Dynamic analysis and early warning platform for smart park

Abstract

The invention discloses a smart park dynamic analysis and early warning platform, and particularly relates to the field of smart park danger prediction, the smart park dynamic analysis and early warning platform comprises an acquisition and sorting module, a feature extraction module, a matching and mapping module, a training classification module, a prediction generation module and a sorting intervention module, arranging the time sequences into a multivariable time sequence set with a unified structure according to equipment categories and timestamps; a multi-variable time sequence fusing multi-equipment operation states and environment information is constructed, disturbance characteristics are extracted, a propagation path is constructed, a risk classification model and a risk prediction map are combined, and a dynamic risk prediction and hierarchical intervention mechanism facing a multi-system linkage and nonlinear disturbance scene is formed. The problem that an existing smart park energy system lacks a dynamic danger prediction capability oriented to a complex disturbance situation and cannot realize advanced identification and intervention of nonlinear energy risks is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart park danger prediction, more specifically, the present application relates to a smart park dynamic analysis early warning platform. BACKGROUND

[0002] In the current energy management system of the smart park, the danger prediction capability is seriously dependent on the static load model and the rule-driven mechanism, and the abnormality recognition is mainly realized by comparing the preset threshold value with the fixed period data, however, such mechanism lacks the dynamic modeling capability for the suddenness and nonlinearity events, especially in the park scene with highly coupled energy systems, it is difficult to effectively handle complex situations such as temporary power surge, environmental feedback lag or multi-system concurrent abnormality, in the prior art, although there are monitoring modules for power fluctuation and running state, the response mechanism is mostly post-compensation, lacking of feedforward warning capability;

[0003] Further, the energy subsystems in the smart park, such as air conditioners, elevators, lighting, charging piles, etc., are often in a multi-point asynchronous running state, under such heterogeneous structure, the energy consumption mode has high dynamicity and uncertainty, the existing rule setting method cannot cover the edge state scenes such as cross load transfer, fault cascade propagation or short period energy consumption abnormality, resulting in serious delay or false alarm when dealing with the imbalance of energy system operation, at the same time, the multi-source data fusion capability is weak, which also makes it difficult for the danger prediction system to distinguish between normal fluctuation and potential threat, ultimately limiting the response accuracy and stability of the whole system;

[0004] Under this background, the core problem is that the existing smart park energy system lacks dynamic danger prediction capability for multi-system linkage and complex disturbance situation, and cannot realize the early identification and intervention of non-linear energy risk. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a smart park dynamic analysis early warning platform, by constructing a multivariate time series integrating the running state of multiple devices and environmental information, extracting disturbance features and constructing propagation paths, combining a risk classification model and a danger prediction map, a set of dynamic danger prediction and hierarchical intervention mechanism for multi-system linkage and non-linear disturbance scene is formed, so as to solve the problems that the existing smart park energy system lacks dynamic danger prediction capability for complex disturbance situation and cannot realize the early identification and intervention of non-linear energy risk.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a smart park dynamic analysis early warning platform, comprising a collection and integration module, a feature extraction module, a matching and mapping module, a training and classification module, a prediction generation module and a sorting and intervention module.

[0007] The collection and sorting module obtains multi-device operation and environment state data, and sorts them into a multi-variable time series set with unified structure according to device categories and time stamps;

[0008] The feature extraction module extracts a disturbance feature group composed of disturbance amplitude, disturbance rate and disturbance frequency by performing Fourier transform and sliding window segmentation on the multi-variable time series set, and generates a standard disturbance vector sequence after normalization processing;

[0009] The matching and mapping module pairs and counts vectors with the same time stamp and a Euclidean distance lower than a threshold in the standard disturbance vector sequence, and constructs a disturbance propagation path set according to device identifiers and pairing frequencies;

[0010] The training and classification module inputs the disturbance propagation path set and the standard disturbance vector sequence into a classification calculation process, trains a disturbance risk classification model, and establishes a mapping relationship between the disturbance feature group and the risk label;

[0011] The prediction generation module inputs the current standard disturbance vector into the disturbance risk classification model, matches the risk label with the propagation path node number, and generates a hazard prediction map containing node index and risk number;

[0012] The sorting and intervention module sorts the nodes in the hazard prediction map according to risk number and path connection degree, generates load unloading instructions, running cycle resetting instructions and visual alarm signals in sequence, and writes execution data into the time series set.

[0013] In a preferred embodiment, the collection and sorting module obtains the unit time energy consumption value, voltage change value, current response value, and load proportion value of the air conditioning device, lighting device, elevator drive device and electric vehicle charging device, and the corresponding environment temperature value and environment humidity value for each sampling, and then labels the running state data and environment state data of each device according to the sampling time, and outputs the synchronous data group of device data and environment data;

[0014] Each set of running state data and environment state data in the synchronous data group is matched at the field level according to the time stamp, and an aligned data structure containing sampling time, device type, device number, running state parameter and environment state parameter is constructed through field correspondence, and a device-level binding data block with unified field order and complete dimension is outputted;

[0015] The device-level binding data block is subjected to device type uniform coding, field position regularization and time axis alignment operation, and all data blocks are arranged according to the set time granularity and field structure, and a multi-variable time series set with consistent field structure, clear device identifier and continuous time is outputted.

[0016] In a preferred embodiment, the feature extraction module obtains the time series segments corresponding to each type of equipment by grouping the multivariate time series set by equipment number, then performs Fourier transform on each segment under a fixed window length to extract the frequency domain signals containing the frequency distribution of energy consumption changes, the frequency distribution of voltage changes and the frequency distribution of current response, and outputs the frequency domain signal set corresponding to the equipment;

[0017] The continuous sections in the frequency domain signal set are obtained by setting a time sliding window with a fixed width and a fixed overlap ratio, then the intercept operation is performed in the order of window movement to form a sequence of frequency domain sub-segments with continuous time indexes, and the frequency domain time distribution set containing all window segments is output.

[0018] In a preferred embodiment, in the feature extraction module, the disturbance amplitude is obtained by calculating the difference between the maximum and minimum values of the signal amplitude of each sub-segment in the frequency domain time distribution set, the disturbance rate is calculated by calculating the amplitude change slope between adjacent segments, and the disturbance frequency is located by locating the main frequency point of each sub-segment, then the three disturbance quantities are combined to form a disturbance feature data group, and the disturbance feature group set is output.

[0019] The disturbance amplitude, disturbance rate and disturbance frequency in the disturbance feature group set are respectively subjected to linear normalization calculation according to the preset upper and lower limits, and all feature data groups are rearranged according to the time index to generate a vector format expression with consistent fields and stable order, and a standard disturbance vector sequence is output.

[0020] In a preferred embodiment, the matching mapping module constructs a disturbance vector combination group at the corresponding sampling time by extracting the data with the same timestamp in the standard disturbance vector sequence, and takes any two disturbance vectors in each group as a calculation object;

[0021] The Euclidean distance calculation is performed on each calculation object to form a Euclidean distance matrix containing each pair of disturbance vectors, and the judgment conditions include that the difference between the corresponding disturbance amplitudes in each vector pair is less than the preset disturbance difference threshold, and the value of the Euclidean distance calculation is lower than the set distance similarity threshold; if both conditions are met at the same time, it is identified as an effective disturbance vector pair and output to the disturbance vector pair set; otherwise, the disturbance vector combination group at the timestamp is reconstructed and the Euclidean distance calculation is performed again;

[0022] The equipment number and equipment type of each disturbance vector pair in the disturbance vector pair set are taken as matching key values, the pairing frequency statistics and timestamp position record are performed, and a disturbance linkage statistical record table containing the paired equipment number, pairing frequency and sampling time index is generated.

[0023] The device number pairs with the pairing frequency reaching the specified lower limit in the disturbance linkage statistical record table are constructed as a directed node pair, and a direction label is established according to the sampling time index, and a disturbance propagation path set containing node numbers, connection weights and time label fields is generated.

[0024] In a preferred embodiment, the training classification module constructs a joint feature input group containing path features and disturbance features by extracting node numbers, connection weights and time labels from the disturbance propagation path set, and combining the disturbance amplitude, disturbance rate and disturbance frequency of the corresponding time stamp in the standard disturbance vector sequence;

[0025] The joint feature input group is subjected to a training set division operation, and a classification label set is generated according to the historical running state label or artificial annotation result corresponding to each path, and a training data pair between the feature group and the risk label is constructed;

[0026] The training data pair is input into the classification calculation process, and feature encoding, weight initialization and iterative optimization operations are performed to train the disturbance risk classification model and form a mapping structure between the input features and the output labels;

[0027] The trained disturbance risk classification model is subjected to a verification operation, the classification accuracy of the joint feature input group on the test data is calculated, and the loss function convergence rate based on the risk label prediction bias and the training round change rate is solved;

[0028] If the classification accuracy is higher than the recognition accuracy lower limit and the convergence rate exceeds the loss decrease rate threshold, it is confirmed that the model meets the training effectiveness requirement, the disturbance risk classification model is output and input into the prediction generation module to perform the risk map generation operation; if any of the judgment conditions is not met, the feature encoding and weight initialization are re-performed on the current training data pair, and a new disturbance risk classification model is constructed.

[0029] In a preferred embodiment, the prediction generation module inputs the standard disturbance vector at the current time into the disturbance risk classification model, performs risk classification calculation, and outputs the risk category number corresponding to the standard disturbance vector;

[0030] According to the time index and device number of the standard disturbance vector, the corresponding node number and connection path information are extracted from the disturbance propagation path set, and an index relationship set of device nodes and risk propagation paths is constructed;

[0031] The risk category number and the index relationship set are subjected to a pairing operation, combined in the order of node numbers to form a risk node set, and each device node is attached with its corresponding risk category number;

[0032] Combine the device number, node number, time index and risk category number in the risk node set to construct a data graph with consistent field structure, and output a hazard prediction graph containing node information, time information and risk level.

[0033] In a preferred embodiment, the ranking intervention module constructs a ranking feature set containing risk intensity parameters and structure coupling parameters by extracting the corresponding risk category number of each node in the hazard prediction graph and the connection degree value of the node in the disturbance propagation path set;

[0034] Perform a double-factor ranking operation on the ranking feature set, arrange from high to low according to the level of the risk category number, and if the risk category numbers are the same, arrange from high to low according to the connection degree value, to generate a high-priority device node queue after ranking;

[0035] Match each node in the high-priority device node queue with its device number and current running state in turn, and generate corresponding load offloading instructions, running cycle resetting instructions and visual alarm signals in combination with the risk category number;

[0036] After executing the load offloading instructions, running cycle resetting instructions and visual alarm signals generated by each node, extract the instruction success markers, resetting feedback values and alarm trigger records corresponding to the node, and write them into a multivariate time series set consistent with the time index of the node to construct data update records matching the intervention actions.

[0037] Technical effects and advantages of the present application:

[0038] 1. The present application establishes a dynamic association graph among multiple devices by constructing a disturbance feature vector sequence and a propagation path set, and realizes risk identification by using a classification model, so that the system has the ability to handle energy abnormal scenarios with multiple device linkage, frequent disturbances and response lag, and makes up for the defects of traditional static threshold-based mechanisms that cannot cope with sudden nonlinear risks;

[0039] 2. The present application integrates the running state parameters and environmental data of various devices into a time series set with consistent structure by setting a unified time granularity and field order, ensuring complete data dimension and clear identification, facilitating subsequent modeling and analysis, and solving the problem of difficult fusion of multi-source asynchronous data in existing systems;

[0040] 3. The present application generates a path set containing physical connection and dynamic propagation characteristics by statistically pairing similar disturbance vectors within the same timestamp after extracting disturbance features, combining device identification and pairing frequency, so that the system can restore the regularity and influence range of risk transmission between devices;

[0041] 4、The scheme forms the mapping structure between the disturbance feature and the risk label by training the generated disturbance risk classification model with the disturbance vector and the path information as the joint input, so that the model can perform hierarchical risk prediction and distinguish different risk forms and evolution paths, and improve the classification accuracy and prediction depth;

[0042] 5、The scheme combines the risk number of the node in the risk prediction graph and the path connection degree to perform sorting analysis and intervention output, realizes the hierarchical control instruction generation mechanism driven by the risk prediction result, and enables the system to output corresponding intervention strategies for different risk levels. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The system module diagram of the present application.

[0044] Figure 2 The collection and feature extraction flowchart of the present application.

[0045] Figure 3 The matching mapping flowchart of the present application.

[0046] Figure 4 The training and classification flowchart of the present application.

[0047] Figure 5 The prediction generation flowchart of the present application.

[0048] Figure 6 The sorting intervention flowchart of the present application. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0050] With reference to the drawings in the description Figures 1-6 An intelligent park dynamic analysis and early warning platform according to an embodiment of the present application includes a collection and arrangement module, a feature extraction module, a matching mapping module, a training and classification module, a prediction generation module, and a sorting intervention module.

[0051] The collection and arrangement module obtains multi-device operation and environment state data, and arranges them into a multivariate time series set with a unified structure according to device categories and time stamps.

[0052] The feature extraction module extracts a disturbance feature group composed of disturbance amplitude, disturbance rate and disturbance frequency by performing Fourier transform under a fixed window length and sliding window segmentation on the multivariate time series set, and generates a standard disturbance vector sequence after normalization processing;

[0053] The matching mapping module pairs and counts vector pairs with the same timestamp and a Euclidean distance below a threshold in the standard disturbance vector sequence, and constructs a disturbance propagation path set according to the device identifier and the pairing frequency;

[0054] The training classification module inputs the disturbance propagation path set and the standard disturbance vector sequence into a classification calculation process, trains a disturbance risk classification model, and establishes a mapping relationship between the disturbance feature group and the risk label;

[0055] The prediction generation module inputs the current standard disturbance vector into the disturbance risk classification model, matches the risk label with the propagation path node number, and generates a hazard prediction map containing node index and risk number;

[0056] The sorting intervention module sorts the nodes in the hazard prediction map according to the risk number and path connection degree, generates load unloading instructions, running cycle resetting instructions and visual alarm signals in turn, and writes execution data into the time series set.

[0057] The acquisition and integration module acquires the unit time energy consumption value, voltage change value, current response value and load proportion value of the air conditioning equipment, lighting equipment, elevator drive device and electric vehicle charging device, as well as the corresponding environmental temperature value and environmental humidity value, and then labels the running state data and environmental state data of each device according to the sampling time, and outputs the synchronous data group of device data and environmental data;

[0058] Each set of running state data and environmental state data in the synchronous data group is matched at the field level according to the timestamp, and an aligned data structure containing sampling time, device type, device number, running state parameter and environmental state parameter is constructed through field correspondence, and a device-level binding data block with uniform field order and complete dimension is output;

[0059] The device type uniform coding, field position regularization and time axis alignment operation are performed on the device-level binding data block, all data blocks are arranged according to the set time granularity and field structure, and a multivariate time series set with consistent field structure, clear device identifier and continuous time is output.

[0060] The feature extraction module groups the multivariate time series set by device number, obtains the time series segment corresponding to each type of device, then performs Fourier transform on each segment under a fixed window length to extract the frequency domain signals containing energy change frequency distribution, voltage change frequency distribution and current response frequency distribution, and outputs the frequency domain signal set corresponding to the device;

[0061] The frequency domain signal set is obtained by setting a time sliding window with fixed width and fixed overlap ratio, then the interception operation is performed in the order of window movement to form a frequency domain sub-segment sequence with continuous time index, and the frequency domain time distribution set containing all window segments is output.

[0062] In the feature extraction module, the disturbance amplitude is obtained by calculating the difference between the maximum and minimum values of the signal amplitude of each sub-segment in the frequency domain time distribution set, the disturbance rate is calculated by calculating the amplitude change slope between adjacent segments, and the disturbance frequency is located by locating the main frequency point of each sub-segment, then the three disturbance quantities are combined to form a disturbance feature data group, and the disturbance feature group set is output;

[0063] The disturbance amplitude, disturbance rate and disturbance frequency in the disturbance feature group set are respectively calculated by linear normalization according to the preset upper and lower limits, and all feature data groups are rearranged according to the time index to generate a vector format expression with consistent fields and stable order, and a standard disturbance vector sequence is output.

[0064] The matching mapping module extracts data with the same timestamp from the standard disturbance vector sequence, constructs a disturbance vector combination group at the corresponding sampling time, and takes any two disturbance vectors in each group as a calculation object;

[0065] Each calculation object is calculated by Euclidean distance to form a Euclidean distance matrix containing each disturbance vector pair, and the judgment conditions include that the difference between the corresponding disturbance amplitudes in each vector pair is less than the preset disturbance difference threshold, and the value of Euclidean distance calculation is lower than the set distance similarity threshold; if both conditions are met, it is identified as an effective disturbance vector pair and output to the disturbance vector pair set; otherwise, the disturbance vector combination group at the timestamp is reconstructed and the Euclidean distance calculation is performed again;

[0066] The device number and device type of each disturbance vector pair in the disturbance vector pair set are taken as matching key values, the pairing frequency statistics and timestamp position record are performed, and a disturbance linkage statistical record table containing paired device numbers, pairing frequencies and sampling time indexes is generated;

[0067] The device number pairs in the disturbance linkage statistical record table whose pairing frequency reaches a specified lower limit are constructed as directed node pairs, and the direction label is established according to the sampling time index, and a disturbance propagation path set containing node number, connection weight and time label field is generated.

[0068] The training classification module extracts node numbers, connection weights, and time labels from the set of perturbation propagation paths, combines the perturbation amplitude, perturbation rate, and perturbation frequency corresponding to the time stamp in the standard perturbation vector sequence, and constructs a joint feature input group containing path features and perturbation features;

[0069] The joint feature input group is subjected to a training set division operation, and a classification label set is generated according to the historical running state label or artificial annotation result corresponding to each path, and a training data pair between the feature group and the risk label is constructed;

[0070] The training data pair is input into the classification calculation process, and feature encoding, weight initialization, and iterative optimization operations are performed to train the perturbation risk classification model and form a mapping structure between the input features and the output labels;

[0071] The trained perturbation risk classification model is subjected to a verification operation, the classification accuracy of the joint feature input group on the test data is calculated, and the loss function convergence rate based on the risk label prediction bias and the training round change rate is solved;

[0072] If the classification accuracy is higher than the recognition accuracy lower limit and the convergence rate exceeds the loss decrease rate threshold, it is confirmed that the model meets the training effectiveness requirement, the perturbation risk classification model is output and input into the prediction generation module to perform the risk map generation operation; if any of the judgment conditions is not met, the feature encoding and weight initialization are re-executed on the current training data pair, and a new perturbation risk classification model is constructed;

[0073] It should be noted that for the formula structure involved in the present scheme, the dimensionless term can be used as a proportional or structural adjustment factor. When combined with quantities with units, it only plays a numerical scaling role and does not introduce new physical dimensions, so it will not change or confuse the unit system of the overall expression; such a combination of "dimensionless term and quantity unit term" can be understood as a complex structure expression form commonly used in mathematical and physical modeling, which conforms to the principle of dimensional consistency and has a clear physical interpretation basis;

[0074] Secondly, in the formula structure of the present scheme, if multiple variable terms with different physical units are involved, including but not limited to time, mass, or energy variables, their joint occurrence is to express the cooperative modeling relationship of multiple physical mechanisms. Each variable can be mapped by a function, combined by a ratio, or normalized to form a unified structure with clear units and meanings, and the overall expression conforms to the principle of dimensional consistency and the common norm of engineering modeling;

[0075] If the present scheme involves constants, weights, adjustment factors, threshold parameters, proportionality coefficients, etc., they are all adjustable control parameters for different application environments, whose values depend on the target device configuration, data input characteristics, and performance optimization goals. In the implementation stage, they are set within a reasonable range through model verification, performance constraints, or engineering calibration, etc. Although these parameters do not have a unique value, they have a clear adjustment logic and calculation path, and belong to the deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the scheme has both general adaptability and reproducibility and operability, without affecting its technical clarity and implementability.

[0076] In the training classification module, define is the output value of the perturbation risk classification model, and the perturbation risk classification model represents a classification mapping function trained under all joint perturbation features and graph structure combination inputs. The output result is the risk category number;

[0077]

[0078]

[0079] is the risk label response mapping function, which is used to convert the perturbation graph features and dynamic judgment results into specific classification label output; is the feature aggregation function of perturbation and path graph structure, and the feature aggregation function represents the aggregation integration of tensor coupling after and the path propagation structure function Model training dynamic judgment function Q(t) is used to measure whether the classification training process at time t meets the dual standards of convergence and error stability; is the element-wise product, represents the dynamic coupling operation between the perturbation evolution term and the structure coupling term, which means that at time t, the perturbation feature function and the corresponding path propagation structure function are calculated by element-level dynamic combination to form a dynamic response product;

[0080] A(t) is a disturbance amplitude function representing the current time t in the standard disturbance vector sequence, the value of which is composed of the signal main frequency amplitude extracted after performing Fourier transform on the device running state dimensions such as energy consumption, voltage, current, load, etc.; R(t) is the disturbance rate at the current time t, the disturbance rate represents the first order derivative of the disturbance amplitude A(t) within a local time window, reflecting the change trend of the disturbance intensity; F(t) is the disturbance frequency at the current time t, the disturbance frequency represents the numerical value of the dominant frequency in the disturbance signal, which is obtained by spectral analysis on the standard disturbance vector sequence, and is used to reflect the periodicity of the system state change; the disturbance characteristic function is the joint change rate between the disturbance amplitude and the rate, the joint change rate represents the dynamic response value formed by multiplying the sine function of the disturbance amplitude A(t) and its rate and taking the derivative with respect to time, and is used to reveal the instantaneous response degree of the disturbance state; the disturbance characteristic function is the second order change response of the disturbance frequency and amplitude change, the disturbance characteristic function represents the second order derivative calculation based on the logarithmic growth relationship between the disturbance frequency and the disturbance amplitude, and is used to reveal the curvature level change trend of the disturbance change;

[0081] M is the number of effective device nodes in the disturbance propagation path graph; m is the mth device node in the path graph; is a binary indicator function indicating whether the mth node in the path graph is activated at time t, and the value is 1 if the node is part of the current disturbance path, otherwise 0.

[0082] W m (t) is the connection weight value of the mth node at time t, which represents the connectivity strength between the node and other nodes in the graph structure, and is used to measure the structural importance of the node in the propagation path; D m (t) is the propagation delay between the mth node and the starting node of the propagation path, which represents the time span required for the disturbance signal to propagate from the starting point of the path to the node; L m (t) is the relative depth of the mth node in the graph topology structure, that is, the logical level position of the node from the starting point of the propagation path, and the deeper the level, the greater the value, which is used to model the structural attenuation characteristics of the path propagation; the path propagation structure function represents the dynamic coupling strength formed by combining the activation state, connection strength and propagation delay of each node in the current graph structure through a logarithmic function, which is used to describe the nonlinear propagation ability of the disturbance in the structural path; the path propagation structure function represents the exponential decay derivative of the connection strength with respect to the topological depth in the structural propagation path, which is used to describe the energy attenuation rate of the disturbance in the deep propagation process;

[0083] The perturbation risk classification loss function at the current time t is constructed in the form of cross entropy, which is used to measure the deviation between the model prediction label and the actual label; The derivative of the loss function with respect to time, The error reduction rate in the training process, which is used to evaluate whether the model is in a continuous optimization state; The aggregation feature function The derivative with respect to time, Reflects the stability and change of the perturbation atlas feature in the training process;||·||2 represents the Euclidean 2-norm, which is used to measure the overall change amplitude of the time gradient vector and serves as a normalization benchmark for constructing the dynamic change threshold;

[0084] Further, in the training classification module, define The perturbation risk classification loss value at time t, which is used to measure the overall deviation between the model output and the true risk label, and feedback the dynamic trend in the training process;

[0085]

[0086]

[0087] Where N is the total number of perturbation samples participating in training; y i The true risk label of the i-th perturbation sample, y i Indicates the actual category corresponding to the sample in the perturbation risk classification, which takes the value of the pre-defined risk level number, and the risk level number comes from artificial labeling or historical running state label, which belongs to a fixed binary or multi-classification label set; p i The prediction output of the i-th sample by the perturbation risk classification model, which is the probability of the sample belonging to the label y i After the softmax operation; The reconstruction result of the joint feature vector of the i-th sample by the perturbation risk classification model, As the intermediate expression of the model output layer; v i The original joint feature vector input to the model, which is composed of the perturbation propagation path set and the standard perturbation vector sequence;

[0088] The expression of the reconstruction error of the joint feature, For measuring the reconstruction ability of the model to the original joint feature vector in the encoding-decoding process; λ1 is the adjustment coefficient of the perturbation feature preservation loss term, which is used to control the weight of reconstruction error in the total loss function, and the value of the adjustment coefficient of the perturbation feature preservation loss term is selected by the cross-validation strategy in the training process; λ2 is the adjustment coefficient of the structural disturbance coupling loss term, which is used to adjust the proportion of the structural disturbance in the overall loss; A i (t) is the amplitude component of the i-th disturbance sample; R i (t) is the disturbance rate thereof; F i (t) is the disturbance frequency; M is the number of effective device nodes in the disturbance propagation path graph; is whether node j belongs to the disturbance path graph of the i-th sample; D ij (t) is the propagation delay of the j-th node; W ij (t) is the connection weight; L ij (t) is the topology depth.

[0089] The prediction generation module inputs the standard disturbance vector at the current time into the disturbance risk classification model, performs risk classification calculation, and outputs the risk category number corresponding to the standard disturbance vector;

[0090] According to the time index and device number of the standard disturbance vector, the corresponding node number and connection path information are extracted from the disturbance propagation path set to construct the index relationship set of device nodes and risk propagation paths;

[0091] The risk category number and the index relationship set are paired to form a risk node set in the order of node numbers, and each device node is attached with its corresponding risk category number;

[0092] The device number, node number, time index and risk category number in the risk node set are combined to construct a data graph with consistent field structure, and a dangerous prediction graph containing node information, time information and risk level is outputted for the intervention instruction generation operation of the sorting intervention module.

[0093] The sorting intervention module extracts the corresponding risk category number and connection degree value of each node in the dangerous prediction graph from the disturbance propagation path set to construct a sorting feature set containing risk intensity parameters and structural coupling parameters;

[0094] The sorting feature set is subjected to a double-factor sorting operation, and the risk category numbers are arranged from high to low, and if the risk category numbers are the same, the connection degree values are arranged from high to low, to generate a high-priority device node queue after sorting;

[0095] Each node in the high-priority device node queue is sequentially matched with its device number and current running state to generate corresponding load unloading instructions, running cycle resetting instructions, and visual alarm signals in combination with the risk category number;

[0096] After the load unloading instructions, running cycle resetting instructions, and visual alarm signals generated by each node are executed, the corresponding instruction success markers, resetting feedback values, and alarm trigger records of the node are extracted and written into the multivariate time series set consistent with the time index of the node to construct data update records matching the intervention actions for model iteration input in the training classification module.

[0097] The forming process of the present scheme is a systematic solution path proposed for the key problems of insufficient hazard prediction ability, response lag, and weak multi-source data fusion ability of the energy system in the current smart park. The scheme takes "dynamic analysis and early warning" as the core target and designs and constructs six functional modules of collection and organization, feature extraction, matching and mapping, training classification, prediction generation, and sorting intervention. The modules are connected by a data processing chain to form a dynamic closed-loop process system from bottom-level collection to high-level decision-making, thereby realizing fine modeling and real-time response in the whole process of hazard prediction. The design scheme is based on the current situation that the park energy management system still generally relies on static load models and rule-driven logic, and proposes a hazard prediction mechanism with structure propagation modeling capability and supporting disturbance identification and response control linkage;

[0098] In specific implementation, the scheme starts from the running state of devices and environment, collects the running data of air conditioners, elevators, lighting, and charging devices such as energy consumption, voltage, and current through the collection and organization module, and synchronously obtains the environmental temperature and humidity information at the corresponding time, generates multivariate time series data according to the unified field structure; the feature extraction module extracts disturbance amplitude, disturbance rate, and disturbance frequency and other frequency domain features using Fourier transform under fixed window length and sliding window mechanism to form a standard disturbance vector sequence with time continuity and frequency characteristics, which provides key feature support for subsequent disturbance path structure modeling and hazard prediction;

[0099] The matching mapping module matches the Euclidean distance between the disturbance vectors with the disturbance difference threshold, identifies the possible interference relationship between devices at the same time, and generates a set of disturbance propagation paths with pairing strength and directionality; the training classification module inputs the path structure information and the disturbance vector features jointly, constructs a disturbance risk classification model through supervised learning, and performs model dynamic verification and optimization based on the classification accuracy and the convergence rate of the loss function, then the prediction generation module generates a dangerous prediction map according to the risk judgment result of the current disturbance vector in the model, and labels the high-risk device nodes and their associated structures; the sorting intervention module sorts the nodes according to the risk level and the path coupling degree, converts the prediction results into executable offloading instructions, running cycle resetting commands and visual alarm signals, and writes them into the multivariate time series, completing the complete closed loop of early warning-intervention.

[0100] The design method of the modular series structure mainly considers the following aspects: first, in order to cope with the heterogeneous device structure and the nonlinear disturbance behavior, the unified field expression is needed to improve the fusion computing ability between systems; second, the multi-source dynamic characteristics have complex coupling characteristics, and the frequency domain analysis and structure propagation mechanism need to be introduced to realize effective modeling; third, the dangerous prediction model needs to have training and verification feedback ability to ensure the classification accuracy and stability; fourth, the early warning output needs to be quickly converted into intervention instructions, and can be written into the original data structure in a traceable closed loop, supporting the linkage control demand of the energy management platform, so the scheme has good system implementability and engineering practicability.

[0101] Finally, the above only describes the preferred embodiments of the present application and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A smart park dynamic analysis and early warning platform, comprising a collection and sorting module, a feature extraction module, a matching and mapping module, a training and classification module, a prediction and generation module, and a sorting and intervention module, characterized in that: The collection and sorting module obtains multi-device operation and environment state data, and arranges them into a uniform structure of a multivariate time series set according to device categories and time stamps; The feature extraction module extracts a disturbance feature group composed of disturbance amplitude, disturbance rate and disturbance frequency by performing Fourier transform and sliding window segmentation on the multivariate time series set, and generates a standard disturbance vector sequence after normalization processing; The matching and mapping module pairs and counts vectors with the same time stamp and a Euclidean distance below a threshold value in the standard disturbance vector sequence, and constructs a disturbance propagation path set according to device identifiers and pairing frequencies; The training and classification module inputs the disturbance propagation path set and the standard disturbance vector sequence into a classification calculation process, trains a disturbance risk classification model, and establishes a mapping relationship between the disturbance feature group and the risk label; The prediction and generation module inputs the current standard disturbance vector into the disturbance risk classification model, matches the risk label with the propagation path node number, and generates a hazard prediction graph containing node index and risk number; The sorting and intervention module sorts the nodes in the hazard prediction graph according to risk number and path connection degree, generates load unloading instructions, running cycle resetting instructions and visual alarm signals in sequence, and writes the execution data into the time series set. 2.The dynamic analysis and early warning platform of the smart park according to claim 1, wherein: The collection and sorting module sets a sampling period to obtain the unit time energy consumption value, voltage change value, current response value, and load proportion value of air conditioning equipment, lighting equipment, elevator drive devices, and electric vehicle charging devices, as well as the corresponding environment temperature value and environment humidity value for each sampling, and then labels the running state data and environment state data of each device according to the sampling time, and outputs the synchronous data group of device data and environment data; Each set of running state data and environment state data in the synchronous data group is matched at the field level according to the time stamp, and an aligned data structure containing sampling time, device type, device number, running state parameter, and environment state parameter is constructed through field correspondence, and a device-level binding data block with uniform field order and complete dimension is outputted; The device-level binding data block is subjected to device type uniform coding, field position regularization, and time axis alignment operation, and all data blocks are arranged according to the set time granularity and field structure, and a multivariate time series set with consistent field structure, clear device identifier, and continuous time is outputted. 3.The dynamic analysis and early warning platform of the smart park according to claim 2, characterized in that: The feature extraction module groups the multivariate time series set according to device number, obtains the time series segment corresponding to each type of device, and then performs Fourier transform under a fixed window length on each segment to extract frequency domain signals containing energy consumption change frequency distribution, voltage change frequency distribution, and current response frequency distribution, and outputs the frequency domain signal set of the corresponding device. The continuous segments in the frequency domain signal set are obtained by setting a fixed width and a fixed overlap ratio time sliding window, and then the truncation operation is performed in the window moving order to form a frequency domain sub-fragment sequence with continuous time indexes, and the frequency domain time distribution set containing all window fragments is output. 4.The dynamic analysis and early warning platform of a smart park according to claim 3, characterized in that: In the feature extraction module, the disturbance amplitude is obtained by calculating the difference between the maximum and minimum values of the signal amplitude of each sub-fragment in the frequency domain time distribution set, the disturbance rate is calculated by calculating the amplitude change slope between adjacent fragments, and the disturbance frequency is located by locating the main frequency point of each sub-fragment. Then, the three disturbance quantities are combined to form a disturbance feature data set, and the disturbance feature set is output. The disturbance amplitude, disturbance rate and disturbance frequency in the disturbance feature set are respectively calculated by linear normalization according to the preset upper and lower limits, and all feature data sets are rearranged according to the time index to generate a vector format expression with consistent fields and stable order, and a standard disturbance vector sequence is output. 5.The dynamic analysis and early warning platform of the smart park according to claim 4, characterized in that: The matching mapping module extracts the data with the same timestamp in the standard disturbance vector sequence to construct the disturbance vector combination group at the corresponding sampling time, and takes any two disturbance vectors in each group as a calculation object. Each group of calculation objects is calculated by Euclidean distance to form an Euclidean distance matrix containing each pair of disturbance vectors, and the judgment conditions include that the difference between the corresponding disturbance amplitudes in each vector pair is less than the preset disturbance difference threshold, and the value of the Euclidean distance calculation is lower than the set distance similarity threshold. If both conditions are met, it is identified as an effective disturbance vector pair and output to the disturbance vector pair set. Otherwise, the disturbance vector combination group at the timestamp is reconstructed and the Euclidean distance calculation is performed again. The device number and device type of each disturbance vector pair in the disturbance vector pair set are taken as matching key values, and the pairing frequency statistics and timestamp position record are performed to generate a disturbance linkage statistical record table containing paired device numbers, pairing frequencies and sampling time indexes. The device number pairs with pairing frequencies reaching the specified lower limit in the disturbance linkage statistical record table are constructed as directed node pairs, and the direction label is established according to the sampling time index to generate a disturbance propagation path set containing node numbers, connection weights and time label fields. 6.The dynamic analysis and early warning platform of the smart park according to claim 5, characterized in that: The training classification module extracts the node number, connection weight and time label from the disturbance propagation path set, combines the disturbance amplitude, disturbance rate and disturbance frequency corresponding to the timestamp in the standard disturbance vector sequence, and constructs a joint feature input group containing path features and disturbance features. The joint feature input group is divided into training sets, and the classification label set is generated according to the historical running state label or artificial annotation result of each path to construct the training data pair between the feature group and the risk label. The training data pair is input into the classification calculation process to perform feature encoding, weight initialization and iterative optimization operations, train the disturbance risk classification model, and form the mapping structure between the input features and the output labels. The trained disturbance risk classification model is verified to calculate the classification accuracy of the joint feature input group on the test data, and the loss function convergence rate based on the risk label prediction bias and the training round change rate is solved. If the classification accuracy is higher than the lower limit of the recognition accuracy and the convergence rate exceeds the loss decrease rate threshold, it is confirmed that the model meets the training validity requirement, and the perturbation risk classification model is output and input to the prediction generation module to perform risk map generation operation; if any judgment condition is not met, the feature encoding and weight initialization are re-executed for the current training data pair, and a new perturbation risk classification model is constructed.

7. The dynamic analysis and early warning platform for a smart park according to claim 6, characterized in that: The prediction generation module inputs the standard perturbation vector at the current time into the perturbation risk classification model, performs risk classification calculation, and outputs the risk category number corresponding to the standard perturbation vector; According to the time index and device number of the standard perturbation vector, the corresponding node number and connection path information are extracted from the perturbation propagation path set to construct the index relationship set of device nodes and risk propagation paths; The risk category number and the index relationship set are paired and combined to form a risk node set in the order of node number, and the corresponding risk category number is attached to each device node; The device number, node number, time index and risk category number in the risk node set are combined to construct a data graph with consistent field structure, and a hazard prediction graph containing node information, time information and risk level is output. 8.The dynamic analysis and early warning platform of the smart park according to claim 7, characterized in that: The sorting intervention module extracts the corresponding risk category number and connection degree value of each node in the hazard prediction graph from the perturbation propagation path set to construct a sorting feature set containing risk intensity parameters and structure coupling parameters; The sorting feature set is sorted by double factors, arranged from high to low according to the risk category number, and if the risk category number is the same, arranged from high to low according to the connection degree value, to generate a high priority device node queue after sorting; Each node in the high priority device node queue is matched with its device number and current running state to generate corresponding load unloading instructions, running cycle reset instructions and visual alarm signals combined with the risk category number; After executing the load unloading instructions, running cycle reset instructions and visual alarm signals generated by each node, the instruction success flag, reset feedback value and alarm trigger record corresponding to the node are extracted and written into the multivariate time series set consistent with the time index of the node to construct data update records matching the intervention actions.

Images