A method for predicting short-term power output of distributed small hydropower stations

Abstract

The present invention relates to a distributed method for predicting short-term power generation output of small hydropower stations. The method analyzes the hydrological patterns of the watershed based on historical power and rainfall information of the small hydropower stations. A distributed framework structure and large-scale data mining calculation methods are used to establish calibration parameters for the correlation between rainfall and power generation of the power station. A chaotic time series method is then used to predict the power generation of the power station over the next three days using a future watershed rainfall forecast sequence. A similarity machine learning discriminant method is then used to decompose the daily power generation process, ultimately yielding a three-day output process forecast sequence for the small hydropower station. This method accurately describes the intrinsic relationship between rainfall and power generation for small hydropower stations without water condition monitoring. Furthermore, a recurrent neural network and similarity data mining are used to decompose the daily power generation process, achieving a daily output process forecast for the power station without water condition monitoring. This method provides a method for compiling planned output for power stations without water condition monitoring for refined small hydropower scheduling.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation capacity prediction, in particular to a method for predicting short-term power generation output of a distributed small hydropower station. Background Art

[0002] With the continuous development of the social economy, the electricity load of residents and enterprises has increased rapidly, and the imbalance between power generation and consumption among regions has become increasingly prominent. During peak periods of electricity load, it is necessary to fully utilize all available resources to meet peak load demands. With the development of refined power dispatching, small hydropower has the characteristics of small installed capacity and large total volume, and the need to arrange its peak load has gradually become prominent. However, due to the low construction standards of small hydropower stations, most of them do not have water regime monitoring stations. Conventional power forecasting methods for power stations are ineffective in such cases. Therefore, it is necessary to carry out power generation output forecasting for small hydropower stations without water regime monitoring to meet the needs of formulating output plans for small hydropower stations and refining power generation.

[0003] For power station output prediction without water condition monitoring stations, there are many related factors that affect its power generation. Traditional prediction methods are difficult to express the relationship between various related factors. Therefore, simplifying the complexities and combining predictions with current data mining related technologies can solve this type of data analysis and prediction problem from another perspective. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method for predicting the short-term power generation output of a distributed small hydropower station, which can solve the technical problems raised in the background technology.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for predicting short-term power generation output of a distributed small hydropower station according to the present invention comprises the following steps:

[0007] Step A: Based on the power generation of small hydropower stations without water condition monitoring [y1, y2, y3…yi] and the basin rainfall [r1, r2, r3…ri] information, a time series correlation relationship is constructed. The data is classified using data mining and cluster analysis methods, and the least squares method is used for parameter calibration to establish the correlation between rainfall and power generation of the power station:

[0008] y * =R(r1,r2,r3...r i ) (1)

[0009] In formula (1), y* is the target power, r is the rainfall sequence, and R is the calibration function;

[0010] Step B: Using the chaotic time series method, based on the basin rainfall information [r1, r2, r3…ri], a rolling forecast is made for the daily power consumption of the small hydropower station over the next three days. The obtained rainfall sequence is classified according to time, and a matching relationship formula between rainfall and power generation is found to predict the daily power consumption process.

[0011] y i =Input(r i ) (2)

[0012] In formula (2), yi is the predicted daily electricity on the i-th day, the input function is the conversion function of the input rainfall into electricity, and ri is the predicted basin rainfall data on the i-th day;

[0013] Step C: Using a recurrent neural network method, find similar processes for daily electricity consumption, decompose the daily electricity consumption into output processes, and obtain the daily power generation output process;

[0014] [O1,O2...O 96 ]=Gene(y t-2 ,y t-1 ,y t ,t) (3)

[0015] In formula (3), O1…O 96 is the daily power generation process, y is the predicted power sequence, and t is the date of the daily power to be decomposed. In order to improve the prediction accuracy, y t Relying on the power consumption process predicted in the previous period;

[0016] Step D: The predicted output process is subjected to a sliding average method to eliminate the burr process of the prediction sequence caused by actual load fluctuations, thereby obtaining a smooth curve that can be used for prediction.

[0017] The principle of the present invention is as follows: basin hydrological law analysis is conducted based on historical power consumption information and rainfall information of small hydropower stations, and large-scale data mining calculation methods are used in a MapReduce distributed framework structure to establish calibration parameters for the correlation between rainfall and power station power consumption. Then, a chaotic time series method is used to predict the power station power consumption for the next three days through a future basin rainfall prediction sequence. Then, a similarity machine learning discrimination method is used to decompose the daily power consumption process on a day-by-day basis, and finally a prediction sequence for the output process of the small hydropower station for the next three days is obtained.

[0018] The present invention has the following beneficial effects: A distributed small hydropower station short-term power output forecasting method utilizes information technology, such as big data analysis, to accurately describe the intrinsic relationship between rainfall and power generation at small hydropower stations without hydrological monitoring. Furthermore, it uses recurrent neural networks and similarity data mining to decompose the daily power generation process, enabling daily power generation process forecasting for stations without hydrological monitoring. This provides a method for compiling planned output for stations without hydrological monitoring, facilitating refined small hydropower scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0020] Figure 1 It is a specific implementation flow chart of the present invention. DETAILED DESCRIPTION

[0021] like Figure 1 As shown in the figure, a distributed small hydropower station short-term power output prediction method of this embodiment analyzes the hydrological laws of the watershed based on the historical power information and rainfall information of the small hydropower station, adopts the MapReduce distributed framework structure and uses large-scale data mining calculation methods to establish the correlation parameters between rainfall and power station power, and then uses the future watershed rainfall prediction sequence to predict the power station power for the next three days using a chaotic time series method. Then, using the similarity machine learning discrimination method, the daily power is decomposed into a day-by-day process, and finally a power output process prediction sequence for the small hydropower station for the next three days is obtained.

[0022] There are four steps in total:

[0023] Step A: Based on the power generation of small hydropower stations without water condition monitoring [y1, y2, y3…yi] and basin rainfall [r1, r2, r3…ri], a time series correlation relationship is constructed, where y1, y2, y3…yi in the power generation [y1, y2, y3…yi] are the power generation in different time periods, and r1, r2, r3…ri are the basin rainfall in different time periods. At the same time, the time nodes in the two time periods are unified. The two fluctuation curves under different time axes can be analyzed by using the CoFlux algorithm to obtain the time series correlation relationship between the power generation of small hydropower stations [y1, y2, y3…yi] and the basin rainfall [r1, r2, r3…ri];

[0024] Data mining and clustering analysis are then used to classify the data. Common methods for data analysis using data mining include classification, regression analysis, clustering, association rules, feature, variation and deviation analysis, and web page mining. These methods each mine data from different perspectives. In this embodiment, the power generation [y1, y2, y3…yi] and basin rainfall [r1, r2, r3…ri] information is classified primarily using a combination of feature, variation, and deviation analysis. Specifically, the rainfall series is classified according to specific value ranges, degree of variation, and deviation from the average value.

[0025] After classification, the least squares method is used for parameter calibration. The results of the deviation analysis can be directly used for least squares calculations. The least squares method is a familiar and optimized method. It mainly minimizes the square of the error and the matching function of the best data. By matching the power [y1, y2, y3…yi] and the basin rainfall [r1, r2, r3…ri] information by minimizing the square of the error and the best data, a correlation between rainfall and power station power is established:

[0026] y * =R(r1,r2,r3...r i ) (1)

[0027] In formula (1), y* is the target power, r is the rainfall sequence, and R is the calibration function.

[0028] Step B: Using the chaotic time series method, the daily electricity consumption of the small hydropower station in the next three days is predicted based on the basin rainfall information [r1, r2, r3…ri] in a discrete and random manner instead of in chronological order. The predicted value is y*.

[0029] Classify the acquired rainfall sequence by time, for example, by rainfall amount. The corresponding time when the rainfall amount is greater than a threshold is classified. From the perspective of rainfall amount, a matching relationship formula between rainfall and power generation of power stations is found to predict the daily power generation process.

[0030] y i =Input(r i ) (2)

[0031] In formula (2), yi is the predicted daily electricity on the i-th day, the input function is the conversion function of the input rainfall into electricity, and ri is the predicted basin rainfall data on the i-th day;

[0032] Step C: Using the recurrent neural network (RNN) method, we can automatically determine whether there is a predicted relationship between rainfall and power generation in step B through machine learning, i.e., y i =Input(r i), and find similar processes for daily electricity. The recurrent neural network can substitute the previous historical data into the later prediction operation, compare it with the actual value, and continuously correct the relationship function to finally obtain a more accurate prediction result; decompose the output process of daily electricity to obtain a discrete daily power generation output process prediction;

[0033] [O1,O2...O 96 ]=Gene(y t-2 ,y t-1 ,y t ,t) (3)

[0034] In formula (3), O1…O 96 is the daily power generation process, y is the predicted power sequence, and t is the date of the daily power to be decomposed. In order to improve the prediction accuracy, y t Relying on the power consumption process predicted in the previous period;

[0035] Step D: The predicted output process is subjected to the sliding average method to eliminate the burr process of the prediction sequence caused by the actual load fluctuation, and a smooth curve is obtained. The short-term power generation output of the small hydropower station in the future time period is predicted by the smooth curve.

[0036] This method, a distributed small hydropower station short-term power output forecasting method, utilizes information technology such as big data analysis to accurately describe the intrinsic relationship between rainfall and power generation for small hydropower stations without hydrological monitoring. It also uses recurrent neural networks and similarity data mining to decompose the daily power generation process, enabling daily power generation process forecasting for stations without hydrological monitoring. This provides a method for compiling planned output for stations without hydrological monitoring, facilitating refined small hydropower scheduling.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting short-term power generation output of a distributed small hydropower station, characterized by: The following steps are involved: Step A: Based on the power generation of small hydropower stations without water condition monitoring [y1,y2,y3…y i ] and basin rainfall [r1, r2, r3…r i ] information, where the power generation [y1,y2,y3…y i ] in y1, y2, y3...y i is the power generation in different time periods, and r1, r2, r3...r i The rainfall in the basin in different time periods is unified by the time nodes in the two time periods to obtain the power generation of the small hydropower station [y1, y2, y3…y i ] and basin rainfall [r1, r2, r3…r i ], and use data mining cluster analysis method to classify data, use the least square method to calibrate parameters, and minimize the square of the error and the most appropriate data for the electricity [y1,y2,y3…y i ] and basin rainfall [r1,r2,r3…r i ] information to establish a correlation between rainfall and power generation of power stations: y * =R(r1,r2,r3...r i ) (1) In formula (1), y * is the target power, r is the rainfall sequence, and R is the rating function; Step B: Using the chaotic time series method, according to the basin rainfall information [r1, r2, r3…r i ], rolling forecast of the daily electricity process of the small hydropower station in the future time period T, classify the obtained rainfall sequence according to time, find the matching relationship formula between rainfall and power station electricity, and predict the daily electricity process; y i =Input(r i ) (2) In formula (2), y i is the predicted daily electricity consumption on the i-th day, the Input function is the conversion function of the input rainfall into electricity consumption, r i is the predicted basin rainfall data for day i; Step C: Using a recurrent neural network method, through machine learning, automatically determine whether there is a predicted relationship between rainfall and power generation in step B. Find similar processes for daily power generation, decompose the daily power generation process, and obtain the daily power generation output process. [O1,O2...O 96 ]=Gene(y t-2 ,and t-1 ,and t ,t) (3) In formula (3), O1…O 96 is the daily power generation process, y is the predicted power sequence, and t is the date of the daily power to be decomposed. In order to improve the prediction accuracy, y t Relying on the power consumption process predicted in the previous period; Step D: The predicted output process is subjected to the sliding average method to eliminate the prediction sequence burr process caused by actual load fluctuations, and a prediction smooth curve is obtained. The short-term power generation output of the small hydropower station in the future time period is predicted by the smooth curve.

2. A method for predicting short-term power generation output of a distributed small hydropower station according to claim 1, characterized in that: In step B, the future time period T is 3 days, and the rainfall sequence classification is performed by combining feature, change and deviation analysis.

Images