Lake time series data prediction method based on unmanned aerial vehicle laser radar point cloud

Abstract

The invention discloses a lake time series data prediction method based on unmanned aerial vehicle laser radar point cloud, and is applied to the technical field of lake data prediction. Comprising the following steps: collecting point cloud data of a lake and surrounding terrains and water levels through a laser radar carried by an unmanned aerial vehicle; converting the point cloud data into a DEM elevation map; re-discretizing the DEM elevation map to obtain a new discrete point cloud picture; lake topographic feature points are screened; combining the elevation data of the lake topographic feature points with the elevation data of the water level to obtain a database and performing preprocessing; training a time sequence prediction model based on a database, and performing multivariable seasonal trend prediction; and visualizing the elevation data of the prediction result. According to the method, multivariable seasonal data can be predicted, and finally, lake geological structure change, water level fluctuation and water reserve can be predicted through visualization, so that development of multiple works such as lake region development and geological structure research recording is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of lake data prediction, and more specifically to a lake time series data prediction method based on unmanned aerial vehicle laser radar point cloud. Background Art

[0002] Lakes are an important component of surface water and play an important role in maintaining hydrological balance and maintaining a healthy environment. However, climate change and human interference in the lake environment have caused huge changes in the morphological characteristics of lakes, which have had adverse negative effects on the ecological environment, greatly affected the availability of freshwater resources, led to a series of regional ecological and environmental problems, and restricted the sustainable development of the region. Changes in lake parameters usually indicate environmental changes. Therefore, it is of great practical significance to extract lake water information, reflect changes in lake water volume, understand its impact, and propose solutions in a timely manner. The existing lake water storage estimation methods are mainly based on traditional water level station measurement and remote sensing technology. The overall engineering volume is large, the maintenance cost of fixed station instruments is high, and the accuracy of remote sensing data is low. In many application scenarios, many lakes do not have professional water level stations and supporting equipment, and the cost of personnel and materials is high. Therefore, how to provide a lake time series data prediction method based on drone lidar point cloud is a problem that technicians in this field urgently need to solve. Summary of the invention

[0003] In view of this, the present invention provides a lake time series data prediction method based on UAV lidar point cloud. The lake block point cloud data of the UAV lidar can be used to establish a digital elevation model of the lake block and measure the water level. The time series prediction model is used to predict the changes in the lake basin and surrounding geological structure, and to estimate and predict the lake water reserves and water level time series data.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A lake time series data prediction method based on UAV laser radar point cloud includes the following steps:

[0006] S1. Collect point cloud data of the lake, surrounding terrain and water level using a drone equipped with a lidar;

[0007] S2, converting the point cloud data of the lake and surrounding terrain and water level into a DEM elevation map;

[0008] S3, re-discretizing the DEM elevation map to obtain a new discrete point cloud map;

[0009] S4, screening lake terrain feature points in the new discrete point cloud map;

[0010] S5, combining the elevation data of the lake terrain feature points with the elevation data of the water level to obtain a database, and preprocessing the database;

[0011] S6. Train a time series prediction model based on the database, and use the trained time series prediction model to perform multivariate seasonal trend prediction;

[0012] S7. Visualize the elevation data of the prediction results.

[0013] Optionally, S1 specifically includes: using a drone equipped with a lidar combined with an inertial measurement unit and a GPS module to collect point cloud data of the lake and surrounding terrain and water level at certain time intervals, and performing multi-cycle measurements.

[0014] Optionally, S4 is specifically:

[0015] S41, performing sparse grid equal distance screening on the new discrete point cloud image;

[0016] S42, further screening by joint time domain filtering method;

[0017] S43, calculating the gradient value corresponding to each data point, and setting a dynamic threshold value of the gradient modulus for screening.

[0018] Optionally, S41 specifically includes: selecting one data point from every four data points as a feature point to reduce the point cloud density.

[0019] Optionally, S42 specifically includes: calculating the coefficient of variation CV of the Z coordinate of the data point:

[0020]

[0021]

[0022] In the formula, X i represents the elevation data Z of the ith period of the geographical location point X, N is the total number of measurement periods, and ε represents X i The mean of the samples, S is the sample standard deviation, CV < 10% indicates that the data are very concentrated, 10% ≤ CV < 30% indicates that the data are relatively concentrated, 30% ≤ CV < 50% indicates that the data are moderately dispersed, and CV ≥ 50% indicates that the data are highly dispersed. Data points with CV ≥ 10% are selected and enter S43.

[0023] Optionally, S43 specifically includes: calculating the gradient value corresponding to each data point, calculating the angle between the gradient vector and the unit vector perpendicular to the XY plane, setting a dynamic threshold of the gradient modulus, taking data points exceeding the dynamic threshold range of the gradient modulus as inflection points, and repeatedly setting the dynamic threshold for multiple rounds of screening.

[0024] Optionally, the preprocessing in S5 includes outlier processing and data set expansion. The outlier processing adopts one of the isolation forest algorithm, the local anomaly factor algorithm, the standard deviation-based detection algorithm, and the interquartile range method. The data set expansion method includes time series interpolation, synthetic data generation, and data enhancement.

[0025] It can be seen from the above technical solution that compared with the prior art, the present invention provides a lake time series data prediction method based on UAV lidar point cloud, which has the following beneficial effects: the present invention can predict multivariate seasonal data, and finally visualize it to predict changes in the lake's geological structure, water level fluctuations and water storage size, and help carry out lake area development, geological structure research and recording and other tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0027] Figure 1 It is a flow chart of the lake time series data prediction method of the present invention;

[0028] Figure 2 This is a schematic diagram of characteristic points before screening in the present invention;

[0029] Figure 3 This is a schematic diagram of feature points after screening in the present invention;

[0030] Figure 4 The DEM elevation map of the first period in the embodiment of the present invention;

[0031] Figure 5 is a point cloud map of the first cycle in an embodiment of the present invention;

[0032] Figure 6 A 3D geological structure map of the first period in an embodiment of the present invention;

[0033] Figure 7 It is a DEM elevation map of a new period in the future in an embodiment of the present invention;

[0034] Figure 8 A point cloud map of a new period in the future in an embodiment of the present invention;

[0035] Fig. 9 A 3D geological structure map of a new period in the future in the embodiment of the present invention;

[0036] Fig.10It is a schematic diagram of the water area in a new cycle in the future according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] The embodiment of the present invention discloses a lake time series data prediction method based on drone laser radar point cloud, such as Figure 1 As shown, the following steps are included:

[0039] S1. Collect point cloud data of the lake, surrounding terrain and water level using a drone equipped with a lidar;

[0040] S2, converting the point cloud data of the lake and surrounding terrain and water level into a DEM elevation map;

[0041] S3, re-discretize the DEM elevation map to obtain a new discrete point cloud map; discretization can maintain the single value of elevation while sparse point cloud for subsequent processing;

[0042] S4, screening lake terrain feature points in the new discrete point cloud map;

[0043] In order to predict terrain data and reduce the amount of calculation to improve the efficiency of the time series prediction model, it is necessary to perform multi-step lake terrain feature point screening. The feature points before and after screening are as follows: Figure 2 and Figure 3 As shown;

[0044] S5, combining the elevation data of the lake terrain feature points with the elevation data of the water level to obtain a database, and preprocessing the database;

[0045] S6. Train a time series prediction model based on the database, and use the trained time series prediction model to perform multivariate seasonal trend prediction;

[0046] S7. Visualize the elevation data of the prediction results.

[0047] Time series, also known as dynamic series, refers to a series of values ​​of the same statistical indicator arranged in the order of their occurrence. The main purpose of time series analysis is to predict the future based on existing historical data. Depending on the observation time, the time in the time series can be years, quarters, months or any other time form. In an embodiment of the present invention, the lake time series data is the time series data of the coordinate information of the lake and its vicinity, the lake water level, and the lake water storage.

[0048] Furthermore, S1 is specifically as follows: at certain time intervals, a drone equipped with a laser radar is used in conjunction with an inertial measurement unit and a GPS module to collect point cloud data of the lake and surrounding terrain and water level, and perform multi-cycle measurements. The laser radar point cloud has high accuracy, and the error range can still reach the centimeter level in deep water areas. At the same time, a long hovering drone equipped with a laser radar is used to collect point cloud data of lake area terrain, and the range radius can reach hundreds of meters.

[0049] In the embodiment of the present invention, when a drone equipped with a laser radar is used for multi-cycle measurement, geological structure and water level sampling at long time intervals is required, mainly considering the slowness of geological structure changes and the convenience of collection. Taking half a month as an example, the long period of time not only ensures the significance of geological structure changes, but also ignores temporary water level anomalies caused by short-term factors such as weather.

[0050] During the data collection process of multiple cycles, the terrain height at the same longitude and latitude or the same geographical location will undergo different displacements on the z-axis over time. The time-varying law of these displacements can be used to predict the geological structure of the lake. Here, the change in the z-axis coordinate at the same coordinate (X, Y) is considered to be the displacement of the (X, Y) point on the z-axis.

[0051] Digital Elevation Model (DEM) is a method of digitally expressing the earth's surface morphology. It uses a limited amount of terrain elevation data to construct a ground model in the form of an ordered numerical matrix. It is a type of Digital Terrain Model (DTM). DTM covers many aspects of terrain, including but not limited to the spatial distribution of attributes such as elevation, slope, aspect, and slope change rate. In contrast, DEM focuses on providing zero-order single-dimensional information of terrain height, while other terrain features can be obtained through further analysis of DEM data.

[0052] The data structure of digital elevation model mainly includes two forms: regular grids (Regular Grids) and irregular triangulated networks (Triangulated Irregular Networks, TINs):

[0053] 1. Regular grid: This format is based on the Gaussian projection system, which arranges terrain points at fixed intervals on the XY plane to form an array consisting of coordinate values ​​and their corresponding elevation values. The position of a given point P{i,j} can be determined by its row number i and column number j in the DEM and the basic information stored in the DEM file. The advantage of a regular grid is that it has lower data storage requirements and supports effective data compression technology, thereby simplifying data processing and maintenance;

[0054] 2. Irregular triangulated network: TIN is a method of describing terrain features using a non-uniformly distributed triangular network. In TIN, each node represents an original measurement data point, ensuring the accurate reproduction of terrain features, especially for modeling complex terrain. However, TIN has a high data density. In addition to saving three-dimensional coordinates, it also needs to record the connection relationship between nodes, which makes it more suitable for processing large-scale remote sensing image data.

[0055] In the embodiment of the present invention, in order to facilitate point cloud data processing, a regular grid is used for subsequent calculations.

[0056] The resolution of DEM is one of the key parameters for evaluating its terrain representation capabilities, and it also affects the choice of its applicable scenarios. DEM resolution specifically refers to the basic unit that constitutes the model - the grid size. As discretized geographic spatial information, each (X, Y) position actually corresponds to a small grid with a specific elevation value, and the size of these grids is the DEM resolution. Generally speaking, the smaller the resolution value, the higher the detail retention ability and the more detailed surface description, but it will also lead to a significant increase in the data size. Therefore, when creating or selecting a DEM, it is necessary to weigh the relationship between accuracy and data management according to actual needs in order to achieve the best application effect.

[0057] In an embodiment of the present invention, the discretization adopts the equal-width binning method, which is a simple and intuitive data discretization method, which divides the continuous data interval into several sub-intervals evenly, and the width of each sub-interval is equal, and the interval width is adjusted according to the accuracy requirement. Since most DEM images maintain the continuity of elevation data by interpolation, it is necessary to discretize them again to facilitate subsequent data processing. By selecting different interval widths, the DEM image is divided into several blocks, and the elevation value corresponding to the center point of the block is taken as the characteristic elevation value of the block. Through this discretization process, a preliminary uniformly distributed point cloud is obtained.

[0058] Furthermore, S4 is specifically:

[0059] S41, performing sparse grid equal distance screening on the new discrete point cloud image;

[0060] S42, further screening by joint time domain filtering method;

[0061] S43, calculating the gradient value corresponding to each data point, and setting a dynamic threshold value of the gradient modulus for screening.

[0062] In order to predict terrain data and reduce the amount of calculation to improve the efficiency of the time series prediction model, a multi-step screening of lake terrain feature points is required.

[0063] Furthermore, S41 is specifically as follows: one data point is selected from every four data points as a feature point to reduce the point cloud density.

[0064] Further, S42 is specifically: calculating the coefficient of variation CV of the Z coordinate of the data point:

[0065]

[0066] In the formula, X i represents the elevation data Z of the ith period of the geographical location point X, N is the total number of measurement periods, and ε represents X i The mean of the samples, S is the sample standard deviation, CV < 10% indicates that the data are very concentrated, 10% ≤ CV < 30% indicates that the data are relatively concentrated, 30% ≤ CV < 50% indicates that the data are moderately dispersed, and CV ≥ 50% indicates that the data are highly dispersed. Data points with CV ≥ 10% are selected and enter S43.

[0067] Furthermore, S43 is specifically as follows: calculating the gradient value corresponding to each data point, calculating the angle between the gradient vector and the unit vector perpendicular to the XY plane, setting a dynamic threshold of the gradient modulus, taking data points exceeding the dynamic threshold range of the gradient modulus as inflection points, and repeatedly setting the dynamic threshold for multiple rounds of screening.

[0068] In the embodiment of the present invention, the dynamic threshold of the gradient modulus is selected according to the terrain. For example, a relatively small angle such as 15° / 30° can be selected to reduce the number of inflection points, thereby reducing the time complexity of the model. Subsequently, following this screening logic, multiple rounds of data point screening are performed to reduce the amount of calculation while retaining the main geological structure of the lake.

[0069] The joint time domain filtering method adopted by S42 can screen out terrain feature points with obvious time-varying displacement, and the time-varying size of the coefficient of variation exceeds the threshold, while the inflection point gradient in S43 screens out its terrain feature points. Simply put, many points in some blocks simultaneously show the same trend of time-varying displacement, and the overall trend is smooth. At this time, you only need to know how the edge points of this part of the block change to describe the time-varying displacement of the entire plane.

[0070] In the embodiment of the present invention, since different inflection points may occur between multiple cycles, a weight factor is introduced here to determine the logical order of the prediction model to further optimize the time complexity of the model:

[0071]

[0072] Where N is the total amount of data and the total number of measurement cycles, k i is the total number of occurrences of the i-th inflection point in all cycles.

[0073] By arranging the weight factors in reverse order and giving priority to calculating the important inflection points with the largest number of inflection points, the main characteristic points of the predicted geological structure can be obtained as quickly as possible.

[0074] Save the filtered point cloud elevation values, and together with the water level elevation data, we get all the time-varying elevation data sets. Use the time series prediction model to make the next prediction. The time series prediction model is a statistical or machine learning method used to predict future values ​​based on past data points. These models are widely used in finance, economics, weather forecasting, sales forecasting and other fields. Various types of time series prediction models can be used according to different needs and data characteristics. Due to the different complexity of different lake terrains, the optimal model can be selected for various terrains through cross-validation. Taking LSTM as an example, an attention mechanism module can also be added to further enhance the accuracy of time series prediction.

[0075] Furthermore, the preprocessing in S5 includes outlier processing and data set expansion. The outlier processing adopts one of the isolation forest algorithm, local anomaly factor algorithm, standard deviation-based detection algorithm, and interquartile range method. The data set expansion method includes time series interpolation, synthetic data generation, and data enhancement.

[0076] Outliers refer to data points that deviate significantly from the rest of the observations. These data points may be caused by measurement errors, recording errors, or unconventional events. Their existence may have a negative impact on model training, leading to overfitting or decreased generalization ability of the model. Therefore, effectively identifying and processing outliers is an important part of time series analysis. Since lake waters may not always be closed, and there may be artificial changes or large-scale dam breaches, if abnormal point clouds appear in two adjacent detections, it is necessary to pre-process the point cloud blocks and not include them in the prediction model. If the original abnormal point cloud blocks stabilize again in the new detection and there is a new small-scale evolution process, they will be re-included in the time series prediction model.

[0077] Dataset augmentation refers to the technique of increasing the size of the training dataset by adding additional data points, which is particularly important for avoiding overfitting and improving the generalization ability of the model. Since sampling at long intervals will inevitably lead to relatively small amounts of data, certain methods are needed to increase the augmented dataset.

[0078] In an embodiment of the present invention, time series interpolation can fill missing values ​​or create new data points by using techniques such as linear interpolation and spline interpolation. This method can maintain the original trend and seasonal characteristics of the time series. Synthetic data generation uses models such as ARIMA and LSTM to generate new time series data based on existing data; this method can create new samples with similar statistical properties to the original data, which helps to enhance the learning ability of the model. Data enhancement is similar to data enhancement technology in the field of image processing. It can modify the original data by adding noise, changing the time scale, etc. to generate new training samples.

[0079] After preprocessing the data, the dataset and various time series models are used for training and prediction. Through visualization, the predicted DEM elevation map or point cloud map for multiple future periods can be obtained. The changes in water reserves and the predicted map of water area can also be further predicted through calculation. The data here are multivariate and have seasonal trends, so the characteristics of the dataset must be considered when selecting the time series prediction model to ensure the prediction effect.

[0080] After the visualization operation matches the LSTM predicted elevation value with the corresponding two-dimensional position coordinates, a three-dimensional spatial coordinate data is formed. These spatial coordinates are redrawn to obtain the predicted point cloud map and DEM map, which is convenient for intuitive analysis. In the embodiment of the present invention, the DEM elevation map, point cloud map and 3D geological structure map of the first cycle obtained by the visualization simulation are shown in Figure 1. Figure 4 , Figure 5 and Figure 6 As shown in the figure, the DEM elevation map, point cloud map and 3D geological structure map of the next cycle are as follows: Figure 7 , Figure 8 and Fig. 9 As shown in the figure, the water area of ​​the new cycle in the future will be as follows Fig.10 As shown, the coordinate zero point in the figure is a selected fixed observation point. Since the multi-period time series terrain data needs to coordinate the position information of each feature point to simplify the calculation, a point corresponding to a specific longitude and latitude is selected as a fixed zero point as a coordinate reference. At the same time, the coordinate axis needs to be fixed. In the embodiment of the present invention, the east direction is selected as the positive direction of the x-axis and the north direction is selected as the positive direction of the y-axis.

[0081] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0082] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A lake time series data prediction method based on UAV laser radar point cloud, characterized in that: The following steps are involved: S1. Collect point cloud data of the lake, surrounding terrain and water level using a drone equipped with a lidar; S2, converting the point cloud data of the lake and surrounding terrain and water level into a DEM elevation map; S3, re-discretizing the DEM elevation map to obtain a new discrete point cloud map; S4, screening lake terrain feature points in the new discrete point cloud map; S5, combining the elevation data of the lake terrain feature points with the elevation data of the water level to obtain a database, and preprocessing the database; S6. Train a time series prediction model based on the database, and use the trained time series prediction model to perform multivariate seasonal trend prediction; S7. Visualize the elevation data of the prediction results.

2. The lake time series data prediction method based on UAV laser radar point cloud according to claim 1 is characterized in that: S1 is as follows: At certain time intervals, a drone equipped with a lidar combined with an inertial measurement unit and a GPS module is used to collect point cloud data of the lake, surrounding terrain and water level, and perform multi-cycle measurements.

3. The lake time series data prediction method based on UAV laser radar point cloud according to claim 1 is characterized in that: S4 is specifically: S41, performing sparse grid equal distance screening on the new discrete point cloud image; S42, further screening by joint time domain filtering method; S43, calculating the gradient value corresponding to each data point, and setting a dynamic threshold value of the gradient modulus for screening.

4. The lake time series data prediction method based on UAV laser radar point cloud according to claim 3 is characterized in that: S41 is specifically: selecting one data point from every four data points as a feature point to reduce the point cloud density.

5. The lake time series data prediction method based on UAV laser radar point cloud according to claim 3 is characterized in that: S42 is specifically: calculating the coefficient of variation CV of the Z coordinate of the data point: Where, X i represents the elevation data Z of the ith period of the geographical location point X, N is the total number of measurement periods, and ε represents X i The mean of the samples, S is the sample standard deviation, CV < 10% indicates that the data are very concentrated, 10% ≤ CV < 30% indicates that the data are relatively concentrated, 30% ≤ CV < 50% indicates that the data are moderately dispersed, and CV ≥ 50% indicates that the data are highly dispersed. Data points with CV ≥ 10% are selected and enter S43.

6. The lake time series data prediction method based on UAV laser radar point cloud according to claim 3 is characterized in that: S43 is specifically as follows: calculating the gradient value corresponding to each data point, calculating the angle between the gradient vector and the unit vector perpendicular to the XY plane, setting a dynamic threshold of the gradient modulus, taking data points exceeding the dynamic threshold range of the gradient modulus as inflection points, and repeatedly setting the dynamic threshold for multiple rounds of screening.

7. The lake time series data prediction method based on UAV laser radar point cloud according to claim 1 is characterized in that: The preprocessing in S5 includes outlier processing and data set expansion. The outlier processing adopts one of the isolation forest algorithm, local anomaly factor algorithm, standard deviation-based detection algorithm, and interquartile range method. The data set expansion methods include time series interpolation, synthetic data generation, and data enhancement.

Images