Method and system for predicting power load of power internet of things
The power load forecasting method of the Internet of Things for power systems, which integrates multiple data sources and adaptively selects models, solves the problem of low accuracy in power load forecasting in existing technologies and achieves accurate forecasting in complex power grid environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-26
- Publication Date
- 2026-04-10
AI Technical Summary
Existing power load forecasting methods suffer from low prediction accuracy in complex power grid environments due to their limited data dimensions and poor model adaptability, especially in real power grid environments where accuracy and realism are insufficient.
By receiving power grid operation data and environmental data, performing preprocessing and calculations, and using multiple data fusion methods to train the power grid load forecasting model, including RTU and PMU data acquisition, meteorological sensor data acquisition, adaptive selection and weighted fusion of multiple models, a power grid load forecasting system is constructed.
It enables accurate prediction of grid load in complex grid environments, improving prediction precision and accuracy.
Smart Images
Figure CN121840581A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of smart grid, and particularly relates to a power load prediction method and system for power internet of things. BACKGROUND
[0002] Power resource allocation and scheduling is an important work of the power dispatch department and is an important part of the power system. Accurate and effective power load prediction is the basis for scientific scheduling of power backup energy and is an important guarantee for safe, stable and economic operation of the power system.
[0003] The existing power load prediction method has low prediction accuracy due to problems such as single data dimension and poor model self-adaptive ability, and especially in the face of complex real power grid environment, the prediction accuracy and authenticity are insufficient. SUMMARY
[0004] Technical purpose: In view of the defects in the prior art, the present application discloses a power load prediction method and system for power internet of things, which realizes the improvement of prediction accuracy through the calculation of multiple data fusion and the construction of model self-adaptive selection.
[0005] Technical scheme: In order to achieve the above technical purpose, the present application adopts the following technical scheme.
[0006] A power load prediction method for power internet of things, the method comprising: receiving power grid operation data, preprocessing the power grid operation data to obtain processed power grid operation data, performing power grid operation performance processing calculation based on the processed power grid operation data to obtain a power grid operation performance coefficient; wherein the power grid operation data comprises operating voltage, operating current and vibration frequency; receiving environmental data, performing environmental impact calculation based on the environmental data to obtain an environmental impact coefficient, inputting the environmental impact coefficient and the power grid operation performance coefficient into a pre-trained power grid load prediction model, and outputting to obtain a power grid load prediction result, wherein the environmental data comprises temperature data, humidity data and precipitation data, the pre-trained power grid load prediction model is trained based on historical load data and historical environmental data, wherein the historical load data comprises active power and reactive power.
[0007] A power load prediction system for power internet of things, for realizing the above-mentioned power load prediction method for power internet of things, comprising: a data processing module for receiving power grid operation data, preprocessing the power grid operation data to obtain processed power grid operation data, performing power grid operation performance processing calculation based on the processed power grid operation data to obtain a power grid operation performance coefficient; wherein the power grid operation data comprises operating voltage, operating current and vibration frequency; The power grid load prediction module is used for receiving environmental data, inputting the environmental data and the power grid operation performance coefficient into a pre-trained power grid load prediction model, and outputting a power grid load prediction result, wherein the environmental data includes temperature data, humidity data and precipitation data, and the pre-trained power grid load prediction model is trained based on historical load data, wherein the historical load data includes active power and reactive power.
[0008] Beneficial effects: After collecting power grid operation data and environmental data, the power grid operation performance coefficient and the environmental influence coefficient are obtained by respectively performing power grid operation performance processing calculation and environmental influence calculation, the power grid load prediction model is obtained based on historical load data, and the power grid load prediction is performed on the environmental influence coefficient and the power grid operation performance coefficient based on the power grid load prediction model, so that the load of the power grid is accurately predicted by combining various types of data and determining the power grid performance and the influence of the environment, and the prediction accuracy of the power grid prediction is improved. BRIEF DESCRIPTION OF DRAWINGS
[0009] Figure 1 is a method flowchart of the present application; Figure 2 is a system structure schematic diagram of the present application. DETAILED DESCRIPTION
[0010] In order for those skilled in the art to better understand the present application, the technical solutions in the present application will be described clearly and completely in conjunction with 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. EMBODIMENT
[0011] As shown in the accompanying drawings, Figure 1 The power load prediction method of the power internet of things in the embodiment includes the following steps: S101, receiving power grid operation data, preprocessing the power grid operation data to obtain processed power grid operation data, performing power grid operation performance processing calculation based on the processed power grid operation data to obtain a power grid operation performance coefficient; wherein the power grid operation data includes operating voltage, operating current and vibration frequency; Specifically, in order to obtain accurate power grid operation data and historical load data, the data collection means adopted by the present application is as follows: In order to collect power grid operation data, the application can select to collect by RTU or PMU, wherein the RTU is installed behind the transformer substation of the power grid, receives the input signal of the transformer substation, then samples and calculates to obtain the effective value of voltage, current and vibration frequency, wherein the sampling frequency of the RTU is 2-10 seconds interval, and the reliability and real-time performance of the collected data are relatively strong; the PMU is installed on the key node of power transmission of the transformer substation, then high-speed sampling is performed, and the phasor value of voltage, current and vibration frequency is calculated by algorithm, and the corresponding voltage, current and frequency phasor value is time-stamped, and the sampling frequency is very high, which can reach 25 frames per second, and the sampled data is suitable for dynamic analysis, and compared with the two collection methods, the sampling speed of the PMU is faster than that of the RTU, but the sampling accuracy and real-time performance are slightly weak, therefore, finally the application adopts the RTU as the collection tool for subsequent use; The process of preprocessing the power grid operation data and historical load data includes: The power grid operation data and historical load data are subjected to data cleaning and data normalization, wherein the data cleaning includes processing of missing values, abnormal values and blank data, the missing values are processed by linear interpolation method, and the abnormal values are processed by Z-score method: The Z-score value of each data is calculated, the data with a Z-score value greater than a preset threshold value is regarded as an abnormal value and is deleted, and the IQR method is combined for auxiliary detection, the data within the set IQR range is regarded as an abnormal value and is removed, so that the abnormal value of the data is processed and removed; The blank data is supplemented by a multiple imputation method, for continuous variables in the blank data, the missing values are predicted based on a pre-established linear regression model, and for classification variables in the blank data, the KNN method is used to supplement the missing values according to similar samples; The data normalization process is the normalization processing of the data by the minimum and maximum value method.
[0012] Based on the processed power grid operation data, the power grid operation performance processing calculation is performed to obtain the power grid operation performance coefficient, including: the processed power grid operation data is marked, the marked power grid operation data is combined with a preset related standard coefficient and a related influence coefficient to perform power grid operation performance processing calculation, and the power grid operation performance coefficient is obtained; The preset related standard coefficient includes a preset standard voltage coefficient, a preset standard current coefficient and a preset standard frequency coefficient. The marking process of the power grid operation data includes: Mark the running voltage as Vi, mark the running current as Li, and mark the vibration frequency as Pi; in the formula, i is the collection number of the collected power grid operation data, and i=1, 2, 3,..., n, n is the total number of the collected power grid operation data; Based on the marked power grid operation data, the power grid operation performance processing calculation is performed, and the calculation formula is as follows: , In the formula, Xi is the power grid operation performance coefficient, V0 is the preset standard voltage coefficient, L0 is the preset standard current coefficient, P0 is the preset standard frequency coefficient, q1 is the voltage influence coefficient, q2 is the current influence coefficient, and q3 is the frequency influence coefficient.
[0013] Further, in the specific implementation process, the preset standard voltage coefficient, the preset standard current coefficient, and the preset standard frequency coefficient are obtained by daily collection of the running voltage, the running current, and the vibration frequency, and by multiple simulation calculations and data mean taking. In this embodiment, the voltage influence coefficient, the current influence coefficient, and the frequency influence coefficient are obtained by the application through daily acquisition of the running voltage, the running current, and the vibration frequency, and are calculated based on comprehensive evaluation of external factors, including human factors, machine detection, and environmental factors.
[0014] S102, receive environment data, perform environment influence calculation based on the environment data to obtain an environment influence coefficient, input the environment influence coefficient and the power grid operation performance coefficient into a pre-trained power grid load prediction model, and output to obtain a power grid load prediction result, wherein the environment data includes temperature data, humidity data, and precipitation data, the pre-trained power grid load prediction model is trained based on historical load data and historical environment data, and the historical load data includes active power and reactive power.
[0015] Specifically, the environment data of the application is collected by a meteorological sensor installed to obtain temperature data, humidity data, and precipitation data in the vicinity of the substation in real time. The process of performing environment influence calculation based on the environment data to obtain the environment influence coefficient includes: Mark the temperature data as T, mark the humidity data as S, and mark the precipitation data as J; Use the formula to calculate the environment influence coefficient H. In the formula, alpha, beta and gamma are preset weight coefficients, which are obtained based on the weight proportion of the influence of temperature data, humidity data and precipitation data on the environment respectively; exp() is an exponential function, and ln() is a logarithmic function. The nonlinear influence of temperature, humidity and precipitation on power load is simulated by using the mathematical properties of the exponential function and the logarithmic function.
[0016] The active power and the reactive power are obtained by collecting the historical load data through the record of the intelligent electric meter. In order to improve the accuracy of power grid load prediction, the power grid load prediction model pre-trained by the application can be constructed based on a time series benchmark model and a deep learning model, wherein the time series benchmark model is XGboost or LightGBM, and the deep learning model can use a multi-head attention mechanism plus transformer or an LSTM model. When the time series benchmark model is applied, it has excellent structured data processing capability. The deep learning model can capture long-term dependencies and sequence dynamic modeling to improve the ability to capture dynamic data. Finally, the application selects the XGboost model of the time series benchmark model plus the LSTM model of the deep learning model as the basic model for subsequent training. The XGboost model of the time series benchmark model and the LSTM model of the deep learning model are connected in parallel, and the output of the power grid load prediction model is weighted and fused. Parallel training, parallel prediction and weighted fusion are realized by parallel connection. Compared with the series connection method, the parallel connection method can reduce the error accumulation problem (for example, if the first model has an error, the second model will also have an error, which will increase the prediction error). Collect and obtain historical load data and historical environment data, and use the historical load data and the historical environment data as a data set. The true value obtained during model training is the historical load data, including active power and reactive power.
[0017] The historical load data and the historical environment data correspond to each other, and the mapping relationship is generated by corresponding training of the model to realize the training process of the model.
[0018] The training set is input into the XGboost model and the LSTM model, respectively, and the Bayesian optimization method is used to search for the optimal hyperparameter combination. After multiple iterations, the trained XGboost model and the LSTM model are output. Determine the preset weight based on the test set, test the trained XGboost model and LSTM model based on the preset weight, input the preset standard threshold value, when the value predicted by the model at this time exceeds the preset standard threshold value range (15%), it means that the allocation weight at this time is unreasonable, then the weight coefficients of the XGboost model and the LSTM model need to be updated, and finally the obtained model is used as the power grid load prediction model.
[0019] The preset weight is the initial weight set, and the initial weight is obtained by allocating based on the performance indicators (mean square error, mean absolute error, etc.) of the two models; The process of weighted fusion is as follows: through the prediction of the test set by the XGboost model and the LSTM model in the time series reference model, the determined preset weight is obtained, then the new data is input into the XGboost model and the LSTM model in the time series reference model respectively to obtain two prediction values, the two prediction values are multiplied by the corresponding initial weight respectively and summed to obtain the fused prediction value, and the fused prediction value is compared with the preset standard threshold value, when it exceeds the preset standard threshold value range (15%), the weight is adaptively updated; The preset standard threshold value is obtained by calculating the historical power grid performance coefficient based on historical power grid data multiple times, and then calculating the mean value of the historical power grid performance coefficient multiple times; the historical power grid data includes historical voltage data, historical current data and historical vibration frequency data.
[0020] As shown in the accompanying Figure 2 The embodiment also discloses a power Internet of Things power load prediction system, which comprises: A data processing module 11 is configured to receive power grid operation data, pre-process the power grid operation data to obtain processed power grid operation data, and perform power grid operation performance processing calculation based on the processed power grid operation data to obtain a power grid operation performance coefficient; wherein the power grid operation data includes operating voltage, operating current and vibration frequency. A power grid load prediction module 12 is configured to receive environmental data, input the environmental data and the power grid operation performance coefficient into a pre-trained power grid load prediction model, and output a power grid load prediction result, wherein the environmental data includes temperature data, humidity data and precipitation data, the pre-trained power grid load prediction model is trained based on historical load data and historical environmental data, and the historical load data includes active power and reactive power.
[0021] The "first", "second" and the like in the names mentioned in the embodiments of the application are only used for name identification, and do not represent the first and second in order.
[0022] From the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the above-mentioned example methods can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solutions of the present application can be embodied in the form of a software product. The computer software product can be stored in a storage medium, and the storage medium can be various types of memories, such as random access memory (RAM), read only memory (ROM), flash memory, etc., such as read only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to cause a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of the present application.
[0023] The above only describes the preferred embodiments of the present application, and it should be noted that those of ordinary skill in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements should also be considered within the scope of protection of the present application.
Claims
1. A power load forecasting method for the Internet of Things (IoT), characterized in that, The methods include: The system receives power grid operation data, preprocesses the data to obtain processed power grid operation data, and performs power grid operation performance calculations based on the processed data to obtain power grid operation performance coefficients. The power grid operation data includes operating voltage, operating current, and vibration frequency. The system receives environmental data, calculates the environmental impact coefficient based on the data, inputs the environmental impact coefficient and the power grid operation performance coefficient into a pre-trained power grid load prediction model, and outputs the power grid load prediction result. The environmental data includes temperature data, humidity data and precipitation data. The pre-trained power grid load prediction model is trained based on historical load data and historical environmental data. The historical load data includes active power and reactive power.
2. The power load forecasting method for the power Internet of Things according to claim 1, characterized in that, The process of performing power grid operation performance processing and calculation based on the processed power grid operation data includes: The power grid operation data is labeled, and the labeled power grid operation data is combined with preset relevant standard coefficients and relevant influence coefficients to calculate the power grid operation performance coefficient. Among them, the preset relevant standard coefficients include preset standard voltage coefficient, preset standard current coefficient and preset standard frequency coefficient; The preset relevant influence coefficients include voltage influence coefficient, current influence coefficient, and frequency influence coefficient.
3. The power load forecasting method for the power Internet of Things according to claim 2, characterized in that: The process of marking power grid operation data includes: The operating voltage is labeled as Vi, the operating current as Li, and the vibration frequency as Pi; where i is the number of times the power grid operation data is collected, and i = 1, 2, 3, ..., n, where n is the total number of times the power grid operation data is collected; The calculation formulas for power grid operation performance processing are as follows: ， In the formula, Xi is the power grid operation performance coefficient, V0 is the preset standard voltage coefficient, L0 is the preset standard current coefficient, P0 is the preset standard frequency coefficient, q1 is the voltage influence coefficient, q2 is the current influence coefficient, and q3 is the frequency influence coefficient.
4. The power load forecasting method for the power Internet of Things according to claim 1, characterized in that, The process of calculating environmental impact coefficients based on environmental data includes: Label temperature data as T, humidity data as S, and precipitation data as J; Using formula The environmental impact factor H was calculated. In the formula, α, β, and γ are all preset weighting coefficients, exp() is an exponential function, and ln() is a logarithmic function.
5. The power load forecasting method for the power Internet of Things according to claim 1, characterized in that, Historical load data and historical environmental data are matched one-to-one.
6. The power load forecasting method for the power Internet of Things according to claim 1, characterized in that, The pre-trained power grid load forecasting model is constructed based on a time-series benchmark model and a deep learning model. The time-series benchmark model and the deep learning model are connected in parallel, and the output is weighted and fused to output the forecasting results of the power grid load forecasting model.
7. The power load forecasting method for the power Internet of Things according to claim 6, characterized in that, The time series benchmark model uses the XGboost model, and the deep learning model uses the LSTM model.
8. The power load forecasting method for the power Internet of Things according to claim 7, characterized in that, The training process of a pre-trained power grid load forecasting model includes: Historical load data and historical environment data are collected and used as a dataset. The dataset is then divided into a training set and a test set. The training set is input into the XGboost model and the LSTM model respectively, and the optimal hyperparameter combination is searched using the Bayesian optimization method. After multiple iterations, the trained XGboost model and LSTM model are finally output. The preset weights are determined based on the test set. The trained XGboost model and LSTM model are tested based on the preset weights. The weight coefficients of the XGboost model and LSTM model are updated by inputting a preset standard threshold. The final model is used as the power grid load prediction model. The preset standard threshold is obtained by averaging the historical power grid performance coefficients calculated multiple times based on historical power grid data.
9. The power load forecasting method for the power Internet of Things according to claim 1, characterized in that, The process of preprocessing power grid operation data includes: Perform data cleaning and normalization on power grid operation data and historical load data; Data cleaning includes handling missing values, outliers, and blank data. Outlier handling is based on the Z-score method, and the process is as follows: Calculate the Z-score value for each data point, and delete data points with Z-score values greater than a preset threshold as outliers. Combine this with the IQR method for auxiliary detection, and remove data points within the set IQR range as outliers.
10. A power Internet of Things (IoT) power load forecasting system, used to implement the power IoT power load forecasting method as described in any one of claims 1 to 9, characterized in that, include: The data processing module is used to receive power grid operation data, preprocess the power grid operation data to obtain processed power grid operation data, and perform power grid operation performance processing calculations based on the processed power grid operation data to obtain power grid operation performance coefficients; wherein, the power grid operation data includes: operating voltage, operating current and vibration frequency; The power grid load forecasting module receives environmental data, inputs the environmental data and power grid operation performance coefficients into a pre-trained power grid load forecasting model, and outputs the power grid load forecasting results. The environmental data includes temperature data, humidity data, and precipitation data. The pre-trained power grid load forecasting model is trained based on historical load data, which includes active power and reactive power.