A method, system, device and storage medium for early warning of umbrella opening risk

Abstract

The application discloses a parachute opening risk early warning method, system, device and storage medium, the method comprises the following steps: in the parachute descent process, the air pressure, three-axis acceleration, three-axis angular velocity and descent speed are pretreated; the pretreated descent speed is subjected to three-layer discrete wavelet multi-scale decomposition, LSTM prediction and speed fusion; according to the enhanced speed sequence, the pretreated air pressure, three-axis acceleration and three-axis angular velocity, a parachute descent state sequence is generated; the parachute descent state sequence is input into a preset parachute descent state prediction model, and a parachute bag pressure difference prediction value is output, and the preset parachute descent state prediction model is constructed by combining an EVO algorithm, a CNN and a BiLSTM; based on the parachute bag pressure difference prediction value, the parachute descent risk state is estimated, and the parachute opening risk early warning is carried out. The application realizes real-time acquisition of multi-dimensional flight state data of parachute descent, fusion of spatial features and time sequence information, high-precision parachute bag pressure difference prediction and parachute opening safety early warning, and improvement of the safety and reliability of the parachute descent task.

Description

Technical Field

[0001] This invention relates to the field of parachute safety technology, and in particular to a method, system, device and storage medium for early warning of parachute deployment risks. Background Technology

[0002] Parachuting personnel or supplies are subject to complex aerodynamic disturbances during their descent, including changes in lateral wind speed, fluctuations in descent velocity, increased turbulence, and wake vortices. The formation of aerodynamic vortices can lead to risks such as changes in pressure gradients, asymmetrical inflation of the canopy, entanglement of parachute lines, or delayed deployment of the main parachute, which can severely impact parachute stability and deployment reliability.

[0003] Existing methods for eddy current prediction and parachute deployment safety assurance mainly rely on computational fluid dynamics (CFD) numerical simulations, wind tunnel tests, or empirical models. However, CFD computation is costly and struggles to meet real-time requirements; wind tunnel tests cannot cover real parachute drop conditions; and traditional monitoring methods based on single sensors are significantly affected by attitude changes and vibration interference, making it difficult to accurately characterize the turbulent evolution process. Furthermore, existing machine learning models are mostly static, unable to effectively extract the temporal dynamic characteristics of the parachute drop flow field, thus hindering the early identification of hazardous aerodynamic environments.

[0004] Therefore, how to achieve high-precision prediction of parachute pressure difference and thus identify and warn of parachute opening risks in advance has become an urgent problem to be solved.

[0005] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention The main objective of this invention is to provide a method, system, device, and storage medium for early warning of parachute opening risks, aiming to address the technical problem of how to achieve high-precision prediction of parachute pressure differences and thus identify and warn of parachute opening risks in advance.

[0006] To achieve the above objectives, the present invention provides a method for early warning of parachute deployment risk, the method comprising: During the parachute descent, air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals are collected, and the air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals are preprocessed. The preprocessed descent velocity sequence is subjected to three-level discrete wavelet multi-scale decomposition, LSTM prediction and velocity fusion to obtain an enhanced velocity sequence. A parachute state sequence is generated based on the enhanced velocity sequence, the pre-processed air pressure sequence, the triaxial acceleration sequence, and the triaxial angular velocity sequence. The parachute state sequence is input into a preset parachute state prediction model, and the predicted value of the parachute pressure difference is output. The preset parachute state prediction model is constructed by a combination of EVO algorithm, CNN and BiLSTM. Based on the predicted parachute pressure difference, the parachute drop risk status is estimated, and an early warning of parachute deployment risk is issued based on the parachute drop risk status.

[0007] Optionally, the preprocessing of the air pressure signal, the triaxial acceleration signal, the triaxial angular velocity signal, and the descent velocity signal includes: The air pressure signal is smoothed by a moving average. The triaxial acceleration signal and the triaxial angular velocity signal are filtered by a second-order Butterworth low-pass filter, respectively. The descent speed signal is subjected to median filtering. The smoothed air pressure signal, the filtered triaxial acceleration signal, the triaxial angular velocity signal, and the descent velocity signal were standardized respectively.

[0008] Optionally, the step of performing three-level discrete wavelet multi-scale decomposition, LSTM prediction, and velocity fusion on the preprocessed descent velocity sequence to obtain an enhanced velocity sequence includes: The preprocessed descent velocity sequence is subjected to three-level discrete wavelet multi-scale decomposition and denoising. The denoised multi-scale decomposition components are input into the corresponding LSTM branches to output trend prediction values, mesoscale perturbation prediction values, and fast perturbation prediction values. The trend prediction, mesoscale disturbance prediction, and fast disturbance prediction are weighted and fused to obtain an enhanced velocity sequence.

[0009] Optionally, the step of performing three-level discrete wavelet multi-scale decomposition and denoising on the preprocessed descent velocity sequence includes: The preprocessed descent velocity sequence is subjected to the first layer of discrete wavelet decomposition to obtain the first layer of low-frequency approximate components and the first layer of high-frequency detail components. The first-level low-frequency approximation component is subjected to the second-level discrete wavelet decomposition to obtain the second-level low-frequency approximation component and the second-level high-frequency detail component. The second-level low-frequency approximation component is subjected to a third-level discrete wavelet decomposition to obtain the third-level low-frequency approximation component and the third-level high-frequency detail component. The first layer low-frequency approximation component, the first layer high-frequency detail component, the second layer low-frequency approximation component, the second layer high-frequency detail component, the third layer low-frequency approximation component, and the third layer high-frequency detail component are denoised by using the soft thresholds corresponding to each component.

[0010] Optionally, the weighted fusion of the trend prediction value, the mesoscale disturbance prediction value, and the fast disturbance prediction value includes: The trend prediction value, the mesoscale disturbance prediction value, and the fast disturbance prediction value are weighted and fused using a weighted fusion formula. The weighted fusion formula is as follows:

[0011]

[0012] In the formula, To enhance the velocity sequence, , and For parameters, This is a trend forecast value. These are predicted values ​​for mesoscale disturbances. These are the predicted values ​​for rapid disturbances.

[0013] Optionally, the step of inputting the parachute state sequence into a preset parachute state prediction model and outputting the predicted parachute pressure difference includes: The parachute state sequence is input into a preset parachute state prediction model; The EVO algorithm is used to automatically search for CNN structures and then combine the searched CNN structures into the current optimal CNN. The parachute state sequence is subjected to sliding convolution operation by the current optimal CNN to obtain the local spatial feature sequence of air disturbance and pressure change; The overall temporal features are obtained by performing feature analysis on the local spatial feature sequence using BiLSTM. Based on the overall temporal features, a global temporal feature vector is obtained through an attention layer, and the predicted value of the parachute pressure difference is obtained through a two-layer fully connected network based on the global temporal feature vector.

[0014] Optionally, obtaining the global temporal feature vector through an attention layer based on the overall temporal features includes: Based on the overall temporal characteristics, the attention weights for each time step are calculated through an attention layer; The attention weights at each time step and the features at each time step within the overall temporal feature matrix are weighted and summed to obtain the global temporal feature vector.

[0015] Furthermore, to achieve the above objectives, the present invention also proposes an umbrella opening risk warning system, the umbrella opening risk warning system comprising: The data processing module is used to collect air pressure signals, triaxial acceleration signals, triaxial angular velocity signals and descent velocity signals during the parachute descent, and to preprocess the air pressure signals, triaxial acceleration signals, triaxial angular velocity signals and descent velocity signals. The velocity enhancement module is used to perform three-level discrete wavelet multi-scale decomposition, LSTM prediction and velocity fusion on the preprocessed descent velocity sequence to obtain the enhanced velocity sequence. The data integration module is used to generate a parachute state sequence based on the enhanced velocity sequence, the preprocessed air pressure sequence, the triaxial acceleration sequence, and the triaxial angular velocity sequence. The differential pressure prediction module is used to input the parachute state sequence into a preset parachute state prediction model and output the predicted value of the parachute pressure difference. The preset parachute state prediction model is constructed by a combination of EVO algorithm, CNN and BiLSTM. The risk warning module is used to estimate the parachute drop risk status based on the predicted parachute pressure difference value, and to issue a parachute deployment risk warning based on the parachute drop risk status.

[0016] Furthermore, to achieve the above objectives, the present invention also proposes an umbrella opening risk warning device, the device comprising: a memory, a processor, and an umbrella opening risk warning program stored in the memory and executable on the processor, the umbrella opening risk warning program being configured to implement the steps of the umbrella opening risk warning method described above.

[0017] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing a parachute opening risk warning program, wherein when the parachute opening risk warning program is executed by a processor, it implements the steps of the parachute opening risk warning method described above.

[0018] During the parachute descent, this invention first acquires air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals. These signals are then preprocessed. Next, the preprocessed descent velocity sequence undergoes three-level discrete wavelet multi-scale decomposition, LSTM prediction, and velocity fusion to obtain an enhanced velocity sequence. Based on this enhanced velocity sequence, the preprocessed air pressure sequence, triaxial acceleration sequence, and triaxial angular velocity sequence, a parachute descent state sequence is generated. This sequence is then input into a preset parachute descent state prediction model, which outputs a parachute pack pressure difference prediction value. This model is constructed using a combination of the EVO algorithm, CNN, and BiLSTM. Finally, based on the predicted parachute pack pressure difference value, the parachute descent risk status is estimated, and a parachute deployment risk warning is issued according to the predicted parachute descent risk status. This invention constructs a nine-dimensional time-series input using multi-source data such as air pressure, descent velocity, triaxial acceleration, and triaxial angular velocity. After preprocessing, a unified feature sequence is formed. An Energy Valley Optimizer (EVO) algorithm is used to adaptively optimize the convolutional neural network (CNN) structure to extract local spatial features of the parachute descent flow field. Then, a Bidirectional Long Short-Term Memory (BiLSTM) network is combined to capture the temporal dependencies in the eddy evolution process. A key turbulent segment is highlighted through an attention mechanism to achieve high-precision prediction of eddy intensity and parachute pressure difference, thereby identifying risk states in real time and providing early warning. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of the umbrella opening risk warning device in the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the first embodiment of the umbrella opening risk warning method of the present invention; Figure 3 This is a structural block diagram of the first embodiment of the umbrella opening risk warning system of the present invention.

[0020] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0022] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of the umbrella opening risk warning device in the hardware operating environment involved in the embodiment of the present invention.

[0023] like Figure 1As shown, the umbrella opening risk warning device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. Optionally, the memory 1005 may also be a storage system independent of the aforementioned processor 1001.

[0024] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the umbrella opening risk warning device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0025] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and an umbrella opening risk warning program.

[0026] exist Figure 1 In the parachute opening risk warning device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the parachute opening risk warning device of the present invention can be set in the parachute opening risk warning device, and the parachute opening risk warning device calls the parachute opening risk warning program stored in the memory 1005 through the processor 1001 and executes the parachute opening risk warning method provided in the embodiment of the present invention.

[0027] This invention provides a method for early warning of umbrella opening risks, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the umbrella opening risk warning method of the present invention.

[0028] In this embodiment, the umbrella opening risk warning method includes the following steps: S1. During the parachute descent, the system collects air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals, and preprocesses the air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals.

[0029] It is easy to understand that the executing entity of this embodiment can be a system for early warning of umbrella opening risks with functions such as data processing, network communication and program operation, or other computer devices with similar functions. This embodiment does not limit it.

[0030] It should be understood that if supplies are being parachuted, barometers, three-axis accelerometers, three-axis gyroscopes, and descent velocity meters need to be fixedly installed on the parachute and supplies; if personnel are being parachuted, barometers, three-axis accelerometers, three-axis gyroscopes, and descent velocity meters need to be fixedly installed on the parachute and the paratrooper's chest harness.

[0031] It should also be noted that the embedded data acquisition module synchronously samples all sensors at a fixed frequency of 200Hz. The system uses a pulse per second (PPS) timing signal to uniformly synchronize the internal clocks of each sensor for unified calibration, ensuring that the maximum deviation between the sampling timestamps of each sensor is less than 1ms, thereby ensuring the timing consistency of multi-channel data.

[0032] Furthermore, the preprocessing method for the air pressure signal, triaxial acceleration signal, triaxial angular velocity signal, and descent velocity signal is as follows: the air pressure signal is smoothed by moving average; the triaxial acceleration signal and triaxial angular velocity signal are filtered by a second-order Butterworth low-pass filter; the descent velocity signal is filtered by median filtering; and the smoothed air pressure signal, the filtered triaxial acceleration signal, the triaxial angular velocity signal, and the descent velocity signal are standardized.

[0033] In the specific implementation, a second-order Butterworth low-pass filter (cutoff frequency 20Hz) is used to eliminate high-frequency vibrations in the triaxial acceleration signal and triaxial angular velocity signal. At the same time, a 30-point moving average smoothing process is performed on the air pressure signal to suppress DC noise, and the descent velocity signal is filtered by median to remove peak interference.

[0034] Missing timestamps were filled by interpolation, and all channels were resampled at 200Hz. All channels were standardized with the mean μ and standard deviation σ of the training set to obtain normalized sequences with a mean of 0 and a variance of 1, namely the preprocessed air pressure sequence, triaxial acceleration sequence, triaxial angular velocity sequence and descent velocity sequence.

[0035] S2, the preprocessed descent velocity sequence is subjected to three-level discrete wavelet multi-scale decomposition, LSTM prediction and velocity fusion to obtain the enhanced velocity sequence.

[0036] It should be noted that the preprocessed descent velocity sequence (i.e., the standardized descent velocity sequence) is subjected to three-level discrete wavelet multi-scale decomposition using Haar wavelet (db1). The three-level decomposition automatically separates the descent velocity signal into three subsequences according to different time scales by performing "low-pass filtering + high-pass filtering + downsampling" operations in each level.

[0037] Furthermore, the preprocessed descent velocity sequence is subjected to three-level discrete wavelet multi-scale decomposition and denoising; the denoised multi-scale decomposition components are input into the corresponding LSTM branches to output trend prediction values, mesoscale perturbation prediction values ​​and fast perturbation prediction values; the trend prediction values, mesoscale perturbation prediction values ​​and fast perturbation prediction values ​​are weighted and fused to obtain the enhanced velocity sequence.

[0038] Furthermore, the three-level discrete wavelet multi-scale decomposition and denoising process for the preprocessed descent velocity sequence is as follows: the preprocessed descent velocity sequence is subjected to a first-level discrete wavelet decomposition to obtain a first-level low-frequency approximation component and a first-level high-frequency detail component; the first-level low-frequency approximation component is subjected to a second-level discrete wavelet decomposition to obtain a second-level low-frequency approximation component and a second-level high-frequency detail component; the second-level low-frequency approximation component is subjected to a third-level discrete wavelet decomposition to obtain a third-level low-frequency approximation component and a third-level high-frequency detail component; and denoising is performed on the first-level low-frequency approximation component, the first-level high-frequency detail component, the second-level low-frequency approximation component, the second-level high-frequency detail component, the third-level low-frequency approximation component, and the third-level high-frequency detail component using soft thresholds corresponding to each component.

[0039] In the specific implementation, in the first-level decomposition, the descent velocity sequence V(t) is subjected to low-pass filtering and high-pass filtering and downsampling to obtain the first-level low-frequency approximation component A1(t) and the first-level high-frequency detail component D1(t). D1(t) represents local rapid changes and turbulent disturbances during the descent process.

[0040] In the second-level decomposition, the low-frequency component A1(t) from the previous level is subjected to the same low-pass filtering, downsampling, and high-pass filtering processes to obtain the second-level low-frequency approximation component A2(t) and the second-level high-frequency detail component D2(t). This high-frequency component represents the mesoscale perturbation characteristics.

[0041] In the third-level decomposition, A2(t) is again subjected to low-pass and high-pass filtering and downsampling to obtain the third-level low-frequency approximation component A3(t) and the third-level high-frequency detail component D3(t). Among them, A3(t) represents the overall trend change of descent speed, reflecting the slow dynamics during the parachute descent process.

[0042] The six components formed by multi-scale decomposition are denoised using soft thresholding to eliminate noise interference. In this embodiment, the thresholds for the three high-frequency components are set to T1=0.12, T2=0.15, and T3=0.18, respectively, to suppress the impact of vibration and high-frequency noise on prediction accuracy.

[0043] After denoising, the three scale components are input into three LSTM branches. The third low-frequency component A3(t) is input into a trend prediction LSTM (two-layer structure, 64 hidden units per layer) to predict the overall trend change of descent velocity, characterizing the slow velocity evolution during parachute descent, and outputting the trend prediction value. The second high-frequency component D2(t) is input into a mesoscale disturbance prediction LSTM (two-layer structure, 48 hidden units per layer) to capture mesoscale disturbances, such as aerodynamic disturbances and airflow changes, and outputs the mesoscale disturbance prediction value. The first high-frequency component D1(t) is input into a fast disturbance prediction LSTM (single-layer 32-unit structure) to characterize fast aerodynamic disturbances such as turbulence, capture rapid fluctuations and turbulent peaks in descent velocity, and output the fast disturbance prediction value.

[0044] Furthermore, the trend prediction value, the mesoscale disturbance prediction value, and the fast disturbance prediction value are weighted and fused using a weighted fusion formula; Weighted fusion formula:

[0045]

[0046] In the formula, To enhance the velocity sequence, , and These are parameters (i.e., fixed values ​​obtained through automatic optimization during training). This is a trend forecast value. These are predicted values ​​for mesoscale disturbances. These are the predicted values ​​for rapid disturbances.

[0047] It should also be noted that the fusion factor is defined. , , These are three trainable scalar parameters, with initial values ​​of 0.5, 0.3, and 0.2 respectively. These three parameters are automatically updated through backpropagation during network training, eventually converging to the optimal fusion weights.

[0048] The enhanced velocity sequence is used to replace the original descent velocity input to more accurately characterize the variation of descent velocity at different time scales.

[0049] S3. Generate a parachute descent state sequence based on the enhanced velocity sequence, the pre-processed air pressure sequence, the triaxial acceleration sequence, and the triaxial angular velocity sequence.

[0050] The enhanced velocity sequence, together with the barometric pressure sequence, triaxial acceleration sequence, and triaxial angular velocity sequence, forms a fixed-structure nine-dimensional input sequence (i.e., the parachute landing state sequence). Each frame is organized as follows:

[0051] In the formula, This is a sequence of parachute landing states. It is a pressure sequence, ( ) is a triaxial acceleration sequence, ( () is a triaxial angular velocity sequence.

[0052] S4, input the parachute state sequence into the preset parachute state prediction model, and output the predicted value of the parachute pressure difference. The preset parachute state prediction model is constructed by combining the EVO algorithm, CNN and BiLSTM.

[0053] Further, the parachute state sequence is input into a preset parachute state prediction model; the CNN structure is automatically searched using the EVO algorithm, and the searched CNN structures are combined into the current optimal CNN; the parachute state sequence is subjected to sliding convolution operation by the current optimal CNN to obtain the local spatial feature sequence of air disturbance and pressure change; the local spatial feature sequence is analyzed by BiLSTM to obtain the overall temporal features; based on the overall temporal features, the attention weights of each time step are calculated through the attention layer; the attention weights of each time step and the features of each time step in the overall temporal feature matrix are weighted and summed to obtain the global temporal feature vector, and the predicted value of parachute pressure difference is obtained through a two-layer fully connected network based on the global temporal feature vector.

[0054] In the specific implementation, the kernel size and number of channels can be set. The Energy Valley Optimization (EVO) algorithm is used to search for the optimal combination of convolutional network structures in the hyperparameter space of the set kernel size (3, 5, 7) and the set number of channels (16, 32, 64). Finally, the CNN structure is determined, for example, consisting of three one-dimensional convolutional layers with kernel sizes of 5, 3, and 3, and the number of channels of 32, 32, and 64, respectively. Each convolutional layer is followed by a ReLU activation function and a max pooling layer.

[0055] The searched CNN structures are combined to form the current optimal CNN. This optimal CNN is then used to perform sliding convolution operations on the nine-dimensional input sequence to extract local spatial features of air disturbances and pressure changes during parachute descent. The feature sequence output by the CNN is fed into a temporal network consisting of two layers of bidirectional LSTM (64 units per layer) to extract the forward-backward dependencies of aerodynamic disturbances. The BiLSTM, through joint modeling of forward and backward hidden states, accurately captures the temporal dependency structure of aerodynamic disturbances changing over time during parachute descent, generating an overall temporal feature representation matrix H(t).

[0056] An attention layer is applied to the BiLSTM output sequence, and the attention weights at each time step are calculated:

[0057] In the formula, The attention weights are for time step t. The features at each time step within the overall time series feature matrix are... These are trainable parameters.

[0058] The global temporal feature vector is obtained by weighted summation:

[0059] In the formula, F is the global time-series feature vector, which is used to highlight key segments such as enhanced aerodynamic vortices, sudden pressure changes, and abnormal descent speeds.

[0060] The global temporal feature vector after attention fusion is input into a two-layer fully connected network (128→64→2) to output the predicted value of parachute pressure difference ΔP. The output layer uses a linear activation function to ensure that the predicted value is consistent with the range of the actual physical quantity.

[0061] S5. Based on the predicted value of the parachute pressure difference, estimate the parachute drop risk status and issue a parachute deployment risk warning based on the parachute drop risk status.

[0062] In practice, the risk status of the parachute landing also needs to be determined according to the risk status assessment rules.

[0063] The risk assessment rules are as follows: A state is considered dangerous when ΔP ≥ 120 Pa.

[0064] The condition is considered to be in a critical state when 80Pa≤ΔP<120Pa.

[0065] All other situations are considered safe.

[0066] To avoid false alarms, a three-window consistency check is used to reduce false alarms. In this embodiment, the window size is 0.5 seconds (100 sampling points), and the final warning result is output when three consecutive windows show the same level of judgment.

[0067] In a dangerous situation, a red warning is output and the time is recorded; in a critical situation, a yellow warning is output to prompt the paratrooper to adjust their attitude; in a safe situation, a green indicator is maintained. In practice, the parachute can be pre-deployed or assisted in attitude adjustment based on the warning signal.

[0068] In this embodiment, during the parachute descent, air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals are first acquired. These signals are then preprocessed. Next, the preprocessed descent velocity sequence undergoes three-layer discrete wavelet multi-scale decomposition, LSTM prediction, and velocity fusion to obtain an enhanced velocity sequence. Based on this enhanced velocity sequence, the preprocessed air pressure sequence, triaxial acceleration sequence, and triaxial angular velocity sequence, a parachute descent state sequence is generated. This sequence is then input into a preset parachute descent state prediction model, which outputs a parachute pack pressure difference prediction value. This model is constructed using a combination of the EVO algorithm, CNN, and BiLSTM. Finally, the parachute descent risk status is estimated based on the predicted parachute pack pressure difference value, and a parachute deployment risk warning is issued accordingly. This embodiment uses multi-source data such as air pressure, descent velocity, triaxial acceleration, and triaxial angular velocity to form a nine-dimensional time-series input. After preprocessing, a unified feature sequence is formed. The energy valley optimization algorithm is used to adaptively optimize the convolutional neural network (CNN) structure to extract the local spatial features of the parachute descent flow field. Then, a bidirectional long short-term memory network is combined to capture the temporal dependencies in the eddy evolution process. The attention mechanism is used to highlight key turbulent segments, thereby achieving high-precision prediction of eddy intensity and parachute pressure difference, and thus identifying risk states in real time for early warning.

[0069] Reference Figure 3 , Figure 3 This is a structural block diagram of the first embodiment of the umbrella opening risk warning system of the present invention.

[0070] like Figure 3 As shown, the umbrella opening risk early warning system proposed in this embodiment of the invention includes: The data processing module 3001 is used to collect air pressure signals, triaxial acceleration signals, triaxial angular velocity signals and descent velocity signals during the parachute descent, and to preprocess the air pressure signals, triaxial acceleration signals, triaxial angular velocity signals and descent velocity signals. The velocity enhancement module 3002 is used to perform three-level discrete wavelet multi-scale decomposition, LSTM prediction and velocity fusion on the preprocessed descent velocity sequence to obtain the enhanced velocity sequence. Data integration module 3003 is used to generate a parachute state sequence based on the enhanced velocity sequence, the preprocessed air pressure sequence, the triaxial acceleration sequence, and the triaxial angular velocity sequence; The differential pressure prediction module 3004 is used to input the parachute state sequence into a preset parachute state prediction model and output the predicted value of the parachute pressure difference. The preset parachute state prediction model is constructed by a combination of EVO algorithm, CNN and BiLSTM. The risk warning module 3005 is used to estimate the parachute drop risk status based on the predicted parachute pressure difference value, and to issue a parachute deployment risk warning based on the parachute drop risk status.

[0071] Other embodiments or specific implementations of the umbrella opening risk warning system of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0072] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0073] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0074] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0075] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for early warning of umbrella opening risk, characterized in that, The method includes the following steps: During the parachute descent, air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals are collected, and the air pressure signals, triaxial acceleration signals, triaxial angular velocity signals, and descent velocity signals are preprocessed. The preprocessed descent velocity sequence is subjected to three-level discrete wavelet multi-scale decomposition, LSTM prediction and velocity fusion to obtain an enhanced velocity sequence. A parachute state sequence is generated based on the enhanced velocity sequence, the pre-processed air pressure sequence, the triaxial acceleration sequence, and the triaxial angular velocity sequence. The parachute state sequence is input into a preset parachute state prediction model, and the predicted value of the parachute pressure difference is output. The preset parachute state prediction model is constructed by a combination of EVO algorithm, CNN and BiLSTM. Based on the predicted parachute pressure difference, the parachute drop risk status is estimated, and an early warning of parachute deployment risk is issued based on the parachute drop risk status.

2. The method as described in claim 1, characterized in that, The preprocessing of the air pressure signal, the triaxial acceleration signal, the triaxial angular velocity signal, and the descent velocity signal includes: The air pressure signal is smoothed by a moving average. The triaxial acceleration signal and the triaxial angular velocity signal are filtered by a second-order Butterworth low-pass filter, respectively. The descent speed signal is subjected to median filtering. The smoothed air pressure signal, the filtered triaxial acceleration signal, the triaxial angular velocity signal, and the descent velocity signal were standardized respectively.

3. The method as described in claim 1, characterized in that, The process of performing three-level discrete wavelet multi-scale decomposition, LSTM prediction, and velocity fusion on the preprocessed descent velocity sequence to obtain an enhanced velocity sequence includes: The preprocessed descent velocity sequence is subjected to three-level discrete wavelet multi-scale decomposition and denoising. The denoised multi-scale decomposition components are input into the corresponding LSTM branches to output trend prediction values, mesoscale perturbation prediction values, and fast perturbation prediction values. The trend prediction, mesoscale disturbance prediction, and fast disturbance prediction are weighted and fused to obtain an enhanced velocity sequence.

4. The method as described in claim 3, characterized in that, The process of performing three-level discrete wavelet multi-scale decomposition and denoising on the preprocessed descent velocity sequence includes: The preprocessed descent velocity sequence is subjected to the first layer of discrete wavelet decomposition to obtain the first layer of low-frequency approximate components and the first layer of high-frequency detail components. The first-level low-frequency approximation component is subjected to the second-level discrete wavelet decomposition to obtain the second-level low-frequency approximation component and the second-level high-frequency detail component. The second-level low-frequency approximation component is subjected to a third-level discrete wavelet decomposition to obtain the third-level low-frequency approximation component and the third-level high-frequency detail component. The first layer low-frequency approximation component, the first layer high-frequency detail component, the second layer low-frequency approximation component, the second layer high-frequency detail component, the third layer low-frequency approximation component, and the third layer high-frequency detail component are denoised by using the soft thresholds corresponding to each component.

5. The method as described in claim 3, characterized in that, The weighted fusion of the trend prediction value, the mesoscale disturbance prediction value, and the fast disturbance prediction value includes: The trend prediction value, the mesoscale disturbance prediction value, and the fast disturbance prediction value are weighted and fused using a weighted fusion formula. The weighted fusion formula is as follows: In the formula, To enhance the velocity sequence, , and For parameters, This is a trend forecast value. These are predicted values ​​for mesoscale disturbances. These are the predicted values ​​for rapid disturbances.

6. The method as described in claim 1, characterized in that, The step of inputting the parachute state sequence into a preset parachute state prediction model and outputting the predicted parachute pressure difference value includes: The parachute state sequence is input into a preset parachute state prediction model; The EVO algorithm is used to automatically search for CNN structures and then combine the searched CNN structures into the current optimal CNN. The parachute state sequence is subjected to sliding convolution operation by the current optimal CNN to obtain the local spatial feature sequence of air disturbance and pressure change; The overall temporal features are obtained by performing feature analysis on the local spatial feature sequence using BiLSTM. Based on the overall temporal features, a global temporal feature vector is obtained through an attention layer, and the predicted value of the parachute pressure difference is obtained through a two-layer fully connected network based on the global temporal feature vector.

7. The method as described in claim 6, characterized in that, The process of obtaining the global temporal feature vector through an attention layer based on the overall temporal features includes: Based on the overall temporal characteristics, the attention weights for each time step are calculated through an attention layer; The attention weights at each time step and the features at each time step within the overall temporal feature matrix are weighted and summed to obtain the global temporal feature vector.

8. A system for early warning of umbrella opening risks, characterized in that, The system includes: The data processing module is used to collect air pressure signals, triaxial acceleration signals, triaxial angular velocity signals and descent velocity signals during the parachute descent, and to preprocess the air pressure signals, triaxial acceleration signals, triaxial angular velocity signals and descent velocity signals. The velocity enhancement module is used to perform three-level discrete wavelet multi-scale decomposition, LSTM prediction and velocity fusion on the preprocessed descent velocity sequence to obtain the enhanced velocity sequence. The data integration module is used to generate a parachute state sequence based on the enhanced velocity sequence, the preprocessed air pressure sequence, the triaxial acceleration sequence, and the triaxial angular velocity sequence. The differential pressure prediction module is used to input the parachute state sequence into a preset parachute state prediction model and output the predicted value of the parachute pressure difference. The preset parachute state prediction model is constructed by a combination of EVO algorithm, CNN and BiLSTM. The risk warning module is used to estimate the parachute drop risk status based on the predicted parachute pressure difference value, and to issue a parachute deployment risk warning based on the parachute drop risk status.

9. A device for early warning of umbrella opening risks, characterized in that, The device includes: a memory, a processor, and an umbrella opening risk warning program stored in the memory and executable on the processor, the umbrella opening risk warning program being configured to implement the steps of the umbrella opening risk warning method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a parachute opening risk warning program, which, when executed by a processor, implements the steps of the parachute opening risk warning method as described in any one of claims 1 to 7.

Images