsequence group has a causal relationship in any causal skeleton diagram, then based on the difference between the standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range, the causal relationship direction index of any two monitoring data sequences in any causal skeleton diagram is obtained.

[0061] Specifically, for any lag time within the lag time range, if any lag time is greater than or equal to 0, the target subsequence group ( and The standardized cross-correlation coefficient at any lag time is denoted as the positive correlation coefficient, i.e., the target subsequence group ( and The causal relationship at any of the said lag times is If any lag time is less than 0, then the standardized cross-correlation coefficient of the target subsequence group at any lag time is recorded as the negative correlation coefficient, i.e., the target subsequence group ( and The causal relationship at any of the said lag times is ;

[0062] The standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range are divided into a set of positive correlation coefficients and a set of negative correlation coefficients. The sum of the maximum absolute values ​​of the positive correlation coefficients in the positive correlation coefficient set and the negative correlation coefficients in the negative correlation coefficient set is obtained to obtain the total correlation coefficient. The difference between the maximum absolute values ​​of the positive correlation coefficients in the positive correlation coefficient set and the negative correlation coefficients in the negative correlation coefficient set is obtained to obtain the correlation coefficient difference. The ratio of the correlation coefficient difference to the total correlation coefficient is obtained to obtain the causal relationship direction index of any two monitoring data sequences in any causal skeleton diagram.

[0063] In one embodiment, taking monitoring data sequence X and monitoring data sequence Y and the i-th causal skeleton diagram as an example, the calculation formula for the causal relationship direction index of monitoring data sequence X and monitoring data sequence Y in the i-th causal skeleton diagram is as follows:

[0064] ;

[0065] in, The causal relationship direction indicator for monitoring data sequence X and monitoring data sequence Y in the i-th causal skeleton diagram; The maximum absolute value of the set of positive correlation coefficients is the positive correlation coefficient. The negative correlation coefficient is the largest absolute value of the set of negative correlation coefficients.

[0066] It should be noted that, The larger the value, the more likely it is to be a target subsequence group ( and The causal relationship at the corresponding lag time is: The greater the certainty of the direction, The larger the value, the more likely it is to be a target subsequence group ( and The causal relationship at the corresponding lag time is: The greater the certainty of the direction, The value of is [-1, 1], where a positive value indicates that the causal relationship direction between monitoring data sequence X and monitoring data sequence Y in the i-th causal skeleton diagram is . A negative value indicates that the causal relationship between monitoring data sequence X and monitoring data sequence Y in the i-th causal skeleton diagram is in the direction of... The larger the value (i.e.) The larger the value, the greater the certainty of the direction of the causal relationship.

[0067] Following the method for obtaining the causal relationship direction index of monitoring data sequence X and monitoring data sequence Y in the i-th causal skeleton diagram, obtain the causal relationship direction index of monitoring data sequence X and monitoring data sequence Y in each causal skeleton diagram.

[0068] Furthermore, by combining different time scales, the causal relationship direction index of any two monitoring data sequences in each causal skeleton diagram is analyzed to obtain the overall causal relationship direction index between any two monitoring data sequences, comprehensively reflecting the final causal direction and the direction confidence of the final causal direction between any two monitoring data sequences. The method for obtaining the overall causal relationship direction index between any two monitoring data sequences is as follows:

[0069] Bootstrap resampling is performed on the subsequences corresponding to any two monitoring data sequences at each time scale to obtain the bootstrap confidence of any two monitoring data sequences at each time scale. Bootstrap resampling is an existing technology and will not be elaborated here. The formula for calculating the overall causal direction index between any two monitoring data sequences is as follows:

[0070] ;

[0071] in, This is the overall directional index of the causal relationship between any two monitoring data sequences; The j-th causal relationship direction indicator for any two monitoring data sequences; The Bootstrap confidence level at the time scale corresponding to the j-th causal relationship direction indicator between any two monitoring data sequences; The number of causal relationship direction indicators for any two monitoring data sequences; Set a preset consistency sensitivity coefficient; For the causal skeleton diagram corresponding to the j-th causal relationship direction index of any two monitoring data sequences, the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value between the subsequences corresponding to the two monitoring data sequences is defined. For any two monitoring data sequences, in each causal skeleton diagram with causal relationship directional indicators, the median of the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value between the corresponding subsequences is used; For any two monitoring data sequences, the maximum value of the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value between the corresponding subsequences in each causal skeleton diagram with causal relationship direction indicators is (the lag time here only represents the numerical value. If the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value of the three causal skeleton diagrams is 6, -13, and 2, then the maximum value is 13 and the minimum value is 2). For any two monitoring data sequences, the minimum lag time corresponding to the standardized cross-correlation coefficient between the corresponding subsequences in each causal skeleton diagram with causal relationship directional indicators is the maximum absolute value between them. It is an exponential function with the natural constant as the base.

[0072] It should be noted that if the number of causal relationship direction indicators between any two monitoring data sequences is greater than or equal to 2, it indicates that any two monitoring data sequences have a causal relationship at different time scales. This indicates the robustness of statistical estimation at the i-th time scale. The larger the value, the greater the reliability of the causal relationship orientation index obtained in the i-th time scale; This represents the lag time difference between the lag time corresponding to the maximum absolute value standardized cross-correlation coefficient between the target subsequences at the i-th time scale and the baseline lag time (i.e., the median of the lag time corresponding to the maximum absolute value standardized cross-correlation coefficient between the corresponding subsequences of any two monitoring data sequences in each causal skeleton diagram). The smaller the value, the greater the consistency between the lag time corresponding to the maximum absolute value standardized cross-correlation coefficient among the target subsequences at the i-th time scale and the baseline lag time, and the greater the reliability of the causal relationship direction index obtained at the i-th time scale; a preset consistency sensitivity coefficient is used. The larger the value, the better. The greater the impact, the more likely it is to be affected. This embodiment sets... There are no restrictions here; settings can be made according to the specific implementation scenario.

[0073] Thus, the overall causal direction index between any two monitoring data sequences is obtained.

[0074] Step S104: Based on the data correlation between any two monitoring data sequences, obtain the causal relationship strength index between any two monitoring data sequences; based on the total causal relationship direction index and the causal relationship strength index, obtain the optimized causal relationship between any two monitoring data sequences.

[0075] The overall causal direction index comprehensively reflects the final causal direction between any two monitoring data sequences and the confidence level of the final causal direction, but it cannot represent the strength of the influence between any two monitoring data sequences, that is, the degree of change, i.e., how many units the result variable will change with the dependent variable when the causal variable changes by one unit.

[0076] Therefore, in this embodiment, based on the data correlation between any two monitoring data sequences, a causal relationship strength index between any two monitoring data sequences is obtained through curve fitting. This quantifies the influence strength of the causal relationship between any two monitoring data sequences, providing priority guidance for the subsequent construction of the Bayesian network structure model, and ensuring that agricultural management resources can focus on the key factors with the greatest impact.

[0077] The method for obtaining the causal relationship strength index between any two monitoring data sequences by curve fitting based on the data correlation between the two sequences is as follows:

[0078] The maximum value of the standardized cross-correlation coefficient between the maximum absolute value of any two monitoring data sequences in each causal skeleton diagram is obtained and denoted as the maximum standardized cross-correlation coefficient. The time scale corresponding to the maximum standardized cross-correlation coefficient is denoted as the target time scale, and the lag time corresponding to the maximum standardized cross-correlation coefficient is denoted as the target lag time.

[0079] The causal relationship direction index of any two monitoring data sequences in the causal skeleton diagram at the target time scale is a vector. Based on the causal relationship direction index of any two monitoring data sequences in the causal skeleton diagram at the target time scale, the target subsequences corresponding to the two monitoring data sequences in the target time scale are divided into causal subsequences and result subsequences. For example, if the target subsequence group corresponding to monitoring data sequence X and monitoring data sequence Y in the causal skeleton diagram at the target time scale is ( and The causal relationship is as follows: Then the target subsequence Let the cause subsequence be denoted as the target subsequence. Let this be the result subsequence;

[0080] Remove the last monitoring data point with the target lag time from the cause subsequence. For example, if the target lag time is... Then in the cause subsequence Remove the last one From the monitoring data, a causal variable sequence is obtained. Then, the monitoring data with the target lag time are removed from the resulting subsequence. For example, if the target lag time is... Then in the result subsequence Before removal From monitoring data, a sequence of result variables is obtained;

[0081] The least squares method is used to linearly fit the causal variable sequence and the result variable sequence to obtain a fitting curve. That is, a two-dimensional coordinate system is constructed with the causal variable sequence as the horizontal axis and the result variable sequence as the vertical axis. The least squares method is used to fit the coordinate points in the two-dimensional coordinate system to obtain the fitting curve. The slope of the fitting curve is obtained as an indicator of the strength of the causal relationship between any two monitoring data sequences. The method of linear fitting using the least squares method and the method of obtaining the slope of the fitting curve are existing technologies and will not be described in detail here.

[0082] The overall causal direction index comprehensively reflects the final causal direction and the directional confidence of the final causal direction between any two monitoring data sequences. The causal strength index quantifies the influence strength of the causal relationship between any two monitoring data sequences. Furthermore, by combining the overall causal direction index and the causal strength index, the optimized causal relationship between any two monitoring data sequences is obtained, which is used to construct a Bayesian network structure model and output a complete causal map with direction, confidence, and effect strength.

[0083] The method for obtaining the optimized causal relationship between any two monitoring data sequences based on the overall causal direction index and the causal strength index is as follows:

[0084] The direction in the overall causal direction index is recorded as the causal direction between any two monitoring data sequences. The value in the overall causal direction index is recorded as the causal confidence level between any two monitoring data sequences. The causal strength index is recorded as the effect strength between any two monitoring data sequences. The causal direction, causal confidence level, and effect strength between any two monitoring data sequences constitute the optimized causal relationship between them. For example: fertilization → growth, causal confidence level 70%, effect strength 3; watering → humidity, causal confidence level 85%, effect strength 2.

[0085] Thus, the optimized causal relationship between any two monitoring data sequences is obtained.

[0086] Step S105: Utilize the optimized causal relationship between every two monitoring data sequences to construct a Bayesian network structure model for any crop, which is used for intelligent monitoring of the future growth stages of any crop.

[0087] According to the method for obtaining the optimized causal relationship between any two monitoring data sequences, the optimized causal relationship between each pair of monitoring data sequences is obtained. Using the optimized causal relationship between each pair of monitoring data sequences, a Bayesian network structure model of the crop is constructed for intelligent monitoring of the future growth stage of the crop.

[0088] Using the optimized causal relationship between each pair of monitoring data sequences to construct a Bayesian network structure model is an existing technology, which is briefly described here: (1) Network structure determination: The optimized causal relationship between each pair of monitoring data sequences is used as the skeleton structure of the Bayesian network. The optimized causal relationship with high effect strength is used as the core connection of the network structure, and the optimized causal relationship with low effect strength is used as the optional connection of the network structure, so as to ensure that the network structure contains the main causal path and maintains computational efficiency; (2) Conditional probability table parameter learning: Using the multidimensional monitoring data sequence, based on the maximum likelihood estimation or Bayesian parameter learning method, the conditional probability distribution of each node is calculated. For the optimized causal relationship with high effect strength, higher weight is given in the parameter learning to ensure that the network structure accurately reflects the key causal mechanism and obtains a trained Bayesian network structure model.

[0089] After obtaining the Bayesian network structure model, the crop is monitored in real time during its future growth stage. The multidimensional monitoring data is input into the obtained Bayesian network structure model. When an abnormal state of the crop is detected, the posterior probability of each potential cause is calculated through network inference, thereby realizing intelligent supervision and source tracing diagnosis based on causal relationships and completing the intelligent supervision of the future growth stage of the crop.

[0090] It is worth noting that the main objective of this invention is to leverage the multi-scale periodicity characteristics of agricultural monitoring data, integrate and analyze the causal relationships of monitoring data sequences at different time scales, obtain optimized causal relationships, construct a Bayesian network structure model, and improve the reliability of causal relationships within the Bayesian network structure model. Utilizing a Bayesian network structure model to achieve intelligent monitoring and source tracing diagnosis based on causal relationships is existing technology and will not be elaborated upon here.

[0091] In addition, after each growth cycle of the crop, the multidimensional monitoring data monitored during that growth cycle can be added to the initial multidimensional monitoring data sequence to update and optimize the Bayesian network structure model, making the causal relationship in the Bayesian network structure model more reliable.

[0092] In summary, in this embodiment of the invention, a causal skeleton diagram of all subsequences at each time scale is constructed to reflect the causal relationship between any two monitoring data sequences at different time scales; a causal relationship direction index is obtained to preliminarily determine the causal direction and directional certainty of any two monitoring data sequences in each causal skeleton diagram, with a larger value indicating greater directional certainty; a total causal relationship direction index is obtained, which, combined with different time scales, comprehensively reflects the final causal direction and directional confidence of any two monitoring data sequences; a causal relationship strength index is obtained to quantify the actual impact of the causal relationship between any two monitoring data sequences, with a larger index indicating greater causal effect strength; and an optimized causal relationship is obtained to construct a Bayesian network structure model, outputting a complete causal map with direction, confidence, and effect strength, ultimately achieving intelligent monitoring and source tracing diagnosis of crop causal relationships.

[0093] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for remote monitoring of smart agriculture based on the Internet of Things, characterized in that, The IoT-based smart agriculture remote monitoring method includes: acquiring a multidimensional monitoring data sequence of any crop within a preset historical period; dividing the multidimensional monitoring data sequence into at least two monitoring data sequences according to dimensions; decomposing each monitoring data sequence into subsequences at at least two time scales; constructing a causal skeleton graph of all subsequences at each time scale using a PC algorithm, where nodes in the causal skeleton graph represent subsequences and edges represent causal relationships between nodes; for any two monitoring data sequences, obtaining a causal relationship direction index in each causal skeleton graph based on the data correlation between the corresponding subsequences of the two monitoring data sequences; obtaining a total causal relationship direction index between the two monitoring data sequences based on the causal relationship direction index in each causal skeleton graph; and further obtaining a total causal relationship direction index between the two monitoring data sequences based on the data correlation between the two monitoring data sequences. Based on the correlation, obtain the causal relationship strength index between any two monitoring data sequences; based on the total causal relationship direction index and the causal relationship strength index, obtain the optimized causal relationship between any two monitoring data sequences; using the optimized causal relationship between each pair of monitoring data sequences, construct a Bayesian network structure model for any crop, used for intelligent monitoring of the future growth stage of the crop; obtaining the total causal relationship direction index between any two monitoring data sequences based on the causal relationship direction index in each causal skeleton graph includes: performing Bootstrap resampling on the subsequences corresponding to any two monitoring data sequences at each time scale to obtain the Bootstrap confidence of any two monitoring data sequences at each time scale; the calculation formula for the total causal relationship direction index between any two monitoring data sequences is: ;in, This is the overall directional index of the causal relationship between any two monitoring data sequences; The j-th causal relationship direction indicator for any two monitoring data sequences; The Bootstrap confidence level at the time scale corresponding to the j-th causal relationship direction indicator between any two monitoring data sequences; The number of causal relationship direction indicators for any two monitoring data sequences; Set a preset consistency sensitivity coefficient; For the causal skeleton diagram corresponding to the j-th causal relationship direction index of any two monitoring data sequences, the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value between the subsequences corresponding to the two monitoring data sequences is defined. For any two monitoring data sequences, in each causal skeleton diagram with causal relationship directional indicators, the median of the lag time corresponding to the standardized cross-correlation coefficient of the maximum absolute value between the corresponding subsequences is used; For any two monitoring data sequences, in each causal skeleton diagram with causal relationship directional indicators, the maximum absolute value of the cross-correlation coefficient between the corresponding subsequences is the standardized maximum lag time corresponding to the maximum lag time. For any two monitoring data sequences, the minimum lag time corresponding to the standardized cross-correlation coefficient between the corresponding subsequences in each causal skeleton diagram with causal relationship directional indicators is the maximum absolute value between them. The function is an exponential function with the natural constant as its base. The step of obtaining the optimized causal relationship between any two monitoring data sequences based on the total causal direction index and the causal strength index includes: recording the direction in the total causal direction index as the causal direction between the two monitoring data sequences; recording the value in the total causal direction index as the causal confidence level between the two monitoring data sequences; recording the causal strength index as the effect strength between the two monitoring data sequences; and combining the causal direction, causal confidence level, and effect strength between the two monitoring data sequences to form the optimized causal relationship between the two monitoring data sequences.

2. The method for remote monitoring of smart agriculture based on the Internet of Things according to claim 1, characterized in that, The step of obtaining the causal relationship direction index of any two monitoring data sequences in each causal skeleton graph based on the data correlation between the corresponding subsequences of any two monitoring data sequences in each causal skeleton graph includes: for any causal skeleton graph, denoteing the corresponding subsequences of any two monitoring data sequences in the causal skeleton graph as a target subsequence group; obtaining the lag time range between the target subsequence groups; obtaining the standardized cross-correlation coefficient of the target subsequence group at each lag time within the lag time range based on the data correlation between the target subsequences in the target subsequence group; if the target subsequence group has a causal relationship in any causal skeleton graph, obtaining the causal relationship direction index of any two monitoring data sequences in any causal skeleton graph based on the difference between the standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range.

3. The method for remote monitoring of smart agriculture based on the Internet of Things according to claim 2, characterized in that, The step of obtaining the causal relationship direction index of any two monitoring data sequences in any causal skeleton diagram based on the difference between the standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range includes: for any lag time within the lag time range, if any lag time is greater than or equal to 0, the standardized cross-correlation coefficient of the target subsequence group at any lag time is recorded as a positive correlation coefficient; if any lag time is less than 0, the standardized cross-correlation coefficient of the target subsequence group at any lag time is recorded as a negative correlation coefficient; the ... the step of obtaining the causal relationship direction index of any two monitoring data sequences in any causal skeleton diagram based on the difference between the standardized cross-correlation coefficients of the target subsequence group at each lag time within the lag time range includes: for any lag time within the lag time range, the standardized cross-correlation coefficient of the target subsequence group at each The standardized cross-correlation coefficients at each lag time within the range are divided into a set of positive correlation coefficients and a set of negative correlation coefficients. The sum of the maximum absolute values ​​of the positive correlation coefficients in the positive correlation coefficient set and the negative correlation coefficients in the negative correlation coefficient set is obtained to obtain the total correlation coefficient. The difference between the maximum absolute values ​​of the positive correlation coefficients in the positive correlation coefficient set and the negative correlation coefficients in the negative correlation coefficient set is obtained to obtain the correlation coefficient difference. The ratio of the correlation coefficient difference to the total correlation coefficient is obtained to obtain the causal relationship direction index of any two monitoring data sequences in any causal skeleton diagram.

4. The method for remote monitoring of smart agriculture based on the Internet of Things according to claim 1, characterized in that, The step of obtaining the causal relationship strength index between any two monitoring data sequences based on the data correlation between the two sequences includes: obtaining the maximum value of the maximum absolute standardized cross-correlation coefficient among the subsequences corresponding to each causal skeleton diagram of the two monitoring data sequences, denoted as the maximum standardized cross-correlation coefficient; denoting the time scale corresponding to the maximum standardized cross-correlation coefficient as the target time scale; denoting the lag time corresponding to the maximum standardized cross-correlation coefficient as the target lag time; obtaining the causal variable sequence and the outcome variable sequence corresponding to the two monitoring data sequences based on the target lag time and the causal relationship direction index in the causal skeleton diagram of the two monitoring data sequences at the target time scale; performing linear fitting on the causal variable sequence and the outcome variable sequence to obtain a fitting curve; and obtaining the slope of the fitting curve as the causal relationship strength index between the two monitoring data sequences.

5. The method for remote monitoring of smart agriculture based on the Internet of Things according to claim 4, characterized in that, The step of obtaining the causal variable sequence and the result variable sequence corresponding to the arbitrary two monitoring data sequences based on the target lag time and the causal relationship direction index in the causal skeleton diagram of the arbitrary two monitoring data sequences at the target time scale includes: the causal relationship direction index in the causal skeleton diagram of the arbitrary two monitoring data sequences at the target time scale is a vector; based on the causal relationship direction index in the causal skeleton diagram of the arbitrary two monitoring data sequences at the target time scale, the target subsequence corresponding to the arbitrary two monitoring data sequences at the target time scale is divided into a causal subsequence and a result subsequence; the last target lag time monitoring data in the causal subsequence is removed to obtain the causal variable sequence; the first target lag time monitoring data in the result subsequence is removed to obtain the result variable sequence.

Images