A dynamically adjusted real-time parachute state prediction method and system

Abstract

The application discloses a dynamic adjustment real-time parachute state prediction method, through a plurality of sensors arranged on a parachutist, flight state data are collected in real time and transmitted to a data processing unit wirelessly; the collected data are pretreated, key features are extracted by using a convolutional neural network, and a one-dimensional feature vector is generated; a parachuting task is divided into a plurality of subtasks, each subtask corresponds to different environments or states, initial weight distribution obtained based on historical data is used for feature weighting of each subtask, and the weighted comprehensive feature vector is generated by adjusting the weight in real time; the weighted comprehensive feature vector is subjected to time sequence analysis by using a long short-term memory neural network, and a prediction result for a future time point is generated; the application responds to environment or state changes in real time through a dynamic adjustment mechanism, so that the prediction model is more suitable for complex and changeable parachuting environments.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of parachute safety protection, and particularly relates to a dynamic adjustment real-time parachute state prediction method and system. BACKGROUND

[0002] As a challenging and stimulating high-altitude sport, the safety of parachuting is the focus of attention. In particular, during the parachute descent, the parachutist may encounter a complex and unpredictable airflow environment, among which the vortex is an important influencing factor. The presence of vortexes can cause the parachutist to stay in the air for a longer time, increasing the time affected by the synthetic wind in the air, thus disrupting the planned landing trajectory, and even causing the parachutist to drift out of the landing field, causing landing in complex terrain, which seriously threatens the safety of landing. In addition, the instability and rotation of vortexes are extremely easy to cause the parachute to sway and jolt, and in severe cases, may cause the parachutist to have symptoms such as dizziness and vomiting, and even cause operational errors, endangering the safety of the parachutist.

[0003] In order to predict whether a vortex will be encountered during parachuting, traditional methods mainly rely on static prediction methods based on physical models. These methods usually use fixed parameters and formulas to model and predict key characteristics during parachuting. However, the actual parachuting environment is complex and variable, involving numerous nonlinear factors and dynamic changes, such as nonlinear airflow, temperature changes, rotational fluctuations, etc. These static models often fail to effectively describe these sudden changes or state evolution across time steps, resulting in serious limitations on the accuracy and reliability of the prediction results. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art, and to provide a dynamic adjustment real-time parachute state prediction method and system, which dynamically adjusts the feature weights in real time in response to changes in the environment or state during parachuting, generates a weighted comprehensive feature vector, and makes the prediction model more adaptable to the complex and variable environment during actual parachuting, thereby improving the prediction accuracy and reliability.

[0005] To achieve the above purpose, the technical solution adopted by the present application is:

[0006] A dynamic adjustment real-time parachute state prediction method, comprising the following steps:

[0007] S1, collecting multiple flight state data during parachuting through multiple sensors arranged on the parachutist, and wirelessly transmitting these data to a data processing unit;

[0008] S2, preprocessing the collected data, extracting key features from the preprocessed data using a convolutional neural network, and generating a one-dimensional feature vector representing the parachute environment or state;

[0009] S3, divide the skydiving task into multiple sub-tasks, each corresponding to a different environment or state, use the initial weight distribution obtained from the historical data as the basis, weight the features for each sub-task, and adjust the weight distribution of each feature in real time based on the changes in the environment or state features during the skydiving process, to generate a weighted comprehensive feature vector;

[0010] S4, use a long short-term memory neural network to perform time series analysis on the weighted comprehensive feature vector, analyze the relationship between different time steps, and generate a prediction result for the future time point.

[0011] Step S1 includes:

[0012] By calculating the difference between the pressure sensors of different parts of the body, the direction of the air flow around the skydiver is obtained;

[0013] By air pressure sensor, monitor the environmental pressure during the skydiving process;

[0014] By temperature sensor, monitor the temperature during the skydiving process;

[0015] By attitude angle sensor, measure the body posture of the skydiver, including the tilt angle and the pitch angle;

[0016] By speed sensor, measure the descent speed of the skydiver.

[0017] In step S2, the pre-processing of the collected data includes:

[0018] S2.1, check if there are missing values in the original data collected from each sensor on the skydiver's body, for incomplete data records, discard these records, and based on the rules of historical data, detect and replace abnormal values;

[0019] S2.2, align the data from different sensors to the same time reference through timestamp matching;

[0020] S2.3, for missing points in the data, use dynamic interpolation method to fill in;

[0021] S2.4, select key features from sensor data, including attitude angle, speed, and wind speed, construct a row of these data features at each time point, and form a two-dimensional matrix as time progresses.

[0022] Step S2.3 includes:

[0023] S2.3.1, select the adjacent points g0, g1, g2, g3 of the missing point gt and their corresponding time points, where g1 and g2 are the previous and subsequent valid values of gt respectively;

[0024] S2.3.2, initialize the weights of g1 and g2 as the average value, the calculation formula is:

[0025] w1(t-1) = 0.5, w2(t-1) = 0.5;

[0026] wherein, w1(t-1) and w2(t-1) are the initial weight values of g1 and g2 respectively.

[0027] S2.3.3, calculate the change value using the difference between the feature values of adjacent points:

[0028] Δg1 = |g1-g0|, Δg2 = |g2-g3|;

[0029] S2.3.4, calculate the weight increment of g1 and g2 according to the change value, the calculation method is:

[0030] Δw1 = λfront·Δg1, Δw2 = λrear·Δg2;

[0031] wherein, λfront and λrear are sensitivity parameters set according to historical experience;

[0032] S2.3.5, dynamically update the weights of g1 and g2:

[0033] w1(t) = w1(t-1) + Δw1, w2(t) = w2(t-1) + Δw2;

[0034] S2.3.6, normalize the weights:

[0035] w1(t) = w1(t) / (w1(t) + w2(t)), w2(t) = w2(t) / (w1(t) + w2(t));

[0036] S2.3.7, according to the normalized weights, dynamically interpolate the missing points gt, the calculation formula is:

[0037] gt = w1(t)·g1 + w2(t)·g2.

[0038] In step S2, the pre-processing of the collected data also includes:

[0039] S2.5, extract the minimum value and the maximum value of each feature from the collected original data;

[0040] S2.6, scale each data, the calculation method is:

[0041] x' = (x-min) / (max-min),

[0042] where x is the original data, min is the minimum value of the feature, max is the maximum value of the feature, and x' is the scaled data.

[0043] S2.7, verify the scaled data set to ensure that the data values of each feature are within the range [0, 1].

[0044] Step S3 includes:

[0045] S3.1, divide the state prediction task in the parachute process into multiple sub-tasks f = [f1, f2, …, fn], each sub-task corresponding to different environments or state changes;

[0046] S3.2, inner loop training on sub-tasks, based on historical data to get the initial attention weight α0 = [α1, α2, …, αn];

[0047] S3.3, for the features extracted by the convolutional neural network, use the initial weight obtained from the historical data as the basis to perform feature weighting;

[0048] S3.4, when the environment or state feature changes in the parachute process, the attention weight of the corresponding feature is dynamically adjusted in real time;

[0049] S3.5, through the outer loop, the inner loop results of multiple sub-tasks are summarized and analyzed to extract cross-task common information and adjust the weight again.

[0050] Step S3.4 includes:

[0051] S3.4.1, when the air flow speed changes dramatically, calculate the change value Δf1 of the air flow feature;

[0052] When the relative position and angle of the head and limbs of the parachutist change sharply, calculate the change value Δf2 of the posture feature;

[0053] When the vertical falling speed or lateral drift speed changes suddenly, calculate the change value Δf3 of the speed feature;

[0054] When the rotation rate of the parachutist changes suddenly, calculate the change value Δf4 of the rotation feature;

[0055] S3.4.2, normalize each change value, the calculation formula is as follows:

[0056]

[0057] ; Where max(Δfi) is the maximum change value of all features;

[0058] S3.4.3, calculate the dynamic adjustment value Δαi of each feature based on the change value, the calculation formula is:

[0059]

[0060] wherein, wherein λi is the sensitivity parameter for the ith feature.

[0061] S3.4.4, dynamically adjust the attention weight according to the difference between the current feature f(t) and the historical feature f(t-1), and the calculation formula is:

[0062] α i (t)=α i (t-1)+Δα i

[0063] ; wherein, αi(t-1) is the attention weight of the ith feature at the previous moment;

[0064] S3.4.5, after dynamically adjusting the attention weight of each feature, multiply all the features by the respective dynamic weight and then perform weighted summation to generate a comprehensive weighted feature vector f, and the calculation formula is:

[0065]

[0066] Step S3.5 includes:

[0067] S3.5.1, according to the feature change value of each subtask, judge whether there is a same progress or same retreat task, if the change value of certain two tasks is always same positive or same negative, it belongs to same progress or same retreat task;

[0068] S3.5.2, for the subtasks identified as same progress or same retreat tasks, apply the reduction strategy to reduce the weight difference between these subtasks, which is realized by the following formula:

[0069] α i (t)=α i (t-1)×(1-β)

[0070] α j (t)=α j (t-1)×(1-β);

[0071] wherein, αi(t) and αj(t) are two same progress or same retreat tasks; β is the adjustment ratio, which can be adjusted, and is initially set to 0.2;

[0072] S3.5.3, normalize the adjusted weight, so that the sum of the weights of all subtasks is 1, and the normalization processing calculation formula is as follows:

[0073]

[0074] S3.5.4, after each time step completes the adjustment, enter the next time step t+1, re-detect feature changes and adjust.

[0075] Step S4 comprises:

[0076] S4.1, the weighted integrated feature vector is input into the LSTM network in time step order;

[0077] S4.2, the LSTM network generates eddy current prediction results for future time points according to the input time series features; wherein, the prediction results include identifying and predicting the eddy current region formed around the parachutist, evaluating the intensity of the eddy current and its influence on the parachutist.

[0078] A dynamic adjustment real-time parachute state prediction system comprises:

[0079] A data acquisition module is used to collect multiple flight state data in the parachute process in real time through multiple sensors arranged on the parachutist, and wirelessly transmit these data to the data processing unit;

[0080] A feature extraction module uses a convolutional neural network to extract key features from preprocessed data to generate a one-dimensional feature vector representing the state of the parachutist;

[0081] A dynamic attention adjustment module divides the parachute task into multiple subtasks, each corresponding to a different environment or state, uses the initial weight distribution obtained from historical data as the basis, weights each subtask, and adjusts the weight distribution of each feature in real time based on the changes in the environment or state features during the parachute process to generate a weighted integrated feature vector;

[0082] A time series analysis and prediction module uses a long short-term memory neural network to perform time series analysis on the weighted integrated feature vector, analyzes the relationship between different time steps, and generates prediction results for future time points based on the results of the time series analysis.

[0083] A computer-readable storage medium stores a computer program, wherein the computer program is executed by a processor to implement the method steps described above.

[0084] Compared with the prior art, the beneficial effects of the present application are:

[0085] 1. Through the dynamic adjustment mechanism, the feature weight is dynamically adjusted in real time to respond to the changes in the environment or state during the parachute process, and the weighted integrated feature vector is generated, so that the prediction model is more suitable for the complex and changeable environment in the actual parachute process, thereby improving the prediction accuracy and reliability;

[0086] 2、The state prediction task in the parachute process is divided into multiple sub-tasks, and through the hierarchical cycle training strategy, various situations in the parachute process are captured in more detail, and the adaptability of the model is improved;

[0087] 3、The prediction result generated by the LSTM network includes identifying and predicting the vortex area formed around the parachutist, evaluating the strength of the vortex and the influence on the parachutist, so that the parachutist can more accurately judge the current safety situation and make the next safety decision, thereby reducing the risk in the parachute process. BRIEF DESCRIPTION OF DRAWINGS

[0088] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0089] Figure 1 The overall flowchart of the prediction method in the embodiments of the present application is shown in the figure.

[0090] Figure 2 The method schematic diagram of the one-dimensional feature vector extraction process in the embodiments of the present application is shown in the figure.

[0091] Figure 3 The method schematic diagram of the dynamic attention regulation mechanism in the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0092] The technical solutions of the present application will be described in detail below with reference to the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0093] The size of the serial number of each step in the present application specification does not mean the order of execution. The execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0094] In the description of the present application and the appended claims, the terms "first", "second", "third", etc. are used only to distinguish descriptions and cannot be understood as indicating or implying relative importance. It should also be understood that although the terms "first", "second", etc. are used in the text to describe various elements in some embodiments of the present application, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, the first table can be named as the second table, and similarly, the second table can be named as the first table without departing from the scope of various described embodiments. The first table and the second table are both tables, but they are not the same table.

[0095] In the present application, the reference "one embodiment" or "some embodiments" means that the specific features, structures or characteristics described in connection with the embodiment are included in one or more embodiments of the present application. Therefore, the statements "in one embodiment", "in some embodiments", "in other some embodiments", "in further some embodiments" and the like appearing in different places in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized. The terms "include", "contain", "have" and their variants mean "include but not limited to", unless otherwise specifically emphasized.

[0096] To predict whether a parachuting process will encounter a vortex, the traditional method mainly relies on a static prediction method based on a physical model. These methods usually use fixed parameters and formulas to model and predict key characteristics in the parachuting process. However, the actual parachuting environment is complex and variable, involving many nonlinear factors and dynamic changes, such as nonlinear airflow, temperature changes, rotational fluctuations, etc. These static models often have difficulty in effectively describing these sudden changes or state evolution across time steps, resulting in serious limitations on the accuracy and reliability of the prediction results.

[0097] In view of the above technical problems, as shown in Figure 1 The embodiments of the present application provide a dynamic real-time parachute state prediction method, comprising the following steps:

[0098] S1, collecting multiple flight state data in the parachuting process through multiple sensors arranged on the parachuter, and wirelessly transmitting the data to a data processing unit.

[0099] S2, preprocessing the collected data, extracting key features from the preprocessed data using a convolutional neural network, and generating a one-dimensional feature vector representing the parachuting environment or state.

[0100] S3, divide the parachute task into multiple sub-tasks, each corresponding to a different environment or state, use the initial weight distribution obtained from historical data as the basis, weight each sub-task, and based on the changes in the environment or state characteristics during the parachute process, adjust the weight distribution of each feature in real time to generate a weighted comprehensive feature vector.

[0101] S4, use a long short-term memory neural network to perform time series analysis on the weighted comprehensive feature vector, analyze the relationship between different time steps, and generate a prediction result for the future time point.

[0102] The method of the embodiment dynamically adjusts the feature weight in real time in response to changes in the environment or state during the parachute process, generates a weighted comprehensive feature vector, and makes the prediction model more adaptable to the complex and variable environment during the actual parachute process, thereby improving the prediction accuracy and reliability.

[0103] Step S1 includes:

[0104] The direction of the airflow around the parachutist can be obtained by calculating the difference between the pressure sensors at different parts of the body.

[0105] The environmental pressure during the parachute process can be monitored by the air pressure sensor.

[0106] The temperature during the parachute process can be monitored by the temperature sensor.

[0107] The body posture of the parachutist can be measured by the attitude angle sensor, including the inclination angle and the pitch angle.

[0108] The descent speed of the parachutist can be measured by the speed sensor.

[0109] In step S2, the pre-processing of the collected data includes:

[0110] S2.1, check whether there are missing values in the original data collected from the sensors on the parachutist's body, for the missing data records, if the records are incomplete, discard these records, and based on the rules of historical data, detect and replace abnormal values.

[0111] S2.2, align the data from different sensors to the same time reference through timestamp matching; to eliminate the time misalignment problem caused by different sensor sampling rates or transmission delays.

[0112] S2.3, for missing points in the data, use dynamic interpolation method to fill in, to ensure the continuity and integrity of the data.

[0113] S2.4, select key features from sensor data, including attitude angle, speed, wind speed, and construct these data features at each time point into a row, forming a two-dimensional matrix as time progresses, thereby providing structured data input for subsequent feature extraction and time series analysis.

[0114] Through the above preprocessing process, the processing of missing values and outliers ensures the accuracy of the data, the timestamp matching eliminates the time misalignment problem, and the dynamic interpolation method fills in the missing data points, ensuring the continuity of the data. The selection of key features and the construction of two-dimensional matrix make the data structured, which facilitates the subsequent convolutional neural network for feature extraction and long short-term memory neural network for time series analysis.

[0115] Step S2.3 includes:

[0116] S2.3.1, select the adjacent points g0, g1, g2, g3 of the missing point gt and their corresponding time points, wherein g1 and g2 are the previous and subsequent adjacent valid values of gt respectively;

[0117] S2.3.2, initialize the weights of g1 and g2 as the mean, and the calculation formula is:

[0118] w1(t-1) = 0.5, w2(t-1) = 0.5;

[0119] wherein w1(t-1) and w2(t-1) are the initial weight values of g1 and g2 respectively; it is assumed that the weights are equal in the absence of feature changes.

[0120] S2.3.3, calculate the change amount using the difference between the feature values of adjacent points:

[0121] Δg1 = |g1-g0|, Δg2 = |g2-g3|;

[0122] thereby evaluating the change trend of the data near the missing point.

[0123] S2.3.4, calculate the weight increment of g1 and g2 according to the change amount, and the calculation method is:

[0124] Δw1 = λfront·Δg1, Δw2 = λback·Δg2;

[0125] wherein λfront and λback are sensitivity parameters set according to historical experience, to reflect the importance of data changes in historical experience.

[0126] S2.3.5, dynamically update the weights of g1 and g2:

[0127] w1(t) = w1(t-1) + Δw1, w2(t) = w2(t-1) + Δw2;

[0128] S2.3.6, Normalization weight:

[0129] w1(t)=w1(t) / (w1(t)+w2(t)), w2(t)=w2(t) / (w1(t)+w2(t));

[0130] S2.3.7, Dynamic interpolation of missing points gt according to normalized weight, the calculation formula is:

[0131] gt=w1(t)·g1+w2(t)·g2.

[0132] Using the above dynamic interpolation method, by dynamically adjusting the weight of adjacent points to reflect the trend of change, so that the interpolation result is more in line with the change rule of actual data.

[0133] In step S2, the pre-processing of the collected data also includes:

[0134] S2.5, Extract the minimum and maximum value of each feature from the collected original data;

[0135] S2.6, Scale each data, the calculation method is:

[0136] x'=(x-min) / (max-min),

[0137] Where x is the original data, min is the minimum value of the feature, max is the maximum value of the feature, and x' is the scaled data.

[0138] S2.7, Verify the scaled data set to ensure that the data value of each feature is within the range of [0, 1].

[0139] Through the scaling of the feature value by the above steps, the influence of the difference between different features due to the difference in dimension and value range can be eliminated, and the consistency of the data is improved.

[0140] As shown in Figure 2 In step S2, the pre-processed data is extracted by using convolutional neural network to generate one-dimensional feature vector representing the environment or state of parachute jump, including:

[0141] For each time step of pre-processed data, the first layer of convolution operation is applied, so as to calculate the local feature map.

[0142] After the first convolutional layer, multiple convolutional layers are stacked to gradually extract higher level features.

[0143] Each convolutional layer performs convolution operation on the feature map of the previous layer to extract more abstract and higher level features.

[0144] After each convolutional layer, batch normalization is applied to normalize the distribution of data, speed up the training process, and prevent overfitting of the model.

[0145] Max pooling layers are used to downsample the feature maps output by convolutional layers.

[0146] Max pooling reduces the size of feature maps by selecting the maximum value within local regions, while preserving important information.

[0147] Multiple pooling operations further reduce the dimensionality of feature maps, reducing computational load while preserving key features.

[0148] The feature maps after multiple convolution and pooling operations are flattened into one-dimensional vectors.

[0149] Full connection layers are used to further reduce the high-dimensional feature vectors, generating final one-dimensional feature vectors.

[0150] In the full connection layer, Dropout regularization technique is applied.

[0151] Dropout prevents overfitting of the model by randomly discarding a certain proportion of neurons.

[0152] After multiple convolution, pooling, and dimensionality reduction operations, one-dimensional feature vectors are generated.

[0153] This one-dimensional feature vector contains key features and information in the input data, ready to be input into the subsequent dynamic attention mechanism.

[0154] As shown in Figure 3 , step S3 includes:

[0155] S3.1, divide the state prediction task in the parachute process into multiple subtasks f = [f1, f2, …, fn], each corresponding to different environments or state changes;

[0156] S3.2, perform inner loop training on subtasks, and obtain general initial attention weights α0 = [α1, α2, …, αn] based on historical data;

[0157] S3.3, for the features extracted by the convolutional neural network, use the initial weights obtained from the historical data as the basis for feature weighting;

[0158] S3.4, when the environment or state features change during the parachute process, real-time dynamic adjustment of the attention weights of the corresponding features is performed;

[0159] S3.5, through outer loop, the inner loop results of multiple subtasks are summarized and analyzed, and the common information across tasks is extracted, and the weights are adjusted again.

[0160] By the training strategy of layer-by-layer circulation in the above steps, the overall task is divided into multiple sub-tasks, which can capture various situations in the parachuting process more carefully, so as to more accurately predict the state change.

[0161] Step S3.4 includes:

[0162] S3.4.1, when the air flow speed changes sharply, calculate the change value Δf1 of the air flow feature;

[0163] When the relative position and angle of the head and limbs of the parachutist change sharply, calculate the change value Δf2 of the posture feature;

[0164] When the vertical falling speed or lateral drift speed changes suddenly, calculate the change value Δf3 of the speed feature;

[0165] When the rotation rate of the parachutist changes suddenly, calculate the change value Δf4 of the rotation feature;

[0166] S3.5.2, normalize each change value, and the calculation formula is as follows:

[0167]

[0168] Wherein, max(Δfi) is the maximum change value of all features;

[0169] S3.4.3, calculate the dynamic adjustment value Δαi based on the change value, and the calculation formula is:

[0170]

[0171] Wherein, λi is the sensitivity parameter for the i-th feature.

[0172] S3.4.4, dynamically adjust the attention weight according to the difference between the current feature f(t) and the historical feature f(t-1), and the calculation formula is:

[0173] α i (t)=α i (t-1)+Δα i ;

[0174] Wherein, αi(t-1) is the attention weight of the previous moment;

[0175] S3.4.5, after dynamically adjusting the attention weight of each feature, multiply all features by the respective dynamic weight and then perform weighted summation to generate a comprehensive weighted feature vector f, and the calculation formula is:

[0176]

[0177] In the above steps, the dynamic adjustment mechanism ensures that the system can respond to changes in the environment or state in real time, and the comprehensive weighted feature vector f provides more comprehensive data information.

[0178] Step S3.5 includes:

[0179] S3.5.1, according to the feature change value of each subtask, determine whether there are same or same tasks, wherein if the change value of certain two tasks is always same positive or same negative, it belongs to same or same task.

[0180] Specifically, after obtaining the feature change value of each subtask, further analyze the relationship between these change values, and determine whether there are certain subtasks whose feature change values are always same positive or same negative, i.e. their change trend is consistent. If there is such a correlation, these subtasks are regarded as "same or same task".

[0181] S3.5.2, for the subtasks identified as same or same task, calculate the current weight, and apply the reduction strategy to reduce the weight difference between these subtasks, wherein the reduction strategy is realized by the following formula:

[0182] α i (t)=α i (t-1)×(1-β)

[0183] α j (t)=α j (t-1)×(1-β);

[0184] Wherein, αi(t) and αj(t) are two same or same tasks; β is the adjustment ratio, which can be adjusted, and is initially set to 0.2.

[0185] For the subtasks identified as same or same task, the feature weight of the subtasks needs to be adjusted, and the weight difference between the subtasks needs to be reduced to reflect the commonality or correlation between them.

[0186] S3.5.3, normalize the adjusted weight, so that the sum of the weights of all subtasks is 1. The normalization processing formula is as follows:

[0187]

[0188] Through normalization, it can be ensured that the weight of each subtask is between 0 and 1, and the sum of the weights of all subtasks is 1.

[0189] S3.5.4, after adjusting each time step, enter the next time step t+1, and re-detect the feature change and adjust.

[0190] The whole dynamic adjustment process is a continuous process. At each time step, the feature change value is recalculated according to the current environmental or state characteristics, the same task is determined, the weight is adjusted, and the weighted comprehensive feature vector is generated. Then, the next time step is entered, and the above process is repeated, so as to respond to the environmental or state changes in the parachute process in real time, and improve the accuracy and reliability of the prediction model.

[0191] Step S4 includes:

[0192] S4.1, input the weighted comprehensive feature vector in time step order into the LSTM network;

[0193] S4.2, the LSTM network generates a vortex prediction result for a future time point according to the input time sequence features; wherein the prediction result includes identifying and predicting the vortex region formed around the parachutist, evaluating the intensity of the vortex and the influence on the parachutist.

[0194] By using the LSTM network for time series analysis, the memory unit and the gating mechanism of the LSTM network can process long sequence data and capture long-term dependencies between features, thereby improving the accuracy and reliability of the prediction.

[0195] Through the identification of the vortex region, the evaluation of the intensity and the prediction of the influence on the parachutist, real-time environmental information is provided for the parachutist, and a basis is provided for making the next safety decision.

[0196] Specifically, the training process of the LSTM network is as follows:

[0197] According to the complexity of the data and the demand of the prediction task, the number of layers and the number of neurons of each layer of the LSTM network are designed.

[0198] The forgetting gate, input gate and output gate mechanism are introduced to enhance the network's ability to capture long-term dependencies.

[0199] Extract the key features related to the vortex around the parachutist, such as speed, pressure, temperature, and vortex intensity, and use these features as inputs to the LSTM network.

[0200] During the training process of the LSTM network, the formation, development and dissipation patterns of the vortex in the historical data are learned to establish an understanding of the dynamic changes of the vortex.

[0201] The backpropagation algorithm and the optimizer are used to adjust the network parameters to minimize the prediction error.

[0202] Another aspect of the embodiment of the application provides a real-time parachute state prediction system with dynamic adjustment, comprising:

[0203] The data acquisition module is configured to collect multiple flight state data in a real-time manner during the parachuting process through multiple sensors arranged on the parachutist, and wirelessly transmit the data to the data processing unit.

[0204] The feature extraction module is configured to extract key features from the preprocessed data by using a convolutional neural network, and generate a one-dimensional feature vector representing the state of the parachutist.

[0205] The dynamic attention adjustment module is configured to divide the parachuting task into multiple subtasks, each corresponding to a different environment or state, and perform feature weighting for each subtask based on initial weight distribution obtained from historical data, and adjust the weight distribution of each feature in real time based on changes in the environment or state features during the parachuting process, and generate a weighted comprehensive feature vector.

[0206] The time series analysis and prediction module is configured to perform time series analysis on the weighted comprehensive feature vector by using a long short-term memory neural network, analyze the relationship between different time steps, and generate a prediction result for a future time point based on the result of the time series analysis.

[0207] In another aspect of the embodiments of the present application, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program, wherein the computer program is executed by a processor to implement the method steps described above.

[0208] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A dynamically adjusted real-time parachute landing state prediction method, characterized in that, Includes the following steps: S1. Multiple sensors mounted on the skydiver collect various flight status data in real time during the skydive and wirelessly transmit this data to the data processing unit. S2. Preprocess the collected data, and use a convolutional neural network to extract key features from the preprocessed data to generate a one-dimensional feature vector representing the parachute environment or state. S3. The parachute jump task is divided into multiple sub-tasks, each corresponding to a different environment or state. Using the initial weight allocation obtained from historical data as a basis, features are weighted for each sub-task. Based on the changes in environmental or state features during the parachute jump, the weight allocation of each feature is adjusted in real time to generate a weighted comprehensive feature vector. The sub-task refers to multiple state prediction units divided according to the changes in environmental parameters and the state characteristics of the parachutist during the parachute jump. Each sub-task corresponds to different environmental conditions or attitude states and is used for targeted feature weighting and prediction. The division of sub-tasks is determined based on the parameter change rate and a set threshold. When the change rate of environmental parameters or attitude features exceeds the set threshold, a new sub-task is triggered to achieve adaptive prediction of different states by the model. S4. Use a long short-term memory neural network to perform time series analysis on the weighted comprehensive feature vector, analyze the relationship between different time steps, and generate prediction results for future time points.

2. The method for dynamically adjusting real-time parachute landing status prediction according to claim 1, characterized in that, Step S1 includes: The direction of airflow around the skydiver is obtained by calculating the difference between pressure sensors at different parts of the body. The ambient pressure during the skydive is monitored using an air pressure sensor. Temperature sensors are used to monitor the temperature during the skydive. The attitude angle sensor measures the skydiver's body posture, including tilt angle and pitch angle. The parachutist's descent speed is measured using a speed sensor.

3. The method for dynamically adjusting real-time parachute landing status prediction according to claim 1, characterized in that, Step S2 involves preprocessing the collected data, including: S2.1 Check if there are missing values ​​in the raw data collected from the sensors on the parachutist. If the missing data record is incomplete, discard the record and then detect and replace the outlier based on the rules of historical data. S2.2 Align data from different sensors to the same time reference by matching timestamps; S2.

3. For missing points in the data, use dynamic interpolation to fill them in; S2.4 Select key features from the sensor data, including attitude angle, speed, and wind speed. Construct these data features at each time point into a row, and as time progresses, form a two-dimensional matrix.

4. The method for dynamically adjusting real-time parachute landing status prediction according to claim 3, characterized in that, Step S2.3 includes: S2.3.1 Select the neighboring points g0, g1, g2, and g3 to the left and right of the missing point gt, and their corresponding time points, where g1 and g2 are the valid values ​​before and after gt, respectively. S2.3.

2. Initialize the weights of g1 and g2 to be the mean, calculated using the following formula: w1(t-1)=0.5, w2(t-1)=0.5; Where w1(t-1) and w2(t-1) are the initial weight values ​​of g1 and g2, respectively; S2.3.3 Calculate the change using the difference between the feature values ​​of adjacent points: Δg1=|g1-g0|, Δg2=|g2-g3|; S2.3.4 Calculate the weight increments of g1 and g2 based on the changes. The calculation method is as follows: Δw1=λbefore·Δg1, Δw2=λafter·Δg2; Wherein, λbefore and λafter are sensitivity parameters set based on historical experience; S2.3.5, Dynamically update the weights of g1 and g2: w1(t)=w1(t-1)+Δw1, w2(t)=w2(t-1)+Δw2; S2.3.6, Normalized weights: w1(t)=w1(t) / (w1(t)+w2(t)), w2(t)=w2(t) / (w1(t)+w2(t)); S2.3.

7. Based on the normalized weights, perform dynamic interpolation on the missing points gt. The calculation formula is as follows: gt = w1(t)•g1 + w2(t)•g2.

5. The method for dynamically adjusting real-time parachute landing status prediction according to claim 3, characterized in that, Step S2, the preprocessing of the collected data also includes: S2.5 Extract the minimum and maximum values ​​of each feature from the collected raw data; S2.

6. Scaling is applied to each data point, calculated as follows: x'=(x-min) / (max-min); Where x is the original data, min is the minimum value of the feature, max is the maximum value of the feature, and x' is the scaled data; S2.7 Validate the scaled dataset to ensure that the data value of each feature is within the range of [0,1].

6. The method for dynamically adjusting real-time parachute landing status prediction according to claim 1, characterized in that, Step S3 includes: S3.1 Divide the state prediction task during the parachute process into multiple subtasks f=[f1,f2,…,fn], each subtask corresponding to different environmental or state changes; S3.

2. Perform inner loop training on the subtask and obtain the general initial attention weights α0=[α1,α2,…,αn] based on historical data; S3.3 For features extracted by convolutional neural networks, feature weighting is performed based on the initial weight allocation obtained from historical data; S3.4 When environmental or state characteristics change during the parachute jump, the attention weight of the corresponding characteristics is dynamically adjusted in real time. S3.

5. Summarize and analyze the results of the inner loop of multiple subtasks through the outer loop, extract common information across tasks, and readjust the weights.

7. The method for dynamically adjusting real-time parachute landing status prediction according to claim 6, characterized in that, Step S3.4 includes: S3.4.1 When the airflow velocity changes drastically, calculate the change value Δf1 of the airflow characteristics; When the relative position and angle of the parachutist's head and limbs change drastically, calculate the change value Δf2 of the posture characteristics; When the vertical descent velocity or lateral drift velocity changes suddenly, calculate the change in velocity characteristic Δf3. When the parachutist's rotation rate suddenly changes, calculate the change in rotational characteristic Δf4; S3.4.

2. Normalize each change value using the following formula: ； Where max(Δfi) is the maximum change value of all features within the current time step; S3.4.3 Calculate the dynamic adjustment value Δαi for each feature based on the change value. The calculation formula is as follows: ； Where λi is the sensitivity parameter for the i-th feature; S3.4.

4. Based on the difference between the current feature f(t) and the historical feature f(t−1), the attention weight is dynamically adjusted. The calculation formula is as follows: ； Where αi(t−1) is the attention weight of the i-th feature at the previous time step; S3.4.5 After dynamically adjusting the attention weight of each feature, multiply all features by their respective dynamic weights and then sum them up to generate a weighted feature vector f. The calculation formula is as follows: 。 8. The method for dynamically adjusting real-time parachute landing status prediction according to claim 6, characterized in that, Step S3.5 includes: S3.5.

1. Based on the characteristic change value of each subtask, determine whether there are tasks that advance or retreat together. If the change values ​​of two tasks are always the same positive or negative, then they belong to tasks that advance or retreat together. S3.5.2 For subtasks identified as advancing or retreating together, a reduction strategy is applied to decrease the weight difference between these subtasks, achieved through the following formula: ； Where αi(t) and αj(t) are two tasks that advance or retreat together; β is the adjustment ratio, which is adjustable and is initially set to 0.2; S3.5.

3. Normalize the adjusted weights so that the sum of the weights of all subtasks is 1. The normalization calculation formula is as follows: ； S3.5.4 After each time step is adjusted, proceed to the next time step t+1 to re-detect feature changes and make adjustments.

9. The method for dynamically adjusting real-time parachute landing status prediction according to claim 1, characterized in that, Step S4 includes: S4.1 Input the weighted composite feature vector into the LSTM network in time step order; S4.2 The LSTM network generates eddy current prediction results for future time points based on the input time series characteristics; the prediction results include identifying and predicting the eddy current region formed around the skydiver, assessing the intensity of the eddy current and its impact on the skydiver.

10. A dynamically adjustable real-time parachute landing status prediction system, characterized in that, include: The data acquisition module is used to collect multiple flight status data in real time during the skydiving process through multiple sensors deployed on the skydiver's body, and wirelessly transmit this data to the data processing unit. The feature extraction module uses a convolutional neural network to extract key features from the preprocessed data and generate a one-dimensional feature vector representing the parachutist's state. The dynamic attention adjustment module divides the parachute mission into multiple sub-tasks, each corresponding to a different environment or state. Based on the initial weight allocation obtained from historical data, it weights the features of each sub-task and adjusts the weight allocation of each feature in real time based on the changes in environmental or state features during the parachute process, generating a weighted comprehensive feature vector. The time series analysis and prediction module uses a long short-term memory neural network to perform time series analysis on the weighted comprehensive feature vector, analyzes the relationship between different time steps, and generates prediction results for future time points based on the results of the time series analysis.

Images