An energy-efficient sensor sleep scheduling method for remote state estimation systems

Abstract

The present invention discloses an energy-efficient sensor sleep scheduling method in a remote state estimation system, which includes the following steps: step 1: initializing system parameters, step 2: at the beginning of each time slot when the remote state estimation system is running, the scheduler determines the sensor set currently in the active mode in the system; step 3: the scheduler selects the sensor with the largest Max-Weight value in the active sensor set for scheduling; step 4: the scheduler determines the duration of sleep required for the scheduled sensor in the next cycle by simulating the system operation; step 5: if the data of the scheduled sensor is sent successfully, the ACK returned by the scheduler to the sensor will carry the sleep duration calculated in step 4; step 6: the sensor sets the sleep time according to the sleep duration, switches to the sleep mode, and switches back to the active mode after the sleep ends. This method is far superior to the fixed sleep time method in terms of state estimation error, and has low energy consumption.

Description

Technical Field

[0001] The present invention relates to an energy-efficient sensor sleep scheduling method in a remote state estimation system, and belongs to the field of industrial Internet and sensor scheduling. Background Art

[0002] For the common remote state estimation system in the industrial Internet: it contains multiple sub-closed-loop systems, each of which consists of controlled objects, sensors, controllers, state estimators and other links, and the data of all sensors are sent through a shared communication network. The state estimation process in the system depends on the data provided by the sensors. When the interval of the sensor's sampling data becomes larger, the estimation error of the estimator may increase. Therefore, there is a correlation between the state estimation error and the information age. Due to restrictions such as channel conflicts, not all sensors can send data at the same time in any time slot. It is necessary to formulate a reasonable scheduling method to make the sensors in the network work in an orderly manner. However, this may cause the sensor to be in an invalid active state, that is, it is in an active state but the scheduling method does not arrange it to send data, resulting in energy waste. In this context, we can also introduce a sleep mode for the sensor, hoping to reduce the energy consumption caused by the sensor being in an invalid active state due to channel conflicts and other problems to a certain extent. Considering that the information age is very small when the data is just received, the sensor can be suspended for a period of time after successfully sending the data. This method will not cause too much impact on the information age. In order to balance energy consumption and state estimation error based on information age, it is necessary to consider an effective sensor sleep scheduling method.

[0003] The system contains M linear time-invariant control subsystems that operate independently of each other. Discretize time and divide it into intervals of equal length, and use k∈{1,2,…,K} to represent the current time slot number. For any subsystem i∈{1,2,…,M}, including the controlled object sensor Actuator Estimator∈ i and controller Each time slot allows at most one sensor to send data. The probability of successful data transmission is p. i , the state equation of each subsystem i is:

[0004] x k+1,i =A i x k,i +B i u k,i +w k,i ,

[0005] y k,i =x k,i

[0006] in is the state of system i at time slot k, is the input of system i at time slot k, w k,i is the system noise of system i at time slot k, and the system matrix of system i is The input matrix of system i is Assume that w k,i are independent and identically distributed, with mean 0 and variance Gaussian distribution, y k,i is the output of system i at time slot k, that is, the sensor can directly sample the system state.

[0007] So far, many research results have been achieved in the industrial Internet by optimizing the information age of the system through reasonable scheduling of sensors. However, there are still the following problems: 1) Most studies consider optimizing the information age itself, but there are many key indicators coupled with the information age that can also be optimized from the perspective of information age; 2) In order to improve the freshness of information, the energy consumption of sensors is often ignored, and sensors are kept in active mode to wait for scheduling, resulting in energy waste; 3) Even if some studies introduce a sleep mode for sensors to save energy, this sleep time is pre-set and fixed, and the sensor is not allowed to adjust its sleep time according to the network status. When the sleep time is too long, it may affect the information age, and when the sleep time is too short, it may increase energy consumption. Summary of the invention

[0008] The problem to be solved by the present invention is to provide an effective sensor sleep scheduling method for the remote state estimation system in the industrial Internet, taking into account factors such as packet loss rate and channel conflict in the network, so as to comprehensively optimize the energy consumption in the system and the state estimation error based on information age.

[0009] Technical solution: To achieve the purpose of the present invention, the technical solution adopted by the present invention is: an energy-efficient sensor sleep scheduling method in a remote state estimation system, the method comprising the following steps:

[0010] Step 1: Initialize system parameters, including the system matrix A of system i i 、Input matrix B i , the variance of the system noise Σ i , Energy consumption E of the sensor sending data tran , the energy consumption E of the sensor in active mode in a time slot on , the energy consumption E of the sensor in active mode in a time slot off And the packet receiving rate p i At the same time, a sleep mode is introduced for the sensors in the system. When the sensors are in sleep mode, they stop sampling and sending data to save energy.

[0011] Step 2: At the beginning of each time slot when the remote state estimation system is running, the scheduler determines the set of sensors that are currently in active mode in the system:

[0012]

[0013] Among them U i (k) is the estimator ∈ before the beginning of the kth time slot i The most recent received from the sensor The time when the data to be sent is generated, T i For sensor The currently set sleep time.

[0014] Step 3: The scheduler selects the sensor with the largest Max-Weight value from the active sensor set for scheduling. i The current system state can be estimated based on the last received system state, which depends on the estimator ∈ i Current information age:

[0015]

[0016] where x k,i represents the system state of system i at time slot k, represents the system state of system i at time slot k estimated by the estimator, u k,i represents the system input of system i at time slot k, w k,i represents the system noise of system i at time slot k, Δ i (k) represents the information age of system i at time slot k. The mean square error of state estimation of system i can be described from the perspective of information age:

[0017]

[0018] Where tr(·) is the trace of the matrix. The system mainly aims to optimize the mean square error of state estimation:

[0019]

[0020] Where K is the system time domain and M is the number of subsystems. For each time slot, the state of the system is:

[0021]

[0022] Define the following positive definite Lyapunov function:

[0023]

[0024] The Lyapunov drift is:

[0025]

[0026] Substituting the Lyapunov function into the above formula, we can get:

[0027]

[0028] Because there are:

[0029]

[0030] So we can get:

[0031]

[0032] Therefore, in order to reduce the mean square error of state estimation based on information age in the system, the above formula needs to be minimized in each time slot. Due to the channel conflict constraint, at most one sensor can be scheduled to send data in each time slot, so the sensor with the largest Max-Weight value can be selected for scheduling:

[0033]

[0034] Step 4: The scheduler determines the duration of sleep time of the scheduled sensor in the next cycle by simulating the operation of the system: By simulating the operation of the sensor in the subsequent time slot, the next MW of the currently scheduled sensor is predicted. i When does the value (k) exceed the MW of all active sensors? i (k) value, so that the current sensor just sleeps until this moment, that is, it is expected that the sensor will wake up just when it needs to be scheduled. The process is as follows:

[0035] 1) The serial number of the currently scheduled sensor is recorded as idx;

[0036] 2) Initialize simulation parameters, let T' i =T i , T' idx =0, U' i (k+1)=U i (k), U' idx (k+1)=k and t=0;

[0037] 3) Start the simulation operation cycle, and set t = t + 1 at the beginning of each cycle;

[0038] 4) If t>1: Let U' i (k+t)=U' i (k+t-1), generate a random number r with a value in [0,1], if r≤ηp tempIdx , then let U' tempIdx (k+t)=k+t-1;

[0039] 5) Update information age: Δ' i (k+t)=k+tU' i (k+t), get the set of sensors that are in active mode at this time: Get the sensor number with the largest Max-Weight value at this time: The simulated value Δ' i (k+t) replaces Δ i (k+t);

[0040] 6) When idx is equal to tempIdx, the simulation process ends and the loop is exited. The scheduled sensor needs to sleep for a period of t-1 in the next cycle.

[0041] Step 5: If the data of the scheduled sensor is sent successfully, the ACK sent by the scheduler to the sensor will carry the sleep duration calculated in step 4.

[0042] Step 6: The sensor sets the sleep time according to the sleep time, switches to the sleep mode, and switches back to the active mode after the sleep mode ends.

[0043] Beneficial effects: The present invention proposes an energy-efficient sensor sleep scheduling method in a remote state estimation system. Compared with the prior art, the present invention not only takes into account factors such as channel conflicts in the remote state estimation system that may cause the sensor to be in an invalid active state and waste energy, but also takes into account that the system should not have a significant impact on the state estimation error after introducing a sleep mode for the sensor. Therefore, the present invention predicts the next best wake-up time of the sensor and dynamically adjusts the sleep time of the sensor, so that each sensor is active at different peaks to avoid channel conflicts, thereby reducing the system state estimation error and saving energy. Experiments show that the method of the present invention is far superior to the fixed sleep time method in terms of state estimation error, and is almost the same as the no-sleep method, but the energy consumption is much lower than that of the no-sleep method. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of a remote state estimation system used in an embodiment of the present invention.

[0045] Figure 2 This is the performance of the method proposed in this patent, the fixed sleep time method, and the no sleep method in terms of the mean square error of system state estimation.

[0046] Figure 3 This is the performance of the method proposed in the patent of this invention, the fixed sleep time method, and the no sleep method in terms of energy consumption. DETAILED DESCRIPTION

[0047] The following will describe the implementation methods of the present invention in detail with reference to the accompanying drawings and embodiments, so that the implementation process of how the present invention applies technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly. It should be noted that as long as there is no conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other, and the technical solutions formed are all within the protection scope of the present invention.

[0048] Embodiment 1: An energy-efficient sensor sleep scheduling method in a remote state estimation system, the method comprising the following steps:

[0049] Figure 1 Schematic diagram of a remote state estimation system used in an embodiment of the present invention, Figure 1 , explaining each step in detail.

[0050] Step 1: Initialize system parameters, including the system matrix A of system i i 、Input matrix B i , the variance of the system noise Σ i , Energy consumption E of the sensor sending data tran , the energy consumption E of the sensor in active mode in a time slot on , the energy consumption E of the sensor in active mode in a time slot off And the packet receiving rate p i At the same time, a sleep mode is introduced for the sensors in the system. When the sensors are in sleep mode, they stop sampling and sending data to save energy.

[0051] Step 2: At the beginning of each time slot when the remote state estimation system is running, the scheduler determines the set of sensors that are currently in active mode in the system:

[0052]

[0053] Among them U i (k) is the estimator ∈ before the beginning of the kth time slot i The most recent received from the sensor The time when the data to be sent is generated, T i For sensor The currently set sleep time.

[0054] Step 3: The scheduler selects the sensor with the largest Max-Weight value from the active sensor set for scheduling. i The current system state can be estimated based on the last received system state, which depends on the current information age of the estimator ∈ i:

[0055]

[0056] where xk,i represents the system state of system i at time slot k, represents the system state of system i at time slot k estimated by the estimator, u k,i represents the system input of system i at time slot k, w k,i represents the system noise of system i at time slot k, Δ i (k) represents the information age of system i at time slot k. The mean square error of state estimation of system i can be described from the perspective of information age:

[0057]

[0058] Where tr(·) is the trace of the matrix. The system mainly aims to optimize the mean square error of state estimation:

[0059]

[0060] Where K is the system time domain and M is the number of subsystems. For each time slot, the state of the system is:

[0061]

[0062] Define the following positive definite Lyapunov function:

[0063]

[0064] The Lyapunov drift is:

[0065]

[0066] Substituting the Lyapunov function into the above formula, we can get:

[0067]

[0068] Because there are:

[0069]

[0070] So we can get:

[0071]

[0072] Therefore, in order to reduce the mean square error of state estimation based on information age in the system, the above formula needs to be minimized in each time slot. Due to the channel conflict constraint, at most one sensor can be scheduled to send data in each time slot, so the sensor with the largest Max-Weight value can be selected for scheduling:

[0073]

[0074] Step 4: The scheduler determines the duration of sleep time of the scheduled sensor in the next cycle by simulating the operation of the system: By simulating the operation of the sensor in the subsequent time slot, the next MW of the currently scheduled sensor is predicted. i When does the value (k) exceed the MW of all active sensors? i (k) value, so that the current sensor just sleeps until this moment, that is, it is expected that the sensor will wake up just when it needs to be scheduled. The process is as follows:

[0075] 1) The serial number of the currently scheduled sensor is recorded as idx;

[0076] 2) Initialize simulation parameters, let T' i =T i , T' idx =0, U' i (k+1)=U i (k), U' idx (k+1)=k and t=0;

[0077] 3) Start the simulation operation cycle, and set t = t + 1 at the beginning of each cycle;

[0078] 4) If t>1: Let U' i (k+t)=U' i (k+t-1), generate a random number r with a value in [0,1], if r≤ηp tempIdx , then let U' tempIdx (k+t)=k+t-1;

[0079] 5) Update information age: Δ' i (k+t)=k+tU' i (k+t), get the set of sensors that are in active mode at this time: Get the sensor number with the largest Max-Weight value at this time: The simulated value Δ' i (k+t) replaces Δ i (k+t);

[0080] 6) When idx is equal to tempIdx, the simulation process ends and the loop is exited. The scheduled sensor needs to sleep for a period of t-1 in the next cycle.

[0081] Step 5: If the data of the scheduled sensor is sent successfully, the ACK sent by the scheduler to the sensor will carry the sleep duration calculated in step 4.

[0082] Step 6: The sensor sets the sleep time according to the sleep time, switches to the sleep mode, and switches back to the active mode after the sleep mode ends.

[0083] Embodiment 2:

[0084] Example 2 of the present invention considers a system with the following parameters

[0085]

[0086] Energy consumption in sleep mode: 1.5×10 -3 , 1.35 when active and 2.475 when sending data, the energy consumption unit is 10 -4 J, packet receiving rate p i =i / M, and the results of each simulation are averaged for 2000 times. An energy-efficient sensor sleep scheduling method in a remote state estimation system proposed by the present invention is compared with a sleep time fixed method and a no sleep method.

[0087] Figure 2 The method proposed in this invention is compared with the method with fixed sleep time and the method without sleep time in K∈{5×10 4 ,8×10 4 ,10 5 ,3×10 5} is the performance of the mean square error of system state estimation.

[0088] Figure 3 The method proposed in this invention is compared with the method with fixed sleep time and the method without sleep time in K∈{5×10 4 ,8×10 4 ,10 5 ,3×10 5}The performance of system energy consumption when

[0089] like Figure 2 and Figure 3 As shown, the state estimation mean square error of the method proposed in the present invention is very close to the performance of the no-sleep method, while the state estimation mean square error of the method with a fixed sleep time is close to 2.5 times that of the no-sleep method. This is mainly because the state estimation mean square error increases with the information age exponentially. When the sensor does not provide information data for a long time due to unreasonable sleep time, the state estimation error will be relatively large. At the same time, it can be found that the energy consumption of the no-sleep method is the highest, because this sensor is always in active mode, and the energy consumption of the method proposed in the present invention is between the no-sleep method and the fixed sleep time method. The method proposed in the present invention greatly reduces the impact of the introduced sensor sleep mode on the system state estimation error. The state estimation error performance in the system is much better than the fixed sleep time method, and is almost the same as the no-sleep method, but the energy consumption generated is much lower than that of the no-sleep method.

[0090] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. An energy-efficient sensor sleep scheduling method in a remote state estimation system, characterized in that: The method comprises the following steps: Step 1: Initialize system parameters and introduce a sleep mode for sensors in the system. When the sensors are in sleep mode, they stop sampling and sending data to save energy. Step 2: At the beginning of each time slot when the remote state estimation system is running, the scheduler determines the set of sensors currently in active mode in the system; Step 3: The scheduler selects the sensor with the largest Max-Weight value from the active sensor set for scheduling; Step 4: The scheduler determines the duration that the scheduled sensor needs to sleep in the next cycle by simulating the system operation; Step 5: If the data of the scheduled sensor is sent successfully, the ACK sent by the scheduler to the sensor will carry the sleep duration calculated in step 4. Step 6: The sensor sets the sleep time according to the sleep time, switches to the sleep mode, and switches back to the active mode after the sleep mode ends; In step 1, the system parameters are initialized, including the system matrix A of system i i 、Input matrix B i , the variance of the system noise Σ i , Energy consumption E of the sensor sending data tran , the energy consumption E of the sensor in active mode in a time slot on , the energy consumption of the sensor in sleep mode in a time slot E off And the packet receiving rate p i ,At the same time, a sleep mode is introduced for the sensors in the system. When the sensors are in sleep mode, they stop sampling and sending data to save energy; At the beginning of each time slot of the system operation in step 2, the scheduler determines the set of sensors currently in active mode in the system: Among them U i (k) is the estimator ∈ before the beginning of the kth time slot i The most recent received from the sensor The time when the data to be sent is generated, T i For sensor The currently set sleep duration; In step 3, the scheduler selects the sensor with the largest Max-Weight value from the set of active sensors for scheduling: Estimator∈ i The current system state is estimated based on the last received system state, which depends on the estimator ∈ i Current information age: where x k，i represents the system state of system i at time slot k, represents the system state of system i at time slot k estimated by the estimator, u k，i represents the system input of system i at time slot k, W k，i represents the system noise of system i at time slot k, Δ i (k) represents the information age of system i at time slot k. The mean square error of state estimation of system i is described from the perspective of information age: Where tr(·) is the trace of the matrix, and the system mainly aims to optimize the mean square error of state estimation: Where K is the system time domain, M is the number of subsystems, and for each time slot, the state of the system is: Define the following positive definite Lyapunov function: The Lyapunov drift is: Substituting the Lyapunov function into the above formula, we get: Because there are: So we get: Therefore, in order to reduce the mean square error of state estimation based on information age in the system, each time slot needs to minimize the above formula. Due to the channel conflict constraint, at most one sensor is scheduled to send data in each time slot, and the sensor with the largest Max-Weight value is selected for scheduling: In step 4, the scheduler determines the duration of sleep time of the scheduled sensor in the next cycle by simulating the operation of the system: by simulating the operation of the sensor in the subsequent time slot, the next MW of the currently scheduled sensor is predicted. i (k) value exceeds the MW of all active sensors i (k) is the time when the current sensor is put into sleep mode until this time. That is, it is expected that the sensor will wake up when it needs to be scheduled. The process is as follows: 1) The serial number of the currently scheduled sensor is recorded as idx; 2) Initialize the simulation parameters, let T′ i =T i , T′ idx =0, U′ i (k+1)=U i (k), U′ idx (k+1)=k and t=0; 3) Start the simulation operation cycle, and set t = t + 1 at the beginning of each cycle; 4) If t>1: Let U′ i (k+t)=U′ i (k+t-1), generate a random number r with a value in [0,1], if r≤ηp tempIdx , then let U′ tempIdx (k+t)=k+t-1; 5) Update information age: Δ′ i (k+t)=k+tU′ i (k+t), get the set of sensors that are in active mode at this time: Get the sensor number with the largest Max-Weight value at this time: The simulated value Δ′ is used i (k+t) replaces Δ i (k+t); 6) When idx is equal to tempIdx, the simulation process ends and the loop is exited. The scheduled sensor needs to sleep for a period of t-1 in the next cycle.

Images