Data acquisition method and system based on semi-Markov process

Abstract

The invention provides a data acquisition method and system based on a semi-Markov process, and the method comprises the steps: defining a discrete state space which comprises a set of several data acquisition states set by edge equipment and a set of state conversion events allowed among the data acquisition states; creating a transition probability matrix, wherein matrix elements in the transition probability matrix represent the state transition probability between the data acquisition states; performing data acquisition according to the current data acquisition state to obtain real-time measurement data, and calculating a state transition probability based on the real-time measurement data; updating a transition probability matrix based on the calculated state transition probability; screening the allowed state transition events according to the updated transition probability matrix to obtain a screening result; and executing the state conversion event in the screening result to realize the switching of the corresponding data acquisition state. On the premise that low energy consumption is met, low cost and high intelligence are achieved to the maximum extent.

Description

Technical Field

[0001] The present invention relates to the technical fields of the Internet of Things and device monitoring, and in particular, to a data acquisition method and system based on a semi-Markov process. Background Art

[0002] With the rise of the concept of smart cities, a large number of edge devices are applied to the status monitoring and data acquisition tasks of public infrastructure such as water, electricity, and gas. In the urban gas pipeline transmission and distribution scenario, edge devices with explosion-proof requirements need to be powered by internal batteries. This power supply mode poses a severe challenge to the energy management of edge devices.

[0003] More specifically, for urban gas pipelines buried underground, gas companies will set inspection wells on the ground in accordance with relevant technical specifications. Gas companies can install data monitoring and management terminal units (edge devices) underground to monitor multiple parameters such as methane concentration, temperature, humidity, and water level underground. For the sake of energy conservation, traditional underground data monitoring and management terminal units will use a relatively low frequency (for example, once every 15 minutes) for data acquisition. The anti-interference ability of edge devices is poor at this acquisition frequency: when the acquisition interval is set to 15 minutes, one piece of abnormal data will contaminate a 30-minute time zone. If the contaminated data is directly applied to intelligent algorithms such as machine learning (for example, monitoring methane leakage events), the performance of the algorithm will be significantly reduced, resulting in a large number of false alarms or missed alarms. If a higher acquisition frequency is set (for example, once every 15 seconds), the existing battery technology will limit the battery life of edge devices, resulting in frequent battery replacement and increased device maintenance costs.

[0004] Low energy consumption, low cost, and high intelligence, these three elements constitute an impossible triangle for edge devices (indicating that the three elements cannot be satisfied simultaneously under the existing technology). For edge devices in the urban gas pipeline transmission and distribution scenario, how to maximize low cost and high intelligence on the premise of meeting low energy consumption is the main technical problem in this field. Summary of the Invention

[0005] Aiming at the problems existing in the prior art, the present application provides a data acquisition method and system based on a semi-Markov process, aiming to provide an adaptive data acquisition strategy for edge devices with low energy consumption.

[0006] A data acquisition method based on a semi-Markov process, which is applied to the underground data acquisition of edge devices, includes:

[0007] Step A1, defining a discrete state space, including a set of various data acquisition states set by the edge device and a set of allowed state transition events between the data acquisition states;

[0008] Step A2, create a transition probability matrix, where the matrix elements in the transition probability matrix represent the state transition probabilities between data collection states;

[0009] Step A3, perform data collection based on the currently relied-upon data collection state to obtain real-time measurement data, and calculate the state transition probability based on the real-time measurement data;

[0010] Step A4, update the transition probability matrix based on the state transition probability calculated in Step A3;

[0011] Step A5, screen the allowed state transition events according to the updated transition probability matrix to obtain a screening result;

[0012] Step A6, execute the state transition events in the screening result to achieve the switching of the corresponding data collection state, and use the converted data collection state as the currently relied-upon data collection state to continue Step A3.

[0013] Further, in Step A1, the edge device sets three data collection states: normal state, wake-up state, and emergency alarm state;

[0014] The sampling frequency in the normal state is less than the sampling frequency in the wake-up state, and the sampling frequency in the wake-up state is less than the sampling frequency in the emergency alarm state.

[0015] Further, the set of allowed state transition events includes:

[0016] The state transition event from the normal state to the wake-up state;

[0017] The state transition event from the wake-up state to the emergency alarm state;

[0018] The state transition event from the wake-up state to the normal state;

[0019] The state transition event from the emergency alarm state to the wake-up state.

[0020] Further, Step A3 includes:

[0021] Step A31, create a historical state transition event data set;

[0022] Step A32, according to the historical state transition event data set, determine the value of the standard deviation parameter in the probability density function by statistical methods;

[0023] Step A33, integrate the probability density function with the determined value of the standard deviation parameter to obtain the cumulative distribution function of the normal distribution;

[0024] Step A34, calculate the state transition probability based on the cumulative distribution function of the normal distribution.

[0025] Furthermore, in step A3, the soft trigger threshold in the probability density function is further optimized, and the optimization process includes:

[0026] Step B31, define a comprehensive loss function, which is a weighted sum function of the probability of state transition error, the probability that a state should transition but does not, and the expected time delay between state transitions;

[0027] Step B32, minimize the comprehensive loss function based on a preset optimization algorithm, and obtain the minimized threshold of the minimized comprehensive loss function as the soft trigger threshold in the probability density function.

[0028] Furthermore, in step A5, for each state transition probability corresponding to an allowed state transition event in the transition probability matrix, it is judged whether it is greater than the corresponding probability threshold, and the state transition events with state transition probabilities greater than the corresponding probability thresholds are used as the screening results.

[0029] Furthermore, step A5 includes:

[0030] Step A51, for each state transition probability corresponding to an allowed state transition event in the transition probability matrix, judge whether it is greater than the corresponding probability threshold;

[0031] Step A52, screen out the state transition events with state transition probabilities greater than the corresponding probability thresholds;

[0032] Step A53, extract the corresponding additional trigger conditions according to the screened state transition events;

[0033] Step A54, when the real-time measurement data meets the additional trigger conditions, use the screened state transition events as the screening results.

[0034] Furthermore, in step A6, the state transition events in the screening results are only executed after the state residence time exceeds a preset minimum residence time.

[0035] A data acquisition system based on a semi-Markov process, integrated in an edge device, for executing the foregoing data acquisition method based on a semi-Markov process, includes:

[0036] A state space definition module, used to define a discrete state space, including a set of multiple data acquisition states set by the edge device and a set of allowed state transition events between data acquisition states;

[0037] A probability matrix creation module, connected to the state space definition module, used to create a transition probability matrix, and the matrix elements in the transition probability matrix represent the state transition probabilities between data acquisition states;

[0038] A conversion probability calculation module, configured to perform data acquisition according to the current data acquisition status to obtain real-time measurement data, and calculate the state conversion probability based on the real-time measurement data;

[0039] A probability matrix update module, respectively connected to the probability matrix creation module and the conversion probability calculation module, configured to update the conversion probability matrix based on the calculated state conversion probability;

[0040] A conversion event screening module, connected to the probability matrix update module, configured to screen the allowed state conversion events according to the updated conversion probability matrix to obtain a screening result;

[0041] A conversion event execution module, connected to the conversion probability calculation module and the conversion event screening module, configured to execute the state conversion events in the screening result to implement the switching of the corresponding data acquisition status. After that, the conversion probability calculation module uses the switched data acquisition status as the current data acquisition status for data acquisition.

[0042] The beneficial technical effects of the present invention are as follows: A multi-data acquisition status switching method based on a semi-Markov process of the present invention provides a low-power consumption edge device adaptive data acquisition strategy, dynamically responds to sensor readings for status switching, maximizes low cost and high intelligence on the premise of meeting low power consumption, and has strong robustness and adaptability. Description of the Drawings

[0043] Figures 1-4 It is a step flowchart of a data acquisition method based on a semi-Markov process of the present invention;

[0044] Figure 5 It is a module schematic diagram of a data acquisition system based on a semi-Markov process of the present invention. Detailed Embodiments

[0045] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0047] Next, the present invention will be further described in conjunction with the drawings and specific embodiments, but it is not a limitation of the present invention.

[0048] See Figure 1, the present invention provides a data acquisition method based on a semi - Markov process, which is applied to downhole data acquisition of edge devices. It is characterized by including:

[0049] Step A1, define a discrete state space, including a set of various data acquisition states set by the edge device and a set of allowed state transition events between data acquisition states;

[0050] Step A2, create a transition probability matrix, where the matrix elements in the transition probability matrix represent the state transition probabilities between data acquisition states;

[0051] Step A3, perform data acquisition according to the currently based data acquisition state to obtain real - time measurement data, and calculate the state transition probability based on the real - time measurement data;

[0052] Step A4, update the transition probability matrix based on the state transition probability calculated in Step A3;

[0053] Step A5, screen the allowed state transition events according to the updated transition probability matrix to obtain a screening result;

[0054] Step A6, execute the state transition events in the screening result to realize the conversion of the corresponding data acquisition state, and use the converted data acquisition state as the currently based data acquisition state to continue Step A3.

[0055] If there are allowed state transition events in the screening result, then perform state transition according to the allowed state transition events in the screening result in Step A6. If there are no allowed state transition events in the screening result, it means to maintain the current state without conversion.

[0056] A multi - data acquisition state switching method based on a semi - Markov process of the present invention provides a low - energy - consumption edge device adaptive data acquisition strategy, dynamically responds to sensor readings for state switching, maximizes low - cost and high - intelligence under the premise of meeting low - energy - consumption, and has strong robustness and adaptability.

[0057] Further, in Step A1, the edge device sets three data acquisition states: normal state, wake - up state, and emergency alarm state;

[0058] The sampling frequency in the normal state is less than the sampling frequency in the wake - up state, and the sampling frequency in the wake - up state is less than the sampling frequency in the emergency alarm state. Step A1, define a discrete state space, define a set S of three types of data acquisition states:

[0059] S = {S1, S2, S3}

[0060] S1 represents the normal state, with a corresponding sampling frequency of f1. Preferably, f1 is, for example, 15 minutes / frame.

[0061] S2 represents the wake-up state, with a corresponding sampling frequency of f2. Preferably, f2 is, for example, 20 seconds / frame.

[0062] S3 represents the emergency alarm state, with a corresponding sampling frequency of f3. Preferably, f3 is, for example, 10 seconds / frame.

[0063] Furthermore, the set of allowed state transition events includes:

[0064] The state transition event from the normal state to the wake-up state;

[0065] The state transition event from the wake-up state to the emergency alarm state;

[0066] The state transition event from the wake-up state to the normal state;

[0067] The state transition event from the emergency alarm state to the wake-up state.

[0068] The set ε of allowed state transition events includes four types of allowed state transition events:

[0069] ε = {E 12 , E 23 , E 21 , E 32}

[0070] where E ij = S i → S j , indicating a transition from the data acquisition state S i to the data acquisition state S j .

[0071] In step S2, a probabilistic framework is used to define a soft trigger, rather than a deterministic transition based on a strict threshold, and a transition probability matrix P is defined:

[0072]

[0073] where P ij represents the probability of the data acquisition state transitioning from S i to state S j . The transition probability is derived from the cumulative distribution functions (CDFs) to represent a gradual change. The magnitude of the transition probability depends on the sensor measurement data x of the edge device.

[0074] Preferably, the sensor reading x is the LEL (Lower Explosion Limit) of the methane concentration sensor. The measurement data LEL (Lower Explosion Limit) of the methane concentration sensor refers to the lowest concentration of the combustible gas methane in the air, below which the gas will not explode.

[0075] In step A3, when the sensor measurement data x of the edge device approaches the threshold μ ij the probability P i of transitioning from state S j to state S ij increases.

[0076] P ij = F(x; μ ij , σ ij )

[0077] Specifically, F(x; μ, σ) is the cumulative distribution function of the normal distribution, obtained by integrating the probability density function (PDF).

[0078] Specifically, μ ij is the soft trigger threshold, which can be set manually or obtained by statistical or machine learning methods. σ ij is the standard deviation parameter used to capture measurement noise.

[0079] See Figure 2 , further, step A3 includes:

[0080] Step A31, create a historical state transition event dataset;

[0081] Step A32, based on the historical state transition event dataset, determine the value of the standard deviation parameter in the probability density function by statistical methods;

[0082] Step A33, integrate the probability density function with the determined value of the standard deviation parameter to obtain the cumulative distribution function of the normal distribution;

[0083] Step A34, calculate the state transition probability based on the cumulative distribution function of the normal distribution.

[0084] In step A31, create the historical state transition event dataset D ij :

[0085] D ij = {x(t)|s(t - Δt) = S i , S(t) = S j}

[0086] where x(t) is the function of the sensor measurement data x with respect to time t;

[0087] S(t) is the data acquisition status of the edge device system at time t.

[0088] Assume that the state transition event follows a Gaussian distribution. Then, the probability density function of the transition from state S i to state S j is expressed as follows:

[0089]

[0090] Based on the state transition event dataset D ij , the standard deviation parameter σ ij in the probability density function can be determined by statistical methods.

[0091] In step A4, the matrix elements in the transition probability matrix representing the disallowed state transition events are set to 0.

[0092] See Figure 3 . Further, in step A3, the soft trigger threshold in the probability density function is also optimized. The optimization process includes:

[0093] Step B31, define the comprehensive loss function, which is a weighted summation function of the probability of state transition error, the probability of a state that should have transitioned but did not, and the expected time delay between state transitions;

[0094] Step B32, minimize the comprehensive loss function based on a preset optimization algorithm, and obtain the minimized threshold of the minimized comprehensive loss function as the soft trigger threshold in the probability density function.

[0095] As a preferred embodiment of the present invention, in step A3, the soft trigger threshold μ ij is optimized through the set comprehensive loss function.

[0096] Specifically, for the soft trigger thresholds μ 12 and μ 23 , the defined comprehensive loss function L(μ 12 , μ 23 ) is:

[0097] L(μ 12 , μ 23 ) = αL false + βL missed + γL delay

[0098] where L false represents the probability of data acquisition state transition error, and is calculated by the following formula:

[0099]

[0100] Among them, L missed represents the probability that the data acquisition state conversion should be performed but has not been converted, and is calculated by the following formula:

[0101]

[0102] Among them, L delay represents the expected time delay for the conversion between data acquisition states, and is calculated by the following formula:

[0103] L delay = E[T 12 |x≥μ 12 + E[T 23 |x≥μ 23

[0104] E represents the symbol of mathematical expectation;

[0105] T 12 represents the time taken to convert from data acquisition state S1 to state S2;

[0106] T 23 represents the time taken to convert from data acquisition state S2 to state S3;

[0107] α, β, and γ are weight coefficients used to adjust the relative importance of the corresponding terms.

[0108] f 12 and f 23 are probability density functions.

[0109] A gradient-based optimization algorithm (such as the SGD algorithm, Adam algorithm), or a heuristic method (such as the GA algorithm, SA algorithm) can be used to minimize the comprehensive loss function L(μ 12 , μ 23 ), thereby solving and

[0110]

[0111] The values of the parameters to be optimized are updated through multiple rounds of iteration:

[0112]

[0113] Among them, η is the learning rate, and k represents the number of iterations.

[0114] ​The transition probabilities corresponding to the disallowed state transition events in the transition probability matrix are 0. For example, the transitions from the normal state to the emergency alarm state S1→S3 and from the emergency alarm state to the normal state S3→S1 are both disallowed state transitions. At any given time t, the dynamically calculated transition probability matrix is represented as follows:

[0115]

[0116] The state transition probability P ij is a function of the current x(t) and can be calculated based on the real-time measurement data x of the sensor:

[0117]

[0118]

[0119] P 21 (t) = 1 - F 12 (x(t))

[0120] P 32 (t) = 1 - F 23 (x(t))

[0121] To reduce the computational load and improve the real-time performance of the edge device, F 12 (x) and F 23 (x) can be pre-computed and stored as a table, and then the corresponding F 12 (x) and F 23 (x) can be obtained by looking up the table with the real-time measurement data x.

[0122] An effective interpolation method can be used to evaluate the values of the cumulative distribution functions (CDFs).

[0123] In step A3, after calculating the transition probabilities, in step A4, the transition probabilities are updated.

[0124] Furthermore, in step A5, it is judged one by one whether the state transition probabilities corresponding to the allowed state transition events in the transition probability matrix are greater than the corresponding probability thresholds, and the state transition events with state transition probabilities greater than the corresponding probability thresholds are used as the screening results.

[0125] In step A5, the probability threshold τ ij is used to determine whether to perform a state transition and what kind of state transition to perform. For each allowed state transition event E ij a corresponding probability threshold τ ij is set. Only when the actual transition probability P ij exceeds the corresponding probability threshold τ ij will a state transition occur, that is, it will be selected as the screening result.

[0126] For the set ε = {E 12 , E 23 , E 21 , E 32} of allowed state transition events, the following four determinations need to be performed:

[0127] If P 12 (t) > τ 12 , then S1 → S2;

[0128] If P 21 (t) > τ 21 , then S2 → S1;

[0129] If P 23 (t) > τ 23 , then S2 → S3;

[0130] If P 32 (t) > τ 32 , then S3 → S2;

[0131] The probability threshold τ ij ranges from 0 to 1 and can be set empirically.

[0132] For example, the probability threshold τ ij can be set to 0.8. If the actual conversion probability P ij is greater than 0.8, then the state transition event E ij is executed.

[0133] See Figure 4 , furthermore, step A5 includes:

[0134] Step A51, one by one, determine whether the state transition probability corresponding to the allowed state transition event in the conversion probability matrix is greater than the corresponding probability threshold;

[0135] Step A52, screen out the state transition events whose state transition probabilities are greater than the corresponding probability thresholds;

[0136] Step A53, extract the corresponding additional trigger conditions according to the screened state transition events;

[0137] Step A54, when the real-time measurement data satisfies the additional trigger conditions, use the screened state transition events as the screening results.

[0138] If only relying on the probability threshold to judge whether to perform a state transition, it may cause frequent state switching due to small fluctuations. Therefore, as a preferred embodiment of the present invention, a damping term δ ijAs an additional triggering condition, on the premise of judging the state transition by the probability threshold, it is also necessary to meet the additional triggering condition to trigger the state transition event.

[0139]

[0140]

[0141] When the condition of P ij > τ ij is satisfied:

[0142] If then S1 → S2;

[0143] If then S2 → S1;

[0144] If then S2 → S3;

[0145] If then S3 → S2;

[0146] By increasing the additional triggering condition threshold for the conversion from the low - energy - consumption data acquisition state to the high - energy - consumption data acquisition state, the disturbance can be suppressed. It is also possible not to change the additional triggering condition threshold for the conversion from the high - energy - consumption data acquisition state to the low - energy - consumption data acquisition state to quickly return to the low - energy - consumption data acquisition state. As an example: Set it to 1.5, Set it to 0, Set it to 1.5, Set it to 0.

[0147] Furthermore, in step A6, the state transition event in the screening result is executed only after the state residence time exceeds the preset minimum residence time.

[0148] In step A6, if the allowed state transition event in the screening result of step A5 is immediately executed, it may lead to a very short time interval between two adjacent state transitions and frequent state transitions. To avoid frequent state transitions, the minimum residence time T d .

[0149] The minimum residence time T d is introduced to implement a time - based state hysteresis conversion strategy.

[0150] When step 6 executes the screened state transition event, record the time stamp of the state transition and update the parameter t last-transition . t last-transition represents the time stamp of the most recent state transition.

[0151] The state residence time Tcurr Indicates the duration of the current data acquisition state, that is, the difference between the current timestamp t and the timestamp at the most recent state transition. The calculation formula is as follows:

[0152] T curr = t - t last-transition

[0153] When there are allowed state transition events in the screening results, the actual execution of the state transition event needs to be triggered when the state residence event is greater than or equal to the minimum residence time, that is, T curr ≥ T d .

[0154] Preferably, the minimum residence time is 15 minutes. Preferably, a corresponding minimum residence time can be set for each different data acquisition state.

[0155] In step A6, a finite-time stability condition is also set, that is, a residence time upper limit is set for some data acquisition states. The mathematical expression is:

[0156] T curr (S i ) ≤ T max (S i )

[0157]

[0158] as t → T max (S i )

[0159] For energy-saving considerations, a residence time upper limit is set for the state residence time of the data acquisition state S2 that adopts high-frequency sampling. The expression is:

[0160]

[0161] T curr (S2) ≤ β2·T max (S2)

[0162] β2 ∈ (0, 1]

[0163] For energy-saving considerations, a residence time upper limit is set for the state residence time of the data acquisition state S3 that adopts high-frequency sampling. The expression is:

[0164]

[0165] T curr (S3) ≤ β3·T max (S3)

[0166] β3 ∈ (0, 1]

[0167] α 21 、α 32 is the decay rate, and β2 and β3 are adjustment factors.

[0168] The finite-time stability condition is set to limit the maximum duration of the high-energy consumption state, that is, to save power. After reaching the time upper limit, it will be forced to transfer to the low-energy consumption state.

[0169] The present invention also provides a data acquisition system based on a semi-Markov process, integrated in an edge device, for performing the foregoing data acquisition method based on a semi-Markov process, including:

[0170] A state space definition module (1) for defining a discrete state space, including a set of multiple data acquisition states set by the edge device and a set of allowed state transition events between the data acquisition states;

[0171] A probability matrix creation module (2), connected to the state space definition module (1), for creating a transition probability matrix, and the matrix elements in the transition probability matrix represent the state transition probabilities between the data acquisition states;

[0172] A transition probability calculation module (3) for performing data acquisition based on the currently relied-upon data acquisition state to obtain real-time measurement data, and calculating the state transition probability based on the real-time measurement data;

[0173] A probability matrix update module (4), respectively connected to the probability matrix creation module (2) and the transition probability calculation module (3), for updating the transition probability matrix based on the calculated state transition probability;

[0174] A transition event screening module (5), connected to the probability matrix update module (4), for screening the allowed state transition events according to the updated transition probability matrix to obtain a screening result;

[0175] A transition event execution module (6), connected to the transition probability calculation module (3) and the transition event screening module (5), for executing the state transition events in the screening result to realize the switching of the corresponding data acquisition states, and then the transition probability calculation module (3) uses the switched data acquisition state as the currently relied-upon data acquisition state for data acquisition.

[0176] The above are only preferred embodiments of the present invention, and do not limit the implementation manners and protection scope of the present invention. For those skilled in the art, it should be realized that all equivalent replacements and obvious changes made by using the description and drawings of the present invention should be included in the protection scope of the present invention.

Claims

1. A data acquisition method based on a semi-Markov process, applied to downhole data acquisition of edge devices, characterized in that: include: Step A1, defining a discrete state space, including a set of multiple data acquisition states set by the edge device and a set of state transition events allowed between data acquisition states; Step A2, creating a transition probability matrix, wherein the matrix elements in the transition probability matrix represent the state transition probability between data acquisition states; Step A3, collecting data according to the current data collection state to obtain real-time measurement data, and calculating the state transition probability based on the real-time measurement data; Step A4, updating the transition probability matrix based on the state transition probability calculated in step A3; Step A5, screening the allowed state transition events according to the updated transition probability matrix to obtain screening results; Step A6, executing the state conversion event in the screening result to realize the switching of the corresponding data collection state, and taking the converted data collection state as the current data collection state to continue the step A3.

2. A data collection method based on a semi-Markov process as claimed in claim 1, characterized in that: In the step A1, the edge device sets three data collection states: normal state, awakening state and emergency alarm state; The sampling frequency of the normal state is lower than the sampling frequency of the awake state, and the sampling frequency of the awake state is lower than the sampling frequency of the emergency alarm state.

3. A data collection method based on a semi-Markov process as claimed in claim 2, characterized in that: The set of state transition events allowed includes: The state transition event of the normal state transitioning to the awake state; the state transition event of the awake state transitioning to the emergency alarm state; The state transition event of the awake state transitioning to the normal state; The state transition event of the emergency alarm state transitioning to the awake state.

4. A data collection method based on a semi-Markov process as claimed in claim 1, characterized in that: The step A3 comprises: Step A31, creating a historical state transition event data set; Step A32, determining the value of the standard deviation parameter in the probability density function by a statistical method according to the historical state transition event data set; Step A33, obtaining a cumulative distribution function of a normal distribution according to the integral of the probability density function with the value of the determined standard deviation parameter; Step A34, calculating the state transition probability based on the cumulative distribution function of the normal distribution.

5. A data collection method based on a semi-Markov process as claimed in claim 4, characterized in that: In step A3, the soft trigger threshold in the probability density function is also optimized, and the optimization process includes: Step B31, defining a comprehensive loss function, wherein the comprehensive loss function is a weighted sum function of the probability of state transition error, the probability that a state should be transitioned but is not transitioned, and the expected time delay of transition between states; Step B32, minimizing the comprehensive loss function based on a preset optimization algorithm, and obtaining a minimization threshold of the minimized comprehensive loss function as a soft trigger threshold in the probability density function.

6. A data collection method based on a semi-Markov process as claimed in claim 1, characterized in that: In step A5, it is determined one by one whether the state transition probabilities corresponding to the allowed state transition events in the transition probability matrix are greater than the corresponding probability threshold, and the state transition events whose state transition probabilities are greater than the corresponding probability threshold are taken as the screening results.

7. A data collection method based on a semi-Markov process as claimed in claim 6, characterized in that: The step A5 comprises: Step A51, determining one by one whether the state transition probabilities corresponding to the state transition events allowed in the transition probability matrix are greater than the corresponding probability thresholds; Step A52, screening out the state transition events whose state transition probabilities are greater than the corresponding probability thresholds; Step A53, extracting corresponding additional trigger conditions according to the screened state transition events; Step A54: when the real-time measurement data meets the additional trigger condition, the filtered state transition event is used as the filtering result.

8. A data collection method based on a semi-Markov process as claimed in claim 1, characterized in that: In the step A6, the state transition event in the screening result is executed only after the state stay time exceeds the preset minimum stay time.

9. A data acquisition system based on a semi-Markov process, integrated in the edge device, characterized in that: A method for data collection based on a semi-Markov process as described in any one of claims 1 to 8, comprising: A state space definition module is used to define a discrete state space, including a set of multiple data acquisition states set by edge devices and a set of state transition events allowed between data acquisition states; A probability matrix creation module, connected to the state space definition module, for creating a transition probability matrix, wherein the matrix elements in the transition probability matrix represent the state transition probability between data acquisition states; A transition probability calculation module, used to collect data according to the current data collection state to obtain real-time measurement data, and calculate the state transition probability based on the real-time measurement data; A probability matrix updating module, connected to the probability matrix creating module and the transition probability calculating module respectively, and used for updating the transition probability matrix based on the calculated state transition probability; A transition event screening module, connected to the probability matrix updating module, for screening the allowed state transition events according to the updated transition probability matrix to obtain a screening result; The conversion event execution module is connected to the conversion probability calculation module and the conversion event screening module, and is used to execute the state conversion event in the screening result to realize the switching of the corresponding data collection state. After that, the conversion probability calculation module uses the switched data collection state as the current data collection state for data collection.

Images