Self-optimization data acquisition method and device based on environmental state perception

Abstract

The invention discloses a self-optimization data acquisition method and device based on environmental state perception. In the scheme, a plurality of candidate sampling frequencies are determined based on current environment state information; screening the plurality of candidate sampling frequencies based on the current environment state information to obtain a target sampling frequency; and performing data acquisition by using the target sampling frequency to obtain a data acquisition result. Compared with the prior art in which fixed sampling frequency is adopted, data redundancy or key information loss is easily caused and the data quality of a data acquisition result is influenced under the condition of frequent environmental fluctuation or sudden event, and a threshold triggering mechanism is used for performing data acquisition, so that the sampling rate is increased only during data sudden change, progressive data change is easily ignored, and the data quality is influenced. And once the triggering condition is met, a large number of similar data points may be generated, the density of effective data of the data acquisition result is reduced, and the data quality of the data acquisition result is influenced, so that the method has obvious advantages.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data collection, in particular to a self-optimizing data collection method and device based on environmental state perception. BACKGROUND

[0002] With the rapid development of the Internet of Things, intelligent sensors and environmental monitoring, data collection technology has become a key link to realize intelligent perception and monitoring, mainly through data collection equipment to obtain environmental information.

[0003] Most of the existing data collection equipment uses a fixed sampling frequency, which is simple to implement, but in the case of frequent environmental fluctuations or event bursts, it is easy to cause data redundancy or loss of critical information, affecting the data quality of the data collection results. Although some existing solutions attempt to collect data through a threshold trigger mechanism, by setting one or more thresholds to determine when to increase the sampling rate to capture important data changes, this mechanism only increases the sampling rate when the data mutates, which can easily ignore gradual data changes, and once the trigger condition is met, it can generate a large number of similar data points, reducing the density of effective data in the data collection results, thereby affecting the data quality of the data collection results.

[0004] How to improve the data quality of the data collection results is a technical problem to be solved. SUMMARY

[0005] Based on the above problems, the present application provides a self-optimizing data collection method and device based on environmental state perception, aiming to improve the data quality of the data collection results.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] The first aspect of the present application provides a self-optimizing data collection method based on environmental state perception, which comprises:

[0008] determining a plurality of candidate sampling frequencies based on current environmental state information;

[0009] selecting the plurality of candidate sampling frequencies based on the current environmental state information to obtain a target sampling frequency;

[0010] collecting data using the target sampling frequency to obtain a data collection result.

[0011] Optionally, the method further comprises:

[0012] constructing time series data based on the current environmental state information;

[0013] predict a sampling frequency at a next moment based on the time series data, to obtain a sampling frequency prediction result;

[0014] determine the plurality of candidate sampling frequencies according to a preset rule based on the sampling frequency prediction result.

[0015] Optionally, the prediction of the sampling frequency at the next moment based on the time series data, to obtain the sampling frequency prediction result, comprises:

[0016] inputting the time series data into a long short-term memory network, the long short-term memory network predicting the sampling frequency at the next moment to obtain the sampling frequency prediction result; the long short-term memory network being obtained after training.

[0017] Optionally, the filtering of the plurality of candidate sampling frequencies based on the current environment state information, to obtain the target sampling frequency, comprises:

[0018] constructing an environment state vector based on the current environment state information;

[0019] inputting the environment state vector into a reinforcement learning model, the reinforcement learning model filtering the plurality of candidate sampling frequencies according to a preset reward function to obtain the target sampling frequency; the reinforcement learning model being configured to calculate a reward function value of each candidate sampling frequency based on the environment state vector, and sort the plurality of candidate sampling frequencies based on the reward function value to obtain the target sampling frequency; the reinforcement learning model being obtained after training; the preset reward function being set according to power consumption and data error of a data acquisition device.

[0020] Optionally, the reinforcement learning model comprises a deep Q network, and the inputting of the environment state vector into the reinforcement learning model, the reinforcement learning model filtering the plurality of candidate sampling frequencies according to the preset reward function to obtain the target sampling frequency, comprises:

[0021] inputting the environment state vector into the deep Q network, the deep Q network filtering the plurality of candidate sampling frequencies according to the preset reward function to obtain the target sampling frequency; the deep Q network being configured to calculate a Q value of each candidate sampling frequency based on the environment state vector, and sort the plurality of candidate sampling frequencies based on the Q value to obtain the target sampling frequency.

[0022] Optionally, after the data acquisition result is obtained by using the target sampling frequency, the method further comprises:

[0023] obtaining a power consumption value and an error value corresponding to the data acquisition result;

[0024] based on the power consumption value and the error value, a reward function value is calculated;

[0025] based on the reward function value, parameters of the long short-term memory network and parameters of the reinforcement learning model are updated.

[0026] The second aspect of the present application provides a self-optimizing data acquisition device based on environment state perception, which comprises:

[0027] A candidate sampling frequency determination module is configured to determine a plurality of candidate sampling frequencies based on current environment state information.

[0028] A target sampling frequency acquisition module is configured to filter the plurality of candidate sampling frequencies based on the current environment state information to obtain a target sampling frequency.

[0029] A data acquisition module is configured to perform data acquisition using the target sampling frequency to obtain a data acquisition result.

[0030] Optionally, the candidate sampling frequency determination module is specifically configured to:

[0031] construct time series data based on the current environment state information;

[0032] predict a sampling frequency at a next time based on the time series data to obtain a sampling frequency prediction result;

[0033] determine the plurality of candidate sampling frequencies according to a preset rule based on the sampling frequency prediction result.

[0034] The third aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and when the program is run by a processor, the self-optimizing data acquisition method based on environment state perception is realized.

[0035] The fourth aspect of the present application provides a processor for running a computer program, and the program performs the self-optimizing data acquisition method based on environment state perception when running.

[0036] Compared with the prior art, the present application has the following beneficial effects:

[0037] The self-optimization data collection method based on environment state perception provided in the application determines a plurality of candidate sampling frequencies based on current environment state information. The plurality of candidate sampling frequencies are screened based on the current environment state information to obtain a target sampling frequency. The sampling frequency is screened according to the environment state information, and accurate matching between the sampling frequency and the environment is achieved, so that the situation that the data collection result is distorted due to the use of too high or too low sampling frequency is effectively avoided. Data collection is performed by using the target sampling frequency to obtain a data collection result. Data collection is performed by using the target sampling frequency, which can reduce data redundancy, so that the data collection result has higher information density, thereby improving the data quality of the data collection result. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0039] Figure 1 A flowchart of a self-optimization data collection method based on environment state perception provided in an embodiment of the application;

[0040] Figure 2 A flowchart of another self-optimization data collection method based on environment state perception provided in an embodiment of the application;

[0041] Figure 3 A structural schematic diagram of a self-optimization data collection device based on environment state perception provided in an embodiment of the application. DETAILED DESCRIPTION

[0042] As described above, most of the current data collection devices use fixed sampling frequency, which is simple to implement, but in the case of frequent environmental fluctuations or sudden events, it is easy to cause data redundancy or loss of key information, affecting the data quality of the data collection result. Although some existing solutions attempt to collect data through a threshold triggering mechanism, by setting one or more thresholds to determine when to increase the sampling rate to capture important data changes, this mechanism only increases the sampling rate when the data mutates, which easily ignores gradual data changes, and once the triggering condition is met, a large number of similar data points may be generated, reducing the density of effective data in the data collection result, thereby affecting the data quality of the data collection result.

[0043] In view of the above problems, the application provides a self-optimizing data collection method and device based on environmental state perception. Based on current environmental state information, a plurality of candidate sampling frequencies are determined. Based on the current environmental state information, the plurality of candidate sampling frequencies are screened to obtain a target sampling frequency. Data collection is performed using the target sampling frequency to obtain a data collection result.

[0044] In order for those skilled in the art to better understand the scheme of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] Referring to Figure 1 , the figure is a flowchart of a self-optimizing data collection method based on environmental state perception provided by an embodiment of the present application. As Figure 1 indicated, the method comprises the following steps:

[0046] S101, based on current environmental state information, a plurality of candidate sampling frequencies are determined.

[0047] In an implementable embodiment:

[0048] Based on the current environmental state information, time series data is constructed.

[0049] Based on the time series data, the sampling frequency at the next moment is predicted to obtain a sampling frequency prediction result.

[0050] Based on the sampling frequency prediction result, the plurality of candidate sampling frequencies are determined according to a preset rule.

[0051] The environmental state information includes but is not limited to temperature, humidity, illumination, battery power, noise level, and time stamp, etc. The design method of the preset rule is not limited here, for example, it can be designed by using proportional offset method or discrete mapping method, etc., and it can also use signal fidelity and anti-aliasing rules, hardware capability constraint rules, or dynamic response adaptability rules, etc.

[0052] According to the preset rule, the sampling frequency prediction result is converted into a plurality of candidate sampling frequencies, which can avoid blind enumeration and improve the quality of the candidate sampling frequencies.

[0053] By constructing time series for prediction, the trend of environmental change can be effectively identified, the integrity of key data is ensured, and the quality of the candidate sampling frequencies is improved. The plurality of candidate sampling frequencies provided provide high-quality input for the subsequent screening process.

[0054] Optionally, the sampling frequency of the next moment is predicted based on the time series data, to obtain a sampling frequency prediction result, comprising:

[0055] The time series data is input into a long short-term memory network, which predicts the sampling frequency of the next moment to obtain the sampling frequency prediction result; the long short-term memory network is obtained after training.

[0056] The structure of the long short-term memory network is not limited here.

[0057] The long short-term memory network (Long-Term Dependencies, LSTM) models the time series corresponding to the environmental state information, effectively captures long-term dependencies and dynamic trends using its gating mechanism, actively adjusts rather than remedies after the fact, and can ensure that key data is not missed. And LSTM supports multivariate time series, realizes comprehensive decision-making of multi-dimensional perception, and improves the quality of candidate sampling frequencies.

[0058] S102, based on the current environmental state information, the plurality of candidate sampling frequencies are screened to obtain a target sampling frequency.

[0059] In an optional implementation:

[0060] Based on the current environmental state information, an environmental state vector is constructed;

[0061] The environmental state vector is input into a reinforcement learning model, and the reinforcement learning model screens the plurality of candidate sampling frequencies according to a preset reward function to obtain a target sampling frequency.

[0062] The reinforcement learning model is used to calculate the reward function value of each candidate sampling frequency based on the environmental state vector, and sort the plurality of candidate sampling frequencies based on the reward function value to obtain the target sampling frequency; the reinforcement learning model is obtained after training; and the preset reward function is set according to the power consumption and data error of the data acquisition device.

[0063] The environmental state vector can provide context awareness information, and screening using context awareness information is beneficial to decision-making in complex environments.

[0064] The reinforcement learning model includes a model based on a deep Q network (Deep Q-Network, DQN) or proximal policy optimization (Proximal Policy Optimization, PPO).

[0065] The reinforcement learning model takes each candidate frequency as an action, evaluates the long-term benefits of each action according to the current environment state, and outputs an optimal action as the target sampling frequency. By using the reinforcement learning model for screening, the optimal trade-off between power consumption and data quality is achieved, taking into account energy saving and fidelity, and realizing the intelligentization, self-adaptation and optimization of the data acquisition process.

[0066] Optionally, the reinforcement learning model includes a deep Q network, the environment state vector is input into the reinforcement learning model, the reinforcement learning model screens the plurality of candidate sampling frequencies according to a preset reward function to obtain the target sampling frequency, and the method comprises the following steps:

[0067] The environment state vector is input into the deep Q network, and the deep Q network screens the plurality of candidate sampling frequencies according to the preset reward function to obtain the target sampling frequency.

[0068] The deep Q network is used to calculate the Q value of each candidate sampling frequency based on the environment state vector, and sort the plurality of candidate sampling frequencies based on the Q value to obtain the target sampling frequency.

[0069] The deep Q network combines a deep neural network with Q-learning, and can learn an optimal strategy in a high-dimensional state space. The deep Q network adapts to a discrete action space, and is directly efficient in decision-making for discrete values such as candidate sampling rates. Moreover, this model can be deployed on edge devices to meet the low-power and real-time requirements of the data acquisition process.

[0070] S103, data acquisition is performed using the target sampling frequency to obtain a data acquisition result.

[0071] In an implementable embodiment, after data acquisition is performed using the target sampling frequency to obtain a data acquisition result, the method further comprises the following steps:

[0072] An error value and a power consumption value corresponding to the data acquisition result are obtained.

[0073] Based on the power consumption value and the error value, a reward function value is calculated.

[0074] Based on the reward function value, the parameters of the long short-term memory network and the parameters of the reinforcement learning model are updated.

[0075] The embodiment of the application provides a self-optimizing data collection method based on environment state perception, determines a plurality of candidate sampling frequencies based on current environment state information. The plurality of candidate sampling frequencies are screened based on the current environment state information to obtain a target sampling frequency. The sampling frequency is screened according to the environment state information, accurate matching between the sampling frequency and the environment is realized, and the situation that the data collection result is distorted due to too high or too low sampling frequency is effectively avoided. Data collection is performed by using the target sampling frequency to obtain a data collection result. The data collection is performed by using the target sampling frequency, data redundancy is reduced, the data collection result has higher information density, and therefore the data quality of the data collection result is improved.

[0076] In order to further improve the self-optimizing data collection method based on environment state perception described in the above embodiment, on the basis of the above embodiment, the step of updating the parameters of the long short-term memory network and the parameters of the reinforcement learning model based on the data collection result is added.

[0077] Referring to Figure 2 , the figure is a flowchart of another self-optimizing data collection method based on environment state perception provided by the embodiment of the application. As Figure 2 shown, the method comprises the following steps:

[0078] S201, based on current environment state information, a plurality of candidate sampling frequencies are determined by using a long short-term memory network.

[0079] Exemplarily, based on the current environment state information, temperature data in the past 10 minutes is obtained, based on the temperature data in the past 10 minutes, a temperature data sequence :

[0080] ;

[0081] Wherein, the sequence length of the temperature data sequence is 10, and the 10 data points from t-9 to the current time t are included.

[0082] The temperature data sequence is input into the long short-term memory network, the number of layers of the long short-term memory network is 2, and the number of hidden units is 64. The long short-term memory network outputs a hidden state :

[0083] ;

[0084] Wherein, includes weight parameters and bias parameters of the LSTM and the like.

[0085] The predicted sampling frequency of the next time point is output by a mapping function : :​

[0086] ;

[0087] where W is the weight matrix, b is the bias term, is the predicted next time sampling frequency.

[0088] The mean square error (MSE) is used to measure the deviation between the predicted sampling frequency and the actual optimal sampling frequency:

[0089] ;

[0090] where, is the actual optimal sampling frequency, is the LSTM predicted sampling frequency, and N is the number of samples.

[0091] LSTM can capture long-term environmental trends, remember key events, and predict information density peaks in advance. By modeling the environmental state information with LSTM, the quality of the candidate sampling frequency is improved.

[0092] S202, based on the current environmental state information, using a reinforcement learning model to filter the plurality of candidate sampling frequencies, to obtain a target sampling frequency.

[0093] For example, the plurality of candidate sampling frequencies are 1Hz, 5Hz and 10Hz, and the environmental fluctuation degree is calculated by the following formula:

[0094] ;

[0095] where, is the state variable at the current time t; is the standard deviation, which measures the environmental volatility; is the i-th data point, all average value; n is the number of data points in the sliding window.

[0096] The agent decides the sampling frequency to be selected next according to the current environmental fluctuation. For example, a high frequency is selected for sampling when the fluctuation is large, and a low frequency is selected for sampling when the fluctuation is small.

[0097] The action space is obtained from the plurality of candidate sampling frequencies The reward function is calculated by the following formula:

[0098] ;

[0099] where, is the sampling frequency corresponding power consumption, is Weighting coefficients; Sampling frequency The resulting prediction error yes The weighting coefficients. If Indicating high-frequency sampling, such as 10Hz, results in higher power consumption but smaller prediction errors; if Indicating low-frequency sampling, such as 1Hz, results in larger prediction errors but lower power consumption.

[0100] In a dynamically changing environment, reinforcement learning models take into account both power consumption and accuracy, and automatically select the optimal sampling frequency to achieve a balance between low power consumption and high data quality.

[0101] S203. Data is collected using the target sampling frequency to obtain the data collection results.

[0102] Sampling according to the target sampling frequency enables on-demand sampling, which reduces data redundancy and results in higher information density in the data collection results, thereby improving the data quality of the data collection results.

[0103] S204. Based on the data acquisition results, update the parameters of the Long Short-Term Memory Network and the parameters of the Reinforcement Learning Model.

[0104] In one feasible implementation:

[0105] Obtain the power consumption value and error value corresponding to the data acquisition results.

[0106] The reward function value is calculated based on the power consumption value and the error value.

[0107] Based on the reward function value, the parameters of the Long Short-Term Memory network and the parameters of the reinforcement learning model are updated.

[0108] For example, the Q table is updated using the following formula:

[0109] ;

[0110] in, It is in state Take action below The expected cumulative reward, It's the learning rate. It is the reward obtained at the current moment. A discount factor is used to measure the importance of future rewards. It's a new state The maximum Q value in.

[0111] Based on the data acquisition result, the parameters of the long short-term memory network and the parameters of the reinforcement learning model are updated, the long short-term memory network can more accurately capture the trend of environmental changes, reduce prediction bias, and improve the quality of candidate sampling frequency. The reinforcement learning model can learn a more optimal power consumption and precision trade-off strategy to continuously improve and dynamically optimize overall performance.

[0112] Another self-optimizing data acquisition method based on environmental state perception is provided in the embodiments of the present application. Based on current environmental state information, a long short-term memory network is used to determine a plurality of candidate sampling frequencies. The long short-term memory network is used to model the environmental state information to determine the candidate sampling frequencies, thereby improving the quality of the candidate sampling frequencies. Based on the current environmental state information, a reinforcement learning model is used to screen the plurality of candidate sampling frequencies to obtain a target sampling frequency. The reinforcement learning model is used to screen the sampling frequency, thereby achieving accurate matching of the sampling frequency and the environment and effectively avoiding the situation that using too high or too low sampling frequency causes distortion of the data acquisition result. The target sampling frequency is used for data acquisition to obtain a data acquisition result. Using the target sampling frequency for data acquisition can reduce data redundancy, so that the data acquisition result has higher information density, thereby improving the data quality of the data acquisition result. Based on the data acquisition result, the parameters of the long short-term memory network and the parameters of the reinforcement learning model are updated, the long short-term memory network can more accurately capture the trend of environmental changes, reduce prediction bias, and improve the quality of candidate sampling frequency. The reinforcement learning model can learn a more optimal power consumption and precision trade-off strategy to continuously improve and dynamically optimize overall performance.

[0113] Based on the data acquisition method introduced in the foregoing embodiments, the present application also provides a self-optimizing data acquisition device based on environmental state perception. Figure 3 The structure of the device is shown in FIG. 1. As shown in FIG. 1, the data acquisition device includes: Figure 3

[0114] A candidate sampling frequency determination module 301 is configured to determine a plurality of candidate sampling frequencies based on current environmental state information.

[0115] A target sampling frequency acquisition module 302 is configured to screen the plurality of candidate sampling frequencies based on the current environmental state information to obtain a target sampling frequency.

[0116] A data acquisition module 303 is configured to use the target sampling frequency to perform data acquisition to obtain a data acquisition result.

[0117] Optionally, the candidate sampling frequency determination module is specifically configured to:

[0118] Based on the current environmental state information, time series data is constructed.​

[0119] predict a sampling frequency of a next moment based on the time series data, to obtain a sampling frequency prediction result;

[0120] determine the plurality of candidate sampling frequencies according to a preset rule based on the sampling frequency prediction result.

[0121] Optionally, the sampling frequency prediction based on the time series data comprises:

[0122] inputting the time series data into a long short-term memory network, the long short-term memory network being configured to predict the sampling frequency of the next moment to obtain the sampling frequency prediction result; the long short-term memory network being trained.

[0123] Optionally, the target sampling frequency obtaining module is specifically configured to:

[0124] construct an environment state vector based on the current environment state information;

[0125] input the environment state vector into a reinforcement learning model, the reinforcement learning model being configured to filter the plurality of candidate sampling frequencies according to a preset reward function to obtain the target sampling frequency; the reinforcement learning model being configured to calculate a reward function value of each candidate sampling frequency based on the environment state vector, and sort the plurality of candidate sampling frequencies based on the reward function value to obtain the target sampling frequency; the reinforcement learning model being trained; the preset reward function being set according to power consumption and data error of the data acquisition device.

[0126] Optionally, the reinforcement learning model comprises a deep Q network, and the inputting of the environment state vector into the reinforcement learning model comprises:

[0127] inputting the environment state vector into the deep Q network, the deep Q network being configured to filter the plurality of candidate sampling frequencies according to the preset reward function to obtain the target sampling frequency; the deep Q network being configured to calculate a Q value of each candidate sampling frequency based on the environment state vector, and sort the plurality of candidate sampling frequencies based on the Q value to obtain the target sampling frequency.

[0128] Optionally, the apparatus further comprises a feedback module.

[0129] The feedback module is configured to obtain a power consumption value and an error value corresponding to the data acquisition result.

[0130] Based on the power consumption value and the error value, a reward function value is calculated;

[0131] Based on the reward function value, parameters of the long short-term memory network and parameters of the reinforcement learning model are updated.

[0132] In addition, the embodiment of the present application further provides a computer readable storage medium, the computer readable storage medium stores a computer program, when the program is run by a processor, a self-optimizing data acquisition method based on environment state perception as any manner of the method embodiment is realized.

[0133] In addition, the embodiment of the present application further provides a processor, the processor is used for running a computer program, the program is run and executes the self-optimizing data acquisition method based on environment state perception as any manner of the preceding method embodiment is introduced.

[0134] It should be noted that each embodiment in the specification adopts a progressive manner for description, and the same and similar parts between each embodiment can be referred to each other, and each embodiment focuses on the different places from other embodiments. Especially, for the device embodiment, since it is basically similar to the method embodiment, it is described more simply, and the related parts can be referred to the part of the method embodiment. The device embodiment described above is only schematic, and the units described as separate components can be or can not be physically separated, and the components prompted as units can be or can not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. According to the actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment scheme. Those skilled in the art can understand and implement without creative labor.

[0135] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this, any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be limited by the protection scope of the claims.

Claims

1. A self-optimizing data acquisition method based on environmental state perception, characterized in that, include: Based on the current environmental status information, multiple candidate sampling frequencies are determined; Based on the current environmental state information, the multiple candidate sampling frequencies are filtered to obtain the target sampling frequency; Data is collected using the target sampling frequency to obtain the data collection results.

2. The method according to claim 1, characterized in that, The determination of multiple candidate sampling frequencies based on the current environmental state information includes: Based on the current environmental state information, time series data is constructed; Based on the time series data, the sampling frequency at the next moment is predicted to obtain the sampling frequency prediction result; Based on the sampling frequency prediction results, the multiple candidate sampling frequencies are determined according to preset rules.

3. The method according to claim 2, characterized in that, The step of predicting the sampling frequency for the next time step based on the time series data to obtain the sampling frequency prediction result includes: The time series data is input into a Long Short-Term Memory (LSTM) network, which predicts the sampling frequency for the next time step, thus obtaining the sampling frequency prediction result; the LSM network is obtained after training.

4. The method according to claim 1, characterized in that, The step of filtering the multiple candidate sampling frequencies based on the current environmental state information to obtain the target sampling frequency includes: Based on the current environmental state information, an environmental state vector is constructed; The environmental state vector is input into a reinforcement learning model, which filters the multiple candidate sampling frequencies according to a preset reward function to obtain the target sampling frequency. The reinforcement learning model is used to calculate the reward function value of each candidate sampling frequency based on the environmental state vector, and sort the multiple candidate sampling frequencies based on the reward function values ​​to obtain the target sampling frequency. The reinforcement learning model is obtained after training. The preset reward function is set according to the power consumption and data error of the data acquisition device.

5. The method according to claim 4, characterized in that, The reinforcement learning model includes a deep Q-network. The environment state vector is input into the reinforcement learning model, and the model filters the multiple candidate sampling frequencies according to a preset reward function to obtain the target sampling frequency, including: The environmental state vector is input into a deep Q-network, which filters the multiple candidate sampling frequencies according to the preset reward function to obtain the target sampling frequency. The deep Q-network is used to calculate the Q value of each candidate sampling frequency based on the environmental state vector, and sort the multiple candidate sampling frequencies based on the Q values ​​to obtain the target sampling frequency.

6. The method according to claims 3 and 4, characterized in that, After acquiring data using the target sampling frequency and obtaining the data acquisition results, the method further includes: Obtain the power consumption value and error value corresponding to the data acquisition results; The reward function value is calculated based on the power consumption value and the error value; Based on the reward function value, the parameters of the Long Short-Term Memory network and the parameters of the reinforcement learning model are updated.

7. A self-optimizing data acquisition device based on environmental state perception, characterized in that, include: The candidate sampling frequency determination module is used to determine multiple candidate sampling frequencies based on the current environmental state information. The target sampling frequency acquisition module is used to filter the multiple candidate sampling frequencies based on the current environmental state information to obtain the target sampling frequency; The data acquisition module is used to acquire data using the target sampling frequency and obtain the data acquisition results.

8. The apparatus according to claim 7, characterized in that, The candidate sampling frequency determination module is specifically used for: Based on the current environmental state information, time series data is constructed; Based on the time series data, the sampling frequency at the next moment is predicted to obtain the sampling frequency prediction result; Based on the sampling frequency prediction results, the multiple candidate sampling frequencies are determined according to preset rules.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the self-optimizing data acquisition method based on environmental state awareness as described in any one of claims 1-6.

10. A processor, characterized in that, Used to run a computer program, which, when running, executes the self-optimizing data acquisition method based on environmental state awareness as described in any one of claims 1-6.

Images