Data processing method for improving measurement stability of activity meter

Abstract

The invention discloses a data processing method for improving the measurement stability of an activity meter. The method comprises the following steps: 1) establishing a discrete time state space model of an activity meter measurement system; 2) establishing an observation model of an activity meter measurement system; 3) dynamically adjusting the covariance Q of the process noise or the covariance R of the observation noise according to the sliding window statistical value or variance of the innovation: if the innovation is continuously greater than a set threshold value, increasing Q or decreasing R to enable a filter to quickly track the current signal change; when the signal is stable, Q is reduced, and a smoother signal is obtained; 4) predicting the optimal state at the moment t according to the optimal state estimated at the moment t-1, and substituting the optimal state into the observation model to obtain an observation value Zt at the moment t; 5, the Zt and the optimal state at the moment t are fused, Kalman gain updating is calculated, the optimal state estimated at the moment t is obtained and serves as an activity value after filtering, and 6, a series of activity values are output.

Description

Technical Field

[0001] This invention belongs to the field of radiometric measurement signal processing technology, and relates to a data processing method for improving the measurement stability of activity meters. It is used to optimize the nuclide activity measurement data of activity meters and is applicable to scenarios with high requirements for both measurement stability and response speed, such as medical radiometric detection and nuclear industry environmental monitoring. Background Technology

[0002] The core measurement principle of an activity meter is as follows: the radiation emitted by a radioactive nuclide ionizes the gas in the ionization chamber, generating a weak current. This current is converted into a voltage signal by a high-resistance amplifier, and then calculated using a calibration coefficient to obtain the nuclide activity value. Signal processing is crucial for improving measurement accuracy, and existing technologies mainly focus on noise suppression and data smoothing. In existing technologies, activity meter data processing primarily employs two schemes, the specific implementation processes of which are as follows.

[0003] Option 1: Fixed Window Arithmetic Mean Method 1. Acquire voltage signals at a fixed sampling frequency of 1-2Hz; 2. Collect N data points consecutively (usually N=5-20), and assign all data points the same weight; 3. Calculate the arithmetic mean of N data points and output it as the current activity measurement value; 4. Repeat the above steps to achieve continuous measurement.

[0004] Option 2: Fixed-parameter Kalman filtering method 1. Establish a simple state-space model, with fixed process noise covariance Q and observation noise covariance R. 2. Acquire signals at a lower sampling frequency (2-5Hz) and perform data fusion using a Kalman filter formula with fixed Q and R parameters; 3. Output the filtered estimated value. Throughout the process, the Q and R parameters are not adjusted as the signal changes.

[0005] The existing technical solutions have the following main technical drawbacks.

[0006] Disadvantages of Option 1 (Fixed Window Arithmetic Mean Method) 1. Significant Response Lag: Because historical and latest data are treated equally in influencing the results, the system output is "anchored" by a large amount of past information, making it unable to quickly respond to real changes in the signal. This directly leads to a significant lag between the displayed value and the true value. This problem is particularly prominent when measuring short-half-life nuclides, as their activity decays rapidly over time, and the lag effect introduces huge measurement errors, seriously affecting the accuracy of quantitative analysis.

[0007] 2. Information Processing Passivation: While fixed-weight averaging smooths out random noise, it indiscriminately smooths out the gradual trends and subtle features inherent in the signal itself. This "information passivation" phenomenon makes the measurement system extremely insensitive to slow drifts or early changing trends, resulting in the loss of the ability to preserve details that are crucial for process monitoring and early warning.

[0008] 3. Rigid noise suppression: Its noise suppression capability depends entirely on the length of the average window, which is a rigid trade-off—lengthening the window enhances smoothness but exacerbates lag, while shortening the window speeds up the response but results in insufficient noise suppression. The system cannot dynamically self-optimize based on the actual statistical characteristics of the signal and noise.

[0009] Disadvantages of Scheme 2 (Fixed Parameter Kalman Filtering) 1. Fixed parameters lead to poor adaptability: The preset Q and R parameters cannot cope with dynamic changes in the signal (such as rapid activity decay and sudden noise changes), the filtering smoothness is insufficient when the signal is stable, and the tracking response is lagging when the signal changes suddenly. 2. Mismatch between sampling frequency and processing efficiency: Low sampling frequency limits the capture of signal details, and fixed parameter calculation cannot fully utilize the optimal estimation characteristics of Kalman filtering. Summary of the Invention

[0010] To address the problems existing in the prior art, the purpose of this invention is to provide a data processing method that improves the stability of activity meter measurements.

[0011] This invention employs an adaptive Kalman filter algorithm to process the high-speed raw sampling data of the activity meter in real time, achieving a balance between noise suppression and dynamic signal tracking.

[0012] The technical solution of this invention is as follows: A data processing method for improving the stability of activity meter measurements includes the following steps: 1) Establish the discrete-time state-space model of the activity meter measurement system: X_t = F * X_{t-1} + W_t, where X_t is the true activity value to be estimated at time t, F is the state transition matrix, and W_t is the process noise with covariance Q; 2) Establish the observation model of the activity meter measurement system: Z_t = H * X_t + V_t; where Z_t represents the actual measurement value of the activity meter at time t, H is the observation matrix, and V_t is the observation noise with covariance R. 3) Based on the sliding window statistics or variance of the innovation, dynamically adjust the covariance Q of the process noise or the covariance R of the observation noise: if the innovation is continuously greater than the set threshold, increase Q or decrease R to make the filter quickly track the current signal change; when the signal is stable, decrease Q to obtain a smoother signal output. 4) Substitute the estimated optimal state X_{t-1|t-1} at time t-1 into X_{t|t-1} = F * X_{t-1|t-1} to predict the optimal state X_{t|t-1} at time t; substitute X_{t|t-1} into the observation model Z_t = H * X_t + V_t to obtain the observed value Z_t = H * X_{t|t-1} + V_t at time t; then perform prior error covariance prediction to determine the uncertainty P_{t|t-1} of the optimal state X_{t|t-1} = F * P_{t-1|t-1} * F^T + Q; P_{t-1|t-1} is the posterior error covariance at time t-1, i.e., the uncertainty of the estimated optimal state X_{t-1|t-1} at time t-1. 5) Fuse the actual activity meter measurement Z_t at time t with the optimal state X_{t|t-1} at time t, and calculate the Kalman gain K_t = P_{t|t-1} * H^T * (H * P_{t|t-1} * H^T + R)^{-1}, where K_t is the dynamic weighting factor; then perform a posterior state update to obtain the estimated optimal state X_{t|t} = X_{t|t-1} + K_t * (Z_t - H * X_{t|t-1}), which is used as the filtered activity value; then perform a posterior error covariance update to obtain the uncertainty P_{t|t} = (I - K_t * H) * P_{t|t-1} of the optimal state X_{t|t}; 6) The discrete-time state-space model outputs the activity values ​​X_{1|1}, X_{2|2}, ..., X_{t|t} at time t and before. 7) Iterate through steps 3) to 6) until the set conditions are met and then terminate.

[0013] Preferably, the innovation is the difference between the observed value Z_t and the predicted value H * X_{t|t-1}, Z_t - H * X_{t|t-1}.

[0014] Preferably, if the innovation is greater than 1.5 times the Q value for more than 3 sampling periods, then Q is increased or R is decreased.

[0015] Preferably, if the signal is stable, Q is reduced to a value of 0.1 to 0.2 times.

[0016] Preferably, a model that fits the decay law is constructed using the state transition matrix F=e^(-λΔt) to measure short-half-life nuclides; λ is the nuclide decay constant, and Δt is the sampling interval.

[0017] Preferably, in the activity meter scenario, the observation matrix H=1, meaning that the measured value and the true value have a linear relationship.

[0018] Preferably, if the activity is stable within a set short period of time, the state transition matrix F=1.

[0019] The advantages of this invention are as follows: 1. Dynamic Tracking and Rapid Response: Through an exponential decay memory mechanism, the influence of historical data rapidly diminishes over time, resulting in higher confidence levels for new observations. The filter closely tracks instantaneous signal changes, significantly eliminating display lag and adapting to dynamic measurement scenarios such as those involving short-half-lived nuclides.

[0020] 2. Optimal Signal-to-Noise Separation and Detail Preservation: Based on the optimal estimation of system model and noise statistical characteristics, it can intelligently distinguish between real signal dynamics and random noise. While achieving excellent smoothing effect, it keenly preserves the trend changes and subtle features of the signal, avoiding "information dulling" and providing high-fidelity stable output.

[0021] 3. Adaptive and Self-Optimizing Performance: The algorithm is a dynamic feedback system that adaptively adjusts the confidence distribution between model predictions and the latest observations by updating the estimation error covariance in real time. Continuous self-optimization ensures that the system approaches the optimal estimate under different operating conditions, improving measurement reliability and reading intuitiveness. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0024] The method flow of this invention is as follows: Figure 1 As shown, the main steps include the following.

[0025] Step 1: System modeling and state definition, establishing a discrete-time state-space model for the activity meter measurement system.

[0026] State variable definition: The true activity value to be estimated at time t is defined as the system state variable X_t.

[0027] State transition model: Establish the state equation X_t = F * X_{t-1} + W_t. F is the state transition matrix, which simplifies to F=1 if the activity is stable within a short time. When measuring short-lived nuclides, the state transition matrix F=e^(-λΔt) (λ is the nuclide decay constant, Δt is the sampling interval) can be used to construct a model that fits the decay law. W_t is the process noise with covariance Q, representing the uncertainty of the model prediction (such as small fluctuations in the decay rate).

[0028] Step 2: Establish the observation model and establish the system observation equation Z_t = H * X_t + V_t.

[0029] Observed variable Z_t: represents the direct measurement reading of the activity meter at time t.

[0030] Observation matrix H: maps the true state of the system to the observation space. In the activity meter scenario, H is usually set to 1, that is, the measured value and the true value are linearly related.

[0031] Observation noise V_t: covariance R, representing the random error and statistical fluctuations of the measuring instrument itself.

[0032] Step 3: Adaptive weight adjustment. This step is the key difference between this method and fixed parameter filtering, and it realizes the "exponential decay weight" of historical data.

[0033] Dynamically adjust Q or R: By calculating the sliding window statistics or variance of the innovation (the difference between the observed value Z_t and the predicted value H * X_{t|t-1} Z_t - H * X_{t|t-1}), the signal change characteristics are monitored in real time, and the process noise covariance Q or the observation noise covariance R is dynamically adjusted.

[0034] Achieving exponential decay: When the innovation is consistently large (e.g., the innovation is greater than 1.5 times the Q value for more than 3 sampling periods), it indicates that the signal is changing rapidly. The algorithm automatically increases Q (to trust the new observation data more) or decreases R, which is equivalent to reducing the weight of historical estimates, allowing the filter to quickly track the current changes. When the signal (the activity meter reading, i.e., the activity value) is stable, Q is decreased to 0.1~0.2 times the Q value to obtain a smoother output.

[0035] Step 4: Prediction Step (Time Update) Before obtaining the new round of observations Z_t, a prediction is made based on the optimal estimate from the previous time step. Substituting X_{t|t-1} into the observation model Z_t = H * X_t + V_t, we obtain the observation value at time t: Z_t = H * X_{t|t-1} + V_t.

[0036] Prior state prediction: X_{t|t-1} = F * X_{t-1|t-1}, that is, predicting the current state based on the optimal estimate of the previous time step.

[0037] Prior error covariance prediction: P_{t|t-1} = F * P_{t-1|t-1} * F^T + Q, predicts the uncertainty of the estimate at the current time.

[0038] Step 5: Update step (measurement update). After obtaining the new measurement value Z_t, the predicted value X_{t|t-1} is optimally fused with the measured value Z_t.

[0039] Kalman gain calculation: K_t = P_{t|t-1} * H^T * (H * P_{t|t-1} * H^T + R)^{-1}, where K_t is a dynamic weighting factor that determines the degree of confidence in the new measurement value Z_t.

[0040] Posterior state update: X_{t|t} = X_{t|t-1} + K_t * (Z_t - H * X_{t|t-1}), the residual between the predicted and measured values ​​is weighted and corrected by Kalman gain to obtain the optimal estimated value X_{t|t} (filtered activity value) at the current time.

[0041] Posterior error covariance update: P_{t|t} = (I - K_t * H) * P_{t|t-1}, update the uncertainty of the estimated value to prepare for the next round of calculation.

[0042] Step 6: Smooth Output and Performance Evaluation High-fidelity output: The final output activity value is a series of optimal estimates X_{1|1}, X_{2|2}, ..., X_{t|t}, which effectively suppresses random noise while retaining the ability to track changes in the real signal, thus solving the lag problem of the fixed averaging method.

[0043] Improved stability and reliability: Through iterative iteration, operators obtain readings with significantly reduced fluctuations that reflect the true trend in a timely manner, reducing reading errors and the risk of misjudgment, and improving measurement reliability and user experience.

[0044] Although specific embodiments of the invention have been disclosed for illustrative purposes to aid in understanding and implementing the invention, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the invention should not be limited to the content disclosed in the preferred embodiments, and the scope of protection claimed by the invention is defined by the claims.

Claims

1. A data processing method for improving the stability of activity meter measurements, comprising the following steps: 1) Establish the discrete-time state-space model of the activity meter measurement system: X_t = F * X_{t-1} + W_t, where X_t is the true activity value to be estimated at time t, F is the state transition matrix, and W_t is the process noise with covariance Q; 2) Establish the observation model of the activity meter measurement system: Z_t = H * X_t + V_t; where Z_t represents the actual measurement value of the activity meter at time t, H is the observation matrix, and V_t is the observation noise with covariance R. 3) Based on the sliding window statistics or variance of the innovation, dynamically adjust the covariance Q of the process noise or the covariance R of the observation noise: if the innovation is continuously greater than the set threshold, increase Q or decrease R to make the filter quickly track the current signal change; when the signal is stable, decrease Q to obtain a smoother signal output. 4) Substitute the estimated optimal state X_{t-1|t-1} at time t-1 into X_{t|t-1} = F * X_{t-1|t-1} to predict the optimal state X_{t|t-1} at time t; substitute X_{t|t-1} into the observation model Z_t = H * X_t + V_t to obtain the observed value Z_t = H * X_{t|t-1} + V_t at time t; then perform prior error covariance prediction to determine the uncertainty P_{t|t-1} of the optimal state X_{t|t-1} = F * P_{t-1|t-1} * F^T + Q; P_{t-1|t-1} is the posterior error covariance at time t-1, i.e., the uncertainty of the estimated optimal state X_{t-1|t-1} at time t-1. 5) Fuse the actual activity meter measurement Z_t at time t with the optimal state X_{t|t-1} at time t, and calculate the Kalman gain K_t = P_{t|t-1} * H^T * (H * P_{t|t-1} * H^T + R)^{-1}, where K_t is the dynamic weighting factor; then perform a posterior state update to obtain the estimated optimal state X_{t|t} = X_{t|t-1} + K_t* (Z_t - H * X_{t|t-1}), which is used as the filtered activity value; then perform a posterior error covariance update to obtain the uncertainty P_{t|t} = (I - K_t * H) * P_{t|t-1} of the optimal state X_{t|t}; 6) The discrete-time state-space model outputs the activity values ​​X_{1|1}, X_{2|2}, ..., X_{t|t} at time t and before. 7) Iterate through steps 3) to 6) until the set conditions are met and then terminate.

2. The method according to claim 1, characterized in that, The new information is the difference between the observed value Z_t and the predicted value H * X_{t|t-1}, Z_t - H * X_{t|t-1}.

3. The method according to claim 1, characterized in that, If the new information is greater than 1.5 times the Q value for more than 3 sampling periods, then increase Q or decrease R.

4. The method according to claim 1, characterized in that, If the signal is stable, reduce Q to 0.1 to 0.2 times the Q value.

5. The method according to claim 1, characterized in that, A model that fits the decay law is constructed using the state transition matrix F=e^(-λΔt) to measure short-half-life nuclides; λ is the nuclide decay constant, and Δt is the sampling interval.

6. The method according to claim 1, characterized in that, In the activity meter scenario, the observation matrix H=1, meaning that the measured value and the true value have a linear relationship.

7. The method according to claim 1, characterized in that, When the activity is stable within a short period of time, the state transition matrix F=1.

Images