A smart campus AIOT big data visual analysis method

Abstract

The application relates to the technical field of Internet of Things data processing and artificial intelligence application, in particular to a smart campus AIOT big data visual analysis method, which comprises the following steps: collecting heterogeneous behavior data; performing Z-Score standardization processing on the heterogeneous behavior data to generate dimensionless standard scores; constructing a behavior state matrix; generating compliance entropy; calculating an AI intervention confidence score; comparing the AI intervention confidence score with a preset dynamic decision threshold; if the AI intervention confidence score is higher than the dynamic decision threshold, performing AI precise induction intervention; otherwise, performing space mapping calculation on the compliance entropy to generate an administrative intervention resource allocation amount; calculating the difference between the number of false positives and the system expected false positives; using the updated data entropy penalty factor for AI intervention confidence score calculation at the next time step; and the application solves the problem of high false positive rate caused by the lack of logical verification in the prior art, and ensures the robustness and reliability of early warning decision.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) data processing and artificial intelligence (AI) application technology, specifically to a smart campus AIoT big data visualization and analysis method. Background Technology

[0002] Big data analytics in smart campus IoT is a key element in improving campus management efficiency and security, directly impacting the real-time nature and accuracy of student behavior monitoring. This system encompasses multi-source heterogeneous data from consumption, access control, and utilities, enabling risk warnings and resource allocation through data collection and processing. Effectively integrating multimodal data, accurately identifying abnormal behavior, and optimizing intervention strategies have become key research areas in smart education. Existing technologies primarily rely on single-dimensional threshold judgments or simple statistical rules, failing to fully utilize the logical connections between heterogeneous data for comprehensive decision-making. They also struggle to address issues of inconsistent physical dimensions and inherent sensor noise interference, resulting in insufficient consistency verification capabilities for multi-source data. Furthermore, static decision-making logic is ill-suited to the dynamic evolution of adversarial behavior, unable to distinguish between device errors and genuine anomalies, and lacks a closed-loop adaptive adjustment mechanism based on false alarm feedback. In complex and ever-changing campus scenarios, these technologies suffer from high false alarm rates, insufficiently targeted intervention measures, and unreasonable allocation of administrative resources. Therefore, a solution is urgently needed to address the problems existing in current technologies. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a smart campus AIoT big data visualization and analysis method. Specifically, the technical solution of this invention includes:

[0004] S1. Collect heterogeneous behavioral data from campus IoT gateways. The heterogeneous behavioral data includes consumption data, access control data, and water and electricity data. Perform Z-Score standardization on the heterogeneous behavioral data to generate dimensionless standard scores. Construct a behavioral state matrix based on the dimensionless standard scores.

[0005] S2. Obtain the preset sensor inherent noise coefficient; based on the sensor inherent noise coefficient, calculate the consistency coefficient between different modal data in the behavior state matrix; input the consistency coefficient into the entropy evolution model to generate the entropy.

[0006] S3. Obtain the behavioral risk value output by the external AI prediction module; obtain the current data entropy penalty factor; calculate the AI ​​intervention confidence score based on the behavioral risk value, compliance entropy, and data entropy penalty factor;

[0007] S4. Compare the AI ​​intervention confidence score with the preset dynamic decision threshold; if the AI ​​intervention confidence score is higher than the dynamic decision threshold, execute AI-guided precise intervention; if the AI ​​intervention confidence score is lower than or equal to the dynamic decision threshold, perform spatial mapping calculation on the obedience entropy to generate the administrative intervention resource allocation amount.

[0008] S5. Obtain the feedback of the number of false alarms after manual verification; calculate the difference between the feedback of the number of false alarms and the system's expected number of false alarms; update the data entropy penalty factor based on the difference; use the updated data entropy penalty factor to calculate the AI ​​intervention confidence score in the next time step.

[0009] Optionally, S1 specifically includes:

[0010] S11. Obtain historical baseline data of the student population from the previous semester from the cloud database. The historical baseline data includes the historical mean and standard deviation.

[0011] S12. Using historical mean and standard deviation, map consumption data to standard scores in the monetary dimension; map access control data to standard scores in the time frequency dimension; map water and electricity data to standard scores in the usage dimension.

[0012] S13. Generate a behavioral state matrix using the standard scores for the combined amount dimension, the standard scores for the time frequency dimension, and the standard scores for the usage dimension.

[0013] S2 specifically includes:

[0014] S21. Determine the inherent noise figure of the sensor based on the equipment's factory calibration data;

[0015] S22. Based on the sensor's inherent noise figure and the preset system sensitivity adjustment gain, the consistency coefficient between any two types of dimensionless standard scores is calculated using the weighted Euclidean distance algorithm.

[0016] S23. Set the dimensionless standard score corresponding to the access control data as the benchmark mode; calculate the consistency coefficient between the dimensionless standard score of other modes and the benchmark mode;

[0017] S24. Input the consistency coefficients corresponding to the baseline mode into the entropy evolution model.

[0018] Optionally, the entropy evolution model in S2 adopts the following logic:

[0019] S25. Based on the confidence weights of different modalities, the logarithmic values ​​of the consistency coefficients are weighted and summed to generate a static quality term;

[0020] S26. Perform differential calculation on the rate of change of the consistency coefficient over time, and generate a dynamic trend term by combining it with the preset trend-weighted time constant.

[0021] S27. Combine the static quality term and the dynamic trend term to generate the degree entropy.

[0022] Optionally, S3 specifically includes:

[0023] S31. Obtain the preset baseline risk sensitivity coefficient and entropy value;

[0024] S32. In response to the entropy of the degree of obedience being less than the entropy baseline, the data entropy penalty factor is set to zero; in response to the entropy of the degree of obedience being greater than or equal to the entropy baseline, the data entropy penalty factor remains unchanged.

[0025] S33. Calculate the product of the behavioral risk value and the basic risk sensitivity coefficient, and use it as the numerator of the utility function;

[0026] S34. Calculate the difference between the obedience entropy and the entropy baseline; calculate the product of the difference and the data entropy penalty factor; add one to the product and use it as the denominator of the utility function.

[0027] S35. Calculate the ratio of the numerator to the denominator of the utility function to generate the confidence score for AI intervention.

[0028] Optionally, the specific allocation of administrative intervention resources generated in S4 includes:

[0029] S41. Obtain the physical coordinates of the individual students that generate the degree entropy;

[0030] S42. Using the Gaussian kernel density estimation method, the discrete degree entropy is mapped to a continuous spatial entropy field based on physical coordinates.

[0031] S43. Identify regions where the field value in the spatial entropy field exceeds a preset critical threshold.

[0032] S44. Based on the preset intervention density coefficient, perform double integral calculation on the entropy value in the area exceeding the critical threshold to generate the administrative intervention resource allocation amount.

[0033] Optionally, S5 specifically includes:

[0034] S51. Calculate the number of false alarms by subtracting the system's expected number of false alarms from the number of false alarms reported.

[0035] S52. Multiply the result by the preset learning rate to generate the correction increment;

[0036] S53. The correction increment is added to the current data entropy penalty factor to generate the updated data entropy penalty factor.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. This invention eliminates differences in physical dimensions and constructs a unified behavioral state matrix by standardizing heterogeneous behavioral data such as consumption, access control, and water and electricity. By combining the inherent noise coefficient of the sensor to calculate the consistency coefficient between different modes, it effectively distinguishes between equipment hardware errors and real behavioral anomalies. This solves the problem of difficulty in multi-source data fusion caused by inconsistent physical dimensions and noise interference in the prior art, and improves the accuracy of heterogeneous data analysis.

[0039] 2. This invention constructs a degree entropy evolution model that includes static quality items and dynamic trend items, which can simultaneously quantify the logical self-consistency and the rate of change of consistency coefficient over time of multi-source data. This model not only evaluates the static quality of data, but also captures the dynamic evolution trend of adversarial behavior, overcoming the limitations of existing technologies that rely on static threshold judgment, and significantly enhancing the system's ability to identify and warn of complex abnormal behaviors and logical deception behaviors.

[0040] 3. This invention introduces a game theory-based decision-making mechanism, which integrates the behavioral risk value predicted by external artificial intelligence with the obedience degree entropy to calculate the intervention confidence score, and compares it with the dynamic decision threshold calculated in real time based on the global risk volatility. This dual verification mechanism effectively suppresses the risk of misjudgment by a single artificial intelligence model, solves the problem of high false alarm rate caused by lack of logical verification in the prior art, and ensures the robustness and reliability of early warning decisions.

[0041] 4. This invention establishes a closed-loop adaptive adjustment mechanism based on false alarm feedback, dynamically updating the data entropy penalty factor according to the number of false alarms, thus realizing the self-learning of model weights; at the same time, it uses Gaussian kernel density estimation to map discrete entropy values ​​into a continuous spatial field, guiding the precise allocation of administrative intervention resources, solving the problems of weak targeting of intervention measures and unreasonable resource allocation in existing technologies, and improving the intelligence and automation level of campus safety management. Attached Figure Description

[0042] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0043] Figure 1 This is a flowchart illustrating a smart campus AIoT big data visualization and analysis method according to the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0045] Example 1:

[0046] Please see Figure 1 A smart campus AIoT big data visualization and analysis method includes the following steps:

[0047] S1. Collect heterogeneous behavioral data from campus IoT gateways. The heterogeneous behavioral data includes consumption data, access control data, and water and electricity data. Perform Z-Score standardization on the heterogeneous behavioral data to generate dimensionless standard scores. Construct a behavioral state matrix based on the dimensionless standard scores.

[0048] S2. Obtain the preset sensor inherent noise coefficient; based on the sensor inherent noise coefficient, calculate the consistency coefficient between different modal data in the behavior state matrix; input the consistency coefficient into the entropy evolution model to generate the entropy.

[0049] S3. Obtain the behavioral risk value output by the external AI prediction module; obtain the current data entropy penalty factor; calculate the AI ​​intervention confidence score based on the behavioral risk value, compliance entropy, and data entropy penalty factor;

[0050] S4. Compare the AI ​​intervention confidence score with the preset dynamic decision threshold; if the AI ​​intervention confidence score is higher than the dynamic decision threshold, execute AI-guided precise intervention; if the AI ​​intervention confidence score is lower than or equal to the dynamic decision threshold, perform spatial mapping calculation on the obedience entropy to generate the administrative intervention resource allocation amount.

[0051] S5. Obtain the feedback of the number of false alarms after manual verification; calculate the difference between the feedback of the number of false alarms and the system's expected number of false alarms; update the data entropy penalty factor based on the difference; use the updated data entropy penalty factor to calculate the AI ​​intervention confidence score in the next time step.

[0052] This embodiment discloses a smart campus AIoT big data visualization and analysis method. This method achieves closed-loop processing of heterogeneous campus behavioral data by combining side-channel consistency verification with game theory decision-making mechanism.

[0053] Heterogeneous behavioral data is collected from campus IoT gateways. This heterogeneous behavioral data specifically refers to student behavior records from different sources and with different physical dimensions, including consumption data, access control data, and water and electricity data. Z-Score standardization is performed on the heterogeneous behavioral data to eliminate differences in physical dimensions and generate dimensionless standard scores. A behavioral state matrix is ​​constructed based on the dimensionless standard scores.

[0054] Obtain the preset sensor intrinsic noise coefficient; this coefficient is a dimensionless constant characterizing the measurement accuracy limit of a specific sensor model; based on the sensor intrinsic noise coefficient, calculate the consistency coefficient between different modal data in the behavior state matrix, which is used to quantify the temporal coupling degree of different behavioral logics; input the consistency coefficient into the entropy evolution model to generate the entropy; this entropy not only reflects the static quality of the data, but also includes the dynamic trend characteristics of adversarial behavior;

[0055] The behavioral risk value output by the external AI prediction module is obtained. The external AI prediction module shares heterogeneous behavioral data from the same source as step S1 of this method as input, and uses a deep neural network, such as LSTM or Transformer, to predict the probability value of abnormal behavior occurring at the current moment based on historical behavioral sequences as the behavioral risk value; and the current data entropy penalty factor; the data entropy penalty factor is a dimensionless variable used to dynamically adjust the weights of the AI ​​model. The initial value of the data entropy penalty factor is preset to 1.0, and it is dynamically updated according to the false alarm feedback mechanism in the subsequent step S5; based on the behavioral risk value, obedience entropy, and data entropy penalty factor, the AI ​​intervention confidence score is calculated; this score is a decision variable characterizing the degree of confidence of the system in the AI ​​prediction results;

[0056] The AI ​​intervention confidence score is compared with a preset dynamic decision threshold. If the AI ​​intervention confidence score is higher than the dynamic decision threshold, precise AI-guided intervention is executed. This precise AI-guided intervention specifically includes: sending a warning message to the student's linked mobile terminal, or generating an instruction to temporarily reduce the data collection sampling interval of the student's corresponding IoT gateway to obtain high-frequency behavioral data, or sending an instruction to the access control system to lock access to a specific area. If the AI ​​intervention confidence score is lower than or equal to the dynamic decision threshold, spatial mapping calculation is performed on the obedience entropy to generate the allocation of administrative intervention resources, thereby guiding the precise deployment of offline physical intervention resources.

[0057] Obtain the number of false alarms after manual verification; calculate the difference between the number of false alarms reported and the system's expected number of false alarms; update the data entropy penalty factor based on this difference; and use the updated data entropy penalty factor to calculate the AI ​​intervention confidence score for the next time step.

[0058] Example 2:

[0059] S1 specifically includes:

[0060] S11. Obtain historical baseline data of the student population from the previous semester from the cloud database. The historical baseline data includes the historical mean and standard deviation.

[0061] S12. Using historical mean and standard deviation, map consumption data to standard scores in the monetary dimension; map access control data to standard scores in the time frequency dimension; map water and electricity data to standard scores in the usage dimension.

[0062] S13. Generate a behavioral state matrix using the standard scores for the combined amount dimension, the standard scores for the time frequency dimension, and the standard scores for the usage dimension.

[0063] This embodiment illustrates the specific logic of data standardization processing;

[0064] Historical baseline data of the student population from the previous semester was obtained from a cloud database; this historical baseline data includes the historical mean. with standard deviation ;

[0065] Using historical averages with standard deviation The following mapping calculations are performed: Subtract the historical mean of the consumption category from the consumption data and divide by the standard deviation of the consumption category to map to the standard score of the amount dimension; Subtract the historical mean of the frequency category from the access control data and divide by the standard deviation of the frequency category to map to the standard score of the time frequency dimension; Subtract the historical mean of the consumption category from the water and electricity data and divide by the standard deviation of the consumption category to map to the standard score of the usage dimension.

[0066] By combining the standard scores from the monetary dimension, the time frequency dimension, and the usage dimension, a behavioral state matrix is ​​generated. .

[0067] Example 3:

[0068] S2 specifically includes:

[0069] S21. Determine the inherent noise figure of the sensor based on the equipment's factory calibration data;

[0070] S22. Based on the sensor's inherent noise figure and the preset system sensitivity adjustment gain, the consistency coefficient between any two types of dimensionless standard scores is calculated using the weighted Euclidean distance algorithm.

[0071] S23. Set the dimensionless standard score corresponding to the access control data as the benchmark mode; calculate the consistency coefficient between the dimensionless standard score of other modes and the benchmark mode;

[0072] S24. Input the consistency coefficients corresponding to the baseline mode into the entropy evolution model.

[0073] This embodiment illustrates the calculation logic of the consistency coefficient and the calibration method of the sensor's inherent noise figure;

[0074] Determine the sensor's inherent noise figure based on the device's factory calibration data. Specifically, a calibration dataset independent of the runtime data is selected. This dataset contains samples of measurement errors from sensors in a silent state. Through calculation Standard deviation Using the historical standard deviation of the corresponding sensor category in Example 2 Normalize it and take As the inherent noise figure of the sensor, if If it is zero, then... Set to a preset minimum positive number, such as This formula ensures that when historical data fluctuations are small, the noise figure is low. This decreases accordingly, thereby increasing the system's sensitivity to minor deviations in that mode, conforming to anomaly detection logic, and ensuring... It only characterizes the device's hardware features and covers 99.7% of the inherent noise range;

[0075] Based on the sensor's inherent noise figure Adjust the gain with the preset system sensitivity. The consistency coefficient between any two classes of dimensionless standard scores is calculated using the weighted Euclidean distance algorithm. The calculation formula is as follows:

[0076]

[0077] in, , These are the normalized real-time values ​​of the behavioral features, derived from the behavioral state matrix generated in the preceding steps. , For the first , The inherent normalized noise figure of the sensor is a dimensionless constant calculated based on the device factory calibration data and the standard deviation under silent conditions in step S21. The sensitivity adjustment gain of the system is used to control the steepness of the decision function, and its value ranges from 0.5 to 5.0; in this embodiment, The preferred value is 1.0 to balance the sensitivity of the function to small differences;

[0078] The dimensionless standard score corresponding to the access control data is directly set as the reference modality (ref), because access control data usually has high physical entity reliability; the consistency coefficient between the dimensionless standard scores of other modalities and the reference modality is calculated. The consistency coefficients corresponding to the baseline mode are input into the degree entropy evolution model.

[0079] Example 4:

[0080] The obedience-entropy evolution model in S2 adopts the following logic:

[0081] S25. Based on the confidence weights of different modalities, the logarithmic values ​​of the consistency coefficients are weighted and summed to generate a static quality term;

[0082] S26. Perform differential calculation on the rate of change of the consistency coefficient over time, and generate a dynamic trend term by combining it with the preset trend-weighted time constant.

[0083] S27. Combine the static quality term and the dynamic trend term to generate the degree entropy.

[0084] It should be noted that the obedience entropy in this invention borrows the concept of information entropy to measure the uncertainty of the logical consistency state of multimodal data and characterize the degree of disorder in the logical association of the Internet of Things system.

[0085] This embodiment illustrates the internal logic of the entropy evolution model;

[0086] Confidence weights based on different modalities The logarithmic values ​​of the consistency coefficients are weighted and summed to generate a static quality term; this term quantifies the degree of multi-source logical consistency of the data at the current moment.

[0087] The rate of change of the consistency coefficient over time is differentiated and calculated, combined with a preset trend-weighted time constant. Generate dynamic trend items; It has a time dimension to cancel out the reciprocal time dimension generated by the differentiation operation, ensuring that the physical meaning is consistent;

[0088] By combining static quality terms and dynamic trend terms, a degree entropy-compliant function is generated. Its calculation model is as follows:

[0089]

[0090] in, For example, take a preset numerical stability constant. This is used to prevent logarithmic calculation overflow caused by the consistency coefficient approaching zero; differential term In discrete-time systems, first-order backward difference computation is used, i.e. , The system's preset data sampling and update cycle; For the first The confidence weights of the modal class satisfy the following conditions: ; This represents the total number of modes involved in the calculation. This is the consistency coefficient for real-time input.

[0091] Example 5:

[0092] S3 specifically includes:

[0093] S31. Obtain the preset baseline risk sensitivity coefficient and entropy value;

[0094] S32. In response to the entropy of the degree of obedience being less than the entropy baseline, the data entropy penalty factor is set to zero; in response to the entropy of the degree of obedience being greater than or equal to the entropy baseline, the data entropy penalty factor remains unchanged.

[0095] S33. Calculate the product of the behavioral risk value and the basic risk sensitivity coefficient, and use it as the numerator of the utility function;

[0096] S34. Calculate the difference between the obedience entropy and the entropy baseline; calculate the product of the difference and the data entropy penalty factor; add one to the product and use it as the denominator of the utility function.

[0097] S35. Calculate the ratio of the numerator to the denominator of the utility function to generate the confidence score for AI intervention.

[0098] This embodiment illustrates the calculation process of the confidence score for AI intervention;

[0099] Obtain the preset basic risk sensitivity coefficient With entropy baseline ; The upper limit of the entropy value that the system can tolerate is set by the system administrator based on historical false alarm rate statistics;

[0100] Response to entropy Less than the entropy baseline Define the temporary penalty coefficients involved in this utility function calculation. Zero, but maintaining the data entropy penalty factor for global storage. The value remains unchanged, and no zeroing operation is performed; in response to the requirement that the entropy is greater than or equal to the entropy baseline, the temporary penalty coefficient is set. Equal to the current global storage data entropy penalty factor ;

[0101] Calculate behavioral risk value With basic risk sensitivity coefficient The product of these is used as the numerator of the utility function; the entropy is calculated. With entropy baseline The difference; calculate this difference and the data entropy penalty factor. The product of ; add one to this product and use it as the denominator of the utility function;

[0102] Calculate the ratio of the numerator to the denominator of the utility function to generate a confidence score for AI intervention. Its mathematical model is as follows:

[0103]

[0104] in, Probability of behavioral risk prediction derived from external AI prediction modules; This represents the ReLU logical operator, ensuring that the penalty mechanism is activated only when the entropy value exceeds the limit; further, the dynamic decision threshold involved in step S4 above... It is not a fixed constant, but is calculated in real time based on the current global risk volatility of the system. The calculation formula is as follows:

[0105]

[0106] in, This is a preset static baseline threshold, for example, a value of 0.8; This is the fluctuation sensitivity coefficient, for example, with a value of 0.15; This is the standard deviation of the behavioral risk values ​​of all monitored objects within a sliding time window, such as 30 minutes. This formula is used to automatically raise the threshold for AI intervention and reduce the probability of misjudgment when there are drastic fluctuations in group risk.

[0107] Example 6:

[0108] The specific components of the administrative intervention resource allocation generated in S4 include:

[0109] S41. Obtain the physical coordinates of the individual students that generate the degree entropy;

[0110] S42. Using the Gaussian kernel density estimation method, the discrete degree entropy is mapped to a continuous spatial entropy field based on physical coordinates.

[0111] S43. Identify regions where the field value in the spatial entropy field exceeds a preset critical threshold.

[0112] S44. Based on the preset intervention density coefficient, perform double integral calculation on the entropy value in the area exceeding the critical threshold to generate the administrative intervention resource allocation amount.

[0113] This embodiment illustrates the method for calculating the amount of administrative intervention resources allocated;

[0114] Obtain the physical coordinates of individual students that exhibit degree entropy. Specifically, the consistency coefficients of all modes constituting the current entropy of the student are compared, and the physical location of the sensor terminal corresponding to the mode with the lowest consistency coefficient is selected as the... If the sensor corresponding to this mode is a device that cannot verify the real-time presence of a person, such as online consumption records or water and electricity readings automatically collected by the backend, or if this mode cannot provide real-time physical coordinates, then the student's latest access control record or campus network wireless access point (AP) location within the most recent time window (e.g., within 15 minutes) will be retrieved, and its coordinates will be used as... This ensures that administrative intervention resources are precisely targeted to the physical area where the student is located and where the anomaly is most pronounced;

[0115] Using the Gaussian kernel density estimation method, discrete parameters are expressed based on physical coordinates and entropy. Mapped to a continuous spatial entropy field The calculation formula is as follows:

[0116]

[0117] in, For Gaussian kernel function, To smooth the bandwidth parameter, this step smooths the discrete point data into a dimensionless field intensity distribution.

[0118] Identify when the field value in the spatial entropy field exceeds a preset critical threshold. The area;

[0119] Based on a preset intervention density coefficient The entropy value in the region exceeding the critical threshold is calculated using a double integral to generate the administrative intervention resource allocation amount. The calculation formula is as follows:

[0120]

[0121] in, Defined as the number of human resources required per unit weighted area, its dimensions are... Integral term The calculated result is the weighted risk area, and its dimensions are: Multiplying the two items ensures The output is a dimensionless number of people. Perform an up-rounding operation to generate the final administrative intervention personnel dispatch instructions, ensuring that intervention resources are in integer units and guaranteeing the rigor of physical dimensions.

[0122] Example 7:

[0123] S5 specifically includes:

[0124] S51. Calculate the number of false alarms by subtracting the system's expected number of false alarms from the number of false alarms reported.

[0125] S52. Multiply the result by the preset learning rate to generate the correction increment;

[0126] S53. The correction increment is added to the current data entropy penalty factor to generate the updated data entropy penalty factor.

[0127] This embodiment illustrates the closed-loop update logic of the data entropy penalty factor;

[0128] Calculate the number of false alarms and provide feedback. Subtract the expected false alarms of the system The result;

[0129] Multiply the result by the preset learning rate. Generate correction increments; learning rate This is a dimensionless constant used to control the step size of parameter adjustment;

[0130] The correction increment will be added to the current data entropy penalty factor. Generate an updated data entropy penalty factor The updated formula is as follows:

[0131]

[0132] in, This information comes from feedback records that have been manually verified. These are preset constants for the system.

[0133] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart campus AIoT big data visualization and analysis method, characterized in that, Includes the following steps: S1. Collect heterogeneous behavioral data from campus IoT gateways. The heterogeneous behavioral data includes consumption data, access control data, and water and electricity data. Perform Z-Score standardization on the heterogeneous behavioral data to generate dimensionless standard scores. Construct a behavior state matrix based on dimensionless standard scores; S2. Obtain the preset sensor inherent noise figure; Based on the inherent noise coefficient of the sensor, the consistency coefficient between different modal data in the behavior state matrix is ​​calculated. Input the consistency coefficient into the entropy evolution model to generate the entropy. S3. Obtain the behavioral risk value output by the external AI prediction module; obtain the current data entropy penalty factor; Calculate the confidence score for AI intervention based on behavioral risk value, compliance entropy, and data entropy penalty factor; S4. Compare the AI ​​intervention confidence score with the preset dynamic decision threshold; if the AI ​​intervention confidence score is higher than the dynamic decision threshold, execute AI-guided precise intervention; if the AI ​​intervention confidence score is lower than or equal to the dynamic decision threshold, perform spatial mapping calculation on the obedience entropy to generate the administrative intervention resource allocation amount. S5. Obtain the feedback of the number of false alarms after manual verification; calculate the difference between the feedback of the number of false alarms and the system's expected number of false alarms; update the data entropy penalty factor based on the difference; use the updated data entropy penalty factor to calculate the AI ​​intervention confidence score in the next time step.

2. The smart campus AIoT big data visualization and analysis method according to claim 1, characterized in that, S1 specifically includes: S11. Obtain historical baseline data of the student population from the previous semester from the cloud database. The historical baseline data includes the historical mean and standard deviation. S12. Using historical mean and standard deviation, map consumption data to standard scores in the monetary dimension; map access control data to standard scores in the time frequency dimension; map water and electricity data to standard scores in the usage dimension. S13. Generate a behavioral state matrix using the standard scores for the combined amount dimension, the standard scores for the time frequency dimension, and the standard scores for the usage dimension.

3. The smart campus AIoT big data visualization and analysis method according to claim 1, characterized in that, S2 specifically includes: S21. Determine the inherent noise figure of the sensor based on the equipment's factory calibration data; S22. Based on the sensor's inherent noise figure and the preset system sensitivity adjustment gain, the consistency coefficient between any two types of dimensionless standard scores is calculated using the weighted Euclidean distance algorithm. S23. Set the dimensionless standard score corresponding to the access control data as the benchmark mode; calculate the consistency coefficient between the dimensionless standard score of other modes and the benchmark mode; S24. Input the consistency coefficients corresponding to the baseline mode into the entropy evolution model.

4. The smart campus AIoT big data visualization and analysis method according to claim 1, characterized in that, The obedience-entropy evolution model in S2 adopts the following logic: S25. Based on the confidence weights of different modalities, the logarithmic values ​​of the consistency coefficients are weighted and summed to generate a static quality term; S26. Perform differential calculation on the rate of change of the consistency coefficient over time, and generate a dynamic trend term by combining it with the preset trend-weighted time constant. S27. Combine the static quality term and the dynamic trend term to generate the degree entropy.

5. The smart campus AIoT big data visualization and analysis method according to claim 1, characterized in that, S3 specifically includes: S31. Obtain the preset baseline risk sensitivity coefficient and entropy value; S32. In response to the entropy of the degree of obedience being less than the entropy baseline, the data entropy penalty factor is set to zero; in response to the entropy of the degree of obedience being greater than or equal to the entropy baseline, the data entropy penalty factor remains unchanged. S33. Calculate the product of the behavioral risk value and the basic risk sensitivity coefficient, and use it as the numerator of the utility function; S34. Calculate the difference between the obedience entropy and the entropy baseline; calculate the product of the difference and the data entropy penalty factor; add one to the product and use it as the denominator of the utility function. S35. Calculate the ratio of the numerator to the denominator of the utility function to generate the confidence score for AI intervention.

6. The smart campus AIoT big data visualization and analysis method according to claim 1, characterized in that, The specific components of the administrative intervention resource allocation generated in S4 include: S41. Obtain the physical coordinates of the individual students that generate the degree entropy; S42. Using the Gaussian kernel density estimation method, the discrete degree entropy is mapped to a continuous spatial entropy field based on physical coordinates. S43. Identify regions where the field value in the spatial entropy field exceeds a preset critical threshold. S44. Based on the preset intervention density coefficient, perform double integral calculation on the entropy value in the area exceeding the critical threshold to generate the administrative intervention resource allocation amount.

7. The smart campus AIoT big data visualization and analysis method according to claim 1, characterized in that, S5 specifically includes: S51. Calculate the number of false alarms by subtracting the system's expected number of false alarms from the number of false alarms reported. S52. Multiply the result by the preset learning rate to generate the correction increment; S53. The correction increment is added to the current data entropy penalty factor to generate the updated data entropy penalty factor.

Images