Power load prediction system based on deep learning

Abstract

The invention belongs to the field of power load prediction, relates to a data analysis technology, is used for solving the problem that stage load prediction analysis cannot be performed according to external parameters in the prior art, and particularly relates to a power load prediction system based on deep learning, which comprises a load prediction platform, the load prediction platform is in communication connection with a load analysis module, a feature extraction module, a load prediction module, a prediction management module and a database; the load analysis module is used for performing power utilization load analysis on the power load prediction region: generating an analysis period, dividing the power load prediction region into a plurality of analysis regions, obtaining power utilization intervals of the analysis regions and marking matching days of the power utilization intervals; according to the method, power load analysis can be carried out on a power load prediction region, statistics is carried out on historical power consumption data in an analysis region in a regional analysis mode, and data support is provided for a feature extraction analysis process.

Description

Technical Field

[0001] The present invention belongs to the field of power load forecasting and relates to data analysis technology, and specifically is a power load forecasting system based on deep learning. Background Art

[0002] Electricity load forecasting is a scientific estimation of the amount of electricity required by various industries in social production and people's lives. It is the basis for implementing planned electricity consumption and formulating production plans. It is one of the main items in the design work of industrial and mining enterprises, and is also an important link in ensuring the balance between economic needs and electricity supply.

[0003] The invention patent with announcement number CN104392274B discloses a method for predicting urban short-term electricity load based on electricity load and temperature trends. The prediction method inputs the sudden incremental parameters of the power grid through manual interaction, and participates in the calculation of the urban load prediction model algorithm based on the interactive parameters, so that the sudden factors are quickly integrated into the algorithm in the prediction, solving the problem that the traditional model cannot overcome the impact of sudden factors (major user commissioning, sudden grid connection of large generators) in the short term; however, this method cannot perform phased load forecasting analysis based on external parameters, nor can it verify the accuracy of the prediction results after completing the electricity load forecast, and thus cannot guarantee the accuracy of the electricity load forecast and cannot dynamically optimize the electricity load forecast model.

[0004] In view of the above technical problems, this application proposes a solution. Summary of the invention

[0005] The purpose of the present invention is to provide a power load forecasting system based on deep learning, which is used to solve the problem that the prior art cannot perform phased load forecasting analysis based on external parameters;

[0006] The technical problem to be solved by the present invention is: how to provide a power load forecasting system based on deep learning that can perform phased load forecasting analysis according to external parameters.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A power load forecasting system based on deep learning, comprising a load forecasting platform, wherein the load forecasting platform is communicatively connected with a load analysis module, a feature extraction module, a load forecasting module, a forecasting management module and a database;

[0009] The load analysis module is used to analyze the power load in the area of ​​power load forecasting: generate an analysis cycle, divide the area of ​​power load forecasting into several analysis areas, obtain the power consumption interval of the analysis area and mark the matching day of the power consumption interval; send the power consumption interval and matching day to the feature extraction module through the load forecasting platform;

[0010] The feature extraction module is used to extract and analyze the external features of the power consumption interval and obtain the predicted feature parameters of the power consumption interval; the predicted feature parameters of all power consumption intervals in the analysis area are sent to the load prediction platform through the load prediction platform;

[0011] The load forecasting module is used to perform load forecasting analysis on the area where the power load is forecasted;

[0012] The forecast management module is used to perform forecast accuracy management analysis on the area where the power load is forecasted.

[0013] Furthermore, the specific process of obtaining the electricity consumption interval of the analysis area and marking the matching days of the electricity consumption interval includes: real-time monitoring of the total electricity load of the analysis area every day within the analysis period, forming a power consumption set of the analysis area from the total electricity load of the analysis area on all natural days within the analysis period, forming a power consumption range of the analysis area from the maximum element and the minimum element of the power consumption set, dividing the power consumption range of the analysis area into several electricity consumption intervals, and marking the natural days where the total electricity load is within the electricity consumption interval as matching days of the electricity consumption interval.

[0014] Furthermore, the specific process of the feature extraction module extracting and analyzing the external features of the power consumption interval includes: setting an external parameter i for the analysis area, i=1, 2,..., n, where n is a positive integer, obtaining the value of the external parameter i of the analysis area on the matching day within the power consumption interval and marking it as the analysis value FXi, forming a covering set i of the external parameter i by the analysis values ​​FXi corresponding to all matching days of the external parameter i of the analysis area within the power consumption interval, performing variance calculation on all elements in the covering set i to obtain a covering concentrated value, and generating a switching feature range i of the external parameter i through the covering concentrated value; summing and averaging the total power load of all matching days in the power consumption interval to obtain a load prediction value of the power consumption interval; and forming a prediction feature parameter of the power consumption interval by the switching feature range i and the load prediction value.

[0015] Furthermore, the process of obtaining the switching feature range i of the external parameter i includes: obtaining the coverage concentration threshold through the database, and comparing the coverage concentration value with the coverage concentration threshold: if the coverage concentration value is less than the coverage concentration threshold, the maximum element and the minimum element in the coverage set i constitute the switching feature range i of the external parameter i corresponding to the power consumption interval; if the coverage concentration value is greater than or equal to the coverage concentration threshold, the maximum element and the minimum element in the coverage set are eliminated, and then the coverage concentration value of the coverage set is recalculated, and so on, until the coverage concentration value is less than the coverage concentration threshold.

[0016] Furthermore, the specific process of the load forecasting platform performing load forecasting analysis on the area of ​​power load forecasting includes: generating a forecasting period, obtaining the value of the external parameter i in the analysis area on each natural day of the forecasting period and marking it as the monitoring value JCi, and when the monitoring value JCi is within the corresponding switching feature range i, retrieving the load forecast value of the power consumption interval corresponding to the switching feature range i, and sending the load forecast value to the load forecasting platform for feedback; when the monitoring value JCi cannot match any switching feature range i, it is determined that the forecast has failed, a forecast failure signal is generated, and the forecast failure signal is sent to the forecast management module through the load forecasting platform.

[0017] Furthermore, the specific process of the forecast management module to perform forecast accuracy management analysis on the area of ​​power load forecasting includes: obtaining the forecast performance value YC, the actual performance value SJ and the received data JS of the forecast period; The prediction coefficient YC of the prediction period is obtained, where k1 and k2 are both proportional coefficients, and k1>k2>1. The prediction coefficient YC is used to determine whether the power load prediction accuracy of the analysis area within the prediction period meets the requirements.

[0018] Furthermore, the process of obtaining the predicted performance value YC, the actual performance value SJ and the received data JS includes: obtaining the number of matching days for the analysis area to the power consumption interval within the prediction period and marking it as the matching value, marking the product of the matching value and the load prediction value of the power consumption interval as the predicted performance value YC of the power consumption interval, marking the sum of the actual power load of all matching days for the analysis period to the power consumption interval within the prediction period as the actual performance value SJ, and marking the number of times the prediction failure signal is received within the prediction period as the received data JS.

[0019] Furthermore, the specific process of determining whether the power load prediction accuracy of the analysis area within the prediction period meets the requirements includes: obtaining the prediction threshold YCmax through the database, and comparing the prediction coefficient YC of the prediction period with the prediction threshold YCmax: if the prediction coefficient YC is less than the prediction threshold YCmax, then it is determined that the power load prediction accuracy of the analysis area within the prediction period meets the requirements; if the prediction coefficient YC is greater than or equal to the prediction threshold YCmax, then it is determined that the power load prediction accuracy of the analysis area within the prediction period does not meet the requirements, a load analysis signal is generated and sent to the load analysis module through the load prediction platform. After receiving the load analysis signal, the load analysis module re-divides the power consumption interval and the matching day according to the total power load of each natural day in the analysis area within the prediction period; and re-extracts the prediction characteristic parameters of the power consumption interval for the power load prediction analysis of the analysis area in the next prediction period.

[0020] The present invention has the following beneficial effects:

[0021] 1. The load analysis module can be used to analyze the power load in the area where the power load is predicted. The historical power consumption data in the analysis area can be statistically analyzed in a regional manner. Then, the power consumption interval is generated according to the overall power consumption situation. After the matching day of the power consumption interval is marked, data support is provided for the feature extraction and analysis process.

[0022] 2. The feature extraction module can extract and analyze the external features of the power consumption interval, mark the external parameters that affect the power load, and then analyze the numerical features of the external parameters of all matching days in different power consumption intervals, and then predict the power load based on the extracted prediction feature parameters;

[0023] 3. The load forecasting module can be used to conduct load forecasting analysis on the area of ​​power load forecasting. The numerical value of the external parameters of the analysis area is compared with the switching feature range, and then the load forecast value in the corresponding prediction feature parameter is retrieved to predict the daily power consumption of the analysis area at this stage, and the necessity of switching in the power consumption stage is monitored in real time to achieve phased power load forecasting;

[0024] 4. The forecast management module can be used to perform forecast accuracy management and analysis on the area of ​​power load forecasting, and a comprehensive analysis and calculation of multiple forecast parameters within the forecast period can be performed to obtain the forecast coefficient. The accuracy of the forecast results of the forecast period can then be fed back based on the forecast coefficient. When the accuracy is low, the forecast feature parameters can be re-extracted based on the latest electricity consumption data. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 is a system block diagram of Embodiment 1 of the present invention;

[0027] Figure 2 This is a flow chart of the method of Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0028] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Embodiment 1: Figure 1 As shown, a power load forecasting system based on deep learning includes a load forecasting platform, which is communicatively connected to a load analysis module, a feature extraction module, a load forecasting module, a forecasting management module and a database.

[0030] The load analysis module is used to analyze the power load in the area where the power load is predicted: generate an analysis cycle, divide the area where the power load is predicted into several analysis areas, monitor the total power load of the analysis area every day in real time during the analysis cycle, and form the power consumption set of the analysis area by the total power load of all natural days in the analysis cycle. The maximum element and the minimum element of the power consumption set form the power consumption range of the analysis area. The power consumption range of the analysis area is divided into several power consumption intervals, and the natural days with the total power load in the power consumption interval are marked as matching days of the power consumption interval; the power consumption interval and matching day are sent to the feature extraction module through the load forecasting platform; the historical power consumption data in the analysis area are counted in a regional analysis manner, and then the power consumption interval is generated according to the overall power consumption situation. After the matching day of the power consumption interval is marked, data support is provided for the feature extraction analysis process.

[0031] The feature extraction module is used to extract and analyze the external features of the power consumption interval: set an external parameter i for the analysis area, i = 1, 2, ..., n, n is a positive integer, and the external parameter may include nighttime duration and outdoor temperature, etc.; obtain the value of the external parameter i of the analysis area on the matching day within the power consumption interval and mark it as the analysis value FXi, and the analysis value FXi corresponding to the external parameter i of the analysis area on all matching days within the power consumption interval constitutes a covering set i of the external parameter i, calculate the variance of all elements in the covering set i to obtain the covering concentration value, obtain the covering concentration threshold through the database, and compare the covering concentration value with the covering concentration threshold: if the covering concentration value is less than the covering concentration threshold, the maximum element and the minimum element in the covering set i constitute the switching feature range i of the power consumption interval corresponding to the external parameter i ; If the coverage concentration value is greater than or equal to the coverage concentration threshold, the maximum element and the minimum element in the coverage set are removed, and then the coverage concentration value of the coverage set is recalculated, and so on, until the coverage concentration value is less than the coverage concentration threshold; the total power load of all matching days in the power consumption interval is summed and averaged to obtain the load forecast value of the power consumption interval; the switching feature range i and the load forecast value constitute the prediction feature parameters of the power consumption interval, and the prediction feature parameters of all power consumption intervals in the analysis area are sent to the load forecasting platform through the load forecasting platform; the external characteristics of the power consumption interval are extracted and analyzed, the external parameters that affect the power load are marked, and then the numerical characteristics of the external parameters of all matching days in different power consumption intervals are analyzed, and then the power load is predicted based on the extracted prediction feature parameters.

[0032] The load forecasting module is used to perform load forecasting analysis on the area where power load is forecasted: generate a forecasting cycle, obtain the value of the external parameter i in the analysis area on each natural day of the forecasting cycle and mark it as the monitoring value JCi, when the monitoring value JCi is within the corresponding switching feature range i, retrieve the load forecast value of the power consumption interval corresponding to the switching feature range i, and send the load forecast value to the load forecasting platform for feedback; when the monitoring value JCi cannot match any switching feature range i, it is determined that the forecast has failed, and a forecast failure signal is generated and sent to the forecast management module through the load forecasting platform; perform load forecasting analysis on the area where power load is forecasted, compare and analyze the value of the external parameter of the analysis area with the switching feature range, and then retrieve the load forecast value in the corresponding forecast feature parameter to predict the daily power consumption of the analysis area in this stage, and monitor the necessity of switching in the power consumption stage in real time to achieve staged power load forecasting.

[0033] The prediction management module is used to perform prediction accuracy management and analysis on the area of ​​power load prediction: obtain the number of matching days of the analysis area for the application power interval within the prediction period and mark it as the matching value, mark the product of the matching value and the load prediction value of the power interval as the prediction performance value YC of the power interval, mark the sum of the actual power load of all matching days of the analysis period for the application power interval within the prediction period as the actual performance value SJ, and mark the number of prediction failure signals received within the prediction period as the received data JS; through the formula The prediction coefficient YC of the prediction period is obtained, where k1 and k2 are both proportional coefficients, and k1>k2>1. The prediction threshold YCmax is obtained through the database, and the prediction coefficient YC of the prediction period is compared with the prediction threshold YCmax: if the prediction coefficient YC is less than the prediction threshold YCmax, it is determined that the power load prediction accuracy of the analysis area within the prediction period meets the requirements; if the prediction coefficient YC is greater than or equal to the prediction threshold YCmax, it is determined that the power load prediction accuracy of the analysis area within the prediction period does not meet the requirements, and a load analysis signal is generated and sent to the load analysis through the load forecasting platform. Module, after receiving the load analysis signal, the load analysis module re-divides the power consumption interval and the matching day according to the total power load of the analysis area on each natural day in the prediction period; and re-extracts the prediction characteristic parameters of the power consumption interval for the power load prediction analysis of the analysis area in the next prediction period; performs prediction accuracy management analysis on the area of ​​power load prediction, combines multiple prediction parameters in the prediction period for comprehensive analysis and calculation to obtain the prediction coefficient, and then provides feedback on the accuracy of the prediction results of the prediction period according to the prediction coefficient, and re-extracts the prediction characteristic parameters according to the latest power consumption data when the accuracy is low.

[0034] Embodiment 2: Figure 2As shown, a power load forecasting method based on deep learning includes the following steps:

[0035] Step 1: Conduct power load analysis on the area where power load is predicted: generate an analysis cycle, divide the area where power load is predicted into several analysis areas, and divide the power consumption range of the analysis area into several power consumption intervals;

[0036] Step 2: Extract and analyze the external features of the power consumption interval to obtain the predicted characteristic parameters of the power consumption interval;

[0037] Step 3: Carry out load forecast analysis for the area of ​​power load forecast: Generate a forecast period, and retrieve the load forecast value according to the monitoring value JCi corresponding to the natural day in the forecast period of the analysis area;

[0038] Step 4: Conduct forecast accuracy management analysis on the area of ​​power load forecast and obtain the forecast coefficient YC of the analysis area within the forecast period. Use the forecast coefficient YC to determine whether the forecast accuracy of power load in the analysis area within the forecast period meets the requirements.

[0039] A power load forecasting system based on deep learning generates an analysis cycle when working, divides the area for power load forecasting into several analysis areas, and divides the power consumption range of the analysis area into several power consumption intervals; extracts and analyzes the external characteristics of the power consumption interval to obtain the prediction characteristic parameters of the power consumption interval; generates a forecast cycle, and retrieves the load forecast value according to the monitoring value JCi corresponding to the natural day of the analysis area within the forecast cycle; performs forecast accuracy management analysis on the area for power load forecasting and obtains the prediction coefficient YC of the analysis area within the forecast cycle, and judges whether the power load forecast accuracy of the analysis area within the forecast cycle meets the requirements through the prediction coefficient YC.

[0040] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

[0041] The above formulas are obtained by collecting a large amount of data and performing software simulation to select a formula close to the actual value. The coefficients in the formula are set by technicians in this field according to actual conditions; for example: Formula A technician in this field collects multiple groups of sample data and sets corresponding prediction coefficients for each group of sample data; substitutes the set prediction coefficients and the collected sample data into the formula, any two formulas constitute a binary linear equation system, the calculated coefficients are screened and averaged, and the values ​​of k1 and k2 are obtained to be 3.48 and 2.52 respectively;

[0042] The size of the coefficient is to quantify each parameter to obtain a specific value for subsequent comparison. The size of the coefficient depends on the amount of sample data and the preliminary setting of the corresponding prediction coefficient for each group of sample data by technical personnel in this field; as long as it does not affect the proportional relationship between the parameter and the quantized value, such as the prediction coefficient is proportional to the value of the received data.

[0043] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0044] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A power load forecasting system based on deep learning, characterized in that: It includes a load forecasting platform, which is communicatively connected with a load analysis module, a feature extraction module, a load forecasting module, a forecasting management module and a database; The load analysis module is used to analyze the power load in the area of ​​power load forecasting: generate an analysis cycle, divide the area of ​​power load forecasting into several analysis areas, obtain the power consumption interval of the analysis area and mark the matching day of the power consumption interval; send the power consumption interval and matching day to the feature extraction module through the load forecasting platform; The feature extraction module is used to extract and analyze the external features of the power consumption interval and obtain the predicted feature parameters of the power consumption interval; the predicted feature parameters of all power consumption intervals in the analysis area are sent to the load prediction platform through the load prediction platform; The load forecasting module is used to perform load forecasting analysis on the area where the power load is forecasted; The forecast management module is used to perform forecast accuracy management analysis on the area where the power load is forecasted.

2. The power load forecasting system based on deep learning according to claim 1, characterized in that: The specific process of obtaining the power consumption interval of the analysis area and marking the matching days of the power consumption interval includes: real-time monitoring of the total power load of the analysis area every day within the analysis period, forming the power consumption set of the analysis area by the total power load of the analysis area on all natural days within the analysis period, forming the power consumption range of the analysis area by the maximum element and the minimum element of the power consumption set, dividing the power consumption range of the analysis area into several power consumption intervals, and marking the natural days with the total power load within the power consumption interval as matching days of the power consumption interval.

3. The power load forecasting system based on deep learning according to claim 2, characterized in that: The specific process of the feature extraction module extracting and analyzing the external features of the power consumption interval includes: setting an external parameter i for the analysis area, i=1, 2,..., n, where n is a positive integer, obtaining the value of the external parameter i of the analysis area on the matching day within the power consumption interval and marking it as the analysis value FXi, forming a covering set i of the external parameter i from the analysis value FXi corresponding to all matching days of the external parameter i of the analysis area within the power consumption interval, performing variance calculation on all elements in the covering set i to obtain a covering concentrated value, and generating a switching feature range i of the external parameter i through the covering concentrated value; summing and averaging the total power load of all matching days within the power consumption interval to obtain a load forecast value of the power consumption interval; and forming a forecast feature parameter of the power consumption interval from the switching feature range i and the load forecast value.

4. The power load forecasting system based on deep learning according to claim 3 is characterized in that: The process of obtaining the switching feature range i of the external parameter i includes: obtaining the coverage concentration threshold through the database, and comparing the coverage concentration value with the coverage concentration threshold: if the coverage concentration value is less than the coverage concentration threshold, the maximum element and the minimum element in the coverage set i constitute the switching feature range i of the external parameter i corresponding to the power consumption interval; if the coverage concentration value is greater than or equal to the coverage concentration threshold, the maximum element and the minimum element in the coverage set are eliminated, and then the coverage concentration value of the coverage set is recalculated, and so on, until the coverage concentration value is less than the coverage concentration threshold.

5. The power load forecasting system based on deep learning according to claim 4 is characterized in that: The specific process of the load forecasting platform performing load forecasting analysis on the area of ​​power load forecasting includes: generating a forecasting period, obtaining the value of the external parameter i in the analysis area on each natural day of the forecasting period and marking it as the monitoring value JCi, and when the monitoring value JCi is within the corresponding switching feature range i, retrieving the load forecast value of the power consumption interval corresponding to the switching feature range i, and sending the load forecast value to the load forecasting platform for feedback; when the monitoring value JCi cannot match any switching feature range i, it is determined that the forecast has failed, generating a forecast failure signal and sending the forecast failure signal to the forecast management module through the load forecasting platform.

6. The power load forecasting system based on deep learning according to claim 5, characterized in that: The specific process of the forecast management module to perform forecast accuracy management analysis on the area of ​​power load forecasting includes: obtaining the forecast performance value YC, the actual performance value SJ and the received data JS of the forecast period; The prediction coefficient YC of the prediction period is obtained, where k1 and k2 are both proportional coefficients, and k1>k2>1. The prediction coefficient YC is used to determine whether the power load prediction accuracy of the analysis area within the prediction period meets the requirements.

7. The power load forecasting system based on deep learning according to claim 6, characterized in that: The process of obtaining the predicted performance value YC, the actual performance value SJ and the received data JS includes: obtaining the number of matching days for the analysis area to the power consumption interval within the prediction period and marking it as the matching value, marking the product of the matching value and the load prediction value of the power consumption interval as the predicted performance value YC of the power consumption interval, marking the sum of the actual power load of all matching days for the analysis period to the power consumption interval within the prediction period as the actual performance value SJ, and marking the number of times the prediction failure signal is received within the prediction period as the received data JS.

8. The power load forecasting system based on deep learning according to claim 7, characterized in that: The specific process of determining whether the power load prediction accuracy of the analysis area within the prediction period meets the requirements includes: obtaining the prediction threshold value YCmax through the database, and comparing the prediction coefficient YC of the prediction period with the prediction threshold value YCmax: if the prediction coefficient YC is less than the prediction threshold value YCmax, then it is determined that the power load prediction accuracy of the analysis area within the prediction period meets the requirements; if the prediction coefficient YC is greater than or equal to the prediction threshold value YCmax, then it is determined that the power load prediction accuracy of the analysis area within the prediction period does not meet the requirements, and a load analysis signal is generated and sent to the load analysis module through the load prediction platform. After receiving the load analysis signal, the load analysis module re-divides the power consumption interval and the matching day according to the total power load of each natural day in the analysis area within the prediction period; and re-extracts the prediction characteristic parameters of the power consumption interval for the power load prediction analysis of the analysis area in the next prediction period.

Images