Cluster control method based on group intelligent control

Abstract

The invention relates to the technical field of unmanned aerial vehicle control, and provides a cluster control method based on swarm intelligent control, which comprises the steps of obtaining a grade cluster according to soil dimension data, and further obtaining a deviation degree of the grade cluster according to a difference condition of data in the grade cluster; obtaining a grade area; according to the overlapping relation and the deviation degree between the grade areas, the influence degree of each kind of soil dimension data on the analysis point location is obtained; according to the influence degree of each kind of soil dimension data on the prediction point location and the sensor installation point location, the soil texture similarity is obtained; the theoretical crop height is obtained by combining the crop height measured by the unmanned aerial vehicle according to the soil similarity; according to the difference between the crop height measured by the unmanned aerial vehicle and the theoretical crop height, the height deviation degree and the suspected deviation point location are obtained, and then the unmanned aerial vehicle cluster is controlled. The unmanned aerial vehicle cluster is controlled through the suspected deviation point location and the height deviation degree, and the accuracy of monitoring the crop growth height by the unmanned aerial vehicle is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) control, and particularly to a cluster control method based on swarm intelligence control. Background Art

[0002] The continuous growth of the population and the guarantee of food safety have put forward strict requirements for both the supply scale and quality of agricultural products. The traditional agricultural model highly depends on natural conditions and manual experience, and has problems such as low resource utilization rate, unstable production efficiency, rising labor costs, etc. The excessive use of chemical fertilizers and pesticides has led to soil degradation and environmental pollution. In contrast, smart agriculture realizes the precise monitoring and intelligent decision-making of farmland environment and crop growth through Internet of Things sensors and UAV inspection technology. The data-driven precise irrigation, intelligent fertilization, and pest warning significantly improve the resource utilization efficiency, and the automation equipment reduces the reliance on manpower, forming a sustainable agricultural development model with resource conservation, environmental friendliness, and high output.

[0003] In the large-scale management of smart agriculture, if the UAV flight speed is too fast, the scanning accuracy of the lidar will be affected due to the reduction of point cloud density, the accumulation of motion errors, and the insufficient scanning overlap rate. When flying at high speed, the mismatch between the pulse emission frequency of the lidar and the flight speed will cause the distance between adjacent scanning points to increase, and some plant height details will be lost. At the same time, the attitude angle measurement error of the UAV inertial navigation system in high-dynamic flight will amplify the point cloud position deviation. Especially in undulating terrain or crop-dense areas, it may lead to height data distortion or local missed scans, resulting in inaccurate acquisition of the crop height at some farmland positions, and then leading to misjudgment of the crop growth status. Summary of the Invention

[0004] The present invention provides a cluster control method based on swarm intelligence control to solve the problem that the crop height at some farmland positions obtained by using UAV scanning is inaccurate due to the too fast UAV flight speed. The specific technical solutions adopted are as follows:

[0005] The present invention proposes a cluster control method based on swarm intelligence control, and the method includes the following steps:

[0006] Set a number of analysis points in the farmland. The analysis points include a number of sensor installation points and a number of prediction points, and obtain a number of soil dimension data for each analysis point; use the UAV to scan to obtain the UAV-measured crop height for each analysis point.

[0007] Based on the soil dimension data of the analysis points, several grade clusters of each soil dimension data are obtained; based on the grade clusters of the soil dimension data, the deviation degree of each grade cluster of each soil dimension data is obtained; according to the grade clusters, several grade regions of each soil dimension data are obtained; based on the overlap relationship between the grade regions of the soil dimension data and the deviation degree, the influence degree of each analysis point by each soil dimension data is obtained;

[0008] Based on the similarity relationship of the influence degree of each soil dimension data on the prediction points and the sensor installation points, the soil similarity between each prediction point and each sensor installation point is obtained; based on the soil similarity between the prediction points and the sensor installation points, combined with the measured crop height of the drones at each sensor installation point, the theoretical crop height of each prediction point is obtained; based on the difference relationship between the measured crop height of the drones and the theoretical crop height of each prediction point, the height deviation degree of each prediction point is obtained;

[0009] Based on the height deviation degree, the suspected deviation points are screened; the drone swarm is controlled according to the suspected deviation points and the height deviation degree.

[0010] Preferably, the method for obtaining several grade clusters of each soil dimension data based on the soil dimension data of the analysis points specifically includes:

[0011] For any kind of soil dimension data, the data set composed of the data values of this kind of soil dimension data in all soil comprehensive sensors and the predicted values of all prediction points is denoted as the single-dimension monitoring set of this kind of soil dimension data. The elbow method is used to obtain the optimal number of clusters of the single-dimension monitoring set, and the optimal number of clusters is used as the K value in the K-means clustering algorithm. The elements in the single-dimension monitoring set are clustered into K clusters, which are denoted as the grade clusters of this kind of soil dimension data.

[0012] Preferably, the method for obtaining the deviation degree of each grade cluster of each soil dimension data based on the grade clusters of the soil dimension data specifically includes:

[0013] The calculation method of the deviation degree of the d-th grade cluster of the q-th soil dimension data is:

[0014]

[0015] In the formula, C q,d is the deviation degree of the d-th grade cluster of the q-th soil dimension data; E q,d is the number of elements of the d-th grade cluster of the q-th soil dimension data; E' qis the number of elements in the rank cluster with the largest number of elements among all rank clusters of the q-th soil dimension data; D q,d is the mean of the elements in the d-th rank cluster of the q-th soil dimension data; D' q is the mean of the elements in the rank cluster with the largest number of elements among all rank clusters of the q-th soil dimension data; exp() is the exponential function with the natural constant as the base; norm() is the linear normalization function; || is the absolute value function.

[0016] Preferably, the method for obtaining several rank regions of each soil dimension data according to the rank clusters includes the following specific steps:

[0017] For any rank cluster of any soil dimension data, the area formed by the analysis points corresponding to the elements belonging to this rank cluster in the farmland is recorded as a rank region of this soil dimension data.

[0018] Preferably, the method for obtaining the influence degree of each analysis point by each soil dimension data according to the overlap relationship between the rank regions of the soil dimension data and the deviation degree includes the following specific steps:

[0019] In the f-th rank region of the e-th soil dimension data, the calculation method of the correlation between the e-th soil dimension data and the g-th soil dimension data is:

[0020]

[0021] In the formula, B e,f,g is the correlation between the e-th soil dimension data and the g-th soil dimension data in the f-th rank region of the e-th soil dimension data; F e,f is the number of analysis points in the f-th rank region of the e-th soil dimension data; G e,f,g,h is the total number of analysis points in the rank region where the h-th analysis point in the f-th rank region of the e-th soil dimension data is located in the g-th soil dimension data; G' e,f,g,h is the number of overlapping analysis points between the f-th rank region of the e-th soil dimension data and the rank region where the h-th analysis point in the f-th rank region of the e-th soil dimension data is located in the g-th soil dimension data; C' e,f,h is the deviation degree of the rank cluster corresponding to the data of the h-th analysis point in the f-th rank region of the e-th soil dimension data; C″ e,f,g,h is the deviation degree of the rank cluster where the h-th analysis point in the f-th rank region of the e-th soil dimension data is located in the g-th soil dimension data; || is the absolute value function; ε is a hyperparameter;

[0022] According to the relevance between the soil dimension data of each analysis point and the soil dimension data of all other species in the grade area where the soil dimension data is located, the influence degree of each analysis point by each soil dimension data is obtained.

[0023] Preferably, the method for obtaining the influence degree of each analysis point by each soil dimension data according to the relevance between the soil dimension data of each analysis point and the soil dimension data of all other species in the grade area where the soil dimension data is located includes the following specific methods:

[0024] For any analysis point and any soil dimension data, the mean value of the relevance between the soil dimension data and the soil dimension data of all other species in the grade area where the soil dimension data is located at the analysis point is denoted as the influence degree of the analysis point by the soil dimension data.

[0025] Preferably, the method for obtaining the soil texture similarity between each prediction point and each sensor installation point according to the similarity relationship between the influence degrees of the prediction point and the sensor installation point by each soil dimension data includes the following specific methods:

[0026] The calculation method of the soil texture similarity between the k-th prediction point and the l-th sensor installation point is:

[0027] H k,l =exp(-(MAX(|B′ k -B′ l |)))

[0028] In the formula, H k,l is the soil texture similarity between the k-th prediction point and the l-th sensor installation point; MAX(|B' k -B′ l | is the maximum value of the difference in the influence degrees of the k-th prediction point and the l-th sensor installation point by the same soil dimension data among all soil dimension data; || is the absolute value function; exp() is the exponential function with the natural constant as the base.

[0029] Preferably, the method for obtaining the theoretical crop height of each prediction point according to the soil texture similarity between the prediction point and the sensor installation point and combining the drone-measured crop height of each sensor installation point includes the following specific methods:

[0030] The calculation method of the theoretical crop height of the k-th prediction point is:

[0031]

[0032] In the formula, L k is the theoretical crop height of the k-th prediction point; N is the number of sensor installation points; Hk,l is the soil similarity between the k-th predicted point and the l-th sensor installation point, M l is the UAV-measured crop height at the l-th sensor installation point; softmax() is the weight normalization function.

[0033] Preferably, obtaining the height deviation degree of each predicted point according to the difference relationship between the UAV-measured crop height and the theoretical crop height of each predicted point includes the following specific method:

[0034] For any predicted point, the linearly normalized result of the absolute value of the difference between the theoretical crop height and the UAV-measured crop height of this predicted point is denoted as the height deviation degree of this predicted point.

[0035] Preferably, screening out the suspected deviation points according to the height deviation degree includes the following specific method:

[0036] The predicted points with a height deviation degree greater than the preset deviation threshold are recorded as suspected deviation points.

[0037] The beneficial effects of the present invention are as follows:

[0038] When using UAVs to monitor the height of farmland crops, the height of crop growth in some local areas of the farmland is affected by soil quality. Since the UAVs fly too fast over the farmland, they cannot accurately obtain the height of local areas of the farmland caused by soil quality. Therefore, it is necessary to analyze the soil quality at different positions. The present invention arranges soil comprehensive sensors, analyzes the regional differences in soil dimension data in the soil by using the changes in soil quality in local areas of the farmland, and judges the influence degree of each analysis point by each soil dimension data; for the problem of judging soil similarity, the present invention obtains the soil similarity between each predicted point and each sensor installation point according to the similarity relationship between the influence degrees of each predicted point and each sensor installation point by each soil dimension data; for the problem that the UAVs cannot accurately measure the crop growth height due to their too fast flight speed over the farmland, the present invention combines the soil similarity between the predicted points and the sensor installation points with the UAV-measured crop height of each sensor installation point to obtain the theoretical crop height of each predicted point, and then compares the UAV-measured crop height with the theoretical crop height to obtain the height deviation degree of each predicted point, and judges the possibility that the crop growth in each predicted point is affected by the local area of the farmland due to soil quality. Thus, the present invention screens out the suspected deviation points according to the height deviation degree, controls the UAV swarm according to the suspected deviation points and the height deviation degree, and improves the accuracy of the UAVs in monitoring the crop growth height for the crop growth height data in local farmland areas with local differences in soil quality. Description of the Drawings

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for describing the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0040] Figure 1 Schematic flow chart of a cluster control method based on swarm intelligence control provided by an embodiment of the present invention. Specific embodiments

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0042] Please refer to Figure 1 , which shows a flow chart of a cluster control method based on swarm intelligence control provided by an embodiment of the present invention. The method includes the following steps:

[0043] Step S001: Set a number of analysis points in the farmland, and obtain a number of soil dimension data for each analysis point; use a drone to scan and obtain the measured crop height of each analysis point.

[0044] It should be noted that in order to improve the scanning accuracy during drone scanning while minimizing the impact on the scanning rate, when controlling the drone cluster, it is necessary to control the drone to perform low-speed targeted scanning on the local area where the growth height of the crops deviates from other farmland positions. The direct factor affecting the growth height of the crops is the soil quality of the crops. The difference in soil quality is mainly reflected in the different contents of various soil properties in the farmland. Therefore, soil comprehensive sensors are arranged in the farmland to obtain various soil properties of the farmland.

[0045] Specifically, the farmland that needs to be intelligently managed is divided by a square grid of A×A, and the center point of each square grid is used as a sensor installation point. A soil comprehensive sensor is installed at each sensor installation point. The soil comprehensive sensor integrates a soil humidity sensor, a soil temperature sensor, a soil conductivity sensor, a pH value sensor, a soil nitrogen, phosphorus and potassium sensor, and a soil carbon dioxide sensor; where A is the preset grid width, and in this embodiment, A = 20 meters is used as an example for description;

[0046] After all soil comprehensive sensors are installed, for any one soil comprehensive sensor, collect a set of soil multi-dimensional monitoring data before the UAV takes off. The soil multi-dimensional monitoring data includes soil humidity value, soil temperature value, soil conductivity value, soil pH value, soil nitrogen, phosphorus and potassium content value, and soil carbon dioxide concentration value. Among them, each data value is used as a kind of soil dimension data;

[0047] Based on the A×A square grid, each square grid is further evenly divided into A 2 sub-grids, that is, each square grid is divided into A 2 sub-grids with a side length of 1 meter, and the center point of each sub-grid is used as a prediction point;

[0048] For any kind of soil dimension data, according to all soil comprehensive sensors in this kind of soil dimension data, use ordinary Kriging method to interpolate all prediction points to obtain the predicted value of each prediction point in this kind of soil dimension data; among them, the ordinary Kriging method is a well-known technology, and the specific method will not be introduced here;

[0049] Collectively refer to the sensor installation points and prediction points in the farmland as analysis points;

[0050] Arrange a lidar on the UAV, and use the UAV to scan at a constant speed of 12 m / s to obtain the three-dimensional point cloud data of the farmland, and obtain the measured crop height of the UAV at each analysis point.

[0051] Step S002: According to the soil dimension data of the analysis points, obtain several hierarchical cluster classes for each kind of soil dimension data; according to the hierarchical cluster classes of the soil dimension data, obtain the deviation degree of each hierarchical cluster class of each kind of soil dimension data; according to the hierarchical cluster classes, obtain several hierarchical regions for each kind of soil dimension data; according to the overlapping relationship between the hierarchical regions of the soil dimension data and the deviation degree, obtain the influence degree of each analysis point by each kind of soil dimension data.

[0052] It should be noted that the direct factor affecting the growth of crops is the soil quality. Due to reasons such as slope changes and human factors in some local areas, the soil quality changes. The change in soil quality is mainly manifested as different degrees of changes in each kind of soil dimension data, which in turn affects the growth height of crops. Therefore, it is necessary to judge the influence degree of the crops in each local area of the farmland by each kind of soil dimension data.

[0053] Specifically, for any kind of soil dimension data, the data set consisting of the data values of all soil comprehensive sensors and the predicted values of all predicted points of this kind of soil dimension data is denoted as the single-dimension monitoring set of this kind of soil dimension data. The elbow method is used to obtain the optimal number of cluster classes of this single-dimension monitoring set, and this optimal number of cluster classes is used as the K value in the K-means clustering algorithm. The elements in this single-dimension monitoring set are clustered into K cluster classes, which are denoted as the rank cluster classes of this kind of soil dimension data. Among them, the elbow method and K-means clustering are well-known technologies, and the specific methods will not be introduced here.

[0054] It should be noted that due to the different soil qualities in local areas of the farmland, there will be differences in some soil dimension data between local areas of the farmland with different soil qualities and other farmland areas.

[0055] Specifically, the calculation method of the deviation degree of the d-th rank cluster class of the q-th kind of soil dimension data is as follows:

[0056]

[0057] In the formula, C q,d is the deviation degree of the d-th rank cluster class of the q-th kind of soil dimension data; E q,d is the number of elements in the d-th rank cluster class of the q-th kind of soil dimension data; E' q is the number of elements in the rank cluster class with the largest number of elements among all rank cluster classes of the q-th kind of soil dimension data; D q,d is the mean value of the elements in the d-th rank cluster class of the q-th kind of soil dimension data; D' q is the mean value of the elements in the rank cluster class with the largest number of elements among all rank cluster classes of the q-th kind of soil dimension data; exp() is the exponential function with the natural constant as the base; norm() is the linear normalization function, and the normalization object is (|D q,d -D' q |) of all rank cluster classes of the q-th kind of soil dimension data; || is the absolute value function.

[0058] In the formula, The larger it is, the fewer the elements in the d-th rank cluster class of the q-th kind of soil dimension data are, indicating that the d-th rank cluster class of the q-th kind of soil dimension data is different from the general data of the same kind of soil dimension data; the larger norm(|D q,d -D' q |) is, the greater the difference between the d-th rank cluster class of the q-th kind of soil dimension data and the general data, and the greater the possibility that the d-th rank cluster class of the q-th kind of soil dimension data deviates from the normal value.

[0059] It should be noted that in farmland soil monitoring, for each type of soil dimension data, there are characteristics of local data similarity on farmland. For example, the gradual change of soil quality between farmlands leads to the similarity of soil quality within a local range, but there are differences in soil quality between regions. For example, when the slope of the farmland changes, water accumulates in the low-lying areas, while strong evaporation occurs at the top of the slope, resulting in salinization; the unevenness of human management measures, and farming operations such as fertilization, irrigation, and tillage often have differences in implementation in different areas. Therefore, for each type of soil dimension data, there are characteristics of local data similarity on farmland, that is, for each grade cluster of soil dimension data, there is an aggregated distribution characteristic in the two-dimensional plane space where the farmland is located.

[0060] Specifically, for any grade cluster of any type of soil dimension data, the area formed by the analysis points corresponding to the elements belonging to this grade cluster in the farmland is denoted as a grade area of this type of soil dimension data.

[0061] It should be noted that in farmland soil monitoring, each type of soil dimension data will have an impact on the growth of crops, but the impacts on the growth of crops are not in a linearly additive relationship among soil dimension data. There is a correlation among soil dimension data, and some soil dimension data will jointly affect the growth of crops. The correlation among soil dimension data is mainly reflected in that due to different soil qualities in some local areas, some soil dimension data in this local area are different from those in other positions, resulting in these soil dimension data jointly affecting crop growth. For example, in low-lying and waterlogged areas, high humidity and salt accumulation due to evaporation concentration effects often occur simultaneously, forming a salinization stress area; in plots where organic fertilizers have been applied for a long time, the increase in organic matter content and microbial activity CO2 concentration is often accompanied by an increase in pH buffering capacity, showing a patchy distribution of synergistic improvement. Therefore, according to the overlapping distribution of the grade areas of different soil dimension data, the composition of soil dimension data with different soil qualities is judged to have an impact on crops, and then the correlation among soil dimension data under different soil qualities is obtained.

[0062] Specifically, in the f-th grade area of the e-th type of soil dimension data, the calculation method for the correlation between the e-th type of soil dimension data and the g-th type of soil dimension data is as follows:

[0063]

[0064] In the formula, B e,f,g is the correlation between the e-th type of soil dimension data and the g-th type of soil dimension data in the f-th grade area of the e-th type of soil dimension data; F e,f is the number of analysis points in the f-th grade area of the e-th type of soil dimension data; G e,f,g,his the total number of analysis points in the grade area where the h-th analysis point in the f-th grade area of the e-th soil dimension data is located in the g-th soil dimension data; G' e,f,g,h is the number of analysis points where the f-th grade area of the e-th soil dimension data overlaps with the grade area where the h-th analysis point is located in the g-th soil dimension data; C' e,f,h is the degree of deviation of the grade cluster where the data corresponding to the h-th analysis point in the f-th grade area of the e-th soil dimension data is located; C″ e,f,g,h is the degree of deviation of the grade cluster where the h-th analysis point in the f-th grade area of the e-th soil dimension data is located in the g-th soil dimension data; || is the absolute value function; ε is a hyperparameter to prevent the denominator from being zero.

[0065] In the formula, The larger it is, the greater the degree of overlap between the f-th grade area of the e-th soil dimension data and the grade area where the h-th analysis point is located in the g-th soil dimension data; The larger it is, indicating C' e,f,h and C″ e,f,g,h are both relatively large and close, indicating that both the f-th grade area of the e-th soil dimension data and the grade area where the h-th analysis point is located in the g-th soil dimension data show deviations from the general soil quality; if while it is getting larger, is also getting larger, indicating that both the e-th soil dimension data and the g-th soil dimension data at this analysis point have data offsets due to soil quality reasons; the e-th soil dimension data and the g-th soil dimension data will jointly affect the growth of crops.

[0066] It should be noted that for a local farmland area with soil quality deviation compared to a general farmland, there will also be deviations in some soil dimension data of this local farmland area, that is, it shows a relatively high correlation between some soil dimension data at a single analysis point. These soil dimension data with relatively high correlations jointly affect the growth of crops. Therefore, it is necessary to determine the main soil dimension data that affect the growth of crops in this local farmland area.

[0067] It should be further noted that if a certain soil dimension data is one of the main factors affecting the growth of crops in a local farmland area, then there will be a relatively high correlation between this soil dimension data and the soil dimension data of other main factors. If a certain soil dimension data is not a main factor affecting the growth of crops in a local farmland area, then the correlation between this soil dimension data and other soil dimension data is relatively low.

[0068] Specifically, for any analysis point and any type of soil dimension data, the mean value of the correlation between this type of soil dimension data and all other types of soil dimension data in the grade area where this analysis point is located for this type of soil dimension data is denoted as the degree of influence of this analysis point by this type of soil dimension data.

[0069] Step S003: According to the similarity relationship of the influence degrees of each type of soil dimension data on the prediction point and the sensor installation point, obtain the soil quality similarity between each prediction point and each sensor installation point; according to the soil quality similarity between the prediction point and the sensor installation point, combined with the measured crop height by the unmanned aerial vehicle at each sensor installation point, obtain the theoretical crop height of each prediction point; according to the difference relationship between the measured crop height by the unmanned aerial vehicle and the theoretical crop height of each prediction point, obtain the height deviation degree of each prediction point.

[0070] It should be noted that in large-scale farmland, natural factors (such as microtopography, parent material differences, groundwater fluctuations) and human activities (such as zoned fertilization, uneven irrigation, mechanical compaction) will cause local soil quality variations, and the combinations of soil dimension data with high influence degrees in different soil types (such as saline soil, clay soil, sandy loam) are different, so the soil quality can be distinguished based on this.

[0071] It should be further noted that the height model obtained by using the unmanned aerial vehicle, the direct factor affecting the growth height of crops is the soil quality reason. And since all kinds of soil dimension data at each sensor installation point are actually measured by the comprehensive sensor, it can represent the overall situation of each type of soil dimension data in the farmland within the local range. Therefore, the degree of influence of each comprehensive sensor installation point by each type of soil dimension data can be combined with the measured crop height by the unmanned aerial vehicle, and by comparing the prediction points with the same high influence degree of soil dimension data, the theoretical crop height of the prediction points can be inferred.

[0072] Specifically, the calculation method of the soil quality similarity between the k-th prediction point and the l-th sensor installation point is:

[0073] H k,l =exp(-(MAX(|B′ k -B′ l |)))

[0074] In the formula, H k,l is the soil quality similarity between the k-th prediction point and the l-th sensor installation point; MAX(|B' k -B' lwhere MAX(|B'

[0075] - B' k,l |) is the maximum value of the difference in the degree of influence of the k-th predicted point and the l-th sensor installation point by the same soil dimension data among all soil dimension data; || is the absolute value function; exp( ) is the exponential function with the natural constant as the base. k - B' l The larger H is, the smaller MAX(|B'

[0076] - B'|) is, that is, the difference in the degree of influence between the k-th predicted point and the l-th sensor installation point on the soil dimension data with the largest difference in the degree of influence is still very small. Since the combinations of soil dimension data with high degrees of influence are different in different soil types, the possibility that the soil quality at the k-th predicted point and the l-th sensor is similar is relatively high.

[0077] It should be noted that similar soil quality can often grow crops of similar heights. Since all soil dimension data at each sensor installation point are actually measured by a comprehensive sensor, the accuracy of various soil dimension data at each sensor installation point is relatively high. Therefore, the predicted height of the crops at the predicted point is inferred based on the similarity of the soil quality between the predicted point and the sensor installation point.

[0078]

[0079] Specifically, the calculation method of the theoretical crop height at the k-th predicted point is as follows: k In the formula, L k,l is the theoretical crop height at the k-th predicted point; N is the number of sensor installation points; H l is the similarity of the soil quality between the k-th predicted point and the l-th sensor installation point, and M

[0080] is the crop height measured by the drone at the l-th sensor installation point; softmax() is the weight normalization function, and the normalization object is the similarity of the soil quality between the k-th predicted point and all sensor installation points.

[0081] It should be noted that if there is a deviation between the theoretical crop height at the predicted point and the crop height measured by the drone at the predicted point, it indicates that the soil quality at this predicted point may have undergone local changes.

[0082] Step S004: Screen out suspected deviation points according to the degree of height deviation; control the UAV swarm according to the suspected deviation points and the degree of height deviation.

[0083] It should be noted that the greater the degree of height deviation of the predicted point, the more likely it is that the soil quality at the predicted point has undergone local changes, and the soil quality change at the predicted point affects the growth of crops. It is necessary to improve the monitoring accuracy of the growth status of crops by the UAVs at these predicted points.

[0084] Specifically, the predicted points with a height deviation degree greater than the preset deviation threshold are recorded as suspected deviation points; among them, the preset deviation threshold is 0.7, and this embodiment is described by taking this as an example.

[0085] In the next scan of the farmland by the UAVs, use the particle swarm optimization algorithm to control the flight of the UAV swarm, and adaptively adjust the learning factor c2 and the inertia weight w in the particle swarm optimization algorithm. The specific adjustment process is as follows:

[0086] For any UAV, set the coordinates of the suspected deviation point with the closest Euclidean distance to the UAV as the global optimal position of the particle swarm optimization algorithm when the UAV is flying.

[0087] The calculation method of the dynamic learning factor is:

[0088] c″2 = c2 + α×P

[0089] In the formula, c2 is the preset learning factor, and this embodiment is described by taking c2 = 2 as an example; α is the preset gain coefficient, and this embodiment is described by taking α = 0.1 as an example; P is the degree of height deviation of the suspected deviation point with the closest Euclidean distance to the UAV.

[0090] The calculation method of the dynamic inertia weight is:

[0091] w' = w×e -P

[0092] In the formula, w' is the dynamic inertia weight; w is the preset inertia weight, and this embodiment is described by taking w = 0.9 as an example; e is the natural constant; P is the degree of height deviation of the suspected deviation point with the closest Euclidean distance to the UAV.

[0093] It should be noted that by increasing the learning factor c2, the attraction of the high-deviation area to the UAVs is enhanced, and by reducing the inertia weight w, the inertia of the UAVs is reduced, so that they decelerate near the suspected deviation points and scan finely.

[0094] Furthermore, the adjusted dynamic learning factor c'2 and the dynamic inertia weight w' are used to control using the particle swarm optimization algorithm, and the UWB local communication is used to maintain the formation spacing to avoid collisions. Among them, the particle swarm optimization algorithm and the UWB local communication algorithm are well-known technologies, and the specific methods are not introduced here.

[0095] After the farmland is scanned by the unmanned aerial vehicle, the three-dimensional point cloud data of the crops in the local farmland area with local soil differences is obtained. Furthermore, the accurate crop height at each analysis point is obtained, and corresponding irrigation and fertilization management measures are implemented for the crops at different heights.

[0096] In this embodiment, the exp(-MX) model is used to present the inverse proportional relationship and normalization processing. MX is the input of the model, and the implementer can set the inverse proportional function and the normalization function according to the actual situation.

[0097] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A swarm control method based on swarm intelligence control, characterized in that, The method includes the following steps: Set a number of analysis points in the farmland. The analysis points include a number of sensor installation points and a number of prediction points, and obtain a number of soil dimension data for each analysis point; Use a drone to scan to obtain the drone-measured crop height of each analysis point; According to the soil dimension data of the analysis points, obtain a number of grade cluster classes for each type of soil dimension data; According to the grade cluster classes of the soil dimension data, obtain the deviation degree of each grade cluster class of each type of soil dimension data; According to the grade cluster classes, obtain a number of grade regions for each type of soil dimension data; According to the overlapping relationship between the grade regions of the soil dimension data and the deviation degree, obtain the influence degree of each analysis point by each type of soil dimension data; According to the similarity relationship of the influence degree of each prediction point and each sensor installation point by each type of soil dimension data, obtain the soil texture similarity between each prediction point and each sensor installation point; According to the soil texture similarity between the prediction point and the sensor installation point, combined with the drone-measured crop height of each sensor installation point, obtain the theoretical crop height of each prediction point; According to the difference relationship between the drone-measured crop height and the theoretical crop height of each prediction point, obtain the height deviation degree of each prediction point; Screen out suspected deviation points according to the height deviation degree; Control the drone swarm according to the suspected deviation points and the height deviation degree.

2. The swarm control method based on swarm intelligence control according to claim 1, wherein, The specific method included in obtaining a number of grade cluster classes for each type of soil dimension data according to the soil dimension data of the analysis points is as follows: For any type of soil dimension data, denote the data set composed of the data values of this type of soil dimension data in all soil comprehensive sensors and the predicted values of all prediction points as the single-dimension monitoring set of this type of soil dimension data. Use the elbow method to obtain the optimal number of cluster classes of this single-dimension monitoring set, and use the optimal number of cluster classes as the K value in the K-means clustering algorithm. Cluster the elements in this single-dimension monitoring set into K cluster classes, which are denoted as the grade cluster classes of this type of soil dimension data.

3. The cluster control method based on swarm intelligence control according to claim 1, wherein, The specific method included in obtaining the deviation degree of each grade cluster class of each type of soil dimension data according to the grade cluster classes of the soil dimension data is as follows: The calculation method of the deviation degree of the d-th grade cluster class of the q-th type of soil dimension data is: where C q,d is the deviation degree of the d-th rank cluster of the q-th soil dimension data; E q,d is the number of elements of the d-th rank cluster of the q-th soil dimension data; E' q is the number of elements of the rank cluster with the largest number of elements among all rank clusters of the q-th soil dimension data; D q,d is the mean value of the elements of the d-th rank cluster of the q-th soil dimension data; D' q is the mean value of the elements of the rank cluster with the largest number of elements among all rank clusters of the q-th soil dimension data; exp() is the exponential function with the natural constant as the base; norm() is the linear normalization function; || is the absolute value function.

4. The swarm control method based on swarm intelligence control according to claim 1, characterized in that The specific method included in obtaining a number of grade regions for each type of soil dimension data according to the grade cluster classes is as follows: For any grade cluster class of any type of soil dimension data, denote the area formed by the analysis points corresponding to the elements belonging to this grade cluster class in the farmland as a grade region of this type of soil dimension data.

5. The swarm control method based on swarm intelligence control according to claim 1, characterized in that The specific method included in obtaining the influence degree of each analysis point by each type of soil dimension data according to the overlapping relationship between the grade regions of the soil dimension data and the deviation degree is as follows: In the f-th grade region of the e-th type of soil dimension data, the calculation method of the correlation between the e-th type of soil dimension data and the g-th type of soil dimension data is: In the formula, B e,f,g is the correlation between the e-th soil dimension data and the g-th soil dimension data in the f-th level area of the e-th soil dimension data; F e,f is the number of analysis points in the f-th level area of the e-th soil dimension data; G e,f,g,h is the total number of analysis points in the level area where the h-th analysis point in the f-th level area of the e-th soil dimension data is located in the g-th soil dimension data; G' e,f,g,h is the number of analysis points where the f-th level area of the e-th soil dimension data overlaps with the level area where the h-th analysis point is located in the g-th soil dimension data; C' e,f,h is the degree of deviation of the data corresponding to the h-th analysis point in the f-th level area of the e-th soil dimension data from the level cluster; C″ e,f,g,h is the degree of deviation of the h-th analysis point in the f-th level area of the e-th soil dimension data from the level cluster in the g-th soil dimension data; || is the absolute value function; ε is a hyperparameter; According to the relevance between the soil dimension data of each analysis point in the hierarchical area where the soil dimension data is located and the soil dimension data of all other species, the influence degree of each analysis point by the soil dimension data of each species is obtained.

6. The swarm control method based on swarm intelligence control according to claim 5, wherein The specific method for obtaining the influence degree of each analysis point by the soil dimension data of each species according to the relevance between the soil dimension data of each analysis point in the hierarchical area where the soil dimension data is located and the soil dimension data of all other species is as follows: For any analysis point and any soil dimension data, the mean value of the relevance between the soil dimension data of this species and the soil dimension data of all other species in the hierarchical area where the soil dimension data of this species is located for this analysis point is denoted as the influence degree of this analysis point by the soil dimension data of this species.

7. The swarm control method based on swarm intelligence control according to claim 1, characterized in that The specific method for obtaining the soil texture similarity between each prediction point and each sensor installation point according to the similarity relationship between the influence degrees of each prediction point and each sensor installation point by the soil dimension data of each species is as follows: The calculation method of the soil texture similarity between the k-th prediction point and the l-th sensor installation point is: H k,l = exp(-(MAX(|B′ k - B′ l |))) Where H k,l is the soil similarity between the k-th predicted point and the l-th sensor installation point; MAX(|B' k -B' l |) is the maximum value of the difference in the degree of influence of the k-th predicted point and the l-th sensor installation point by the same soil dimension data among all soil dimension data; || is the absolute value function; exp() is the exponential function with the natural constant as the base.

8. The swarm control method based on swarm intelligence control according to claim 1, characterized in that The specific method for obtaining the theoretical crop height of each prediction point according to the soil texture similarity between the prediction point and the sensor installation point and combining the drone-measured crop height of each sensor installation point is as follows: The calculation method of the theoretical crop height of the k-th prediction point is: where, L k is the theoretical crop height at the k-th prediction point; N is the number of sensor installation points; H k,l is the soil similarity between the k-th prediction point and the l-th sensor installation point, M l is the crop height measured by the UAV at the l-th sensor installation point; softmax() is a weight normalization function.

9. The swarm control method based on swarm intelligence control according to claim 1, characterized in that The specific method for obtaining the height deviation degree of each prediction point according to the difference relationship between the drone-measured crop height and the theoretical crop height of each prediction point is as follows: For any prediction point, the linearly normalized result of the absolute value of the difference between the theoretical crop height and the drone-measured crop height of this prediction point is denoted as the height deviation degree of this prediction point.

10. The swarm control method based on swarm intelligence control according to claim 1, wherein The specific method for screening out suspected deviation points according to the height deviation degree is as follows: The prediction points with a height deviation degree greater than the preset deviation threshold are recorded as suspected deviation points.

Images