A power grid state evaluation and prediction method, device, terminal and medium

Abstract

The application discloses a power grid state evaluation and prediction method, device, terminal and medium, the method comprises the following steps: real-time collection and preprocessing of power distribution network terminal equipment operation data in smart grid, spectrum analysis is carried out, and the characteristics of the signal are analyzed through wavelet transform; a nonlinear system response model based on deep learning is constructed, and the parameters of the response model are optimized; in combination with machine learning, the power grid dynamics are evaluated and predicted according to the results of FFT and wavelet transform and the output of the above response model. The application provides a comprehensive and efficient solution for real-time collection and preprocessing of power grid data in smart grid and evaluation and prediction of smart grid dynamics. Through innovative technical means and algorithms, the application shows significant advantages in the stable and efficient operation of smart grid, especially in analyzing, evaluating and predicting the state characteristics of the power grid.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power systems, in particular to a power grid state evaluation and prediction method, device, terminal and medium. BACKGROUND

[0002] In order to adapt to the acceleration of the integration process of smart grid regulation, the stable operation of distribution network terminal equipment and the dynamic evaluation and prediction of power grid have become the core demand. Under this background, the existing distribution network operation data acquisition system is facing many challenges, especially in monitoring the running state, early warning, and processing the transmission characteristics and response identification of sinusoidal signals, as well as the state evaluation and state prediction of smart grid. SUMMARY

[0003] The present application provides a power grid state evaluation and prediction method, device, terminal and medium, which has the advantage of more accurate state evaluation and state prediction of smart grid.

[0004] The technical scheme of the present application is as follows:

[0005] On the one hand, the present application provides a power grid state evaluation and prediction method, comprising the following steps:

[0006] Step 1: Real-time acquisition of distribution network terminal equipment operation data;

[0007] Step 2: Construct a nonlinear system response model based on deep learning to predict the behavior of the power grid;

[0008] Step 3: Construct a BPNN model, and use the prediction results of the nonlinear system response model in step 2 and the data in step 1 as input to predict the behavior of the power grid;

[0009] The behavior of the power grid described in steps 2 and 3 includes power grid load and / or power grid performance.

[0010] Further, in step 1, the collected data is preprocessed: the collected data is subjected to spectral analysis, and the characteristics of the signal are analyzed by wavelet transform.

[0011] Further, step 1 includes: using a high-precision sensor to collect original power grid signals from the distribution network terminal equipment, expressed in the form of a time function s(t), using a band-pass filter F bp (s(t)) to process the signal s(t), obtaining the filtered signal s filtered (t) = F bp (s(t)), calculating the average value μ and the standard deviation σ of the filtered signal, obtaining Using the Hanning window function w(t) to process the signal, obtaining the windowed signal s windowed (t) = s preThe signal (t)·w(t) is transformed from the time domain to the frequency domain by performing a Fast Fourier Transform (FFT) to obtain the spectrum. Through wavelet transform For the normalized signal s pre (t) performs multi-scale decomposition to capture the instantaneous characteristics and nonlinear changes of the signal.

[0012] Furthermore, step 2 includes: constructing an FbProphet model and an LSTM model; using the FbProphet model to capture the trend and seasonal patterns of time series data, analyzing the trend of power grid data, and predicting future power grid load or performance; and using the LSTM model to capture short-term dependencies in time series data and predict the short-term behavior of the power grid.

[0013] Furthermore, the mathematical expression of the FbProphet model is: y(t)=g(t)+s(t)+h(t)+e(t), where g(t) represents the trend term, which represents the trend of the time series in non-periodic terms; s(t) represents the periodic term; h(t) represents the holiday term, which indicates whether there is a holiday on the current day; and e(t) represents the error term or the residual term.

[0014] Furthermore, the LSTM model includes an input gate, a forget gate, and an output gate, mathematically expressed as follows: i t =σ(U i x t +V i h t-1 +b i );f t =σ(U f x t +V f h t-1 +b f );o t =σ(U0x t +V o h t-1 +b o );c t =f t ⊙c t-1 +i t ⊙tan h(U ia x t +V ia h t-1 +b ia );h t =o t ⊙tan h(c t ); where U and V are weight matrices, b is the bias value of the corresponding gate, the symbol ⊙ represents the Hadamard product, σ represents the activation of the corresponding gate, and is the sigmoid activation function.

[0015] Further, in step 3, the output data of the FbProphet model and the LSTM model in step 2 are taken as the input of the BPNN model, and the data S(f) after FFT transformation and the data W after wavelet transformation processing are taken as the input of the BPNN model. s (t) is also taken as the input of the BPNN model, and the prediction results of the FbProphet model and the LSTM model are fused and output.

[0016] On the other hand, the application provides a power grid state evaluation and prediction device, comprising:

[0017] An acquisition unit is configured to acquire power grid terminal device operation data in real time;

[0018] A first prediction unit is configured to construct a deep learning-based nonlinear system response model to predict power grid behavior.

[0019] A second prediction unit is configured to construct a BPNN model, take the prediction result in the first prediction unit and the data obtained by the acquisition unit as the input to predict power grid behavior.

[0020] In another aspect, the application provides an electronic terminal comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is called and executed by the processor to implement the power grid state evaluation and prediction method as described above.

[0021] In another aspect, the application provides a computer readable medium storing a computer program, and the computer program is called and executed by a computer to implement the power grid state evaluation and prediction method as described above.

[0022] In summary, the application provides a method for real-time collection and analysis of power distribution network terminal device operation data in a smart grid, and evaluation and prediction of power grid status. It uses high-precision sensors to collect power grid signals, ensuring high quality and accuracy of the obtained data, which is crucial for monitoring and managing the power grid. In signal preprocessing, the application uses an advanced band-pass filter for noise removal, effectively eliminating high and low frequency noise, and normalizing the filtered signal to ensure consistency and comparability of data in different environments. In addition, the application improves the frequency spectrum analysis method, using the Hanning window function and fast Fourier transform to improve the resolution and accuracy of the frequency spectrum, so that the dynamic characteristics of power grid operation, such as periodic fluctuations, transient disturbances and nonlinear characteristics, can be analyzed in more detail. In the construction and optimization of response models, the application uses deep learning technology to construct FbProphet and LSTM models based on nonlinear systems, and trains and optimizes them through deep neural networks, which can effectively predict and evaluate the response of the power grid under different conditions, including regular operation and abnormal state. Finally, the application combines the prediction outputs of FbProphet and LSTM models and BPNN model to realize real-time dynamic characteristic evaluation of the power grid, quickly identify and respond to abnormal states and potential risks of the power grid, and predict the state of the power grid, thereby greatly improving the operation efficiency and reliability of the power grid. Overall, the application has significant advantages in improving data collection accuracy, optimizing signal processing, enhancing frequency spectrum analysis, constructing response models, and implementing dynamic evaluation and power grid state prediction, providing important technical support for the stable and efficient operation of the smart grid, especially in processing and analyzing the state of the power grid and predicting the future period of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a schematic diagram of the overall principle of a power grid state evaluation and prediction method in an embodiment of the application.

[0024] Figure 2 is a basic structure diagram of the FbProphet model in an embodiment of the application.

[0025] Figure 3 is a basic structure diagram of the LSTM model in an embodiment of the application.

[0026] Figure 4 is a basic structure diagram of the BPNN model in an embodiment of the application. DETAILED DESCRIPTION

[0027] The specific embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0028] Embodiment: The present application provides a power grid state evaluation and prediction method, as shown in the specific embodiment Figure 1 , comprising the following steps:

[0029] Step 1: Real-time acquisition of power distribution terminal equipment operation data;

[0030] Step 2: Construct a deep learning-based nonlinear system response model to predict power grid behavior;

[0031] Step 3: Construct a BPNN model, input the prediction results of the nonlinear system response model in step 2 and the data in step 1 to predict power grid behavior;

[0032] The power grid behavior described in steps 2 and 3 includes power grid load and / or power grid performance.

[0033] In step 1, the collected data is preprocessed: the collected data is subjected to spectral analysis, and the characteristics of the signal are analyzed by wavelet transform.

[0034] High-precision sensors are used to collect raw power grid signals from power distribution terminal equipment in the form of a time function s(t). The most advanced sensing technologies, such as optical fiber sensors and MEMS sensors, are used to improve the sensitivity and accuracy of signal acquisition. These sensors can work stably in extreme environments, ensuring accurate capture of power grid signals even under high voltage and strong electromagnetic interference conditions.

[0035] A band-pass filter F bp (s(t)) is used to process the signal s(t), obtaining the filtered signal s filtered (t) = F bp (s(t)), effectively eliminating high-frequency and low-frequency noise and retaining key information. This step is crucial for improving the clarity and usability of the signal, ensuring the accuracy and reliability of subsequent analysis.

[0036] The average value μ and the standard deviation σ of the filtered signal are calculated and normalized to obtain In this process, A is used as a signal amplification coefficient to adjust the signal amplitude, ensuring consistency and comparability of data under different equipment and conditions.

[0037] Then, advanced window function technology and zero padding technology are combined to improve the resolution and accuracy of spectral analysis. The Hanning window function w(t) is used to process the signal to reduce spectral leakage at both ends, obtaining the windowed signal s windowed (t) = s pre (t)·w(t), which optimizes the spectral analysis of the signal and improves the resolution and accuracy of the spectrum, where w(t) is the Hanning window function. The Hanning window is applied to each normalized signal segment to reduce spectral leakage, spre (t) is the preprocessed signal.

[0038] Performing a Fast Fourier Transform (FFT) converts the signal from the time domain to the frequency domain. Choosing a suitable FFT algorithm variant for real-time processing, such as a piecewise or parallel FFT algorithm, yields the spectrum. Used to extract and analyze the frequency characteristics of signals, FFT analysis improves spectral resolution and enables more accurate signal spectrum analysis by optimizing the number of FFT points. Spectral analysis provides a detailed understanding of the dynamic characteristics of power grid operation, including periodic fluctuations, transient disturbances, and nonlinear features.

[0039] The characteristics of a signal are analyzed using wavelet transform. For the normalized signal s pre (t) Performs multi-scale decomposition to capture the instantaneous characteristics and nonlinear changes of the signal. The application of wavelet transform can reveal transient events and unconventional fluctuations in power grid signals, providing key information for in-depth analysis of power grid stability and security.

[0040] Step 2 involves constructing a hybrid response model for a nonlinear system based on deep learning and optimizing the parameters of the response model. This includes building an FbProphet model and an LSTM model. The FbProphet model is used to capture trends and seasonal patterns in time-series data to analyze the trends in power grid data and predict future power grid load or performance. The LSTM model is used to capture short-term dependencies in time-series data to predict short-term behavior of the power grid, such as instantaneous load changes.

[0041] The basic structure of the FbProphet model is as follows: Figure 2 As shown, the mathematical expression of the model is: y(t) = g(t) + s(t) + h(t) + e(t), where g(t) represents the trend term, indicating the trend of the time series in non-periodic terms; s(t) represents the periodic term, or seasonal term, which can be in units of hours, days, weeks, or years; h(t) represents the holiday term, indicating whether there is a holiday on that day; and e(t) represents the error term, or residual term. The FbProphet model obtains the predicted value of the time series by fitting these terms and then summing them up.

[0042] The basic structure of the LSTM model is as follows: Figure 3 As shown, its internal unit consists of three key structures: the input gate, the forget gate, and the output gate. The mathematical expressions for these gate pairs are as follows: i t =σ(U i x t +V i h t-1 +b i );ft =σ(U f x t +V f h t-1 +b f );o t =σ(U0x t +V o h t-1 +b o );c t =f t ⊙c t-1 +i t ⊙tan h(U ia x t +V ia h t-1 +b ia );h t =o t ⊙tanh(c t ); where U and V are weight matrices, b is the bias value of the corresponding gate, the symbol ⊙ represents the Hadamard product (the product of corresponding elements of the matrix), σ represents the activation of the corresponding gate, which is the sigmoid activation function, which is responsible for not only opening and closing the corresponding gate, but also supporting the nonlinear function of the LSTM model, and the hyperbolic tangent function (tanh) is responsible for adjusting the output of the corresponding input activation and output gate between -1 and 1.

[0043] First, the forget gate in the internal structure of LSTM is responsible for retaining or forgetting the contents of the previous cell state based on the sigmoid activation function. The amount of content retained or forgotten can be controlled by the weights, i.e.: i t =σ(U i x t +V i h t-1 +b i In the second step of the LSTM operation, the internal cell state of the LSTM is updated by adding the product of the forget gate and the output of the previous cell state to the product of the input gate and the output of the input activation, i.e.: c t =f t ⊙c t-1 +i t ⊙tan h(U ia x t +V ia h t-1 +b ia In the final step of the LSTM operation, the output gate is responsible for passing the output to the next unit in the LSTM structure, which is achieved through the sigmoid activation function, while o t Multiplying the result by the unit with the hyperbolic tangent tanh activation function, which has a regulating and updating effect, yields a final value equal to h.t t o t (t) is the input at this time t-1 and the bias value b and the sigmoid activation function, that is: o t = σ(U0x t +V o h t-1 +b o ), h t = o t ⊙ tanh(c t ). The cell state c t and the output h t will be sent to the next level, the next cycle of LSTM operation will repeat the same process as above, LSTM through the influence of forget gate, input gate, output gate for state C, ultimately decide how much to forget, how much to remember, how much to output at each time point, and finally pass this state down, so as to achieve the effect that it can control not to forget important information far away, and not to take too seriously the nearby unimportant information. The training input data of the two models will be normalized data s pre (t) input to the FbProphet model, which can accurately predict the long-term trend and periodic changes of the power grid data. The normalized data s pre (t), the FFT transformed data S(f) and the wavelet transformed data W s (t) are used as the input of the LSTM model, which can effectively capture the instantaneous changes and complex time-dependent relationships of the power grid data. Here we can choose the data of power grid equipment in a certain place for nearly 7 years to build the model, which needs to be cleaned, standardized and time-labeled. The first 5 years of data are used for model training, and the last two years of data are used for model testing.

[0044] In step 3, the results of FFT and wavelet transform and the output of the response model above are combined with machine learning to evaluate and predict the power grid dynamics. The output data of the FbProphet model and the LSTM model in step 2 are used as the input of the BPNN model, and the FFT transformed data S(f) and the wavelet transformed data W s (t) are also used as the input of the BPNN model. The structure of the BPNN model is shown in Figure 4 Here, the input data needs to be properly processed to make their formats consistent, which can be in the form of vectors. The overall system structure diagram is shown in Figure 1 ​The final prediction result is optimized according to the model. The output of the BPNN is the fusion of the Prophet and LSTM prediction results, providing a prediction considering long-term trends and short-term patterns, and since the FFT transformed data S(f) and wavelet transformed data W s (t) are input, the BPNN can predict short-term and long-term changes in power grid load based on historical and real-time data, and can evaluate the load and abnormal state of the power grid. This step utilizes the high intelligence and automation of machine learning to optimize the entire hybrid model and evaluate and warn the state of the power grid, effectively improving the operation efficiency and reliability of the power grid.

[0045] The principle or implementation steps of the method are as follows:

[0046] Step 1: Signal acquisition and preprocessing

[0047] Step 1.1: Device arrangement, select current transformers and voltage sensors with a precision of up to 0.1%, and a frequency response range of at least 1 kHz, install sensors at key nodes of the distribution network, such as substation outlets and important load points, set the acquisition frequency to at least 100 times per second to capture rapid changes in the power grid, and collect the original signal s(t) of the power grid.

[0048] Step 1.2: Signal preprocessing, use a band-pass filter F bp with a bandwidth of 50Hz to 500Hz to remove high-frequency and low-frequency noise in the signal, and obtain the filtered signal s filtered (t) = F bp (s(t)). Normalization processing, calculate the average value μ and standard deviation σ of the signal, normalize the signal, and obtain the preprocessed signal s Segment the signal for further processing, divide the continuous signal into segments, each with a length of T, to facilitate real-time analysis.

[0049] Step 1.3: Improved Fast Fourier Transform (FFT), apply a window function, use a Hanning window function w(t) to reduce signal leakage at both ends, improve the frequency resolution of FFT, multiply each normalized signal with the Hanning window function to reduce the leakage effect at both ends, and obtain s windowed (t) = s pre (t)·w(t). FFT transform, select a suitable FFT algorithm variant for real-time processing, such as the segmented algorithm or parallel algorithm of Fast Fourier Transform, perform FFT transform on the windowed signal, here you can use special hardware such as FPGA or GPU to accelerate FFT calculation, and obtain the frequency spectrum S(f) By optimizing the FFT point number, the spectral resolution is improved, and the signal spectrum is analyzed more accurately. Combined with the spectrum analysis results and power grid knowledge, the identification ability of power grid events is improved.

[0050] Step 1.4: Signal characteristic decomposition through wavelet transform Perform multi-scale analysis to capture the transient characteristics of the signal.

[0051] Implement Step 2: Train FbProphet model and LSTM model:

[0052] Step 2.1: Build response model, select FbProphet model and LSTM model, suitable for processing time series data, use FbProphet model to capture the trend and seasonal patterns of time series data, analyze the trend of power grid data, and predict future power grid load or performance. Through the LSTM model, capture the short-term dependence relationship in the time series data, and predict the short-term behavior of the power grid, such as instantaneous load changes.

[0053] Step 2.2: Collect historical power grid data, check for missing values in the data, fill or delete records with missing values according to the situation, identify and process outliers, and use statistical methods (such as standard deviation) to identify and replace or delete abnormal data points. Clean the original signal data to ensure data quality. Normalize the cleaned data for preprocessing. Resample the data to the selected time interval, aggregate the data using appropriate methods (such as average, median), and organize the time labels of the data.

[0054] Step 2.3: After preprocessing the data respectively, input them into the FbProphet model and LSTM model, and train the models respectively. Evaluate model performance on independent test sets to ensure generalization ability.

[0055] Implement Step 3: Power grid state analysis and prediction:

[0056] Step 3.1: Build BPNN model, use the prediction results of the FbProphet model and LSTM model in Step 2 as the input of the BPNN, as well as the FFT transformed data S(f) and wavelet transformed data W s (t) as inputs of the BPNN model, and appropriately process each input data to make them consistent in format. Use the prediction results of the FbProphet model and LSTM model in Step 2 as inputs of the BPNN, as well as FFT transformed data S(f) and wavelet transformed data W s (t) to train the model.

[0057] Step 3.2: The output of the BPNN is the fusion of the Prophet and LSTM prediction results, providing a prediction that takes into account both long-term trends and short-term patterns after inputting the FFT-transformed data S(f) and wavelet-transformed data W s After step (t), the BPNN can predict short-term and long-term changes in power grid load based on historical and real-time data, and can assess the load and abnormal state of the power grid accordingly.

[0058] In another specific embodiment, the application environment is a medium-sized urban power distribution network, which includes multiple distribution stations and thousands of end users. The challenges faced are the high-density load changes and the access of renewable energy, and the power distribution network needs to efficiently and accurately monitor and respond to changes in the state of the power grid, and predict the state of the power grid in the future period.

[0059] The specific implementation steps are as follows:

[0060] Step 1: Signal acquisition and preprocessing:

[0061] First, install the equipment, select current transformers and voltage sensors with a precision of up to 0.1% and a frequency response of at least 1 kHz, and install them at key nodes of the power grid, such as substation outlets and important load points. Then, collect real-time operational data from the terminal equipment of the power distribution network, collect voltage and current signals s(t) at each node, and collect hundreds of times per second to capture rapid changes in the power grid. Perform preprocessing operations on the collected data, apply a bandpass filter F bp with a bandwidth of 50Hz to 500Hz to remove high-frequency and low-frequency noise in the signal. Calculate the mean μ and standard deviation σ of the signal, normalize the signal, and obtain the preprocessed signal Apply the Hanning window to each normalized signal segment to reduce spectral leakage and improve the frequency resolution of the FFT, obtaining s windowed (t) = s pre (t)·w(t). Perform FFT on the windowed signal to obtain the spectral characteristics of each signal segment, obtaining the spectrum Signal characteristic decomposition through wavelet transform Perform multi-scale analysis to capture the instantaneous characteristics of the signal.

[0062] Step 2: Train FbProphet model and LSTM model:

[0063] A response model is constructed, FbProphet model and LSTM model are selected, which are suitable for processing time series data, FbProphet model is used to capture the trend and seasonal patterns of time series data, the trend of power grid data is analyzed, and the future power grid load or performance is predicted. LSTM model is used to capture the short-term dependence in time series data, and the behavior of the power grid in the short term is predicted, such as instantaneous load changes. Historical power grid data is collected, missing values in the data are checked, and missing values are filled or deleted according to the situation, abnormal values are identified and processed, statistical methods (such as standard deviation) can be used to identify and replace or delete abnormal data points, the original signal data is cleaned to ensure the quality of the data. The cleaned data is normalized and preprocessed. The data is resampled to the selected time interval, and appropriate methods (such as average, median) are used to aggregate the data. The time label of the data is sorted. After the corresponding preprocessing of the data, it is input into the FbProphet model and the LSTM model, and the models are trained respectively. The model performance is evaluated on an independent test set to ensure the generalization ability.

[0064] Step three: power grid state analysis and prediction

[0065] A BPNN model is constructed, the prediction results of the FbProphet model and the LSTM model in step two are used as the input of the BPNN, and the FFT transformed data S(f) and the wavelet transformed data W s (t) are also used as the input of the BPNN model, and the input data is appropriately processed to make the format consistent. The output of the BPNN is the fusion of the Prophet and LSTM prediction results, which provides a comprehensive consideration of long-term trends and short-term patterns after prediction, and after inputting the FFT transformed data S(f) and the wavelet transformed data W s (t), the BPNN can predict short-term and long-term power grid load changes based on historical and real-time data, and can evaluate the load and abnormal state of the power grid accordingly.

[0066] Implementation effect:

[0067] The monitoring capability is improved, the power grid state is monitored in real time, and the abnormal fluctuations of voltage and current are accurately captured. The early warning system is optimized, and potential power grid failures can be timely warned through analysis of the dynamic characteristics of the power grid. Decision support, through the prediction of the future state of the power grid for a period of time, data support is provided for power grid operation and maintenance, guiding load adjustment and fault elimination. The user experience is improved, ensuring the stability of the power grid and improving the user power supply reliability.

[0068] Through this embodiment, the application demonstrates its effectiveness and practicability in a real urban power distribution network environment, not only improves the operation efficiency of the power grid, but also enhances the response capability to various power grid events.

[0069] Another specific embodiment of the present application provides a power grid state evaluation and prediction device, comprising:

[0070] An acquisition unit, configured to acquire power distribution network terminal device operation data in real time;

[0071] A first prediction unit, configured to construct a deep learning-based nonlinear system response model to predict power grid behavior;

[0072] A second prediction unit, configured to construct a BPNN model, take the prediction result in the first prediction unit and the data obtained by the acquisition unit as input to predict power grid behavior.

[0073] Another specific embodiment of the present application provides an electronic terminal, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is called and executed by the processor to implement the power grid state evaluation and prediction method as described above.

[0074] Another specific embodiment of the present application provides a computer readable medium, which stores a computer program, and the computer program is called and executed by a computer to implement the power grid state evaluation and prediction method as described above.

[0075] The above is only the preferred embodiment of the present application, and it should be pointed out that, for those skilled in the art, without departing from the creative concept of the present application, several modifications and improvements can be made, which all belong to the protection scope of the present application.

Claims

1. A power grid state assessment and prediction method, characterized by, Comprising the following steps: Step 1: Real-time acquisition of power distribution network terminal equipment operation data; preprocessing of the collected data: performing spectral analysis on the collected data, and analyzing the characteristics of the signal through wavelet transform; Step 2: Construct a nonlinear system response model based on deep learning to predict the behavior of the power grid; Step 2 includes: constructing FbPorpet model and LSTM model, using FbProphet model to capture the trend and seasonal patterns of time series data, analyzing the trend of power grid data, and predicting future power grid load or performance; through the LSTM model to capture the short-term dependence relationship in the time series data, to predict the behavior of the power grid in the short term; Step 3: build the BPNN model, take the prediction results of the nonlinear system response model in step 2 and the data in step 1 as input to predict the grid behavior; take the output data of the FbProphet model and the LSTM model in step 2 and the data processed by wavelet transform as the input of the BPNN model, and fuse the prediction results of the FbProphet model and the LSTM model; The behavior of the power grid described in steps 2 and 3 includes power grid load and / or power grid performance.

2. The power grid state assessment and prediction method of claim 1, wherein, Step 1 includes: collecting original power grid signals from power distribution network terminal equipment using high-precision sensors, expressed in the form of time function s(t), using a band-pass filter The signal s(t) is processed to obtain the filtered signal , the average value of the filtered signal is calculated , and the standard deviation is calculated , obtaining , the signal is processed using the Hanning window function w(t) to obtain the windowed signal , performing fast Fourier transform to convert the signal from time domain to frequency domain to obtain the frequency spectrum , through wavelet transform , the normalized signal is decomposed by multiple scales to capture the instantaneous characteristics and nonlinear changes of the signal.

3. The power grid state assessment and prediction method of claim 1, wherein The mathematical expression of the FbProphet model is: y(t) = g(t) + s(t) + h(t) + e(t), where g(t) represents the trend item, which represents the changing trend of the time series over the non-period; s(t) represents the periodic term; h(t) represents the holiday term, which represents whether there is a holiday on the day; e(t) represents the error term or the residual term.

4. The power grid state assessment and prediction method of claim 1, wherein, The LSTM model includes an input gate, a forget gate, and an output gate, and the mathematical expressions are as follows: ; ; ; ; ; wherein U, V are weight matrices, b is the bias value corresponding to the gate, and the symbol represents the Hadamrd product, the symbol represents the activation of the corresponding gate, which is a sigmoid activation function.

5. A power grid state evaluation and prediction apparatus characterized by comprising: Comprising: An acquisition unit for real-time acquisition of power distribution network terminal equipment operation data; Preprocessing of the collected data: performing spectral analysis on the collected data, and analyzing the characteristics of the signal through wavelet transform; A first prediction unit for constructing a nonlinear system response model based on deep learning to predict the behavior of the power grid; constructing FbPorpet model and LSTM model, using FbProphet model to capture the trend and seasonal patterns of time series data, analyzing the trend of power grid data, and predicting future power grid load or performance; through the LSTM model to capture the short-term dependence relationship in the time series data, to predict the behavior of the power grid in the short term; a second prediction unit configured to construct a BPNN model, take the prediction result in the first prediction unit and the data obtained by the acquisition unit as input to predict the power grid behavior; take the output data of the FbProphet model and the LSTM model as input of the BPNN model, take the data processed by the FFT transform and the wavelet transform as input of the BPNN model, and fuse the prediction results of the FbProphet model and the LSTM model to output. and the data processed by the wavelet transform as input of the BPNN model.

6. An electronic terminal, characterized in that The computer readable medium stores a computer program, and the computer program is called and executed by the computer to realize the power grid state evaluation and prediction method of any one of claims 1-4.

7. A computer readable medium characterized by The computer readable medium stores a computer program, and the computer program is called and executed by the computer to realize the power grid state evaluation and prediction method of any one of claims 1-4.

Images