Power load prediction analysis method

Abstract

The invention discloses a power load prediction analysis method, which comprises the following steps: power utilization data acquisition: establishing a power utilization information acquisition system based on a plurality of power utilization acquisition devices, and collecting power utilization data of users through the power utilization acquisition devices; preprocessing the data; comprising data denoising, data filling, data deletion and normalization processing; grid unit division is carried out based on double information of user power utilization characteristics and geographic positions; performing power short-term load prediction on the plurality of power utilization grid units based on the CNN-LSTM combination model, and outputting prediction results; acquiring local power plant generator output condition and system operation mode change data; dynamically matching the electricity demand and supply of the electricity utilization side and the electricity generation side; according to the method, dual information of user power utilization characteristics and geographic space positions is considered, the load prediction accuracy is improved by utilizing a gridding division mode, power generation grid unit dynamic division is performed on the power generation side according to user power utilization demands, and dynamic matching of the power utilization demands and supply is realized.

Description

Technical Field

[0001] The invention belongs to the field of electric power technology, and in particular relates to an electric power load forecasting and analysis method. Background Art

[0002] The power system is mainly used to continuously provide high-quality electricity to various types of users to meet the electricity load requirements of different users, while being safe, reliable and economical. Since the electricity in the power system is supplied in real time, it is consumed as it is generated and cannot be stored. Therefore, improving the accuracy of power load forecasting and achieving efficient power dispatch are important tasks in power planning. Accurate short-term power load forecasting can ensure the safe, economical and reliable operation of the smart grid environment.

[0003] At the same time, with the proposal of the "dual carbon" goals and the continuous growth of the economy, the development of the power grid faces new situations such as significant source-load differentiation, local power supply shortages during peak summer or winter, and increasing customer requirements for quality services. There is an urgent need to improve the accuracy of load forecasts, formulate more reasonable dispatching plans, and achieve efficient matching of electricity consumption and power supply.

[0004] Therefore, in order to solve the above problems, it is necessary to develop a power load forecasting and analysis method. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a method for power load forecasting and analysis, which takes into account the dual information of user power consumption characteristics and geographic spatial location, utilizes a grid division method to improve the accuracy of load forecasting, and dynamically divides the power generation grid units on the power generation side according to user power demand, thereby achieving dynamic matching of power demand and supply, effectively improving the reliability of the power system, and reducing resource waste and environmental pollution.

[0006] The object of the present invention is achieved in this way: a method for predicting and analyzing power load, comprising the following steps:

[0007] S1. Electricity consumption data collection: Building an electricity consumption information collection system based on multiple electricity consumption collection devices, and collecting users' electricity consumption data through electricity consumption collection devices;

[0008] S2, data preprocessing, including data denoising, data filling, data deletion and normalization processing;

[0009] S3, dividing the grid units based on the dual information of user power consumption characteristics and geographical location; using the preprocessed data as input data for initial grouping, the initial grouping uses a clustering algorithm to classify users with similar power consumption patterns into the same group, and the users in the same group classified by the clustering algorithm are secondary grouped into multiple power grid units based on spatial geographical location information;

[0010] S4, perform short-term power load forecasting for multiple power grid units based on the CNN-LSTM combined model and output the forecast results;

[0011] S5. Collect data on the output status of generators in local power plants and changes in the operating mode of the system; build a local power plant power generation reporting system, and the local power plant will report the operating status of the power generation system equipment and the predicted value of the power output power;

[0012] S6. Dynamically match the electricity demand and supply on the electricity consumption side and the power generation side; the dispatching side dynamically divides the power generation grid units on the power supply side based on the divided power consumption grid units, the power load forecast values ​​of the power consumption grid units and the power output power forecast values, and dynamically matches the divided power generation grid units with the power consumption grid units.

[0013] Furthermore, the electricity consumption data in step S1 includes current, voltage, power, electricity consumption period, and electricity consumption information; the electricity consumption collection equipment includes multiple smart meters and sensor equipment.

[0014] Furthermore, the data filling in step S2 includes: for the time period with more than 1 missing days, fill in 0, and record the corresponding missing time point position, and input the missing position information into the model for training; for the time period with 1 missing day, use linear interpolation to fill in the data, and the interpolation function F is shown as follows:

[0015]

[0016] The point (x1, y1) and point (x2, y2) are the left and right points adjacent to the missing time point x.

[0017] Furthermore, the normalization process in step S2 uses Max-min normalization to convert the user's power consumption per unit time into a standard value within the range of [0,1]. The Max-min normalization is shown in the following formula:

[0018]

[0019] Among them, X m is the original data vector, X ′ m The data vector after Max-min normalization.

[0020] Furthermore, the initial grouping in step S3 classifies users with similar electricity consumption patterns into the same group based on the K-means clustering algorithm. Specifically, n user data vectors X i Divide into k clusters C k , where each observation X i All belong to the average value μj The nearest cluster center point, K-means clustering minimizes the intra-cluster variance, which is the inertia function in the following formula:

[0021]

[0022] Furthermore, the short-term power load forecasting for the power grid unit based on the CNN-LSTM combined model in step S4 specifically includes the following steps:

[0023] S41, input the processed data into the model through the input layer;

[0024] S42, extract important information from the power load data using the CNN convolutional layer, reduce the dimension through the pooling layer, and obtain output data;

[0025] S43, output data to the LSTM layer for training;

[0026] S44. The trained output data enters the output layer to obtain the output value.

[0027] Furthermore, in step S5, the power output value is predicted by using a similar day selection algorithm, which divides the real-time weather forecast information with the historical numerical weather forecast information and the historical output power information of the power plant, and then uses a radial basis function neural network prediction model to obtain the predicted value and prediction error of the output power.

[0028] Furthermore, in step S5, the power generation types of the local power plant include new energy power generation and traditional power generation, the new energy power generation specifically includes photovoltaic power generation and waste biomass power generation, and the traditional power generation specifically includes thermal power generation.

[0029] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0030] (1) By analyzing the user's electricity consumption characteristics and performing secondary division of the user's electricity consumption units based on the user's electricity consumption characteristics and geographic location, the electricity grid unit division is made more accurate, effectively improving the accuracy of load forecasting;

[0031] (2) The short-term power load is predicted based on the CNN-LSTM combined model. CNN is responsible for extracting the characteristic factors of the input data, and LSTM is used to receive the output data of CNN for prediction. This can better predict the short-term power load changes and short-term load demand, improve the prediction accuracy, and have higher precision.

[0032] (3) By collecting and forecasting the power generation information of local power plants and dynamically dividing the power generation grid units on the power generation side according to the user's electricity demand, accurate and efficient matching of electricity consumption and power supply can be achieved, effectively improving the reliability of the power system and reducing resource waste and environmental pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0035] like Figure 1 As shown, a method for power load forecasting and analysis comprises the following steps:

[0036] Step S1, electricity consumption data collection: build an electricity consumption information collection system based on multiple electricity consumption collection devices, and collect users' electricity consumption data through electricity consumption collection devices; the electricity consumption data includes current, voltage, power, electricity consumption period, and electricity consumption information; the electricity consumption collection devices include multiple smart meters and sensor devices.

[0037] Step S2: data preprocessing, including data denoising, data filling, data deletion and normalization.

[0038] Specifically, the data filling in step S2 includes: for the time period with more than 1 missing days, fill in 0, and record the corresponding missing time point position, which is input into the model for training as the missing position information; for the time period with 1 missing day, use linear interpolation to fill in the data, and the interpolation function F is shown as follows:

[0039]

[0040] The point (x1, y1) and point (x2, y2) are the left and right points adjacent to the missing time point x.

[0041] Specifically, the normalization process in step S2 uses Max-min normalization to convert the user's power consumption per unit time into a standard value within the range of [0,1]. The Max-min normalization is shown in the following formula:

[0042]

[0043] Among them, X m is the original data vector, X ′ m The data vector after Max-min normalization.

[0044] Step S3, dividing the grid units based on the dual information of the user's power consumption characteristics and geographical location; using the preprocessed data as input data for initial grouping, the initial grouping uses a clustering algorithm to classify users with similar power consumption patterns into the same group, and the users in the same group classified by the clustering algorithm are secondary grouped into multiple power consumption grid units based on spatial geographical location information; specifically, the initial grouping in step S3 classifies users with similar power consumption patterns into the same group based on the K-means clustering algorithm, specifically, n user data vectors X i Divide into k clusters C k , where each observation X i All belong to the average value μ j The nearest cluster center point, K-means clustering minimizes the intra-cluster variance, which is the inertia function in the following formula:

[0045]

[0046] Step S4, performing short-term power load forecasting on multiple power grid units based on the CNN-LSTM combined model and outputting forecast results; specifically, performing short-term power load forecasting on the power grid units based on the CNN-LSTM combined model specifically includes the following steps:

[0047] S41, input the processed data into the model through the input layer;

[0048] S42, extract important information from the power load data using the CNN convolutional layer, reduce the dimension through the pooling layer, and obtain output data;

[0049] S43, output data to the LSTM layer for training;

[0050] S44. The trained output data enters the output layer to obtain the output value.

[0051] Step S5, collecting data on the output status of generators in local power plants and changes in the operating mode of the system; constructing a local power plant power generation reporting system, and the local power plant reports the operating status of the power generation system equipment and the predicted value of the power output power; specifically, the power output power value is predicted by a similar day selection algorithm, the real-time weather forecast information is divided by the numerical weather forecast history information and the historical output power information of the power plant, and then a radial basis neural network prediction model is used to obtain the predicted value and prediction error of the output power; specifically, the power generation types of local power plants include new energy power generation and traditional power generation, the new energy power generation specifically includes photovoltaic power generation and waste biomass power generation, and the traditional power generation specifically includes thermal power generation.

[0052] Step S6: Dynamically match the electricity demand and supply on the electricity consumption side and the power generation side; the scheduling side dynamically divides the power generation grid units on the power supply side based on the divided power grid units, the power load forecast values ​​of the power grid units and the power output power forecast values, and dynamically matches the divided power generation grid units with the power consumption grid units.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting and analyzing power load, characterized in that: The following steps are involved: S1. Electricity consumption data collection: Building an electricity consumption information collection system based on multiple electricity consumption collection devices, and collecting users' electricity consumption data through electricity consumption collection devices; S2, data preprocessing, including data denoising, data filling, data deletion and normalization; S3, dividing the grid units based on the dual information of user power consumption characteristics and geographical location; using the preprocessed data as input data for initial grouping, the initial grouping uses a clustering algorithm to classify users with similar power consumption patterns into the same group, and the users in the same group classified by the clustering algorithm are secondary grouped into multiple power grid units based on spatial geographical location information; S4, perform short-term power load forecasting for multiple power grid units based on the CNN-LSTM combined model and output the forecast results; S5. Collect data on the output status of generators in local power plants and changes in the operating mode of the system; build a local power plant power generation reporting system, and the local power plant will report the operating status of the power generation system equipment and the predicted value of the power output power; S6. Dynamically match the electricity demand and supply on the electricity consumption side and the power generation side; the dispatching side dynamically divides the power generation grid units on the power supply side based on the divided power consumption grid units, the power load forecast values ​​of the power consumption grid units and the power output power forecast values, and dynamically matches the divided power generation grid units with the power consumption grid units.

2. A method for predicting and analyzing power load according to claim 1, characterized in that: The electricity consumption data in step S1 includes current, voltage, power, electricity consumption period, and electricity consumption information; the electricity consumption collection equipment includes multiple smart meters and sensor equipment.

3. The power load forecasting and analysis method according to claim 1, characterized in that: The data filling in step S2 includes: for the time period with more than 1 missing days, fill in 0, and record the corresponding missing time point position, which is input into the model for training as the missing position information; for the time period with 1 missing day, use linear interpolation to fill in the data, and the interpolation function F is shown as follows: The point (x1, y1) and point (x2, y2) are the left and right points adjacent to the missing time point x.

4. The power load forecasting and analysis method according to claim 1, characterized in that: The normalization process in step S2 uses Max-min standardization to convert the user's power consumption per unit time into a standard value within the range of [0,1]. The Max-min standardization is shown in the following formula: Among them, X m is the original data vector, X ′ m The data vector after Max-min normalization.

5. The method for predicting and analyzing power load according to claim 1, characterized in that: In step S3, the initial grouping is based on the K-means clustering algorithm to classify users with similar electricity consumption patterns into the same group. Specifically, n user data vectors X i Divide into k clusters C k , where each observation X i All belong to the average value μ j The nearest cluster center point, K-means clustering minimizes the intra-cluster variance, which is the inertia function in the following formula:

6. The power load forecasting and analysis method according to claim 1, characterized in that: The step S4 specifically includes the following steps: S41, input the processed data into the model through the input layer; S42, extract important information from the power load data using the CNN convolutional layer, reduce the dimension through the pooling layer, and obtain output data; S43, output data to the LSTM layer for training; S44. The trained output data enters the output layer to obtain the output value.

7. The power load forecasting and analysis method according to claim 1, characterized in that: In step S5, the power output value is predicted by using a similar day selection algorithm, which divides the real-time weather forecast information with the historical numerical weather forecast information and the historical output power information of the power plant, and then uses a radial basis function neural network prediction model to obtain the predicted value and prediction error of the output power.

8. The power load forecasting and analysis method according to claim 1, characterized in that: In step S5, the power generation types of the local power plant include new energy power generation and traditional power generation. The new energy power generation specifically includes photovoltaic power generation and waste biomass power generation, and the traditional power generation specifically includes thermal power generation.

Images