Data acquisition method and system based on dynamic data state model configuration
By deploying edge computing nodes in the energy system, collecting and processing operating state parameters in real time, and using the dynamic data state model configuration method, the problem that traditional static models cannot adapt to dynamic changes is solved, high-efficiency management and abnormal detection are achieved, and the operation efficiency and reliability of the energy system are improved.
Patent Information
- Application Number
- CN202510413146.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-03
- Publication Date
- 2025-07-08
AI Technical Summary
Traditional data processing methods rely on static models and cannot adapt to the dynamic changes in the operating state of the energy system, resulting in reduced prediction accuracy and reduced accuracy of abnormal detection. At the same time, the energy consumption management of large-scale sensor networks is difficult to balance, resulting in waste of energy or insufficient data acquisition accuracy.
Using a method based on dynamic data state model configuration, the operating state parameters of the energy system are collected and processed in real time through edge computing nodes, and the data state model is dynamically updated using incremental training algorithms. Combining the time series prediction model and anomaly detection model, the data acquisition strategy of the sensor network is dynamically adjusted, including model compression and pruning optimization.
Real-time monitoring and efficient management of energy systems are realized, operating efficiency and reliability are improved, and can quickly respond to system status changes, detect abnormalities in a timely manner and optimize the energy consumption of the sensor network, and extend the service life.
Smart Images
Figure CN120281082A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data acquisition and analysis, and mainly relates to a data acquisition method and system based on dynamic data state model configuration. Background Art
[0002] The efficient and stable operation of the energy system is of great significance for ensuring the sustainable development of the economy and society. With the rapid progress of Internet of Things (IoT) technology and the continuous construction of smart grids, sensors are increasingly widely used in the energy system. These sensors can real-time monitor various operating parameters of the energy system, such as current, voltage, temperature, and pressure, etc., providing valuable data support for the analysis and management of the system operating state.
[0003] However, traditional data processing methods mainly rely on static models, which are trained under specific conditions and cannot effectively adapt to the dynamic changes of the energy system operating state. For example, when the grid load suddenly increases or the performance of the equipment deteriorates due to aging, static models often cannot capture these changes in time, resulting in a decrease in prediction accuracy and the accuracy of anomaly detection. This limitation severely restricts the improvement of the energy system operation efficiency and the timely discovery of potential problems.
[0004] In addition, the deployment of large-scale sensor networks brings challenges to energy consumption management. Each sensor consumes energy during continuous operation, and it is often difficult to provide a stable power supply in the remote locations of the energy system. Traditional data acquisition methods often struggle to achieve a balance between energy efficiency and performance, either resulting in energy waste or insufficient data acquisition accuracy.
[0005] Therefore, there is an urgent need for a data acquisition method that can achieve real-time monitoring and efficient energy efficiency management, and dynamically adjust the data acquisition strategy of the sensor network, thereby improving the operation efficiency and reliability of the energy system. Summary of the Invention
[0006] In order to solve the above problems existing in the prior art, the present application provides a data acquisition method and system based on dynamic data state model configuration.
[0007] The technical solution of the present application is as follows:
[0008] A data acquisition method based on dynamic data state model configuration, the method includes:
[0009] Real-time collect the operating state parameters of the energy system through the sensor network;
[0010] Build a data state model on the edge computing node based on the operating state parameters, dynamically update the data state model through an incremental training algorithm, and optimize the data state model using model compression and model pruning; wherein, the data state model is a hybrid model architecture, including a time series prediction model and an anomaly detection model;
[0011] Perform state prediction and anomaly detection on the energy system through the real-time updated and optimized data state model, and dynamically adjust the data acquisition strategy of the sensor network according to the prediction results and anomaly detection results output by the data state model.
[0012] As a preferred embodiment of the present invention, the specific method for real-time collecting the operating state parameter data of the energy system through the sensor network is:
[0013] Deploy edge computing nodes in the sensor network. The edge computing nodes are responsible for collecting the operating state parameters of the energy system collected by the sensors and performing data preprocessing. The data preprocessing includes using a low-pass filter to remove high-frequency noise, using wavelet transform to remove noise, and normalizing the data to the interval [0,1]; constructing the preprocessed operating state parameters into a multi-dimensional time series X = [x1, x2,... x t ,..., x T , where x t = [x t1 , x t2 ,..., x tn T is the n operating state parameters at the t-th moment.
[0014] As a preferred embodiment of the present invention, the time series prediction model is an Attention-based Long Short-Term Memory Network (Attention-based LSTM) model, including a forget gate, an input gate, and an output gate, where:
[0015] The forward propagation of the Attention-based LSTM model is expressed by the formula:
[0016] f t = σ(W f [h t-1 , x t + b f );
[0017] i t = σ(W i [h t-1 , x t + b i );
[0018] o t = σ(W o [ht-1 , x t + b o );
[0019]
[0020] Where f t is the output of the forget gate; σ is the sigmoid activation function; W f is the weight matrix of the forget gate; b f is the bias term of the forget gate; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the bias term of the input gate; o t is the output of the output gate; W o is the weight matrix of the output gate; b o is the bias term of the output gate; c t is the cell state at the current time step; is the information update part controlled by the input gate; tanh is the hyperbolic tangent activation function; W c is the weight matrix for cell state update; h t is the hidden state at the current time step; h t-1 is the hidden state at the previous time step; is the Hadamard product;
[0021] The calculation of the attention weights of the Attention-based LSTM model is expressed by the formula:
[0022]
[0023] e t is the attention score; is the transpose of the attention vector v a ; W a is the hidden state weight matrix; U a is the output state weight matrix at the previous time step; s t-1 is the output state at the previous time step; b a is the bias vector;
[0024] Normalize the attention weights, expressed by the formula:
[0025] α t = softmax(e t );
[0026]
[0027] s t = tanh(W s [ht , c t + b s );
[0028] Wherein, α t is the normalized attention weight; s t is the output state at the current time step; W s is the weight matrix; b s is the bias term;
[0029] The final predicted output y of the Attention-based LSTM model t is expressed by the formula as:
[0030] y t = W y · s t + b y ;
[0031] Wherein, W y is the output weight matrix, and b y is the corresponding output bias term.
[0032] As a preferred embodiment of the present invention, the anomaly detection model is an autoencoder model, including an encoder and a decoder, wherein:
[0033] The encoder is used to compress the multi-dimensional time series data x at the current time step t into a low-dimensional representation z, which is expressed by the formula as:
[0034] z = σ(W e x t + b e );
[0035] Wherein, W e is the weight matrix of the encoder; b e is the bias of the encoder;
[0036] The decoder is used to reconstruct the low-dimensional representation z into the original input data which is expressed by the formula as:
[0037]
[0038] Wherein, W d is the weight matrix of the decoder; b d is the bias of the decoder;
[0039] Calculate the reconstruction error e t , which is expressed by the formula as:
[0040]
[0041] If the reconstruction error e t exceeds a preset error threshold, it is determined that there is an abnormality in the energy system and an alarm is triggered.
[0042] As a preferred embodiment of the present invention, optimizing the data-state model by using model compression and model pruning is specifically as follows:
[0043] The model compression is specifically to use singular value decomposition (SVD) to decompose each weight matrix in the data-state model into the product of two low-rank matrices; the model pruning is specifically to calculate the absolute value of each element in each weight matrix in the data-state model, and set the weight values with absolute values less than the preset weight threshold to 0.
[0044] As a preferred embodiment of the present invention, dynamically adjusting the data acquisition strategy of the sensor network includes:
[0045] Using a Bayesian neural network to estimate the variance σ 2 (x t ) of the prediction result in the data-state model, defining a variance threshold τ, and when σ 2 (x t ) > τ, increasing the data acquisition frequency;
[0046] When the anomaly detection model in the data-state model determines that there is an anomaly in the energy system, increasing the data acquisition frequency of the sensors corresponding to the operating state parameters;
[0047] Calculating the performance-power consumption ratio of each sensor, sorting the performance-power consumption ratios, dividing the sensor priorities according to the sorting results, and increasing the sensor sampling frequency according to the sensor priorities.
[0048] As a preferred embodiment of the present invention, calculating the performance-power consumption ratio of each sensor is specifically as follows:
[0049] Defining the energy consumption c i of each sensor and the contribution degree ω i to the prediction of the data-state model, where:
[0050] The energy power consumption of the sensor is expressed by the formula:
[0051] c i = α i · f i + β i · d i + γ i ;
[0052] In the formula, c i is the power consumption of the i-th sensor; f i is the sampling frequency; d i is the amount of transmitted data; α i and βi and γ i are model parameters;
[0053] The total contribution of each sensor is expressed by the formula:
[0054]
[0055] The total contribution is normalized to obtain the relative contribution, which is expressed by the formula:
[0056]
[0057] Calculate the performance - energy consumption ratio r of each sensor i , which is expressed by the formula:
[0058]
[0059] The present invention also provides a data acquisition system based on dynamic data state model configuration. The system includes a multi - source data acquisition module, a data state model construction module, and a dynamic configuration module, wherein:
[0060] The multi - source data acquisition module is used to collect the operation state parameters of the energy system in real - time through a sensor network;
[0061] The data state model construction module constructs a data state model at an edge computing node based on the operation state parameters, dynamically updates the data state model through an incremental training algorithm, and optimizes the data state model by using model compression and model pruning; wherein, the data state model is a hybrid model architecture, including a time - series prediction model and an anomaly detection model; the state of the energy system is predicted and anomaly - detected through the real - time updated and optimized data state model;
[0062] The dynamic configuration module is used to dynamically adjust the data acquisition strategy of the sensor network according to the prediction results and anomaly detection results output by the data state model. The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a data acquisition method based on dynamic data state model configuration as described in any embodiment of the present invention.
[0063] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a data acquisition method based on dynamic data state model configuration as described in any embodiment of the present invention.
[0064] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements a data acquisition method based on dynamic data state model configuration as described in any embodiment of the present invention.
[0065] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0066] 1) The present invention provides a data acquisition method and system based on dynamic data state model configuration. By using edge computing nodes to collect and process the operating state parameters of the energy system in real time and using the incremental training algorithm to dynamically update the data state model, it can quickly respond to system state changes and ensure the real-time nature of data and the dynamic adaptability of the model. Through real-time update and optimization, the present invention can timely respond to changes in the operating state, improve the response speed and operating efficiency.
[0067] 2) The present invention provides a data acquisition method and system based on dynamic data state model configuration. The data state model adopts a hybrid architecture of time series prediction model and anomaly detection model, which can perform state prediction and anomaly detection simultaneously, significantly improving the intelligent level. Through the hybrid model of the present invention, not only can it predict the future operating state, but also it can timely detect abnormal situations and trigger early warnings, thereby enhancing reliability and security.
[0068] 3) The present invention provides a data acquisition method and system based on dynamic data state model configuration. According to the prediction results and anomaly detection results of the data state model, it dynamically adjusts the data acquisition strategy of the sensor network, optimizes the energy consumption of the sensors, and prolongs the service life of the sensor network. The adaptive strategy not only improves the efficiency of data acquisition, but also reduces energy consumption and ensures the long-term stable operation of the acquisition. Description of the Drawings
[0069] Figure 1 is the flowchart of the method of the embodiment of the present invention. Detailed Embodiments
[0070] The following describes the detailed embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.
[0071] The present invention provides the following technical solutions: A data acquisition method and system based on dynamic data state model configuration.
[0072] Embodiment 1:
[0073] As Figure 1As shown, this embodiment provides a data acquisition method based on dynamic data state model configuration, and the method includes:
[0074] S1. Real-time collect the operation state parameters of the energy system through the sensor network;
[0075] S11. The sensor network is specifically a distributed network composed of multiple sensor nodes deployed in the energy system, which can independently collect, process, and transmit data;
[0076] S12. The operation state parameters of the energy system include key parameters such as temperature, pressure, flow rate, and electricity;
[0077] S13. Deploy edge computing nodes in the sensor network. The edge computing nodes are responsible for collecting the operation state parameters of the energy system collected by the sensors and performing data preprocessing to reduce the need for data transmission to the central server. Specifically, the data preprocessing includes using a low-pass filter to remove high-frequency noise, using wavelet transform to remove noise, and normalizing the data to the interval [0,1];
[0078] S14. Construct the operation state parameters after data preprocessing into a multi-dimensional time series X = [x1, x2,... x t ,..., x T , where x t = [x t1 , x t2 ,..., x tn T is the n operation state parameters at the t-th moment;
[0079] S2. Dynamically construct and update the data state model based on the operation state parameters, and use edge computing to optimize the data state model;
[0080] S21. The data state model is a hybrid model architecture, including a time series prediction model and an anomaly detection model, where:
[0081] S211. In this embodiment, the time series prediction model uses an attention-based long short-term memory network Attention-based LSTM model to predict future operation state parameters;
[0082] The basis of the Attention-based LSTM model is still the structure of the standard LSTM, including a forget gate, an input gate, and an output gate. Its forward propagation is expressed by the formula:
[0083] f t = σ(W f [h t-1 , x t +b f );
[0084] i t =σ(W i [h t-1 ,x t +b i );
[0085] o t =σ(W o [h t-1 ,x t +b o );
[0086]
[0087] where f t is the output of the forget gate, which determines how much information to forget from the previous cell state c t-1 at the current time step t; σ is the sigmoid activation function, which outputs a value between 0 and 1; W f is the weight matrix of the forget gate; b f is the bias term of the forget gate; i t is the output of the input gate, which determines how much new information needs to be added to the cell state at the current time step t; W i is the weight matrix of the input gate; b i is the bias term of the input gate; o t is the output of the output gate, which determines how much information of the cell state needs to be output to the hidden state at the current time step t; W o is the weight matrix of the output gate; b o is the bias term of the output gate; c t is the cell state at the current time step, which is also the context vector and is the weighted sum of the hidden states at all time steps, and the weights are determined by the attention mechanism; is the information update part controlled by the input gate; tanh is the hyperbolic tangent activation function, which is used to control the intensity of the information; W c is the weight matrix for cell state update; h t is the hidden state at the current time step; h t-1 is the hidden state at the previous time step; is the Hadamard product;
[0088] Based on the standard LSTM, an attention mechanism is added to calculate the attention weights and apply them to the hidden states. Specifically:
[0089] The calculation of the attention weights is expressed by the formula:
[0090]
[0091] e t is the attention score, representing the current hidden state h t for generating the context vector c t importance; is the attention vector v a transpose; W a is the hidden state weight matrix, used to calculate part of the hidden state; U a is the output state weight matrix at the previous time step; s t-1 is the output state at the previous time step; b a is the bias vector;
[0092] Normalize the attention weights, expressed by the formula:
[0093] α t = softmax(e t );
[0094]
[0095] s t = tanh(W s [h t , c t + b s );
[0096] In the formula, α t is the normalized attention weight, representing the attention distribution at the current time step t; s t is the output state at the current time step; W s is the weight matrix, used to linearly transform the concatenated hidden state h t and the context vector c t into a new vector; b s is the bias term, used to adjust the result of the linear transformation;
[0097] Based on the above steps, the Attention - based LSTM model finally predicts the output y t expressed by the formula:
[0098] y t = W y · s t + b y ;
[0099] In the formula, W y is the output weight matrix, b y is the output corresponding bias term;
[0100] S212. In this embodiment, the anomaly detection model is an autoencoder model, including an encoder and a decoder, where:
[0101] The encoder is used to compress the multi-dimensional time series data x at the current time step t into a low-dimensional representation z, which is expressed by the formula:
[0102] z = σ(W e x t + b e );
[0103] In the formula, W e is the weight matrix of the encoder; b e is the bias of the encoder;
[0104] The decoder is used to reconstruct the low-dimensional representation z into the original input data which is expressed by the formula:
[0105]
[0106] In the formula, W d is the weight matrix of the decoder; b d is the bias of the decoder;
[0107] Calculate the reconstruction error e t , which is expressed by the formula:
[0108]
[0109] If the reconstruction error e t exceeds the preset error threshold, it is determined that the energy system has an anomaly and an alarm is triggered; preferably, the preset threshold is set by statistical methods or experience;
[0110] The data state model in this embodiment can not only predict the future operating state, but also detect in real time whether the energy system has an anomaly, providing double protection;
[0111] S213. In this embodiment, the historical operating state parameters of the energy system are also pre-acquired in advance, a training data set is constructed by using the historical operating state parameters, and the Attention-based LSTM model is trained based on the training data set to optimize the prediction performance; and the historical operating state parameters under normal conditions are used to train the autoencoder model to optimize the reconstruction performance;
[0112] Preferably, the trained data state model is deployed to the edge computing node to achieve local processing and real-time monitoring
[0113] S22. Since the operating state parameters of the energy system may change over time and with the environment (such as load fluctuations, equipment aging, seasonal changes, etc.), and over time, the adaptability of the data-state model to new data will gradually weaken, resulting in a decline in prediction and detection performance, and the abnormal patterns may change with the change of the operating state of the energy system. The data-state model needs to be dynamically adjusted to identify new anomalies. Therefore, this embodiment also includes dynamically updating the data-state model. Specifically:
[0114] Each time new operating state parameters are received, the new operating state parameters are incorporated into the training data set, and the incremental training algorithm is used to update the weight matrix and bias term of the data-state model. Specifically, the mean squared error loss function is used to calculate the loss between the predicted output and the actual output, and the gradient is calculated through backpropagation and the model parameters are updated to avoid retraining the entire model;
[0115] Preferably, the edge computing node is used for model update to reduce the computing pressure on the central server;
[0116] S23. This embodiment also includes optimizing the data-state model on the edge computing node to reduce data transmission delay, improve response speed, and reduce the load on the central server. Specifically:
[0117] The optimization of the data-state model includes model compression and model pruning, where:
[0118] The model compression is to reduce the storage and computing overhead of each weight matrix in the data-state model through low-rank approximation. In this embodiment, specifically, the singular value decomposition (SVD) is used to decompose each weight matrix into the product of two low-rank matrices to reduce the computing overhead;
[0119] The model pruning is to prune unimportant weights by setting a threshold. In this embodiment, specifically, the absolute value of each element in each weight matrix of the data-state model is calculated, and the weight values with absolute values less than the preset weight threshold are set to 0;
[0120] Preferably, in order to further reduce the storage overhead of the sparse matrix after pruning, the sparse matrix representation method (such as CSR or CSC format) is used to only store the non-zero elements and their positions;
[0121] S3. The state of the energy system is predicted and anomalies are detected through the real-time updated and optimized data-state model. According to the prediction results and anomaly detection results output by the data-state model, the data acquisition strategy of the sensor network is dynamically adjusted;
[0122] Dynamically adjusting the data acquisition strategy of the sensor network includes:
[0123] S31. Estimate the variance σ of the prediction result of the Attention-based LSTM model in the data state model using a Bayesian neural network 2 (x t ), define a variance threshold τ. When σ 2 (x t ) > τ, increase the data acquisition frequency;
[0124] S32. When the anomaly detection model in the data state model determines that an anomaly has occurred in the energy system, increase the data acquisition frequency of the sensors corresponding to the operating state parameters;
[0125] S33. Define the energy consumption c i of each sensor and its contribution degree ω i to the prediction of the data state model. Among them:
[0126] The energy consumption of the sensor is related to factors such as the amount of transmitted data and the sampling frequency, and is expressed by the formula:
[0127] c i = α i ·f i + β i ·d i + γ i ;
[0128] In the formula, c i is the power consumption of the i-th sensor; f i is the sampling frequency; d i is the amount of transmitted data; α i , β i and γ i are model parameters, determined by collecting data through actual measurement of the power consumption of the sensor under different working conditions;
[0129] Since x t contains n operating state parameters, that is, n sensor data, the contribution degree of the sensor to the prediction of the data state model is specifically to evenly distribute the attention weight α t to each sensor in x t , that is, the weight of each sensor is Then, for each sensor, its total contribution degree is the sum of its weights at all time steps, and is expressed by the formula:
[0130]
[0131] Normalize the total contribution degree to obtain the relative contribution degree, and express it by the formula:
[0132]
[0133] Calculate the performance - energy consumption ratio r of each sensori , which is expressed by the formula:
[0134]
[0135] Sorted from high to low according to r i Sensors with high performance and low energy consumption have higher priorities. In the case of limited resources in the energy system, the sampling frequency of high-priority sensors is preferentially increased, while the sampling frequency of low-priority sensors is decreased, and the sampling frequency of high-r i sensors is preferentially guaranteed.
[0136] Embodiment 2:
[0137] This embodiment provides a data acquisition system based on dynamic data state model configuration. The system includes a multi-source data acquisition module, a data state model construction module, and a dynamic configuration module, where:
[0138] The multi-source data acquisition module is used to collect the operation state parameters of the energy system in real time through a sensor network;
[0139] The data state model construction module constructs a data state model at an edge computing node based on the operation state parameters, dynamically updates the data state model through an incremental training algorithm, and optimizes the data state model by using model compression and model pruning; among them, the data state model is a hybrid model architecture, including a time series prediction model and an anomaly detection model; the state of the energy system is predicted and anomaly detection is performed through the real-time updated and optimized data state model;
[0140] The dynamic configuration module is used to dynamically adjust the data acquisition strategy of the sensor network according to the prediction results and anomaly detection results output by the data state model.
[0141] Embodiment 3:
[0142] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a data acquisition method based on dynamic data state model configuration as described in Embodiment 1 of the present invention;
[0143] Embodiment 4:
[0144] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements a data acquisition method based on dynamic data state model configuration as described in Embodiment 1 of the present invention.
[0145] It should be noted that the system, electronic device, and computer-readable storage medium of the present invention are all based on the same inventive concept as the method described in Embodiment 1 of the present invention, and will not be elaborated herein.
[0146] The above are only embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.
Claims
1. A data acquisition method based on dynamic data state model configuration, characterized in that, The method includes: Real-time collecting the operation state parameters of the energy system through a sensor network; Based on the operation state parameters, constructing a data state model at an edge computing node, dynamically updating the data state model through an incremental training algorithm, and optimizing the data state model by using model compression and model pruning; wherein, the data state model is a hybrid model architecture, including a time series prediction model and an anomaly detection model; Performing state prediction and anomaly detection on the energy system through the real-time updated and optimized data state model, and dynamically adjusting the data collection strategy of the sensor network according to the prediction result and anomaly detection result output by the data state model.
2. The data acquisition method based on dynamic data state model configuration according to claim 1, wherein Specifically, real-time collecting the operation state parameters of the energy system through a sensor network is as follows: Deploy edge computing nodes in the sensor network. The edge computing nodes are responsible for collecting the operating state parameters of the energy system collected by the sensors and performing data preprocessing. The data preprocessing includes using a low-pass filter to remove high-frequency noise, using wavelet transform to remove noise, and normalizing the data to the interval [0, 1]; constructing the operating state parameters after data preprocessing into a multi-dimensional time series X = [x1, x2,...x t ,...,x T , where x t = [x t1 ,x t2 ,...,x tn T are the n operating state parameters at the t-th moment. 3. The data acquisition method based on dynamic data state model configuration according to claim 2, wherein The time series prediction model is an Attention-based Long Short-Term Memory (Attention-based LSTM) model based on an attention mechanism, including a forget gate, an input gate, and an output gate, where: The forward propagation of the Attention-based LSTM model is expressed by the formula: f t = σ(W f [h t-1 , x t + b f ); i t = σ(W i [h t-1 , x t + b i ); o t = σ(W o [h t-1 , x t + b o ); where f t is the output of the forget gate; σ is the sigmoid activation function; W f is the weight matrix of the forget gate; b f is the bias term of the forget gate; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the bias term of the input gate; o t is the output of the output gate; W o is the weight matrix of the output gate; b o is the bias term of the output gate; c t is the cell state at the current time step; is the information update part controlled by the input gate; tanh is the hyperbolic tangent activation function; W c is the weight matrix for cell state update; h t is the hidden state at the current time step; h t-1 is the hidden state at the previous time step; is the Hadamard product; The calculation of the attention weights of the Attention-based LSTM model is expressed by the formula: e t is the attention score; is the transpose of the attention vector v a ; W a is the hidden state weight matrix; U a is the output state weight matrix at the previous time step; s t-1 is the output state at the previous time step; b a is the bias vector; Normalizing the attention weights is expressed by the formula: α t = softmax(e t ); s t = tanh(W s [h t , c t + b s ); where α t is the normalized attention weight; s t is the output state at the current time step; W s is the weight matrix; b s is the bias term; The final predicted output y of the Attention-based LSTM model t Expressed by the formula as: y t = W y · s t + b y ; Where W y is the output weight matrix, and b y is the corresponding output bias term.
4. A data acquisition method based on dynamic data state model configuration according to claim 2, characterized in that The anomaly detection model is an autoencoder model, including an encoder and a decoder, where: The encoder is used to compress the multi-dimensional time series data x at the current time step t into a low-dimensional representation z, which is expressed by the formula: z = σ(W e x t + b e ); where, W e is the weight matrix of the encoder; b e is the bias of the encoder; The decoder is used to reconstruct the low-dimensional representation z into the original input data Expressed by the formula as: where, W d is the weight matrix of the decoder; b d is the bias of the decoder; Calculate the reconstruction error e t , which is expressed by the formula as follows: If the reconstruction error e t exceeds a preset error threshold, it is determined that there is an abnormality in the energy system and an alarm is triggered.
5. A data acquisition method based on dynamic data state model configuration according to claim 2, characterized in that, Specifically, optimizing the data state model by using model compression and model pruning is as follows: Specifically, the model compression is to decompose each weight matrix in the data state model into the product of two low-rank matrices by using Singular Value Decomposition (SVD); the model pruning is to calculate the absolute value of each element in each weight matrix of the data state model, and set the weight value whose absolute value is less than a preset weight threshold to 0.
6. A data acquisition method based on dynamic data state model configuration according to claim 2, characterized in that, Dynamically adjusting the data collection strategy of the sensor network includes: Estimating the variance σ of the prediction result in the data state model using a Bayesian neural network 2 (x t ), defining a variance threshold τ, and when σ 2 (x t ) > τ, increasing the data acquisition frequency; When the anomaly detection model in the data state model determines that the energy system has an anomaly, increasing the data collection frequency of the sensors corresponding to the operation state parameters; Calculating the performance-energy consumption ratio of each sensor, sorting the performance-energy consumption ratios, dividing the sensor priorities according to the sorting result, and increasing the sensor sampling frequency according to the sensor priorities.
7. A data acquisition method based on dynamic data state model configuration according to claim 6, characterized in that Specifically, calculating the performance-energy consumption ratio of each sensor is as follows: Define the energy consumption c of each sensor i and the contribution degree ω to the prediction of the data state model i where: The energy consumption of the sensor is expressed by the formula: c i = α i · f i + β i · d i + γ i ; where c i is the power consumption of the i-th sensor; f i is the sampling frequency; d i is the amount of transmitted data; α i , β i and γ i are model parameters; The total contribution degree of each sensor is expressed by the formula: Normalizing the total contribution degree to obtain the relative contribution degree is expressed by the formula: Calculate the performance - energy consumption ratio r of each sensor i , which is expressed by the formula as follows:
8. A data acquisition system based on dynamic data state model configuration, characterized in that, The system includes a multi-source data collection module, a data state model construction module, and a dynamic configuration module, where: The multi-source data collection module is used to real-time collect the operation state parameters of the energy system through a sensor network; The data state model construction module constructs a data state model at an edge computing node based on the operation state parameters, dynamically updates the data state model through an incremental training algorithm, and optimizes the data state model by using model compression and model pruning; wherein, the data state model is a hybrid model architecture, including a time series prediction model and an anomaly detection model; performing state prediction and anomaly detection on the energy system through the real-time updated and optimized data state model; The dynamic configuration module is used to dynamically adjust the data collection strategy of the sensor network according to the prediction result and anomaly detection result output by the data state model.
9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a data acquisition method based on the configuration of a dynamic data state model as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a data acquisition method based on the configuration of a dynamic data state model as described in any one of claims 1 to 7.