Multi-object adaptive monitoring method and device based on Internet of Things environment

Abstract

This invention discloses a multi-object adaptive monitoring method and apparatus based on an Internet of Things (IoT) environment, comprising: collecting operating parameters of multiple device objects through a sensor array; predicting the operating parameters of other device objects and constructing a predicted operating parameter space when the first operating parameter is normal; updating the device drift prediction branch according to the cumulative operating time to obtain an update error coefficient, and analyzing the drift degree of other device objects based on the first operating parameter; compensating the operating parameters according to the drift degree and the update error coefficient to construct an actual operating parameter space, and calculating its spatial similarity with the predicted operating parameter space; calculating the monitoring error degree based on the spatial similarity as the monitoring result. This invention solves the technical problems of mixed device parameters, lack of dynamic adaptation and differentiated compensation for sensor drift after long-term operation, and lack of comprehensive evaluation methods for the overall status of multiple devices and multiple parameters in existing IoT monitoring systems.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) monitoring technology, and specifically to a multi-object adaptive monitoring method and apparatus based on an IoT environment. Background Technology

[0002] In IoT monitoring systems for buildings, industries, and industrial parks, existing technologies often employ a centralized architecture that mixes multiple devices of various types and connects them to the same control unit. This architecture leads to mismatches between monitoring points and parameters, resulting in frequent false alarms and missed controls; it is highly dependent on a higher-level platform, meaning a platform failure results in complete loss of control; when a single device fails, the entire control panel or system is forced to shut down; fault location is difficult and troubleshooting is slow, hindering refined management based on device type; and cables are centrally laid to fixed control panels, resulting in long routes and high construction costs.

[0003] Meanwhile, equipment operating characteristics change with increasing cumulative operating time, and existing methods lack the ability to adapt to these changes, leading to a continuous decline in monitoring accuracy. The degree of change in characteristics varies significantly among different devices, yet existing methods treat all devices the same. Furthermore, existing methods analyze monitoring parameters of each device in isolation, lacking a comprehensive assessment tool for the overall operating status of multiple devices and parameters, making it difficult to detect correlational anomalies and systemic risks between devices. Summary of the Invention

[0004] This invention provides a multi-object adaptive monitoring method and device based on an Internet of Things (IoT) environment. This invention solves the technical problems in existing IoT monitoring systems, such as mixed device parameter connections, lack of dynamic adaptation and differentiated compensation for sensor drift after long-term operation, and lack of comprehensive evaluation methods for the overall status of multiple devices and parameters.

[0005] In view of the above problems, the present invention provides a multi-object adaptive monitoring method based on an Internet of Things (IoT) environment, the method comprising:

[0006] By using a sensor array based on an IoT industrial cloud platform, multiple operating parameters of multiple devices in the target environment are collected. After the first operating parameter is determined to be normal, the operating parameters of other devices are predicted and the operating space is constructed based on the first operating parameter of the first device, thus obtaining the predicted operating parameter space. Based on the cumulative running time of multiple device objects, the multiple device drift prediction branches in the device drift prediction intelligence are updated to obtain multiple update error coefficients. Based on the first running parameter, drift analysis of the running parameters of other multiple device objects is performed to obtain multiple drift degrees. Based on multiple drift degrees and multiple update error coefficients, drift compensation is performed on multiple operating parameters to obtain multiple compensated operating parameters. An actual operating parameter space is constructed, and the spatial similarity with the predicted operating parameter space is calculated. The monitoring error is calculated based on spatial similarity to obtain the monitoring error score, which is then used as the monitoring result.

[0007] The present invention also provides a multi-object adaptive monitoring device based on an Internet of Things (IoT) environment, comprising: The predictive operating parameter space construction module is used to collect multiple operating parameters of multiple device objects in the target environment through a sensor array based on the IoT industrial cloud platform. After the first operating parameter is judged to be normal, the module predicts the operating parameters of other device objects and constructs the operating space based on the first operating parameter of the first device object to obtain the predictive operating parameter space. The drift analysis and update module is used to update multiple device drift prediction branches in the device drift prediction intelligence based on the cumulative running time of multiple device objects, to obtain multiple update error coefficients, and to perform drift analysis on the running parameters of other multiple device objects based on the first running parameter, to obtain multiple drift degrees. The compensation and space construction module is used to perform drift compensation on multiple operating parameters based on multiple drift degrees and multiple update error coefficients, obtain multiple compensated operating parameters, construct the actual operating parameter space, and calculate the spatial similarity with the predicted operating parameter space. The evaluation output module is used to calculate the monitoring error based on spatial similarity, and obtain the monitoring error score as the monitoring result.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention uses the operating parameters of a first device to drive the prediction of the operating parameters of other devices, and initiates the prediction process only after the first operating parameter is determined to be normal, ensuring the effectiveness of the prediction benchmark and enabling monitoring to predict deviations in the operating status of downstream devices in advance. By constructing and training an independent operating parameter prediction branch for each downstream device, each branch fully learns the specific operating characteristics of the corresponding device, avoiding prediction bias caused by a single model and enhancing scalability. By dynamically updating and training the initial drift prediction branch based on the cumulative operating time, each branch adaptively adjusts according to the aging of the device, improving the timeliness and accuracy of drift prediction. By calculating the time drift coefficient and the benchmark time drift coefficient using the data adversarial coefficient and the operating time coefficient, and taking the absolute value of the difference between the two divided by the benchmark value as the update error coefficient, a complete quantitative evaluation system for prediction reliability is established. By differentiating the drift degree based on the update error coefficient of each device, devices with high anomalies receive greater compensation while normal devices are compensated in a conventional manner, avoiding the shortcomings of a uniform compensation strategy. Furthermore, by calculating the spatial feature similarity between the compensated multi-device, multi-parameter actual operating parameter space and the predicted operating parameter space, and using spatial similarity as a comprehensive quantitative indicator of operating status, a leap from isolated single-parameter judgment to holistic multi-parameter spatial evaluation is achieved. By calculating negative correlation monitoring errors based on spatial similarity, spatial similarity is converted into monitoring error degree, providing maintenance personnel with an intuitive and unified quantitative monitoring indicator. This facilitates rapid judgment of the overall operating status and provides a reliable data foundation for trend analysis and fault backtracking, thereby comprehensively improving the accuracy of sensor data, differentiated compensation capabilities, and the comprehensiveness of status assessment in IoT monitoring.

[0009] In summary, this invention solves the technical problems of mixed device parameters, lack of dynamic adaptation and differentiated compensation for sensor drift after long-term operation, and lack of comprehensive evaluation methods for the overall status of multiple devices and multiple parameters in existing IoT monitoring systems. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the multi-object adaptive monitoring method based on an Internet of Things environment provided in an embodiment of the present invention. Figure 2 This is a flowchart of the update error coefficient calculation process in the multi-object adaptive monitoring method based on the Internet of Things environment provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the structure of a multi-object adaptive monitoring device based on an Internet of Things environment provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: Predictive operating parameter space construction module 11, drift analysis and update module 12, compensation and space construction module 13, evaluation output module 14. Detailed Implementation

[0012] This invention provides a multi-object adaptive monitoring method and device based on the Internet of Things (IoT) environment, which specifically solves the technical problems in existing IoT monitoring systems such as mixed connection of device parameters, lack of dynamic adaptation and differentiated compensation for sensor drift after long-term operation, and lack of comprehensive evaluation means for the overall status of multiple devices and multiple parameters.

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0014] Example 1, as Figure 1 As shown, this invention provides a multi-object adaptive monitoring method based on an Internet of Things (IoT) environment, the method comprising: S100: Through a sensor array based on an IoT industrial cloud platform, multiple operating parameters of multiple devices in the target environment are collected. After the first operating parameter is determined to be normal, the operating parameters of other devices are predicted and the operating space is constructed based on the first operating parameter of the first device, thus obtaining the predicted operating parameter space.

[0015] In existing centralized automation systems, the mixed connection of monitoring points across various devices leads to mismatches between parameters and commands. Furthermore, these systems are highly dependent on the upper-level platform; a platform failure can trigger a complete loss of control. They lack predictive capabilities, only able to issue post-event alarms for anomalies, unable to proactively predict deviations in equipment operating status based on the process relationships between upstream and downstream devices before a failure occurs. Simultaneously, different devices possess different operating characteristics and parameter dimensions, and existing technologies using single predictive models struggle to adapt to all device types, resulting in insufficient prediction accuracy. Moreover, adding or removing devices in the monitoring scenario requires reconstructing the entire predictive model, leading to poor scalability and high adaptation costs. In addition, existing technologies lack quantitative benchmarks for the expected operating parameters of downstream devices, failing to provide reliable reference for subsequent monitoring error calculations, resulting in insufficient effectiveness of monitoring error assessment. Therefore, how to construct a monitoring and prediction mechanism that is adaptable to multiple device types, possesses proactive predictive capabilities, and is easily scalable, based on the process flow relationships between upstream and downstream devices, to achieve a shift from passive response to proactive prediction and ensure the reliability of subsequent monitoring error assessment, has become a pressing technical problem to be solved.

[0016] Step S100 in the method provided in this embodiment of the invention includes: By using a sensor array based on an IoT industrial cloud platform, multiple operating parameters of multiple devices within the target environment are collected. Extract the first operating parameters of the first device object, wherein the first device object is the upstream device of other device objects; Determine whether the first operating parameter falls within the normal range. If not, conduct monitoring and early warning. If so, input the first operating parameter into the operating parameter prediction agent and output multiple predicted operating parameters of other device objects. A prediction parameter space is constructed based on multiple prediction parameters.

[0017] In this embodiment of the invention, a sensor array based on an IoT industrial cloud platform is first deployed. The IoT industrial cloud platform adopts a distributed data bus architecture, supporting multiple industrial communication protocols such as Modbus, BACnet, and OPCUA, enabling data interfacing with sensors from different manufacturers and of different types. The sensor array is deployed on various equipment objects and their supporting pipes, ducts, electrical lines, and other key components within the target environment, including but not limited to temperature sensors, pressure sensors, flow sensors, humidity sensors, current transformers and voltage transformers, vibration sensors, wind speed sensors, and switch signal acquisition modules. Each sensor continuously collects data at a preset sampling frequency, such as once per second, and uploads the collected data in real time to the data cache of the IoT industrial cloud platform via wired Ethernet, Wi-Fi, or 5G communication methods.

[0018] Furthermore, after collecting the operating parameters of multiple equipment objects, the first operating parameters of the first equipment object are extracted from the time-series dataset. The first equipment object is an upstream device pre-designated according to the process flow topology of the target environment. This upstream device is located at the front end of all other equipment objects in the energy flow or material flow transmission path, and changes in its operating state will have a direct and significant impact on the operating conditions of downstream devices at the physical level. The specific method for determining the first equipment object is as follows: obtain the system process flow diagram or equipment connection topology diagram of the target environment, identify the starting point device for the transmission of energy media such as chilled water, hot water, air, electricity, etc., and mark the starting point device as the first equipment object.

[0019] For example, in a chilled water circulation system, the chiller unit is designated as the first equipment object; in an air delivery system, the main supply fan is designated as the first equipment object; and in a power distribution system, the main distribution cabinet or transformer is designated as the first equipment object. The first operating parameter refers to a key parameter that characterizes the core operating condition of the upstream equipment. For a chiller unit, the first operating parameter may be the chilled water supply temperature or the refrigerant suction pressure; for a main supply fan, the first operating parameter may be the fan operating frequency or the static pressure of the air supply duct; and for a water pump, the first operating parameter may be the outlet pressure or the flow rate.

[0020] It should be noted that other equipment objects refer to controlled equipment in the target environment that is downstream of the first equipment object in the process flow, excluding the first equipment object. For example, downstream equipment of a chiller unit includes terminal heat exchange equipment such as air handling units, fresh air units, and fan coil units; downstream equipment of the main supply fan includes variable air volume terminal units or supply air branch regulating valves on each floor or in each area. In a target environment, the number of first equipment objects can be one or more.

[0021] Furthermore, after extracting the first operating parameter of the first device object, it is determined whether the first operating parameter falls within the preset normal operating range. The normal operating range is a parameter threshold interval set for the first device object, and its setting methods include the following approaches: First, based on the rated operating parameter table provided by the equipment manufacturer, ±10% to ±20% of the rated value is taken as the upper and lower boundaries of the normal range; Second, based on the operating data of the equipment under historical normal operating conditions, the mean μ and standard deviation σ are calculated using statistical process control methods, and the normal range is set as [μ-3σ, μ+3σ]; Third, combining the above two methods, the final threshold interval is determined by domain experts after comprehensive verification.

[0022] If the first operating parameter does not fall within the normal range, it indicates that the upstream equipment itself is in an abnormal operating state. In this case, since the upstream equipment has already malfunctioned, the actual operating parameters of the downstream equipment will inevitably be affected by this malfunction. At this point, predicting the operating parameters of the downstream equipment is meaningless, so the monitoring and early warning mechanism is immediately triggered. The monitoring and early warning mechanism includes: generating early warning information, which includes the device identifier of the first device object, the name of the abnormal parameter, the value of the abnormal parameter, the upper and lower limits of the normal range, and a timestamp; uploading the early warning information to the alarm center of the IoT cloud platform through the communication module, and the alarm center notifying the operation and maintenance personnel through SMS, email, APP push, etc.; at the same time, displaying audible and visual alarm signals on the local human-machine interface to remind on-site personnel to take timely action.

[0023] If the first operating parameter falls within the normal range, it indicates that the upstream equipment is currently operating healthily, meeting the prerequisite for effective prediction of downstream equipment. At this point, the first operating parameter is input into a pre-trained operating parameter prediction agent. The operating parameter prediction agent is a data-driven, multi-output prediction model containing multiple parallel prediction branches. The agent receives the first operating parameter as an input feature vector, performs forward computation through each prediction branch, and outputs the predicted operating parameters for multiple other equipment objects. Predicted operating parameters refer to the expected values ​​of the upstream equipment's operating parameters under normal operating conditions and without faults in the downstream equipment itself. The number and type of predicted operating parameters are determined by the type of downstream equipment and monitoring requirements in the monitoring scenario. For example, for an air conditioning unit, predicted operating parameters may include predicted return air temperature, predicted supply air temperature, predicted coil inlet / outlet temperature difference, and predicted fan operating current.

[0024] Finally, after obtaining multiple predicted operating parameters for other device objects, a predicted operating parameter space is constructed based on these parameters. The predicted operating parameter space is a high-dimensional mathematical space used to characterize the overall expected operating state of all downstream devices under the current upstream operating conditions. Specifically, it is constructed as follows: each predicted operating parameter of each downstream device object is used as a dimension in the space, and all predicted operating parameters are organized into a multi-dimensional vector according to a preset dimensional order. The region where this multi-dimensional vector resides in the high-dimensional real number space is the predicted operating parameter space.

[0025] The operational parameter prediction agent includes multiple operational parameter prediction branches, and the pre-training steps for these multiple operational parameter prediction branches include: Based on the recorded data of multiple device objects under normal operating conditions, sample operating parameters of multiple other device objects are collected from the first sample operating parameter set and labeled to obtain multiple sample predicted operating parameter sets; Construct multiple prediction branches for the operating parameters of other multiple device objects respectively; The first set of operating parameters for the sample and the multiple sets of predicted operating parameters for the sample are used to train and verify the multiple prediction branches of the operating parameters. After the loss converges, the multiple prediction branches of the operating parameters are obtained.

[0026] In this embodiment of the invention, a predictive agent for operating parameters needs to be pre-trained. First, recorded data of multiple device objects under normal operating conditions is obtained from a historical database. The filtering criteria for normal operating conditions include: no fault alarm records for each device object, all operating parameters being within their respective preset normal ranges, the device being in a non-start / stop transient state, and the control system outputting no abnormal commands. Based on these filtering criteria, a clean dataset reflecting the healthy operating status of the devices is extracted from the historical data.

[0027] Secondly, from this clean dataset, the first set of sample operating parameters and sample operating parameters of multiple other device objects are collected. Specifically, for each sampling time t in the historical record, the first operating parameter value x(t) of the first device object at that time is extracted as a sample first operating parameter; at the same time, the actual operating parameter value set {y1(t), y2(t), ..., y...} of multiple other device objects at that time is extracted. n {x(t), y1(t), y2(t), ..., y(t)} is used as the sample running parameter corresponding to x(t). Traversing all sampling times in the historical record, an original sample pair set {x(t), y1(t), y2(t), ..., y(t)} is formed. n (t)|t=1,2,...,T}, where T is the total number of historical samples.

[0028] Then, the collected sample operating parameters are labeled to obtain multiple sets of predicted operating parameters for each sample. The actual operating parameter values ​​y of the downstream equipment at each sampling time are then... i (t) is directly used as the label value of the downstream device at that time, and is used for loss calculation in subsequent supervised learning. For the j-th downstream device, the set of sample prediction running parameters obtained after labeling is { j (t) | t = 1, 2, ..., T}, where j The value of (t) is equal to the actual operating parameter value y at that moment. j (t). Under normal operating conditions, the actual operating parameters of downstream equipment represent the expected normal values ​​that should be achieved under the upstream operating conditions, and therefore can be directly used as tags.

[0029] Furthermore, after completing the collection and labeling of sample data, multiple operational parameter prediction branches are constructed for each of the other multiple device objects. Each operational parameter prediction branch is a prediction model built separately for each downstream device. The branches are structurally independent and do not share parameters, thereby achieving customized modeling for the operational characteristics of different devices.

[0030] In this embodiment of the invention, each running parameter prediction branch preferably employs a feedforward neural network model. The structure of this feedforward neural network model includes: an input layer with the number of nodes equal to the dimension of the input features. In this embodiment, the input features are first running parameters. If the first running parameter is only a single variable, the number of nodes in the input layer is 1; if the first running parameter is a feature vector composed of multiple upstream parameters, the number of nodes in the input layer is equal to the dimension of the feature vector; at least one hidden layer, the number of nodes in which can be determined according to the device type and the scale of the training data, for example, 32, 64, or 128 nodes. The activation function of the hidden layer uses the ReLU function to enhance the nonlinear fitting ability of the model and alleviate the gradient vanishing problem during training; and an output layer with the number of nodes equal to the number of types of running parameters to be predicted for the downstream device. The activation function of the output layer uses a linear activation function to ensure that the predicted output is a continuous real number and its dimensions remain unchanged.

[0031] Furthermore, after constructing the model structure for multiple operational parameter prediction branches, each branch is trained and validated using the obtained first set of sample operational parameters and multiple sets of sample predicted operational parameters, respectively. Taking the j-th operational parameter prediction branch corresponding to the j-th downstream device as an example, its training process is as follows: Using the first set of sample running parameters {x(t)|t=1,2,...,T} as the input feature matrix for this branch, and the labeled set of sample prediction running parameters for this branch { j Let {t}|t=1,2,...,T be the output label matrix. All T samples are divided into training, validation, and test sets according to a preset ratio. Preferably, the training set accounts for 70%, the validation set 15%, and the test set 15%. The preferred partitioning method is stratified sampling with random shuffling to ensure that the distribution characteristics of each dataset are consistent with the entire set.

[0032] Supervised learning was used to iteratively train the network model. The mean squared error loss function was used during training. The Adam adaptive moment estimation algorithm was preferred for optimization, with an initial learning rate of 0.001 and a batch size of 32 or 64 for each iteration. After each iteration, the validation loss value of the current model was calculated using the validation set data, and the trend of this loss value was monitored.

[0033] The criterion for judging loss convergence is: when the single-round decrease of the loss value on the validation set in 20 consecutive iterations is less than the preset threshold, it is determined that the model training converges; or, when the number of training rounds reaches the preset maximum number of training rounds, for example, set to 500 rounds, it is also determined that the model training converges.

[0034] Wherein, the preset threshold is 0.1% of the initial loss value. The initial loss value refers to the loss value calculated on the validation set at the end of the first iteration after the training is started, rather than the loss value in the state where the model parameters are randomly initialized without any training. The purpose of setting 0.1% of the initial loss value as the threshold is to make the convergence condition adaptively match the order of magnitude of the loss value in the early stage of training, and avoid the failure of the fixed threshold due to the difference in the initial loss magnitude of different data sets and different model structures.

[0035] Further, the calculation method of the single-round decrease in 20 consecutive iterations is: ΔL k =L k-1 L k , where L k is the loss value calculated on the validation set after the k-th iteration, and L k-1 is the validation loss value of the previous round. When for any k∈{n-19, n-18, ..., n} (n is the current iteration round), ΔL k <L0×0.1% is satisfied, where L0 is the initial loss value, the convergence condition is met. For example, if the initial loss value L0=12.5, the preset threshold is 0.0125. When the decrease of the validation loss of each round in 20 consecutive rounds compared with the previous round is less than 0.0125, it indicates that the improvement of the loss function by the model parameter update is negligible, and it is difficult to significantly improve the model performance by continuing training, so the convergence is determined.

[0036] After the above convergence condition is triggered, the training is stopped, and all completed iteration rounds are traversed to find the model parameter corresponding to the iteration round with the smallest loss value on the validation set, and the parameter of this round is extracted and saved as the final training result of the operation parameter prediction branch. The purpose of this operation is that the validation loss may fluctuate during the training process, and the model parameters of the last iteration are not necessarily in the globally optimal state. Saving the model parameters when the validation loss is the smallest can ensure that the finally deployed model has the best prediction performance on unknown data.

[0037] After training, the generalization performance of the trained model can be evaluated using test set data. The first set of running parameters from the test set samples is input into the model to obtain the predicted running parameters output by the model. The mean absolute error (MAE) and coefficient of determination (R²) between the predicted and true values ​​are then calculated. When the MAE meets the accuracy requirements of the engineering scenario and the R² is greater than 0.85, it indicates that the model has good generalization ability and can be deployed in a real-world monitoring system. If the evaluation metrics do not meet the above requirements, the model's hyperparameters need to be adjusted or the training samples expanded before retraining.

[0038] Following the same approach described above, each operational parameter prediction branch corresponding to multiple other device objects is independently trained and validated. The training process of each branch is independent and can be performed in parallel or sequentially. After each branch reaches loss convergence, multiple trained operational parameter prediction branches are obtained. These multiple trained operational parameter prediction branches are integrated according to their correspondence with downstream devices to form a complete operational parameter prediction agent. This agent is used to execute the following: when the first operational parameter of the first device object is input, each branch outputs the predicted operational parameters of the corresponding downstream device in parallel.

[0039] This invention uses the operating parameters of a first device to drive the prediction of the operating parameters of other devices, and initiates the prediction process only after the first operating parameter is determined to be normal. This ensures the effectiveness of the prediction benchmark and avoids invalid predictions when upstream devices are already in an abnormal state, thereby significantly improving the predictability and foresight of monitoring. It can predict the deviation of downstream device operating status based on the changing trends of upstream parameters before the device actually deviates from normal operating conditions, providing maintenance personnel with early warning and response time. Furthermore, by constructing a separate operating parameter prediction branch for each downstream device and independently training and validating it using historical data under normal operating conditions, each prediction branch can be fully utilized. By learning the specific operating characteristics of each device, the system avoids prediction bias caused by applying a single model to different devices, significantly improving prediction accuracy and relevance. Furthermore, when adding or removing downstream devices in the monitoring scenario, only the corresponding prediction branches need to be added or deleted without reconstructing the entire prediction model, greatly enhancing the system's scalability and scenario adaptability. In addition, the constructed prediction operating parameter space provides a reliable benchmark for subsequent comparison and analysis between the actual operating parameter space and the prediction operating parameter space, as well as for calculating monitoring errors, ensuring the accuracy and effectiveness of the overall monitoring error assessment. Thus, the system fully realizes the initiative, refinement, specialization, and scalability of the monitoring system.

[0040] S200: Based on the cumulative running time of multiple device objects, update multiple device drift prediction branches in the device drift prediction intelligence to obtain multiple update error coefficients, and perform drift analysis on the running parameters of other multiple device objects based on the first running parameter to obtain multiple drift degrees.

[0041] In existing IoT monitoring systems, sensors experience zero-point drift and sensitivity drift as operating time increases, causing the collected parameters to gradually deviate from their true values. Traditional methods typically assume constant sensor characteristics and fail to consider the impact of cumulative operating time on sensor drift, leading to a continuous decline in the accuracy of monitoring data and the reliability of decision-making over long-term operation. Furthermore, current technologies lack a dynamic mechanism to adaptively predict the drift amplitude of downstream devices based on upstream device operating parameters, and even more so, they lack quantitative evaluation methods for the reliability of updated drift prediction models. This makes it impossible to determine the reliability of updated predictions when device operating time increases and drift characteristics change. In addition, existing drift compensation strategies use a uniform adjustment method for different devices, failing to identify abnormal differences in drift characteristics that deviate from the average level of similar devices, resulting in insufficient compensation for abnormal devices and overcompensation for normal devices. Therefore, how to construct a differentiated drift analysis and compensation mechanism that can dynamically update the drift prediction model based on the cumulative operating time of devices, quantitatively evaluate update errors, and identify the degree of abnormality in drift characteristics of each device, in order to ensure the accuracy of sensor data and the reliability of monitoring decisions under long-term operating conditions, has become an urgent technical problem to be solved.

[0042] Step S200 in the method provided in this embodiment of the invention, as follows: Figure 2 As shown, it includes: An initial device drift prediction agent is obtained, wherein the device drift prediction agent includes multiple device drift prediction branches corresponding to multiple other device objects. Each device drift prediction branch is trained and obtained based on a sample first operating parameter set and a sample drift degree set of labeled device object operating parameters that cause sensing drift. Each sample drift degree includes the magnitude of the sensing drift of the device object's operating parameters under the influence of the corresponding sample first operating parameter. Get the cumulative runtime of multiple device objects other than the first device object; Based on the historical operating data of similar devices from multiple device objects, the sample update first operating parameter set of the first device object and the operating parameter sensing drift amplitude of other multiple device objects are obtained under multiple cumulative operating durations, and multiple sample update drift sets are labeled. The first set of running parameters is updated using samples, and the drift degree set is updated using multiple samples. The drift prediction branches of multiple devices are updated and trained respectively to obtain the updated drift prediction branches of multiple devices, and multiple update error coefficients are obtained. The first operating parameters are input into the updated multiple device drift prediction branches respectively, and multiple drift degrees are output.

[0043] In this embodiment of the invention, an initial device drift prediction agent is pre-constructed and acquired. The device drift prediction agent is a set of predictive models used to predict the degree of sensor drift in the operating parameters of each device. Sensor drift refers to the phenomenon where the measured output value of a sensor or measurement channel slowly shifts over time due to factors such as changes in ambient temperature, component aging, power supply voltage fluctuations, and electromagnetic interference. This phenomenon causes the collected operating parameters to gradually deviate from their true values, and if not identified and compensated for, it will directly affect the accuracy of monitoring decisions.

[0044] The device drift prediction agent comprises multiple device drift prediction branches, each corresponding one-to-one with several other device objects. Each device drift prediction branch is trained based on a sample set of initial operating parameters and a set of sample drift degrees indicating sensor drift in the operating parameters of the labeled device objects. Drift degree refers to the deviation of the sensor measurement value relative to the true value, expressed as a percentage, and is calculated using the formula: Drift degree = (Measured value / Actual value) * ... (True value) / Measured value × 100%. The sign of the drift reflects the direction of the sensor measurement deviation; a positive value indicates that the sensor reading is larger than the true value, and a negative value indicates that the sensor reading is smaller than the true value.

[0045] Furthermore, the cumulative operating time of multiple device objects other than the first device object is obtained. The cumulative operating time refers to the total time each device object has accumulated since it was put into operation or since the last calibration, expressed in hours or days. The cumulative operating time can be obtained by: reading the cumulative operating time register value from the device's local controller's operating log; querying the device's last calibration date and cumulative operating records from the IoT cloud platform's device management database; or accumulating it in real time through the device's associated timer module. The cumulative operating time is a crucial factor affecting the degree of sensor drift. As the device's operating time increases, the aging of sensor components intensifies, the cumulative effect of temperature cycling strengthens, and drift typically becomes more significant.

[0046] Furthermore, based on the historical operating data of similar devices among multiple device objects, the sample update first operating parameter set of the first device object and the operating parameter sensing drift amplitude of other multiple device objects are obtained under multiple cumulative operating durations, and multiple sample update drift degree sets are labeled.

[0047] Specifically, historical records within the same or similar cumulative operating time as the current devices are extracted from the historical database, such as those around 8760 hours, 4380 hours, and 13140 hours. The first operating parameter value of the first device at each sampling moment is used as the sample to update the first operating parameter. Simultaneously, the deviation values ​​between the actual collected operating parameters of multiple other devices at the corresponding moments and the true operating parameters verified by a high-precision calibration device are extracted. This deviation value is the sensing drift amplitude of the operating parameter. The sensing drift amplitude at each sampling moment is labeled to obtain multiple sample update drift degree sets.

[0048] Furthermore, the first set of operating parameters is updated using samples, and the drift degree set is updated using multiple samples. These are then used to update and train multiple device drift prediction branches, resulting in updated multiple device drift prediction branches, and multiple update error coefficients are obtained.

[0049] Update training refers to fine-tuning or incrementally learning the parameters of the initially trained device drift prediction branch using newly acquired sample data reflecting drift characteristics under the current cumulative operating time. Update training can be implemented in the following ways: using the updated first operating parameters of the samples as input features and the updated drift degree of the corresponding samples as output labels, and fine-tuning the original model parameters using supervised learning; or using online learning algorithms to incrementally update the model. After update training, each branch can more accurately predict the degree of device sensor drift under the current cumulative operating time. Multiple update error coefficients are obtained during or after update training. These update error coefficients are used to quantify the prediction deviation of each device drift prediction branch after update training relative to the initial state.

[0050] Finally, the current operating parameters of the first device object are input into the updated multiple device drift prediction branches. After forward calculation, each branch outputs the drift degree of the corresponding device object. The drift degree refers to the predicted magnitude of the sensing drift of the downstream device's operating parameters under the current upstream device operating parameters.

[0051] The initial construction process of the device drift prediction agent is as follows. The device drift prediction agent is a collection of multi-branch parallel prediction models, which contains multiple device drift prediction branches that correspond one-to-one with multiple other device objects. Each branch is independently responsible for predicting the degree of sensor drift of the corresponding device object under the current upstream device operating parameters. The branches are structurally independent and do not share parameters.

[0052] For any downstream device, its corresponding device drift prediction branch is implemented using a feedforward neural network model. The input layer features of this network model are the first operating parameters of the first device. When the first operating parameter is a single variable, the number of input layer nodes is 1; when the first operating parameter is a multi-dimensional feature vector, the number of input layer nodes is equal to the dimension of the feature vector. The number of hidden layers is set to 1 to 3, and the number of nodes in each layer is set to 16 to 64. The ReLU function is used for the activation function of the hidden layers. The number of output layer nodes is 1, which is used to output the predicted drift degree of the device. This predicted drift degree is a continuous real value representing the magnitude of the sensing drift, expressed as a percentage. The linear activation function is used for the output layer to ensure that the predicted drift degree is continuously adjustable within the real number range.

[0053] During the training sample collection and labeling phase, the first operating parameter values ​​of the first equipment object at each sampling time are extracted from the historical database as sample input features. The sampling times should cover various operating conditions that may occur in the upstream equipment to ensure the operating condition coverage of the training samples. The sample drift is used as the output label during model training, and its labeling adopts one or a combination of the following two methods. Method 1 is based on calibration comparison: During regular equipment maintenance or calibration, a high-precision calibration instrument verified by the metrology department is used. At the same time and at the same measuring point, the reading of the calibration instrument is compared with the reading collected by the equipment's built-in sensor. The percentage deviation of the sensor's collected value relative to the measured value of the calibration instrument is calculated, and this percentage deviation is used as the sensor drift. Method 2 is based on historical data inversion: In the absence of regular calibration records, the operating data of the equipment in the initial stage of new operation before significant sensor drift is used as a benchmark. The operating data in the initial stage after equipment operation is extracted as an approximation of the true value. For each subsequent sampling time, the percentage deviation between the sensor's collected value at that time and the benchmark value under the same operating conditions in the initial stage of operation is calculated as the estimated value of the sample drift at that time. After labeling, all sample pairs are randomly shuffled according to the sampling time to form the initial training sample set for this branch.

[0054] During the model training phase, the sample set is divided into training, validation, and test sets according to a preset ratio. Stratified random sampling is used to ensure that the distribution range of the first running parameter in each dataset is consistent with the entire set. The loss function used during training is the mean squared error loss function, which is the average of the squared differences between the model's predicted values ​​and the true labeled values. The optimization algorithm employs the adaptive moment estimation algorithm, with the batch size for each iteration flexibly set according to the total number of samples. Training set samples are input into the network in batches for forward propagation to obtain the prediction drift. After calculating the loss function value, backpropagation is performed, and the optimization algorithm updates the weights and bias parameters of each layer of the network. After each round of traversal of the entire training set samples, the validation loss value of the current model is calculated using the validation set data. Convergence is determined using an early stopping strategy: the model is considered converged when the decrease in loss value on the validation set is less than a preset threshold in each iteration over multiple iterations. The preset threshold is one-thousandth of the initial loss value, which is the loss value calculated on the validation set at the end of the first iteration after training begins. After training stops, all completed iterations are iterated through, and the model parameters corresponding to the iteration with the smallest loss value on the validation set are found and saved as the final training result for that branch. This optimal parameter saving mechanism is used to avoid overfitting problems that may occur in the later stages of training, ensuring that the model has the best generalization performance on unknown data. After training, the model's generalization performance is evaluated using test set data. The mean absolute error and coefficient of determination between the predicted drift and the true labeled drift are calculated. When both the mean absolute error and the coefficient of determination reach the preset performance indicator thresholds, it indicates that the model has good predictive performance and can be used for practical deployment.

[0055] Following the same approach described above, corresponding drift prediction branches are constructed for each downstream device object. The training processes of each branch are independent and can be performed in parallel or sequentially. After all branches are constructed, they are integrated according to their correspondence with the downstream device objects to form a complete initial device drift prediction agent.

[0056] Obtain multiple update error coefficients, including: In each device object's sample first operating parameter set, sample drift set, sample updated first operating parameter set, and sample updated drift set, extract the proportion of samples with inconsistent drift and update drift among the same sample first operating parameters and sample updated first operating parameters, and obtain multiple data adversarial coefficients; Obtain multiple rated runtimes for other device objects; Calculate the ratios of multiple cumulative running times to multiple rated running times to obtain multiple running time coefficients; Calculate the ratios of multiple data adversarial coefficients to multiple runtime coefficients to obtain multiple time drift coefficients, and then calculate the baseline time drift coefficient. Based on multiple time drift coefficients and the baseline time drift coefficient, an update adversarial analysis is performed to calculate multiple update error coefficients.

[0057] In this embodiment of the invention, after constructing the sample first operating parameter set, sample drift degree set, sample updated first operating parameter set, and sample updated drift degree set for each device object, for each device object, sample pairs with the same or similar sample first operating parameter values ​​are searched in the sample dataset used in the initial training phase and the sample dataset used in the updated training phase. For each such sample pair, the initial sample drift degree and the updated sample drift degree are compared to see if they are consistent, that is, whether the absolute value of the difference between the two drift degree values ​​exceeds a preset difference tolerance threshold. The proportion of inconsistent sample pairs to the total number of all comparable sample pairs is counted, and this proportion is used as the data adversarial coefficient of the device object. The data adversarial coefficient reflects the degree of consistency between the initial training data and the updated training data. The higher the coefficient, the more significant the change in device drift characteristics has occurred as the cumulative running time increases.

[0058] The difference tolerance threshold refers to a pre-set boundary value for drift deviation. When the absolute value of the difference between the initial sample drift and the updated sample drift is less than or equal to this difference tolerance threshold, the two drift values ​​are considered to be consistent, meaning that the device drift characteristics have not undergone substantial changes. When the absolute value is greater than this difference tolerance threshold, the two drift values ​​are considered to be inconsistent, meaning that the device drift characteristics have undergone non-negligible changes. The specific value of this difference tolerance threshold can be pre-set by those skilled in the art based on the device type, sensor accuracy level, and the actual tolerance level of the monitoring scenario, preferably between 0.5% and 2%. For example, for a temperature sensor, if the allowable measurement error in engineering is ±0.5℃, corresponding to a drift of approximately ±0.5% to 1%, then the difference tolerance threshold can be set to 0.5%.

[0059] Furthermore, the rated operating time of each device is obtained. Rated operating time refers to the design life or recommended calibration cycle of the equipment under normal operating conditions, as specified by the equipment manufacturer, and is usually expressed in hours. It can be obtained from the equipment nameplate, product manual, or technical parameter table provided by the manufacturer. For each device, the ratio of its current cumulative operating time to its rated operating time is calculated, and this ratio is used as the operating time coefficient for that device. The operating time coefficient reflects the proportion of the equipment's service life to its design life. The closer the coefficient is to or greater than 1, the higher the degree of equipment aging and the greater the uncertainty of sensor drift.

[0060] Furthermore, for each device, the ratio of its data adversarial coefficient to its runtime coefficient is calculated, and this ratio is used as the time drift coefficient for that device. The time drift coefficient comprehensively reflects a normalized indicator of the degree of data adversarial behavior relative to the degree of device aging: under the same runtime coefficient, a larger data adversarial coefficient indicates more drastic changes in drift characteristics; under the same data adversarial coefficient, a smaller runtime coefficient indicates significant changes in drift characteristics occurring within a shorter runtime, which often foreshadows more severe drift anomalies. Based on this, the average time drift coefficient of all device objects is taken to obtain the baseline time drift coefficient, which serves as a reference benchmark for measuring the drift characteristics of each device relative to the overall level.

[0061] Finally, based on multiple time drift coefficients and the baseline time drift coefficient, an update adversarial analysis is performed to calculate multiple update error coefficients, including: The absolute deviation between each time drift coefficient and the reference time drift coefficient is calculated and used as an update error coefficient to reflect the abnormal change in the sensing drift of the corresponding device object, thus obtaining multiple update error coefficients.

[0062] In this embodiment of the invention, an update adversarial analysis is performed based on multiple time drift coefficients and a reference time drift coefficient, and multiple update error coefficients are calculated. Specifically, for each device object, the difference between its time drift coefficient and the reference time drift coefficient is first calculated, and then the difference is divided by the reference time drift coefficient. The absolute value of the result is taken, and the calculated value is used as the update error coefficient reflecting the abnormal change in the sensing drift of the device object.

[0063] Specifically, the update error coefficient measures the relative deviation of each device's time drift coefficient from the benchmark value, rather than the absolute deviation. Using relative deviation rather than absolute deviation as the update error coefficient eliminates the impact of differences in the magnitude of the benchmark time drift coefficient itself across different application scenarios on error assessment. This allows the degree of anomaly corresponding to the same absolute deviation to be reasonably scaled when the benchmark value is large or small, thus ensuring the comparability of update error coefficients between different devices and different scenarios. For example, when the benchmark time drift coefficient is large, even if the absolute difference between the time drift coefficient and the benchmark value is large, if the proportion of this difference to the benchmark value is small, it indicates that the drift characteristics of the device are not significantly deviated, and the calculated update error coefficient is small. Conversely, when the benchmark time drift coefficient is small, even if the absolute difference is not large, if the proportion of this difference to the benchmark value is large, it indicates that the drift characteristics of the device are significantly deviated, and the calculated update error coefficient is large.

[0064] Furthermore, the magnitude of the update error coefficient directly reflects the degree of abnormality of the device's sensing drift characteristics relative to the overall average level. The larger the update error coefficient, the further the device's time drift coefficient deviates from the benchmark value, meaning the higher the degree of abnormality of the device's drift characteristics, and the greater the adjustment weight needs to be given in subsequent drift compensation. The smaller the update error coefficient, the closer the device's drift characteristics are to the overall average level, and compensation can be carried out in a conventional manner.

[0065] This invention acquires sample update data based on the cumulative operating time of each device and dynamically updates and trains the initial drift prediction branch. This allows each prediction branch to adaptively adjust model parameters as the device ages, thereby accurately reflecting the sensor drift characteristics under the current state and significantly improving the timeliness and accuracy of drift prediction. By calculating the data adversity coefficient and the operating time coefficient, a time drift coefficient and a baseline time drift coefficient are further obtained. The absolute deviation between the two is used as the update error coefficient, establishing a complete quantitative evaluation system for drift prediction reliability. This system can objectively determine the reliability of the updated drift prediction branches. Based on the magnitude of the update error coefficient, abnormal devices are given a greater drift compensation weight, while normal devices are compensated in a conventional manner. This achieves a differentiated compensation strategy, comprehensively improving the accuracy of sensor parameter acquisition for multiple devices and the reliability of monitoring decisions under long-term operating conditions.

[0066] S300: Based on multiple drift degrees and multiple update error coefficients, drift compensation is performed on multiple operating parameters to obtain multiple compensated operating parameters, an actual operating parameter space is constructed, and the spatial similarity with the predicted operating parameter space is calculated.

[0067] In existing IoT monitoring systems, the drift characteristics of sensors in different devices vary significantly. The drift degree of some devices may deviate severely from the average level of similar devices. Current technologies employ a uniform drift compensation strategy for all devices, resulting in insufficient compensation for devices with high anomalies and overcompensation for normal devices, leaving significant errors in the compensated parameters. Furthermore, existing technologies lack quantitative verification methods for the compensation effect after completion, making it impossible to determine whether the actual operating parameters after compensation are close to the expected values, or to comprehensively assess the system's operational status from a holistic perspective of multiple devices and parameters. Monitoring parameters for each device are analyzed in isolation, making it difficult to detect correlational anomalies and systemic risks between devices. Therefore, how to construct a method that can perform differentiated drift compensation based on the degree of anomaly in the drift characteristics of each device, and quantitatively assess the overall system operational status by spatially comparing the actual operating status of multiple devices and parameters after compensation with the expected operating status, thereby comprehensively improving the accuracy of drift compensation and the comprehensiveness of system status assessment, has become a pressing technical problem to be solved.

[0068] Step S300 in the method provided in this embodiment of the invention includes: Based on multiple update error coefficients, multiple drift degrees are adjusted for error protection to obtain multiple adjusted drift degrees; Multiple adjustment drift degrees are used to compensate for the drift of multiple operating parameters, resulting in multiple compensated operating parameters; Based on multiple compensated operating parameters, an actual operating parameter space is constructed, and the spatial similarity between the actual operating parameter space and the predicted operating parameter space is calculated.

[0069] In this embodiment of the invention, multiple drift values ​​are first adjusted for error protection based on multiple update error coefficients, resulting in multiple adjusted drift values. Error protection adjustment refers to the process of correcting the predicted drift value of each device using its update error coefficient. Since the update error coefficient reflects the degree of abnormality of the sensing drift characteristics of each device relative to the overall average level, when the update error coefficient deviates from zero, it indicates that the drift characteristics of that device deviate from the overall average level, and the reliability of its predicted drift value decreases. Therefore, protective adjustments are needed during compensation to suppress the impact of unreliable predicted values ​​on the compensation result.

[0070] Specifically, for each device, the adjusted drift is calculated as: Drift × (1 - Update Error Coefficient of the device). When the Update Error Coefficient is zero, the device's drift characteristics are completely consistent with the overall average level, and the adjusted drift is equal to the original drift, with no adjustment made. When the Update Error Coefficient is positive, the adjusted drift is less than the original drift, and the drift compensation range is compressed to avoid overcompensation based on unreliable predictions. When the Update Error Coefficient is negative, the adjusted drift is greater than the original drift, and the drift compensation range is amplified to compensate for insufficient compensation that may result from the device's drift characteristics being below average.

[0071] Furthermore, multiple adjustment drift degrees are used to compensate for the drift of multiple operating parameters, resulting in multiple compensated operating parameters. In this embodiment of the invention, the drift degree is expressed as a percentage, and its sign reflects the direction of the sensor measurement deviation. A positive value indicates that the sensor reading is larger than the true value, and a negative value indicates that the sensor reading is smaller than the true value. Accordingly, this embodiment of the invention uses the measured value × (1 - adjustment drift degree) for drift compensation. This compensation method can automatically adapt the compensation direction according to the positive or negative sign of the drift degree. When the drift degree is positive, (1 - adjustment drift degree) is less than 1, and the measured value is reduced, thereby eliminating the influence of positive deviation on the measurement result; when the drift degree is negative, (1 - adjustment drift degree) is greater than 1, and the measured value is amplified, thereby eliminating the influence of negative deviation on the measurement result. For each device object, its currently acquired operating parameters are multiplied by (1 - adjustment drift degree) to obtain the compensated operating parameters. The compensated operating parameters are the estimated operating parameters that are closer to the true value after eliminating the influence of sensor drift.

[0072] Furthermore, based on multiple compensated operating parameters, an actual operating parameter space is constructed, and the spatial similarity between the actual operating parameter space and the predicted operating parameter space is calculated, including: Based on multiple compensation operation parameters, an actual operation parameter space is constructed; Calculate the spatial feature similarity between the actual operating parameter space and the predicted operating parameter space, and use it as the spatial similarity.

[0073] In this embodiment of the invention, an actual operating parameter space is first constructed based on multiple compensated operating parameters. Specifically, the multiple compensated operating parameters obtained after drift compensation are arranged in the same dimensional order as when constructing the predicted operating parameter space, forming a multidimensional vector. The point corresponding to this multidimensional vector in the high-dimensional real space is the actual operating parameter space point at the current sampling time. The number of dimensions of the actual operating parameter space is equal to the total number of all operating parameters to be monitored for all device objects.

[0074] For example, if the monitoring scenario includes three air conditioning units, and each unit monitors three parameters: return air temperature, supply air temperature, and coil temperature difference, then both the predicted operating parameter space and the actual operating parameter space are nine-dimensional spaces. Each spatial point is represented by a multi-dimensional vector consisting of nine parameter values. The predicted operating parameter spatial points are composed of the predicted values ​​of each parameter, while the actual operating parameter spatial points are composed of the measured values ​​of each parameter after drift compensation.

[0075] Furthermore, the spatial feature similarity between the actual operating parameter space and the predicted operating parameter space is calculated as the spatial similarity. Spatial feature similarity measures the proximity of points in the actual and predicted operating parameter spaces within space. The spatial feature similarity is calculated as follows: a multidimensional spatial distance is calculated between the actual and predicted operating parameter points. The similarity is then calculated based on this multidimensional spatial distance; the closer the distance, the higher the similarity, and the greater the spatial similarity; conversely, the farther the distance, the lower the similarity, and the smaller the spatial similarity.

[0076] Specifically, multidimensional spatial distance can be represented by Euclidean distance, which is calculated as the square root of the sum of the squares of the differences between corresponding dimensions of two multidimensional vectors. After obtaining the Euclidean distance, spatial similarity = 1 / (1 + Euclidean distance), or spatial similarity equals the normalized value of the reciprocal of the Euclidean distance. Regardless of the specific function form used, the core relationship is that spatial similarity is negatively correlated with multidimensional spatial distance; the larger the distance, the smaller the similarity; the smaller the distance, the larger the similarity. The physical meaning of spatial similarity is as follows: the actual operating parameters of each device should be close to the expected values ​​predicted based on the operating parameters of upstream devices. Therefore, the spatial distance between the actual operating parameter spatial points and the predicted operating parameter spatial points is relatively close, resulting in high spatial similarity. When anomalies occur, the actual operating parameters will deviate from the expected values, increasing the distance between the two spatial points and decreasing the spatial similarity. Therefore, spatial similarity can serve as a comprehensive quantitative indicator of operational status.

[0077] This invention differentiates the drift degree based on the update error coefficient of each device, allowing devices with high anomalies to receive greater compensation while normal devices are compensated in a conventional manner. This effectively avoids the problems of insufficient compensation for abnormal devices and overcompensation for normal devices caused by a uniform compensation strategy, significantly improving the targeting and accuracy of drift compensation and making the compensated operating parameters closer to the true values. Simultaneously, by constructing the compensated multi-device, multi-parameter operating parameters into a high-dimensional actual operating parameter space and calculating spatial feature similarity with the predicted operating parameter space, spatial similarity is used as a comprehensive quantitative indicator of the system's operating status. This achieves a leap from isolated judgment of single parameters to overall evaluation of multi-parameter space, enabling the system to comprehensively capture the correlation anomalies between various devices and parameters, providing a reliable data foundation for subsequent monitoring error calculation.

[0078] S400: Calculate the monitoring error based on spatial similarity to obtain the monitoring error score, which is used as the monitoring result.

[0079] While existing IoT monitoring systems can calculate the spatial similarity between the actual and predicted operating parameter spaces, the correlation between this spatial similarity as a similarity index and the operating status is not intuitive. Maintenance personnel cannot easily determine the system's current state level based on spatial similarity alone. Furthermore, there is a lack of means to transform this spatial similarity into a comprehensive monitoring index that facilitates tiered threshold management and tiered alarms. Consequently, spatial comparison results cannot be effectively applied to actual monitoring decisions and early warning responses. Therefore, how to convert spatial similarity into a monitoring error measure that is positively correlated with the degree of system operating error, has an intuitive value range, and is easy to manage with tiered thresholds, thereby achieving effective transformation from spatial comparison results to specific monitoring decisions, has become a pressing technical problem that needs to be solved.

[0080] Step S400 in the method provided in this embodiment of the invention includes: Based on spatial similarity, negative correlation monitoring error is calculated to obtain the monitoring error degree, which is used as the monitoring result.

[0081] In this embodiment of the invention, after calculating the spatial similarity, the monitoring error is calculated based on the spatial similarity to obtain the monitoring error degree, and the monitoring error degree is output as the final monitoring result.

[0082] Monitoring error rate is used to quantitatively assess the degree to which the overall operating status of the entire monitoring system deviates from the expected normal state. There is a negative correlation between spatial similarity and monitoring error rate. Higher spatial similarity indicates that the actual operating status is closer to the predicted operating status, the more normal the operation, and the lower the monitoring error rate; lower spatial similarity indicates that the actual operating status deviates further from the predicted operating status, the more abnormal the operation, and the higher the monitoring error rate. Specifically, monitoring error rate = 1 - spatial similarity.

[0083] Spatial similarity is a value ranging from 0 to 1, where 1 indicates that the actual operating parameter space and the predicted operating parameter space completely overlap, meaning that all operating parameters of all devices are exactly the same as the predicted values. 0 indicates that the two spaces do not overlap at all, meaning that the actual operating parameters deviate greatly from the predicted values. Correspondingly, monitoring error also ranges from 0 to 1. When spatial similarity is 1, monitoring error is 0, indicating an ideal operating state with no errors; when spatial similarity is 0, monitoring error is 1, indicating a worst-case operating state of complete loss of control. In practical applications, monitoring error is usually between 0 and 1, with a higher value indicating a worse overall operating state.

[0084] After obtaining the monitoring error rate, the monitoring error rate is output as the monitoring result. The monitoring result can be displayed in numerical form on the local human-machine interface for on-site maintenance personnel to view, or it can be uploaded to the IoT cloud platform or remote terminal via the communication module, allowing remote management personnel to monitor the overall operating status in real time.

[0085] This invention calculates negative correlation monitoring errors based on spatial similarity, converting spatial similarity into monitoring error degree. This establishes a positive correlation between monitoring error degree and operational error level; the more normal the operation, the lower the monitoring error degree, and the more abnormal the operation, the higher the monitoring error degree. This provides maintenance personnel with an intuitive and unified quantitative monitoring indicator, facilitating rapid assessment of the overall operational status. Furthermore, as a global quantitative indicator integrating all device objects and all operational parameter information, the monitoring error degree comprehensively reflects the overall operational health, achieving an effective transformation from spatial comparison results to intuitive monitoring results at the algorithm level.

[0086] In summary, the embodiments of the present invention have at least the following technical effects: This invention uses the operating parameters of a first device to drive the prediction of the operating parameters of other devices, and initiates the prediction process only after the first operating parameter is determined to be normal, ensuring the effectiveness of the prediction benchmark and enabling monitoring to predict deviations in the operating status of downstream devices in advance. By constructing and training an independent operating parameter prediction branch for each downstream device, each branch fully learns the specific operating characteristics of the corresponding device, avoiding prediction bias caused by a single model and enhancing scalability. By dynamically updating and training the initial drift prediction branch based on the cumulative operating time, each branch adaptively adjusts according to the aging of the device, improving the timeliness and accuracy of drift prediction. By calculating the time drift coefficient and the benchmark time drift coefficient using the data adversarial coefficient and the operating time coefficient, and taking the absolute value of the difference between the two divided by the benchmark value as the update error coefficient, a complete quantitative evaluation system for prediction reliability is established. By differentiating the drift degree based on the update error coefficient of each device, devices with high anomalies receive greater compensation while normal devices are compensated in a conventional manner, avoiding the shortcomings of a uniform compensation strategy. Furthermore, by calculating the spatial feature similarity between the compensated multi-device, multi-parameter actual operating parameter space and the predicted operating parameter space, and using spatial similarity as a comprehensive quantitative indicator of operating status, a leap from isolated single-parameter judgment to holistic multi-parameter spatial evaluation is achieved. By calculating negative correlation monitoring errors based on spatial similarity, spatial similarity is converted into monitoring error degree, providing maintenance personnel with an intuitive and unified quantitative monitoring indicator. This facilitates rapid judgment of the overall operating status and provides a reliable data foundation for trend analysis and fault backtracking, thereby comprehensively improving the accuracy of sensor data, differentiated compensation capabilities, and the comprehensiveness of status assessment in IoT monitoring.

[0087] In summary, this invention solves the technical problems of mixed device parameters, lack of dynamic adaptation and differentiated compensation for sensor drift after long-term operation, and lack of comprehensive evaluation methods for the overall status of multiple devices and multiple parameters in existing IoT monitoring systems.

[0088] Example 2, as Figure 3 As shown, based on the same inventive concept as the multi-object adaptive monitoring method in the Internet of Things environment provided in Embodiment 1, this embodiment of the invention also provides a multi-object adaptive monitoring device in the Internet of Things environment, including: The predictive operating parameter space construction module 11 is used to collect multiple operating parameters of multiple device objects in the target environment through a sensor array based on the Internet of Things industrial cloud platform. After the first operating parameter is judged to be normal, the module predicts the operating parameters of other device objects and constructs the operating space based on the first operating parameter of the first device object to obtain the predictive operating parameter space. The drift analysis and update module 12 is used to update multiple device drift prediction branches in the device drift prediction intelligence body according to the cumulative running time of multiple device objects, to obtain multiple update error coefficients, and to perform drift analysis on the running parameters of other multiple device objects according to the first running parameter, to obtain multiple drift degrees. The compensation and space construction module 13 is used to perform drift compensation on multiple operating parameters based on multiple drift degrees and multiple update error coefficients, obtain multiple compensated operating parameters, construct the actual operating parameter space, and calculate the spatial similarity with the predicted operating parameter space. The evaluation output module 14 is used to calculate the monitoring error based on spatial similarity, and obtain the monitoring error degree as the monitoring result.

[0089] In one embodiment, the predictive runtime parameter space construction module 11 is specifically used for: By using a sensor array based on an IoT industrial cloud platform, multiple operating parameters of multiple devices within the target environment are collected. Extract the first operating parameters of the first device object, wherein the first device object is the upstream device of other device objects; Determine whether the first operating parameter falls within the normal range. If not, conduct monitoring and early warning. If so, input the first operating parameter into the operating parameter prediction agent and output multiple predicted operating parameters of other device objects. A prediction parameter space is constructed based on multiple prediction parameters.

[0090] The operational parameter prediction agent includes multiple operational parameter prediction branches, and the pre-training steps for these multiple operational parameter prediction branches include: Based on the recorded data of multiple device objects under normal operating conditions, sample operating parameters of multiple other device objects are collected from the first sample operating parameter set and labeled to obtain multiple sample predicted operating parameter sets; Construct multiple prediction branches for the operating parameters of other multiple device objects respectively; The first set of operating parameters for the sample and the multiple sets of predicted operating parameters for the sample are used to train and verify the multiple prediction branches of the operating parameters. After the loss converges, the multiple prediction branches of the operating parameters are obtained.

[0091] In one embodiment, the drift analysis and update module 12 is specifically used for: An initial device drift prediction agent is obtained, wherein the device drift prediction agent includes multiple device drift prediction branches corresponding to multiple other device objects. Each device drift prediction branch is trained and obtained based on a sample first operating parameter set and a sample drift degree set of labeled device object operating parameters that cause sensing drift. Each sample drift degree includes the magnitude of the sensing drift of the device object's operating parameters under the influence of the corresponding sample first operating parameter. Get the cumulative runtime of multiple device objects other than the first device object; Based on the historical operating data of similar devices from multiple device objects, the sample update first operating parameter set of the first device object and the operating parameter sensing drift amplitude of other multiple device objects are obtained under multiple cumulative operating durations, and multiple sample update drift sets are labeled. The first set of running parameters is updated using samples, and the drift degree set is updated using multiple samples. The drift prediction branches of multiple devices are updated and trained respectively to obtain the updated drift prediction branches of multiple devices, and multiple update error coefficients are obtained. The first operating parameters are input into the updated multiple device drift prediction branches respectively, and multiple drift degrees are output.

[0092] Obtain multiple update error coefficients, including: In each device object's sample first operating parameter set, sample drift set, sample updated first operating parameter set, and sample updated drift set, extract the proportion of samples with inconsistent drift and update drift among the same sample first operating parameters and sample updated first operating parameters, and obtain multiple data adversarial coefficients; Obtain multiple rated runtimes for other device objects; Calculate the ratios of multiple cumulative running times to multiple rated running times to obtain multiple running time coefficients; Calculate the ratios of multiple data adversarial coefficients to multiple runtime coefficients to obtain multiple time drift coefficients, and then calculate the baseline time drift coefficient. Based on multiple time drift coefficients and the baseline time drift coefficient, an update adversarial analysis is performed to calculate multiple update error coefficients.

[0093] Based on multiple time drift coefficients and a baseline time drift coefficient, an update adversarial analysis is performed, and multiple update error coefficients are calculated, including: The absolute deviation between each time drift coefficient and the reference time drift coefficient is calculated and used as an update error coefficient to reflect the abnormal change in the sensing drift of the corresponding device object, thus obtaining multiple update error coefficients.

[0094] In one embodiment, the compensation and space construction module 13 is specifically used for: Based on multiple update error coefficients, multiple drift degrees are adjusted for error protection to obtain multiple adjusted drift degrees; Multiple adjustment drift degrees are used to compensate for the drift of multiple operating parameters, resulting in multiple compensated operating parameters; Based on multiple compensated operating parameters, an actual operating parameter space is constructed, and the spatial similarity between the actual operating parameter space and the predicted operating parameter space is calculated.

[0095] Based on multiple compensation operation parameters, an actual operation parameter space is constructed; Calculate the spatial feature similarity between the actual operating parameter space and the predicted operating parameter space, and use it as the spatial similarity.

[0096] In one embodiment, the evaluation output module 14 is specifically used for: Based on spatial similarity, negative correlation monitoring error is calculated to obtain the monitoring error degree, which is used as the monitoring result.

[0097] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0099] This specification and accompanying drawings are merely illustrative examples of the invention and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its scope. Therefore, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is intended to include these modifications and modifications.

Claims

1. A multi-object adaptive monitoring method based on an Internet of Things (IoT) environment, characterized in that, The method includes: By using a sensor array based on an IoT industrial cloud platform, multiple operating parameters of multiple devices in the target environment are collected. After the first operating parameter is determined to be normal, the operating parameters of other devices are predicted and the operating space is constructed based on the first operating parameter of the first device, thus obtaining the predicted operating parameter space. Based on the cumulative running time of multiple device objects, update multiple device drift prediction branches in the device drift prediction intelligence to obtain multiple update error coefficients, and perform drift analysis on the running parameters of other multiple device objects based on the first running parameter to obtain multiple drift degrees. Based on multiple drift degrees and multiple update error coefficients, drift compensation is performed on multiple operating parameters to obtain multiple compensated operating parameters. An actual operating parameter space is constructed, and the spatial similarity with the predicted operating parameter space is calculated. The monitoring error is calculated based on spatial similarity to obtain the monitoring error score, which is then used as the monitoring result.

2. The multi-object adaptive monitoring method based on an IoT environment according to claim 1, characterized in that, By using a sensor array based on an IoT industrial cloud platform, multiple operating parameters of various devices within the target environment are collected. After the first operating parameter is determined to be normal, the operating parameters of other devices are predicted and the operating space is constructed based on the first operating parameter of the first device, resulting in a predicted operating parameter space, including: By using a sensor array based on an IoT industrial cloud platform, multiple operating parameters of multiple devices within the target environment are collected. Extract the first operating parameters of the first device object, wherein the first device object is the upstream device of other device objects; Determine whether the first operating parameter falls within the normal range. If not, conduct monitoring and early warning. If so, input the first operating parameter into the operating parameter prediction agent and output multiple predicted operating parameters of other device objects. A prediction parameter space is constructed based on multiple prediction parameters.

3. The multi-object adaptive monitoring method based on an Internet of Things environment according to claim 2, characterized in that, The operational parameter prediction agent includes multiple operational parameter prediction branches, and the pre-training steps for these multiple operational parameter prediction branches include: Based on the recorded data of multiple device objects under normal operating conditions, sample operating parameters of multiple other device objects are collected from the first sample operating parameter set and labeled to obtain multiple sample predicted operating parameter sets; Construct multiple prediction branches for the operating parameters of other multiple device objects respectively; The first set of operating parameters for the sample and the multiple sets of predicted operating parameters for the sample are used to train and verify the multiple prediction branches of the operating parameters. After the loss converges, the multiple prediction branches of the operating parameters are obtained.

4. The multi-object adaptive monitoring method based on an Internet of Things environment according to claim 1, characterized in that, Based on the cumulative running time of multiple device objects, the multiple device drift prediction branches within the device drift prediction intelligence are updated to obtain multiple update error coefficients. Furthermore, based on the first running parameter, drift analysis is performed on the running parameters of other multiple device objects to obtain multiple drift degrees, including: An initial device drift prediction agent is obtained, wherein the device drift prediction agent includes multiple device drift prediction branches corresponding to multiple other device objects. Each device drift prediction branch is trained and obtained based on a sample first operating parameter set and a sample drift degree set of labeled device object operating parameters that cause sensing drift. Each sample drift degree includes the magnitude of the sensing drift of the device object's operating parameters under the influence of the corresponding sample first operating parameter. Get the cumulative runtime of multiple device objects other than the first device object; Based on the historical operating data of similar devices from multiple device objects, the sample update first operating parameter set of the first device object and the operating parameter sensing drift amplitude of other multiple device objects are obtained under multiple cumulative operating durations, and multiple sample update drift sets are labeled. The first set of running parameters is updated using samples, and the drift degree set is updated using multiple samples. The drift prediction branches of multiple devices are updated and trained respectively to obtain the updated drift prediction branches of multiple devices, and multiple update error coefficients are obtained. The first operating parameters are input into the updated multiple device drift prediction branches respectively, and multiple drift degrees are output.

5. The multi-object adaptive monitoring method based on an Internet of Things environment according to claim 4, characterized in that, Obtain multiple update error coefficients, including: In each device object's sample first operating parameter set, sample drift set, sample updated first operating parameter set, and sample updated drift set, extract the proportion of samples with inconsistent drift and update drift among the same sample first operating parameters and sample updated first operating parameters, and obtain multiple data adversarial coefficients; Obtain multiple rated runtimes for other device objects; Calculate the ratios of multiple cumulative running times to multiple rated running times to obtain multiple running time coefficients; Calculate the ratios of multiple data adversarial coefficients to multiple runtime coefficients to obtain multiple time drift coefficients, and then calculate the baseline time drift coefficient. Based on multiple time drift coefficients and the baseline time drift coefficient, an update adversarial analysis is performed to calculate multiple update error coefficients.

6. The multi-object adaptive monitoring method based on an Internet of Things environment according to claim 5, characterized in that, Based on multiple time drift coefficients and a baseline time drift coefficient, an update adversarial analysis is performed, and multiple update error coefficients are calculated, including: The absolute deviation between each time drift coefficient and the reference time drift coefficient is calculated and used as an update error coefficient to reflect the abnormal change in the sensing drift of the corresponding device object, thus obtaining multiple update error coefficients.

7. The multi-object adaptive monitoring method based on an Internet of Things environment according to claim 1, characterized in that, Based on multiple drift degrees and multiple update error coefficients, drift compensation is performed on multiple operating parameters to obtain multiple compensated operating parameters. An actual operating parameter space is constructed, and the spatial similarity between this space and the predicted operating parameter space is calculated, including: Based on multiple update error coefficients, multiple drift degrees are adjusted for error protection to obtain multiple adjusted drift degrees; Multiple adjustment drift degrees are used to compensate for the drift of multiple operating parameters, resulting in multiple compensated operating parameters; Based on multiple compensated operating parameters, an actual operating parameter space is constructed, and the spatial similarity between the actual operating parameter space and the predicted operating parameter space is calculated.

8. The multi-object adaptive monitoring method based on an Internet of Things environment according to claim 7, characterized in that, Based on multiple compensated operating parameters, an actual operating parameter space is constructed, and the spatial similarity between the actual operating parameter space and the predicted operating parameter space is calculated, including: Based on multiple compensation operation parameters, an actual operation parameter space is constructed; Calculate the spatial feature similarity between the actual operating parameter space and the predicted operating parameter space, and use it as the spatial similarity.

9. The multi-object adaptive monitoring method based on an Internet of Things environment according to claim 1, characterized in that, The monitoring error is calculated based on spatial similarity to obtain the monitoring error score, which serves as the monitoring result, including: Based on spatial similarity, negative correlation monitoring error is calculated to obtain the monitoring error degree, which is used as the monitoring result.

10. A multi-object adaptive monitoring device based on an Internet of Things (IoT) environment, characterized in that, The method for performing the multi-object adaptive monitoring method based on an Internet of Things environment according to any one of claims 1 to 9 includes: The predictive operating parameter space construction module is used to collect multiple operating parameters of multiple device objects in the target environment through a sensor array based on the IoT industrial cloud platform. After the first operating parameter is judged to be normal, the module predicts the operating parameters of other device objects and constructs the operating space based on the first operating parameter of the first device object to obtain the predictive operating parameter space. The drift analysis and update module is used to update multiple device drift prediction branches in the device drift prediction intelligence based on the cumulative running time of multiple device objects, to obtain multiple update error coefficients, and to perform drift analysis on the running parameters of other multiple device objects based on the first running parameter, to obtain multiple drift degrees. The compensation and space construction module is used to perform drift compensation on multiple operating parameters based on multiple drift degrees and multiple update error coefficients, obtain multiple compensated operating parameters, construct the actual operating parameter space, and calculate the spatial similarity with the predicted operating parameter space. The evaluation output module is used to calculate the monitoring error based on spatial similarity, and obtain the monitoring error score as the monitoring result.

Images