Multi-source environment monitoring data fusion method based on adaptive Bayesian reasoning

Abstract

The invention discloses a multi-source environment monitoring data fusion method based on adaptive Bayesian reasoning. The method comprises the following steps: S1, collecting an environment monitoring data set; s2, obtaining a preprocessed environment monitoring data set; s3, carrying out uncertainty evaluation on the preprocessed environment monitoring data set; s4, forming an initial prior probability; s5, calculating the posterior probability distribution of each environment monitoring data source through the Bayesian theorem, and dynamically updating the weight of each data source in the environment monitoring data fusion process according to the posterior probability distribution; s6, performing weighted fusion on each environment monitoring data source according to the dynamically updated weight to form fused environment monitoring data; and S7, carrying out post-processing on the fused environment monitoring data. According to the invention, the data fusion process can adapt to environmental changes in real time, so that the accuracy of data fusion is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and particularly to a multi-source environmental monitoring data fusion method based on adaptive Bayesian inference. Background Art

[0002] With the continuous popularization of Internet of Things technology and intelligent sensors, environmental monitoring systems are increasingly widely used in the fields of urban management, pollution warning, and disaster emergency. In the prior art, most environmental monitoring systems mainly rely on multiple sensors distributed in different regions to collect data and use traditional data fusion methods to process the collected data. Usually, data preprocessing, outlier detection, and data completion are performed based on fixed thresholds, simple statistical models, or empirical rules to achieve the monitoring and evaluation of the environmental state. However, there are many deficiencies in the actual application of traditional technologies, especially when dealing with the uncertainty of multi-source data, noise interference, and time synchronization error problems, they seem powerless.

[0003] First of all, existing data fusion technologies often adopt a unified format standard and a fixed time synchronization strategy in the preprocessing link, and it is difficult to fully consider the differences in signal-to-noise ratio, measurement deviation, and timing consistency of the data collected by each sensor. This one-size-fits-all processing method is likely to lead to uneven data quality, thereby affecting the accuracy and reliability of subsequent fusion processing. Secondly, for the uncertainty problems commonly existing in data sources, existing technologies often lack effective evaluation means and cannot accurately quantify the uncertainty degree and data reliability of each data source. Sensors are affected by external environmental interference to generate noise or errors during actual operation, and traditional methods cannot identify and compensate these deviations in a timely manner, resulting in certain deviations in the fused environmental monitoring data.

[0004] In addition, when traditional environmental monitoring data fusion technologies process multi-source data, they usually ignore the dynamic correlation and time-varying characteristics between data sources and only rely on static prior settings for data fusion. In practical application scenarios where data is updated frequently and environmental conditions change drastically, problems such as untimely data response and low fusion accuracy are likely to occur. Especially in scenarios where the sensor layout is complex and there are many data sources, how to dynamically adjust for the measurement accuracy, time synchronization error, and data noise of different data sources has become a difficult problem that needs to be solved urgently by current technologies.

[0005] In summary, the existing environmental monitoring data fusion methods have obvious limitations in data preprocessing, anomaly detection, and dynamic weight adjustment, making it difficult to meet the strict standards for uncertainty and real-time requirements of multi-source monitoring data. In view of the deficiencies in signal-to-noise ratio, time synchronization, and measurement deviation of the data collected by each sensor, it is urgent to develop a new data fusion method to solve the problems of dynamic inadaptability, rigid prior setting, and insufficient response to uncertainty in multi-source data fusion of traditional methods, so as to provide more reliable technical support for the accurate evaluation of environmental monitoring data. Summary of the Invention

[0006] An object of the present invention is to propose a multi-source environmental monitoring data fusion method based on adaptive Bayesian inference, which enables the data fusion process to adapt to environmental changes in real time, thereby improving the accuracy of data fusion.

[0007] A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference according to an embodiment of the present invention includes the following steps:

[0008] S1. Establish a distributed environmental monitoring system and collect an environmental monitoring data set based on the deployment of multiple environmental monitoring sensors;

[0009] S2. Preprocess the environmental monitoring data set to obtain a preprocessed environmental monitoring data set;

[0010] S3. Evaluate the uncertainty of the preprocessed environmental monitoring data set, and calculate the signal-to-noise ratio and data reliability index of each environmental monitoring data source;

[0011] S4. Construct an adaptive Bayesian inference model based on historical environmental monitoring data and the characteristics of the preprocessed environmental monitoring data. The adaptive Bayesian inference model is provided with a dynamic prior adjustment mechanism, and the dynamic prior adjustment mechanism adjusts the initial prior probability distribution of each environmental monitoring data source in real time according to the uncertainty evaluation result to form an initial prior probability;

[0012] S5. Use the adaptive Bayesian inference model with dynamic prior adjustment to fuse the preprocessed environmental monitoring data collected in real time with the initial prior probability, calculate the posterior probability distribution of each environmental monitoring data source through Bayes' theorem, and dynamically update the weight of each data source in the environmental monitoring data fusion process accordingly;

[0013] S6. Perform weighted fusion on each environmental monitoring data source according to the dynamically updated weight to form fused environmental monitoring data;

[0014] S7. Post-process the fused environmental monitoring data to generate environmental monitoring information for environmental quality assessment, pollution warning, disaster emergency response, and smart city management.

[0015] Optionally, S1 includes the following steps:

[0016] S11. Deploy a distributed environment monitoring system within the target environmental area, where the distributed environment monitoring system consists of N environmental monitoring sensors;

[0017] S12. Regularly collect and construct an environmental monitoring data set from the distributed environment monitoring system:

[0018] D raw ={d1, d2,..., d M};

[0019] where di i represents the i-th collected data record, M is the total number of data records, and each data record contains a timestamp t i and monitoring values from different environmental monitoring sensors:

[0020] d i =(t i , T i , H i , P i , C PM,i , C gas,i );

[0021] where t i represents the data record time, T i represents the measurement value of the temperature sensor, H i represents the measurement value of the humidity sensor, P i represents the measurement value of the air pressure sensor, C PM,i represents the particulate matter concentration measurement value, and C gas,i represents the gas concentration measurement value;

[0022] S13. Set the spatial distribution of the environmental monitoring data, define the monitoring area R, and the spatial coordinates of the environmental monitoring sensors within the monitoring area are:

[0023] xi = (xi, yi, zi);

[0024] where (x i , y i , z i ) represents the three-dimensional spatial coordinates of the i-th sensor within the monitoring area R.

[0025] Optionally, S2 includes the following steps:

[0026] S21. Clean the environmental monitoring data set D raw to remove invalid data records and define a valid environmental monitoring data set;

[0027] S22. Unify the format of the effective environmental monitoring data set and set a unified time format t i and the numerical normalization format to obtain a normalized environmental monitoring data set;

[0028] S23. Perform time synchronization, align the time of the normalized environmental monitoring data set, and calculate the statistical mean μ X and standard deviation σ X of the measured values of each sensor in the synchronized environmental monitoring data set, and define the outlier detection rule as:

[0029] X i is regarded as an outlier if and only if |X i - μ X | > λσ X ;

[0030] where λ is the outlier detection threshold. If the measured value X i exceeds the range, then determine that the data point is an outlier and mark it as missing;

[0031] S24. Perform missing data completion. Use time interpolation method to complete the environmental monitoring data with missing values, and finally generate the preprocessed environmental monitoring data set D processed .

[0032] Optionally, the S3 includes the following steps:

[0033] S31. Calculate the signal-to-noise ratio SNR processed of the preprocessed environmental monitoring data set D i :

[0034]

[0035] where is the signal variance of the i-th environmental monitoring data source, is the noise variance of the i-th environmental monitoring data source, and the signal-to-noise ratio SNR i reflects the measurement accuracy of the data source;

[0036] S32. Calculate the measurement deviation D i of each environmental monitoring data source:

[0037]

[0038] where X i,j represents the measured value collected by the i-th environmental monitoring data source at the timestamp t j , μ X is the measurement mean of the data source, and M is the total number of data records of the data source;

[0039] S33. Calculate the time synchronization error E of the environmental monitoring data source sync,i :

[0040]

[0041] where t j is the timestamp of the i-th environmental monitoring data source, t ref,j is the reference time, and the time synchronization error E sync,i reflects the accuracy of the time alignment of the data source;

[0042] S34. Calculate the reliability index R of the data source i :

[0043] R i = w1SNR i + w2(1 - D i ) + w3(1 - E sync,i );

[0044] where w1, w2, w3 are weight parameters;

[0045] S35. Generate the uncertainty evaluation result of the environmental monitoring data based on the data source reliability index R i and define the uncertainty weight U i :

[0046]

[0047] where U i represents the uncertainty weight of the i-th environmental monitoring data source, reflecting the uncertainty degree of the environmental monitoring data source in the data fusion process, θ is a positive adjustment index, reflecting the sensitivity of the uncertainty weight to the change of the data source reliability index R i under low reliability conditions, and κ is a non-negative proportionality coefficient, used to adjust the logarithmic amplification effect of the data source reliability index R i below the threshold.

[0048] Optionally, the S4 includes the following steps:

[0049] S41. Set the adaptive Bayesian inference model M bayes , which is used to fuse the multi-source environmental monitoring data. The adaptive Bayesian inference model consists of a set of state variables X, a prior probability distribution P(X), an environmental monitoring data set D processed and a dynamic adjustment mechanism A dyn :

[0050] M bayes = {X, P(X), D processed , A dyn , P(X|D processed )};

[0051] Among them, X represents the set of environmental state variables, and P(X|D processed ) represents the posterior probability, which is the updated environmental state probability distribution after integrating the latest monitoring data;

[0052] S42. Calculate the initial prior probability P(X i ), and optimize the prior distribution based on the uncertainty weight U i of the data source by using the exponential decay mechanism:

[0053]

[0054] Among them, P(X i ) represents the initial prior probability of the environmental state X i , and α is the adaptive adjustment coefficient;

[0055] S43. Set the dynamic prior adjustment mechanism A dyn , and use the time-weighted optimization strategy to adjust the prior probability of the environmental monitoring data source to obtain the adjusted prior probability P dyn (X i );

[0056] S44. On the basis of the dynamic prior adjustment, use the Markov random field to optimize the correlation between multi-source data, and define the prior probability distribution after collaborative optimization:

[0057]

[0058] Among them, P opt (X) represents the prior probability distribution after collaborative optimization, Z is the normalization factor, ψ(X i , X j ) is the collaborative relationship function between environmental monitoring data sources, and γ is the adjustment parameter to control the uncertainty suppression strength;

[0059] S45. Combine all optimized prior adjustment mechanisms to form the final initial prior probability:

[0060] P init (X i ) = P opt (X i ).

[0061] Optionally, the S43 includes the following steps:

[0062] S431. Set the dynamic prior adjustment mechanism A dyn , and perform time-weighted optimization on the prior probability of the environmental monitoring data source to make it adaptively adjust during the real-time monitoring process. The optimization objective function is as follows:

[0063]

[0064] Among them, is the environmental monitoring data fusion loss function, which measures the error between the current prior probability distribution P(X) and the expected target distribution P target (X). P(X i ) is the current prior probability of the i-th environmental state variable, and P target (X i ) is the prior probability of the environmental state under ideal conditions. represents the uncertainty penalty term, which controls the influence of high-uncertainty data sources on the final probability distribution, and λ is the regularization parameter;

[0065] S432. Calculate the time-weighted prior adjustment amount ΔP(X i ) of the environmental monitoring data source, and define the update rule as follows:

[0066]

[0067] Among them, ΔP(X i ) is the prior adjustment increment of the i-th environmental monitoring data source. is the adjustment term based on gradient optimization, γ1 is the weight adjustment coefficient, T is the time window length, and it is defined to perform dynamic weight adjustment within the past T time steps. w t is the time decay factor;

[0068] S433. Calculate the adaptive weighted data quality correction term ΔP quality (X i ):

[0069]

[0070] Among them, ΔP quality (X i ) is the probability correction term of the i-th environmental monitoring data source based on quality assessment. β is the data quality weight coefficient, which controls the influence of data quality on prior adjustment. P dyn (X i ) - P(X i ) represents the difference between the dynamically adjusted prior probability and the current prior probability, so that data sources with quality higher than the threshold are strengthened during the adjustment process, and the influence of data sources with quality lower than the threshold is weakened;

[0071] S434. Calculate the prior probability P dyn (X i ) of the dynamically adjusted environmental monitoring data source, and define it as follows:

[0072] P dyn (Xi ) = P(X i ) + ΔP(X i ) + ΔP quality (X i );

[0073] Among them, P dyn (X i ) is the prior probability after final dynamic adjustment.

[0074] Optionally, the S5 includes the following steps:

[0075] S51. Based on the final initial prior probability P init (X i ) and the preprocessed environmental monitoring data D processed collected in real time, use Bayes' theorem to calculate the posterior probability distribution:

[0076]

[0077] Among them, P(X i ∣D processed ) represents the posterior probability of the i-th environmental state after obtaining the real-time environmental monitoring data D processed , P(D processed ∣X i ) is the likelihood of observing the data set D i in the environmental state X processed ;

[0078] S52. Calculate the dynamic weight W i of the environmental monitoring data source, and adjust the weight according to the posterior probability and the data uncertainty weight U i :

[0079]

[0080] Among them, W i represents the dynamic weight of the i-th environmental monitoring data source in the data fusion process;

[0081] S53. Calculate the data source credibility correction factor δ i , which is used to dynamically adjust the influence of the data source on the final fusion result:

[0082]

[0083] Among them, δ i represents the credibility correction factor of the i-th environmental monitoring data source, and κ1 is an adjustment coefficient that controls the influence degree of uncertainty penalty;

[0084] S54. Calculate the dynamically updated weight W final,i:

[0085] W final,i =W i ·δ i ;

[0086] Among them, W final,i is the dynamic weight of the environmental monitoring data source in the final data fusion.

[0087] The beneficial effects of the present invention are as follows:

[0088] (1) The present invention introduces an adaptive Bayesian inference model, constructs a fusion model with a dynamic prior adjustment mechanism by combining historical environmental monitoring data and real-time acquisition data, and dynamically adjusts the initial prior probability distribution of each data source by calculating the signal-to-noise ratio, measurement deviation, and time synchronization error indicators of the environmental monitoring data source, enabling the data fusion process to adapt to environmental changes in real time, thereby improving the accuracy of data fusion.

[0089] (2) The present invention proposes a weight optimization mechanism based on uncertainty evaluation. By calculating the signal variance, noise variance, and time synchronization error of each data source, a mathematical model for dynamically adjusting weights is established. The weight optimization mechanism can automatically reduce the influence weight when the reliability of the data source is low, while assigning a higher weight to data sources with high signal-to-noise ratio and high measurement accuracy, thereby significantly improving the stability and anti-interference ability of data fusion.

[0090] (3) On the basis of Bayesian inference, the present invention introduces a Markov random field model to optimize the correlation of multi-source environmental monitoring data, uses the Markov random field to model the collaborative relationship between data sources, and combines the uncertainty weights to jointly optimize the data, ensuring that the probability distribution of the fused environmental state is more accurate. Description of the Drawings

[0091] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0092] Figure 1 is a flowchart of a multi-source environmental monitoring data fusion method based on adaptive Bayesian inference proposed by the present invention. Detailed Embodiments

[0093] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0094] Reference Figure 1, A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference, comprising the following steps:

[0095] S1. Establish a distributed environmental monitoring system, and collect an environmental monitoring data set based on the deployment of multiple environmental monitoring sensors;

[0096] S2. Preprocess the environmental monitoring data set to obtain a preprocessed environmental monitoring data set;

[0097] S3. Evaluate the uncertainty of the preprocessed environmental monitoring data set, and calculate the signal-to-noise ratio and data reliability index of each environmental monitoring data source;

[0098] S4. Construct an adaptive Bayesian inference model based on the characteristics of historical environmental monitoring data and preprocessed environmental monitoring data. The adaptive Bayesian inference model is equipped with a dynamic prior adjustment mechanism. The dynamic prior adjustment mechanism adjusts the initial prior probability distribution of each environmental monitoring data source in real time according to the uncertainty evaluation results to form an initial prior probability;

[0099] S5. Use the adaptive Bayesian inference model with dynamic prior adjustment to fuse the preprocessed environmental monitoring data collected in real time with the initial prior probability, calculate the posterior probability distribution of each environmental monitoring data source through Bayes' theorem, and dynamically update the weight of each data source in the environmental monitoring data fusion process accordingly;

[0100] S6. Perform weighted fusion on each environmental monitoring data source according to the dynamically updated weights to form a fused environmental monitoring data;

[0101] S7. Post-process the fused environmental monitoring data to generate environmental monitoring information for environmental quality assessment, pollution warning, disaster emergency response, and smart city management.

[0102] In this embodiment, S1 includes the following steps:

[0103] S11. Deploy a distributed environmental monitoring system in the target environmental area. The distributed environmental monitoring system consists of N environmental monitoring sensors;

[0104] S12. Regularly collect and construct an environmental monitoring data set from the distributed environmental monitoring system:

[0105] D raw ={d1,d2,...,d M};

[0106] Wherein, d i represents the i-th collected data record, M is the total number of data records, and each data record contains a timestamp t i and monitoring values from different environmental monitoring sensors:

[0107] d i =(t i , T i , H i , P i , C PM,i , C gas,i );

[0108] Among them, t i represents the data recording time, T i represents the measured value of the temperature sensor, H i represents the measured value of the humidity sensor, P i represents the measured value of the barometric pressure sensor, C PM,i represents the measured value of the particulate matter concentration, C gas,i represents the measured value of the gas concentration;

[0109] S13. Set the spatial distribution of the environmental monitoring data, define the monitoring area R, and the spatial coordinates of the environmental monitoring sensors in the monitoring area are:

[0110] xi = (xi, yi, zi);

[0111] Among them, (x i , y i , z i ) represents the three-dimensional spatial coordinates of the i-th sensor in the monitoring area R.

[0112] In this embodiment, S2 includes the following steps:

[0113] S21. Perform data cleaning on the environmental monitoring data set D raw , remove invalid data records, and define a valid environmental monitoring data set;

[0114] S22. Unify the format of the valid environmental monitoring data set, set a unified time format t i and a numerical standardization format to obtain a normalized environmental monitoring data set;

[0115] S23. Perform time synchronization, align the time of the normalized environmental monitoring data set, and calculate the statistical mean μ X and standard deviation σ X of the measured value of each sensor for the synchronized environmental monitoring data set, and define the outlier detection rule as:

[0116] X i is regarded as an outlier if and only if |X i - μ X | > λσ X ;

[0117] Among them, λ is the outlier detection threshold. If the measured value X i exceeds the range, then it is determined that the data point is an outlier and marked as missing;

[0118] S24. Perform missing data completion. For the environmental monitoring data with missing values, time interpolation method is used for completion, and finally the preprocessed environmental monitoring data set D processed .

[0119] In this embodiment, S3 includes the following steps:

[0120] S31. Calculate the signal-to-noise ratio SNR processed of the preprocessed environmental monitoring data set D i :

[0121]

[0122] Among them, is the signal variance of the i-th environmental monitoring data source, is the noise variance of the i-th environmental monitoring data source, and the signal-to-noise ratio SNR i reflects the measurement accuracy of the data source;

[0123] S32. Calculate the measurement deviation D i of each environmental monitoring data source:

[0124]

[0125] Among them, X i,j represents the measured value collected by the i-th environmental monitoring data source at the time stamp t j , μ X is the measurement mean of this data source, and M is the total number of data records of this data source;

[0126] S33. Calculate the time synchronization error E sync,i of the environmental monitoring data source:

[0127]

[0128] Among them, t j is the time stamp of the i-th environmental monitoring data source, t ref,j is the reference time, and the time synchronization error E sync,i reflects the accuracy of the time alignment of the data source;

[0129] S34. Calculate the reliability index R i of the data source:

[0130] R i = w1SNR i + w2(1 - D i ) + w3(1 - Esync,i );

[0131] Among them, w1, w2, and w3 are weight parameters;

[0132] S35. Generate the uncertainty evaluation result of the environmental monitoring data according to the data source reliability index R i Generate the uncertainty evaluation result of the environmental monitoring data, and define the uncertainty weight U i :

[0133]

[0134] Among them, U i represents the uncertainty weight of the i-th environmental monitoring data source, reflecting the uncertainty degree of the environmental monitoring data source in the data fusion process, θ is a positive adjustment index, reflecting the sensitivity of the uncertainty weight to the data source reliability index R i under low reliability conditions, and κ is a non-negative proportionality coefficient used to adjust the logarithmic amplification effect of the data source reliability index R i below the threshold.

[0135] In this embodiment, S4 includes the following steps:

[0136] S41. Set the adaptive Bayesian inference model M bayes , which is used to fuse multi-source environmental monitoring data. The adaptive Bayesian inference model consists of a state variable set X, a prior probability distribution P(X), an environmental monitoring data set D processed and a dynamic adjustment mechanism A dyn :

[0137] M bayes ={X, P(X), D processed , A dyn , P(X|D processed )};

[0138] Among them, X represents the environmental state variable set, and P(X|D processed ) represents the posterior probability, the updated environmental state probability distribution after fusing the latest monitoring data;

[0139] S42. Calculate the initial prior probability P(X i ), and optimize the prior distribution based on the uncertainty weight U i of the data source by using the exponential decay mechanism:

[0140]

[0141] Among them, P(X i ) represents the initial prior probability of the environmental state X i , and α is an adaptive adjustment coefficient;

[0142] S43. Set the dynamic prior adjustment mechanism A dyn , and use the time-weighted optimization strategy to adjust the prior probability of the environmental monitoring data source to obtain the adjusted prior probability P of the environmental state dyn (X i );

[0143] S44. On the basis of dynamic prior adjustment, use the Markov random field to optimize the correlation between multi-source data, and define the prior probability distribution after collaborative optimization:

[0144]

[0145] Among them, P opt (X) represents the prior probability distribution after collaborative optimization, Z is the normalization factor, ψ(X i , X j ) is the collaborative relationship function between environmental monitoring data sources, and γ is the adjustment parameter to control the uncertainty suppression intensity;

[0146] S45. Combine all optimized prior adjustment mechanisms to form the final initial prior probability:

[0147] P init (X i ) = P opt (X i ).

[0148] In this embodiment, S43 includes the following steps:

[0149] S431. Set the dynamic prior adjustment mechanism A dyn , and perform time-weighted optimization on the prior probability of the environmental monitoring data source to make it adaptively adjust during real-time monitoring. The optimization objective function is as follows:

[0150]

[0151] Among them, is the environmental monitoring data fusion loss function, which measures the error between the current prior probability distribution P(X) and the expected target distribution P target (X), P(X i ) is the current prior probability of the i-th environmental state variable, and P target (X i ) is the prior probability of the environmental state under ideal conditions, represents the uncertainty penalty term, which controls the influence of high-uncertainty data sources on the final probability distribution, and λ is the regularization parameter;

[0152] S432. Calculate the time-weighted prior adjustment amount ΔP(X i) The update rules are defined as follows:

[0153]

[0154] Among them, ΔP(X i ) is the prior adjustment increment of the i-th environmental monitoring data source, is the adjustment term based on gradient optimization, γ1 is the weight adjustment coefficient, T is the time window length, and it is defined that dynamic weight adjustment is performed within the past T time steps, w t is the time decay factor;

[0155] S433. Calculate the adaptive weighted data quality correction term ΔP of the environmental monitoring data source quality (X i ):

[0156]

[0157] Among them, ΔP quality (X i ) is the probability correction term of the i-th environmental monitoring data source based on quality assessment, β is the data quality weight coefficient, which controls the influence of data quality on prior adjustment, P dyn (X i ) - P(X i ) represents the difference between the prior probability after dynamic adjustment and the current prior probability, so that the data sources with quality higher than the threshold are strengthened during the adjustment process, and the influence of data sources with quality lower than the threshold is weakened;

[0158] S434. Calculate the prior probability P of the environmental monitoring data source after dynamic adjustment dyn (X i ), which is defined as follows:

[0159] P dyn (X i ) = P(X i ) + ΔP(X i ) + ΔP quality (X i );

[0160] Among them, P dyn (X i ) is the prior probability after the final dynamic adjustment.

[0161] In this embodiment, S5 includes the following steps:

[0162] S51. Based on the final initial prior probability P init (X i ) and the preprocessed environmental monitoring data D processed collected in real time, use Bayes' theorem to calculate the posterior probability distribution:

[0163]

[0164] Among them, P(X i ∣D processed ) represents the posterior probability of the i-th environmental state after obtaining the real-time environmental monitoring data D processed , P(D processed ∣X i ) is the likelihood of observing the data set D i under the environmental state X processed ;

[0165] S52. Calculate the dynamic weight W i of the environmental monitoring data source, and adjust the weight according to the posterior probability and the data uncertainty weight U i :

[0166]

[0167] Among them, W i represents the dynamic weight of the i-th environmental monitoring data source in the data fusion process;

[0168] S53. Calculate the data source credibility correction factor δ i , which is used to dynamically adjust the influence of the data source on the final fusion result:

[0169]

[0170] Among them, δ i represents the data source credibility correction factor of the i-th environmental monitoring data source, and κ1 is an adjustment coefficient to control the influence degree of the uncertainty penalty;

[0171] S54. Calculate the dynamically updated weight W final,i finally used for environmental monitoring data fusion:

[0172] W final,i = W i · δ i ;

[0173] Among them, W final,i is the dynamic weight of the environmental monitoring data source in the final data fusion.

[0174] The present invention introduces an adaptive Bayesian inference model, constructs a fusion model with a dynamic prior adjustment mechanism by combining historical environmental monitoring data and real-time acquisition data, and dynamically adjusts the initial prior probability distribution of each data source by calculating the signal-to-noise ratio, measurement deviation, and time synchronization error indexes of the environmental monitoring data source, so that the data fusion process can adapt to environmental changes in real time, thereby improving the accuracy of data fusion.

[0175] The present invention proposes a weight optimization mechanism based on uncertainty assessment. By calculating the signal variance, noise variance, and time synchronization error of each data source, a mathematical model for dynamically adjusting weights is established. The weight optimization mechanism can automatically reduce the influence weight of a data source when its reliability is low, while assigning a higher weight to a data source with high signal-to-noise ratio and high measurement accuracy, thereby significantly improving the stability and anti-interference ability of data fusion.

[0176] Based on Bayesian inference, the present invention introduces a Markov random field model to optimize the correlation of multi-source environmental monitoring data. The Markov random field is used to model the collaborative relationship between data sources, and the data is jointly optimized in combination with uncertainty weights to ensure that the probability distribution of the fused environmental state is more accurate.

[0177] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered by the protection scope of the present invention.

Claims

1. A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference, characterized in that, It includes the following steps: S1. Establish a distributed environment monitoring system, and collect an environmental monitoring data set based on the deployment of multiple environmental monitoring sensors; S2. Preprocess the environmental monitoring data set to obtain a preprocessed environmental monitoring data set; S3. Conduct uncertainty assessment on the preprocessed environmental monitoring data set, and calculate the signal-to-noise ratio and data reliability index of each environmental monitoring data source; S4. Construct an adaptive Bayesian inference model based on historical environmental monitoring data and the characteristics of the preprocessed environmental monitoring data. The adaptive Bayesian inference model is equipped with a dynamic prior adjustment mechanism, and the dynamic prior adjustment mechanism adjusts the initial prior probability distribution of each environmental monitoring data source in real time according to the uncertainty assessment results to form an initial prior probability; S5. Use the adaptive Bayesian inference model with dynamic prior adjustment to fuse the preprocessed environmental monitoring data collected in real time with the initial prior probability, calculate the posterior probability distribution of each environmental monitoring data source through Bayes' theorem, and dynamically update the weight of each data source in the process of environmental monitoring data fusion accordingly; S6. Perform weighted fusion on each environmental monitoring data source according to the dynamically updated weights to form a fused environmental monitoring data; S7. Post-process the fused environmental monitoring data to generate environmental monitoring information for environmental quality assessment, pollution warning, disaster emergency response, and smart city management.

2. The multi-source environmental monitoring data fusion method based on adaptive Bayesian inference according to claim 1, wherein, The S1 includes the following steps: S11. Deploy a distributed environment monitoring system in the target environmental area. The distributed environment monitoring system consists of N environmental monitoring sensors; S12. Regularly collect and construct an environmental monitoring data set from the distributed environment monitoring system: D raw = {d1, d2,..., d M}; Among them, d i represents the i-th collected data record, M is the total number of data records, and each data record contains a timestamp t i and monitoring values from different environmental monitoring sensors: d i = (t i , T i , H i , P i , C PM,i , C gas,i ); Among them, t i represents the data recording time, T i represents the measured value of the temperature sensor, H i represents the measured value of the humidity sensor, P i represents the measured value of the barometric pressure sensor, C PM,i represents the measured value of the particulate matter concentration, C gas,i represents the measured value of the gas concentration; S13. Set the spatial distribution of the environmental monitoring data, define the monitoring area R, and the spatial coordinates of the environmental monitoring sensors in the monitoring area are: xi = (xi, yi, zi); Among them, (x i , y i , z i ) represents the three-dimensional spatial coordinates of the i-th sensor within the monitoring area R.

3. A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference according to claim 1, wherein The S2 includes the following steps: S21. Clean the environmental monitoring data set D raw to remove invalid data records and define a valid environmental monitoring data set; S22. Unify the format of the effective environmental monitoring data set and set a unified time format t i and the numerical normalization format to obtain a normalized environmental monitoring data set; S23. Perform time synchronization, align the normalized environmental monitoring data set in terms of time, and calculate the statistical mean μ of the measured values of each sensor for the synchronized environmental monitoring data set X and the standard deviation σ X , and define the outlier detection rule as: X i is regarded as an outlier if and only if |X i - μ X | > λσ X ; where λ is the outlier detection threshold. If the measured value X i exceeds the range, the data point is determined to be an outlier and marked as missing; S24. Perform missing data completion. For the environmental monitoring data with missing values, use time interpolation method for completion, and finally generate the preprocessed environmental monitoring dataset D processed .

4. A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference according to claim 3, characterized in that The S3 includes the following steps: S31. Calculate the signal-to-noise ratio SNR of the preprocessing environment monitoring data set D processed i :​ Among them, is the signal variance of the i-th environmental monitoring data source, is the noise variance of the i-th environmental monitoring data source, and the signal-to-noise ratio SNR i reflects the measurement accuracy of the data source; S32. Calculate the measurement deviation D of each environmental monitoring data source i : Among them, X i,j represents the measured value collected by the i-th environmental monitoring data source at timestamp t j , μ X is the mean value of the measurements of this data source, and M is the total number of data records of this data source; S33. Calculate the time synchronization error E of the computing environment monitoring data source sync,i : where t j is the timestamp of the i-th environmental monitoring data source, and t ref,j is the reference time, and the time synchronization error E sync,i reflects the accuracy of the time alignment of the data source; S34. Calculate the reliability index R of the data source i : R i = w1SNR i + w2(1 - D i ) + w3(1 - E sync,i ); Among them, w1, w2, and w3 are weight parameters; S35. According to the data source reliability index R i Generate the evaluation result of the uncertainty of environmental monitoring data, and define the uncertainty weight U i : Among them, U i represents the uncertainty weight of the i-th environmental monitoring data source, reflecting the degree of uncertainty of the environmental monitoring data source in the data fusion process. θ is a positive adjustment index, reflecting the sensitivity of the uncertainty weight to the data source reliability index R i under low reliability conditions. κ is a non-negative proportionality coefficient used to adjust the logarithmic amplification effect of the data source reliability index R i below the threshold.

5. A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference according to claim 1, characterized in that The S4 includes the following steps: S41. Set the adaptive Bayesian inference model M bayes , which is used to fuse multi-source environmental monitoring data. The adaptive Bayesian inference model consists of a set of state variables X, a prior probability distribution P(X), an environmental monitoring data set D processed and a dynamic adjustment mechanism A dyn : M bayes = {X, P(X), D processed , A dyn , P(X|D processed )}; Among them, X represents the set of environmental state variables, and P(X∣D processed ) represents the posterior probability, which is the updated probability distribution of the environmental state after integrating the latest monitoring data; S42. Calculate the initial prior probability P(X i ), and optimize the prior distribution using an exponential decay mechanism based on the uncertainty weight U i of the data source: where P(X i ) represents the initial prior probability of the environmental state X i , and α is an adaptive adjustment coefficient; S43. Set the dynamic prior adjustment mechanism A dyn , and use the time-weighted optimization strategy to adjust the prior probability of the environmental monitoring data source to obtain the adjusted prior probability P of the environmental state dyn (X i ); S44. On the basis of dynamic prior adjustment, use a Markov random field to optimize the correlation between multi-source data, and define a jointly optimized prior probability distribution: where P opt (X) represents the co-optimized prior probability distribution, Z is the normalization factor, and ψ(X i , X j ) is the co-relationship function between environmental monitoring data sources, and γ is the adjustment parameter that controls the intensity of uncertainty suppression; S45. Combine all optimized prior adjustment mechanisms to form a final initial prior probability: P init (X i ) = P opt (X i ).

6. A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference according to claim 5, characterized in that, The S43 includes the following steps: S431. Set the dynamic prior adjustment mechanism A dyn , perform time-weighted optimization on the prior probability of the environmental monitoring data source, enabling it to adaptively adjust during real-time monitoring. The optimization objective function is as follows: Among them, is the environmental monitoring data fusion loss function, which measures the error between the current prior probability distribution P(X) and the expected target distribution P target (X). P(X i ) is the current prior probability of the i-th environmental state variable, and P target (X i ) is the prior probability of the environmental state under ideal conditions. represents the uncertainty penalty term, which controls the influence of high-uncertainty data sources on the final probability distribution, and λ is the regularization parameter; S432. Calculate the time-weighted prior adjustment amount ΔP(X i ) of the computing environment monitoring data source, and define the update rule as follows: Among them, ΔP(X i ) is the prior adjustment increment of the i-th environmental monitoring data source, is the adjustment term based on gradient optimization, γ1 is the weight adjustment coefficient, T is the time window length, and it is defined that dynamic weight adjustment is performed within the past T time steps, w t is the time decay factor; S433. Calculate the adaptive weighted data quality correction term ΔP of the computing environment monitoring data source quality (X i ): Among them, ΔP quality (X i ) is the probability correction term based on quality assessment for the i-th environmental monitoring data source, β is the data quality weight coefficient, which controls the influence of data quality on prior adjustment, P dyn (X i ) - P(X i ) represents the difference between the prior probability after dynamic adjustment and the current prior probability, so that the data sources with quality higher than the threshold are strengthened during the adjustment process, and the influence of the data sources with quality lower than the threshold is weakened; S434. Calculate the prior probability P of the environment monitoring data source after dynamic adjustment dyn (X i ), which is defined as follows: P dyn (X i ) = P(X i ) + ΔP(X i ) + ΔP quality (X i ); Among them, P dyn (X i ) is the prior probability after the final dynamic adjustment.

7. A multi-source environmental monitoring data fusion method based on adaptive Bayesian inference according to claim 5, characterized in that The S5 includes the following steps: S51. Based on the final initial prior probability P init (X i ) and the preprocessed environmental monitoring data D collected in real time processed , the posterior probability distribution is calculated using Bayes' theorem: Among them, P(X i ∣D processed ) represents the posterior probability of the i-th environmental state after obtaining the real-time environmental monitoring data D processed , and P(D processed ∣X i ) is the likelihood of observing the data set D i under the environmental state X processed ; S52. Calculate the dynamic weight W of the data source for computing environment monitoring i , and adjust the weight according to the posterior probability and the data uncertainty weight U i as follows: Among them, W i represents the dynamic weight of the i-th environmental monitoring data source in the data fusion process; S53. Calculate the credibility correction factor δ of the data source i , which is used to dynamically adjust the influence of the data source on the final fusion result: Among them, δ i represents the credibility correction factor of the i-th environmental monitoring data source, and κ1 is an adjustment coefficient to control the influence degree of the uncertainty penalty; S54. Calculate the dynamically updated weight W finally used for environmental monitoring data fusion final,i : W final,i = W i ·δ i ; Among them, W final,i is the dynamic weight of the environmental monitoring data source in the final data fusion.

Images