WO2020118587A1 - 一种负荷预测方法及装置 - Google Patents

一种负荷预测方法及装置 Download PDF

Info

Publication number
WO2020118587A1
WO2020118587A1 PCT/CN2018/120747 CN2018120747W WO2020118587A1 WO 2020118587 A1 WO2020118587 A1 WO 2020118587A1 CN 2018120747 W CN2018120747 W CN 2018120747W WO 2020118587 A1 WO2020118587 A1 WO 2020118587A1
Authority
WO
WIPO (PCT)
Prior art keywords
load prediction
value
initial load
model
prediction value
Prior art date
Application number
PCT/CN2018/120747
Other languages
English (en)
French (fr)
Inventor
刘松
刘鹏
崔亚明
俞石洪
横山隆一
Original Assignee
华北电力大学扬中智能电气研究中心
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华北电力大学扬中智能电气研究中心 filed Critical 华北电力大学扬中智能电气研究中心
Priority to PCT/CN2018/120747 priority Critical patent/WO2020118587A1/zh
Publication of WO2020118587A1 publication Critical patent/WO2020118587A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Definitions

  • the present invention relates to the field of load prediction technology, and in particular to a load prediction method and device.
  • Short-term load forecasting is the basis of power system planning and normal operation. It is related to power generation, dispatching, and decision-making of the power system. Therefore, short-term load forecasting has always been a hot spot for domestic and foreign experts. Improving the accuracy of short-term load forecasting is critical to the operating efficiency, benefit, and safety of the power system.
  • Embodiments of the present invention provide a load prediction method and device to solve the problem of inaccurate load prediction in the prior art.
  • An embodiment of the present invention provides a load forecasting method.
  • the method includes: receiving a first initial load forecast value currently output by at least two single prediction models;
  • the target load prediction value is determined.
  • the method further includes:
  • the determining the target load prediction value based on the BPNN model includes:
  • the target load prediction value is determined according to the first weight of each input layer and the second weight of each output layer in the BPNN model, and each input first initial load prediction value.
  • the training model of the BPNN model includes:
  • the second set of initial load prediction values and the true load value corresponding to the set of second initial load prediction values are input into the BPNN model to train the BPNN model.
  • the method further includes:
  • test load prediction value corresponding to the third initial load prediction value of the group For each set of third initial load prediction values of the preset number in the test set, determine the test load prediction value corresponding to the third initial load prediction value of the group based on the BPNN model;
  • an embodiment of the present invention provides a load forecasting device.
  • the device includes:
  • a receiving module configured to receive the first initial load prediction value currently output by at least two single prediction models
  • the first input module is used to input each first initial load prediction value into a pre-trained back propagation neural network BPNN model
  • the determining module is used to determine the target load prediction value based on the BPNN model.
  • the device further includes:
  • the second input module is configured to input the acquired data within a preset time length to at least two single prediction models.
  • the determination module is specifically configured to determine the target load according to the first weight of each input layer and the second weight of each output layer in the BPNN model, and each input first initial load prediction value Predictive value.
  • the device further includes:
  • the training module is used to input the set of second initial load prediction values and the load real values corresponding to the set of second initial load prediction values into each BPNN model for each set of second initial load prediction values in the training set. For training.
  • the training module is further configured to determine the test load prediction value corresponding to the third initial load prediction value of the group based on the BPNN model for a preset number of each third initial load prediction value in the test set; Test load prediction values and the actual load values corresponding to each test load prediction value, determine the error evaluation value of the BPNN model; determine whether the error evaluation value is less than a preset threshold, and if so, determine the BPNN model training carry out.
  • An embodiment of the present invention provides a load prediction method and device.
  • the method includes: receiving at least two first initial load prediction values currently output by a single prediction model; inputting each first initial load prediction value into a pre-trained completion Back propagation neural network BPNN model; based on the BPNN model, the target load prediction value is determined.
  • the first initial load prediction value is first determined based on a single prediction model, and then each first initial load prediction value is input into the pre-trained BPNN model, based on the BPNN model, combined
  • Each first initial load forecast value determines the target load forecast value, which avoids the problem of a single model that tends to fall into local optimization when the data set has poor convergence, large fluctuations, and is affected by emergencies. Therefore, the determined target load prediction value is made more accurate.
  • Embodiment 1 is a schematic diagram of a load forecasting process provided by Embodiment 1 of the present invention.
  • Embodiment 2 is a schematic diagram of a load prediction process provided by Embodiment 2 of the present invention.
  • FIG. 3 is a schematic structural diagram of a BPNN model provided by Embodiment 3 of the present invention.
  • FIG. 4 is a schematic structural diagram of a load prediction device provided by an embodiment of the present invention.
  • FIG. 1 is a schematic diagram of a load forecasting process provided by an embodiment of the present invention. The process includes the following steps:
  • S101 Receive a first initial load prediction value currently output by at least two single prediction models.
  • the load prediction method provided by the embodiment of the present invention is applied to an electronic device, and the electronic device may be a device such as a PC or a tablet computer.
  • the load forecasting method provided by the embodiment of the present invention includes but is not limited to a short-term load forecasting method applied to the power system.
  • the electronic device stores at least two pre-trained single prediction models, where the single prediction model may be an autoregressive integral moving average ARIMA model and a support vector machine SVM model, and so on.
  • the electronic device Before determining the target load prediction value, the electronic device first receives the first initial load prediction value currently output by at least two single prediction models. The process of determining the first initial load prediction value based on a single prediction model belongs to the prior art, and the process will not be repeated here.
  • the back-propagation neural network BPNN model that has been pre-trained is also stored in the electronic device. After receiving the first initial load prediction value currently received by at least two single prediction models, each first initial load prediction value is input into the pre- The trained BPNN model. Based on the BPNN model, the target load forecast value is determined.
  • the first initial load prediction value is first determined based on a single prediction model, and then each first initial load prediction value is input into the pre-trained BPNN model, based on the BPNN model, combined
  • Each first initial load forecast value determines the target load forecast value, which avoids the problem of a single model that tends to fall into local optimization when the data set has poor convergence, large fluctuations, and is affected by emergencies. Therefore, the determined target load prediction value is made more accurate.
  • the method before receiving the initial load prediction values output by at least two single prediction models, the method further includes:
  • the preset time length may be three days, four days, etc.
  • the data within the preset time length may be the load value within the preset time length, or may be the load value within the preset time length and Weather data within a preset length of time.
  • the weather data may be data such as average temperature and average humidity.
  • the weather data includes the average temperature of the previous three days, the average humidity of the previous three days, the average temperature of the day, the average humidity of the day, and so on.
  • the first initial load prediction value is output based on each single prediction model.
  • FIG. 2 is a schematic diagram of a load forecasting process provided by an embodiment of the present invention.
  • the interference data may affect the accuracy of the final target load forecast
  • the electronic device may store a smaller first threshold and a larger second threshold, use data less than the first threshold and greater than the second threshold in the data as interference data, filter the interference data, and then filter the remaining data
  • the data are entered into a single prediction model.
  • the single prediction model includes ARIMA model, multiple regression model, random forest model and SVM model.
  • the single prediction model in FIG. 2 is only an example, and in the embodiment of the present invention, the type and number of the single prediction model are not limited.
  • the electronic device inputs the acquired data within the preset time length into the ARIMA model, multiple regression model, random forest model, and SVM model, determines the first initial load prediction value based on each single prediction model, and then The load forecast value is input into the BPNN model to determine the target load forecast value.
  • the determining the target load prediction value based on the BPNN model includes:
  • the target load prediction value is determined according to the first weight of each input layer and the second weight of each output layer in the BPNN model, and each input first initial load prediction value.
  • Figure 3 is a schematic diagram of the structure of the BPNN model.
  • the working process of the BPNN model is: first, the multi-layer neural network is used to compare the value of the input signal and the output signal, and the expected output value is used to obtain the mean square error. Finally, the mean square error is back propagated, and the internal weight neurons are adjusted continuously until the error meets the requirements.
  • the BPNN model is composed of three different layers: input layer, hidden layer and output layer, each layer is composed of many neurons, as shown in Figure 3.
  • the first weight between the input layer i neuron and the hidden layer h neuron is Vih
  • the hidden layer is Whj.
  • X i is the input of the i-th neuron in the hidden layer
  • the jth neuron of the output layer receives the input from the hidden layer as ⁇ j :
  • b h is the output of the h-th neuron in the hidden layer.
  • each first initial load prediction value each first weight and each second weight input to the BPNN model, according to the above formula, the target load prediction value can be determined.
  • the training model of the BPNN model includes:
  • the second set of initial load prediction values and the true load value corresponding to the set of second initial load prediction values are input into the BPNN model to train the BPNN model.
  • the initial load prediction value in the training set is used as the second initial load prediction value
  • the electronic device groups the second initial load prediction value, for example, the second initial load output by each single prediction model on the same day Predicted values as a group.
  • the set of second initial load predicted values and the true load values corresponding to the set of second initial load predicted values are input into the BPNN model.
  • the parameters of the BPNN model are adjusted according to the difference between the predicted load value and the true load value output by the BPNN model until the training of the BPNN model is completed.
  • the method further includes:
  • test load prediction value corresponding to the third initial load prediction value of the group For each set of third initial load prediction values of the preset number in the test set, determine the test load prediction value corresponding to the third initial load prediction value of the group based on the BPNN model;
  • a test set is stored in the electronic device and used to check the accuracy of the BPNN model.
  • the initial load prediction value in the test set is used as the third initial load prediction value.
  • a preset number of each set of third initial load prediction values in the test set may be selected. The preset number may be 50, 80, etc.
  • the test load prediction value corresponding to the group of the third initial load prediction value is determined based on the BPNN model. Then, according to each test load prediction value and the actual load value corresponding to each test load prediction value, the error evaluation value of the BPNN model is determined.
  • the error evaluation value may be an absolute average error, an average absolute percentage error, or an average variance.
  • S i is the i-th test load prediction value, It is the true load value corresponding to the i-th test load prediction value.
  • a preset threshold is stored in the electronic device, and the threshold may be a small value, such as 0.1, 0.2, and so on. After determining the error evaluation value of the BPNN model, the electronic device determines whether the error evaluation value is less than a preset threshold, and if so, determines that the BPNN model training is completed. Otherwise, continue training the BPNN model.
  • FIG. 4 is a schematic structural diagram of a load prediction device provided by an embodiment of the present invention, and the device includes:
  • the receiving module 41 is configured to receive the first initial load prediction value currently output by at least two single prediction models
  • the first input module 42 is used to input each first initial load prediction value into a pre-trained back propagation neural network BPNN model
  • the determining module 43 is configured to determine a target load prediction value based on the BPNN model.
  • the device also includes:
  • the second input module 44 is configured to input the acquired data within a preset time length to at least two single prediction models.
  • the determination module 43 is specifically configured to determine the target load prediction value according to the first weight of each input layer and the second weight of each output layer in the BPNN model, and each input first initial load prediction value .
  • the device also includes:
  • the training module 45 is configured to input the set of second initial load prediction values and the load real values corresponding to the set of second initial load prediction values into each BPNN model for each set of second initial load prediction values in the training set.
  • the model is trained.
  • the training module 45 is further configured to determine the test load prediction value corresponding to the third initial load prediction value of the group based on the BPNN model for a preset number of third initial load prediction values of the test set; based on each test The load prediction value and the true load value corresponding to each test load prediction value determine the error evaluation value of the BPNN model; determine whether the error evaluation value is less than a preset threshold, and if so, determine that the BPNN model training is completed.
  • An embodiment of the present invention provides a load prediction method and device.
  • the method includes: receiving at least two first initial load prediction values currently output by a single prediction model; inputting each first initial load prediction value into a pre-trained completion Back propagation neural network BPNN model; based on the BPNN model, the target load prediction value is determined.
  • the first initial load prediction value is first determined based on a single prediction model, and then each first initial load prediction value is input into the pre-trained BPNN model, based on the BPNN model, combined
  • Each first initial load forecast value determines the target load forecast value, which avoids the problem of a single model that tends to fall into local optimization when the data set has poor convergence, large fluctuations, and is affected by emergencies. Therefore, the determined target load prediction value is made more accurate.
  • each flow and/or block in the flowchart and/or block diagram and a combination of the flow and/or block in the flowchart and/or block diagram may be implemented by computer program instructions.
  • These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processing machine, or other programmable data processing device to produce a machine that enables the generation of instructions executed by the processor of the computer or other programmable data processing device
  • These computer program instructions may also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, the instructions
  • the device implements the functions specified in one block or multiple blocks of the flowchart one flow or multiple flows and/or block diagrams.
  • These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operating steps are performed on the computer or other programmable device to produce computer-implemented processing, which is executed on the computer or other programmable device
  • the instructions provide steps for implementing the functions specified in one block or multiple blocks of the flowchart one flow or multiple flows and/or block diagrams.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了一种负荷预测方法及装置，所述方法包括：接收至少两个单一预测模型当前输出的第一初始负荷预测值；将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；基于所述BPNN模型，确定目标负荷预测值。由于在本发明实施例中，在进行负荷预测时，首先基于单一预测模型确定第一初始负荷预测值，然后将每个第一初始负荷预测值输入预先训练完成的BPNN模型，基于BPNN模型，结合每个第一初始负荷预测值确定出目标负荷预测值，避免了单一模型存在的对于那些数据集收敛性差、波动较大和受突发事件影响的情况下，往往会出现陷入局部最优化的问题，因此使得确定的目标负荷预测值更准确。

Description

一种负荷预测方法及装置 技术领域
本发明涉及负荷预测技术领域，尤其涉及一种负荷预测方法及装置。
背景技术
短期负荷预测是电力系统规划和正常运行的基础，它关系到电力系统的发电、调度、和决策等。因此，短期负荷预测的一直是国内外专家研究的热点。提高短期负荷预测的准确度，对于电力系统的运行效率、效益和安全至关重要。
现有技术中进行短期负荷预测时，一般通过时间序列、回归分析、支持向量机等单一的方法进行预测。现有技术中的预测方法对于趋势明显、表现很强收敛性的数据集预测效果较好。但是，单一方法进行预测，对于那些数据集收敛性差、波动较大和受突发事件影响的情况下，现有技术中的预测方法往往会出现陷入局部最优化的问题，导致预测结果准确性较差。
发明内容
本发明实施例提供了一种负荷预测方法及装置，用以解决现有技术中负荷预测不准确的问题。
本发明实施例提供了一种负荷预测方法，所述方法包括：接收至少两个单一预测模型当前输出的第一初始负荷预测值；
将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；
基于所述BPNN模型，确定目标负荷预测值。
进一步地，所述接收至少两个单一预测模型输出的初始负荷预测值之前，所述方法还包括：
将获取到的预设时间长度内的数据分别输入至少两个单一预测模型。
进一步地，所述基于所述BPNN模型，确定目标负荷预测值包括：
根据所述BPNN模型中每个输入层的第一权重和每个输出层的第二权重，以及输入的每个第一初始负荷预测值，确定目标负荷预测值。
进一步地，所述BPNN模型的训练模型包括：
针对训练集中每组第二初始负荷预测值，将该组第二初始负荷预测值，和该组第二初始负荷预测值对应的负荷真实值输入BPNN模型，对所述BPNN模型进行训练。
进一步地，所述方法还包括：
针对测试集中预设数量的每组第三初始负荷预测值，基于所述BPNN模型确定该组第三初始负荷预测值对应的测试负荷预测值；
根据每个测试负荷预测值和每个测试负荷预测值对应的负荷真实值，确定所述BPNN模型的误差评价值；
判断所述误差评价值是否小于预设的阈值，如果是，确定所述BPNN模型训练完成。
另一方面，本发明实施例提供了一种负荷预测装置，所述装置包括：
接收模块，用于接收至少两个单一预测模型当前输出的第一初始负荷预测值；
第一输入模块，用于将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；
确定模块，用于基于所述BPNN模型，确定目标负荷预测值。
进一步地，所述装置还包括：
第二输入模块，用于将获取到的预设时间长度内的数据分别输入至少两个单一预测模型。
进一步地，所述确定模块，具体用于根据所述BPNN模型中每个输入层的第一权重和每个输出层的第二权重，以及输入的每个第一初始负荷预测值，确定目标负荷预测值。
进一步地，所述装置还包括：
训练模块，用于针对训练集中每组第二初始负荷预测值，将该组第二初始负荷预测值，和该组第二初始负荷预测值对应的负荷真实值输入BPNN模型，对所述BPNN模型进行训练。
进一步地，所述训练模块，还用于针对测试集中预设数量的每组第三初始负荷预测值，基于所述BPNN模型确定该组第三初始负荷预测值对应的测试负荷预测值；根据每个测试负荷预测值和每个测试负荷预测值对应的负荷真实值，确定所述BPNN模型的误差评价值；判断所述误差评价值是否小于预设的阈值，如果是，确定所述BPNN模型训练完成。
本发明实施例提供了一种负荷预测方法及装置，所述方法包括：接收至少两个单一预测模型当前输出的第一初始负荷预测值；将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；基于所述BPNN模型，确定目标负荷预测值。
由于在本发明实施例中，在进行负荷预测时，首先基于单一预测模型确定第一初始负荷预测值，然后将每个第一初始负荷预测值输入预先训练完成的BPNN模型，基于BPNN模型，结合每个第一初始负荷预测值确定出目标负荷预测值，避免了单一模型存在的对于那些数据集收敛性差、波动较大和受突发事件影响的情况下，往往会出现陷入局部最优化的问题，因此使得确定的目标负荷预测值更准确。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简要介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域的普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例1提供的负荷预测过程示意图；
图2为本发明实施例2提供的负荷预测流程示意图；
图3为本发明实施例3提供的BPNN模型的结构示意图；
图4为本发明实施例提供的负荷预测装置结构示意图。
具体实施方式
下面将结合附图对本发明作进一步地详细描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
实施例1：
图1为本发明实施例提供的负荷预测过程示意图，该过程包括以下步骤：
S101：接收至少两个单一预测模型当前输出的第一初始负荷预测值。
S102：将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型。
S103：基于所述BPNN模型，确定目标负荷预测值。
本发明实施例提供的负荷预测方法应用于电子设备，该电子设备可以是PC、平板电脑等设备。本发明实施例提供的负荷预测方法包括但不限定是一种应用于电力系统的短期负荷预测方法。
在本发明实施例中，电子设备中保存有预先训练完成的至少两个单一预测模型，其中，单一预测模型可以是自回归积分滑动平均ARIMA模型和支持向量机SVM模型等等。电子设备确定目标负荷预测值之前，首先接收至少两个单一预测模型当前输出的第一初始负荷预测值。其中，基于单一预测模型确定第一初始负荷预测值的过程属于现有技术，在此不再对该过程进行赘述。
电子设备中还保存有预先训练完成的反向传播神经网络BPNN模型，在接收到收至少两个单一预测模型当前输出的第一初始负荷预测值后，将每个第一初始负荷预测值输入预先训练完成的BPNN模型。基于BPNN模型，确定目标负荷预测值。
由于在本发明实施例中，在进行负荷预测时，首先基于单一预测模型确定第一初始负荷预测值，然后将每个第一初始负荷预测值输入预先训练完成的BPNN模型，基于BPNN模型，结合每个第一初始负荷预测值确定出目标负荷预测值，避免了单一模型存在的对于那些数据集收敛性差、波动较大和受突发事件影响的情况下，往往会出现陷入局部最优化的问题，因此使得确定的目标负荷预测值更准确。
实施例2：
在上述实施例的基础上，在本发明实施例中，所述接收至少两个单一预测模型输出的初始负荷预测值之前，所述方法还包括：
将获取到的预设时间长度内的数据分别输入至少两个单一预测模型。
在本发明实施例中，预设时间长度可以是三天、四天等，预设时间长度内的数据可以是预设时间长度内的负荷值，也可以是预设时间长度内的负荷值以及预设时间长度内的天气数据。其中，天气数据可以是平均温度、平均湿度等数据。例如，天气数据包括前三天的平均温度、前三天的平均湿度、当天的平均温度、当天的平均湿度等。
电子设备将获取到的预设时间长度内的数据分别输入至少两个单一预测模型之后，基于每个单一预测模型输出第一初始负荷预测值。
图2为本发明实施例提供的负荷预测流程示意图，如图2所示，由于获取的预设时间长度内的数据中有可能存在干扰数据，干扰数据会影响最终确定的目标负荷预测值的准确性，为了使确定的目标负荷预测值更准确，在将数据分别输入至少两个单一预测模型之前，还需要对数据进行预处理。具体的，电子设备中可以保存较小的第一阈值和较大的第二阈值，将数据中小于第一阈值和大于第二阈值的数据作为干扰数据，将干扰数据滤除，然后将剩余的数据分别输入单一预测模型。如图2所示，单一预测模型包括ARIMA模型、多元回归模型、随机森林模型和SVM模型。需要说明的是，图2中的单一预测模型仅是举例说明，在本发明实施例中，并不对单一预测模型的类型和数量进行限定。电子设备将获取到的预设时间长度内 的数据分别输入ARIMA模型、多元回归模型、随机森林模型和SVM模型，基于每个单一预测模型确定第一初始负荷预测值，然后将每个第一初始负荷预测值输入BPNN模型，确定目标负荷预测值。
实施例3：
在上述各实施例的基础上，在本发明实施例中，所述基于所述BPNN模型，确定目标负荷预测值包括：
根据所述BPNN模型中每个输入层的第一权重和每个输出层的第二权重，以及输入的每个第一初始负荷预测值，确定目标负荷预测值。
图3为BPNN模型的结构示意图，BPNN模型的工作过程是：首先利用多层神经网络将输入信号与输出信号的值进行比较，并用期望的输出值来获得均方误差。最后，将均方误差反向传播，内部权重神经元不断调整直到误差满足要求。BPNN模型由三种不同的组成层：输入层、隐藏层和输出层，每一层都是由许多神经元组成，如图3所示。
假设有d个输入层神经元，有i个输出层神经元，q个隐藏层神经元；输入层第i个神经元与隐藏层第h个神经元之间的第一权重为Vih，隐藏层第h个神经元与输出层第j个神经元之间的第二权重为Whj。
记隐藏层第h个神经元接收到来自于输入层的输入为α h：
Figure PCTCN2018120747-appb-000001
其中，X i为隐藏层第i个神经元的输入；
记输出层第j个神经元接收到来自于隐藏层的输入为β j：
Figure PCTCN2018120747-appb-000002
其中，b h为隐藏层第h个神经元的输出。
根据输入到BPNN模型的每个第一初始负荷预测值、每个第一权重和每个第二权重，根据上述公式，可以确定出目标负荷预测值。
实施例4：
在上述各实施例的基础上，在本发明实施例中，所述BPNN模型的训练模型包括：
针对训练集中每组第二初始负荷预测值，将该组第二初始负荷预测值，和该组第二初始负荷预测值对应的负荷真实值输入BPNN模型，对所述BPNN模型进行训练。
在本发明实施例中，将训练集中的初始负荷预测值作为第二初始负荷预测值，电子设备将第二初始负荷预测值进行分组，例如同一天的每个单一预测模型输出的第二初始负荷预测值作为一组。针对训练集中每组第二初始负荷预测值，将该组第二初始负荷预测值，和该组第二初始负荷预测值对应的负荷真实值输入BPNN模型。根据BPNN模型输出的负荷预测值与负荷真实值的差值对BPNN模型的参数进行调整，直到BPNN模型训练完成。
实施例5：
为了保证BPNN模型的准确性，在上述各实施例的基础上，在本发明实施例中，所述方法还包括：
针对测试集中预设数量的每组第三初始负荷预测值，基于所述BPNN模型确定该组第三初始负荷预测值对应的测试负荷预测值；
根据每个测试负荷预测值和每个测试负荷预测值对应的负荷真实值，确定所述BPNN模型的误差评价值；
判断所述误差评价值是否小于预设的阈值，如果是，确定所述BPNN模型训练完成。
电子设备中保存有测试集，用于检验BPNN模型的准确性。在本发明实施例中，将测试集中的初始负荷预测值作为第三初始负荷预测值。在进行BPNN模型准确性验证时，可以选取测试集中预设数量的每组第三初始负荷预测值，预设数量可以是50、80等。针对每组第三初始负荷预测值，基于BPNN模型确定该组第三初始负荷预测值对应的测试负荷预测值。然后根据每个测试负荷预测值和每个测试负荷预测值对应的负荷真实值，确定BPNN模型的误差评价值。
具体的，在本发明实施例中，误差评价值可以是绝对平均误差、平均 绝对百分误差或者平均方差。
绝对平均误差的计算公式为：
Figure PCTCN2018120747-appb-000003
平均绝对百分误差的计算公式为：
Figure PCTCN2018120747-appb-000004
平均方差的计算公式为：
Figure PCTCN2018120747-appb-000005
式中，S i为第i个测试负荷预测值，
Figure PCTCN2018120747-appb-000006
为第i个测试负荷预测值对应的负荷真实值。
电子设备中保存有预设的阈值，该阈值可以是较小的值，例如0.1、0.2等。电子设备在确定出BPNN模型的误差评价值后，判断误差评价值是否小于预设的阈值，如果是，则确定BPNN模型训练完成。否则，要对BPNN模型继续进行训练。
图4为本发明实施例提供的负荷预测装置结构示意图，所述装置包括：
接收模块41，用于接收至少两个单一预测模型当前输出的第一初始负荷预测值；
第一输入模块42，用于将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；
确定模块43，用于基于所述BPNN模型，确定目标负荷预测值。
所述装置还包括：
第二输入模块44，用于将获取到的预设时间长度内的数据分别输入至少两个单一预测模型。
所述确定模块43，具体用于根据所述BPNN模型中每个输入层的第一权重和每个输出层的第二权重，以及输入的每个第一初始负荷预测值，确定目标负荷预测值。
所述装置还包括：
训练模块45，用于针对训练集中每组第二初始负荷预测值，将该组第 二初始负荷预测值，和该组第二初始负荷预测值对应的负荷真实值输入BPNN模型，对所述BPNN模型进行训练。
所述训练模块45，还用于针对测试集中预设数量的每组第三初始负荷预测值，基于所述BPNN模型确定该组第三初始负荷预测值对应的测试负荷预测值；根据每个测试负荷预测值和每个测试负荷预测值对应的负荷真实值，确定所述BPNN模型的误差评价值；判断所述误差评价值是否小于预设的阈值，如果是，确定所述BPNN模型训练完成。
本发明实施例提供了一种负荷预测方法及装置，所述方法包括：接收至少两个单一预测模型当前输出的第一初始负荷预测值；将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；基于所述BPNN模型，确定目标负荷预测值。
由于在本发明实施例中，在进行负荷预测时，首先基于单一预测模型确定第一初始负荷预测值，然后将每个第一初始负荷预测值输入预先训练完成的BPNN模型，基于BPNN模型，结合每个第一初始负荷预测值确定出目标负荷预测值，避免了单一模型存在的对于那些数据集收敛性差、波动较大和受突发事件影响的情况下，往往会出现陷入局部最优化的问题，因此使得确定的目标负荷预测值更准确。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存 储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
尽管已描述了本发明的优选实施例，但本领域内的技术人员一旦得知了基本创造性概念，则可对这些实施例作出另外的变更和修改。所以，所附权利要求意欲解释为包括优选实施例以及落入本发明范围的所有变更和修改。
显然，本领域的技术人员可以对本发明进行各种改动和变型而不脱离本发明的精神和范围。这样，倘若本发明的这些修改和变型属于本发明权利要求及其等同技术的范围之内，则本发明也意图包含这些改动和变型在内。

Claims (10)

  1. 一种负荷预测方法，其特征在于，所述方法包括：
    接收至少两个单一预测模型当前输出的第一初始负荷预测值；
    将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；
    基于所述BPNN模型，确定目标负荷预测值。
  2. 如权利要求1所述的方法，其特征在于，所述接收至少两个单一预测模型输出的初始负荷预测值之前，所述方法还包括：
    将获取到的预设时间长度内的数据分别输入至少两个单一预测模型。
  3. 如权利要求1所述的方法，其特征在于，所述基于所述BPNN模型，确定目标负荷预测值包括：
    根据所述BPNN模型中每个输入层的第一权重和每个输出层的第二权重，以及输入的每个第一初始负荷预测值，确定目标负荷预测值。
  4. 如权利要求1所述的方法，其特征在于，所述BPNN模型的训练模型包括：
    针对训练集中每组第二初始负荷预测值，将该组第二初始负荷预测值，和该组第二初始负荷预测值对应的负荷真实值输入BPNN模型，对所述BPNN模型进行训练。
  5. 如权利要求4所述的方法，其特征在于，所述方法还包括：
    针对测试集中预设数量的每组第三初始负荷预测值，基于所述BPNN模型确定该组第三初始负荷预测值对应的测试负荷预测值；
    根据每个测试负荷预测值和每个测试负荷预测值对应的负荷真实值，确定所述BPNN模型的误差评价值；
    判断所述误差评价值是否小于预设的阈值，如果是，确定所述BPNN模型训练完成。
  6. 一种负荷预测装置，其特征在于，所述装置包括：
    接收模块，用于接收至少两个单一预测模型当前输出的第一初始负荷预测值；
    第一输入模块，用于将每个第一初始负荷预测值输入预先训练完成的反向传播神经网络BPNN模型；
    确定模块，用于基于所述BPNN模型，确定目标负荷预测值。
  7. 如权利要求6所述的装置，其特征在于，所述装置还包括：
    第二输入模块，用于将获取到的预设时间长度内的数据分别输入至少两个单一预测模型。
  8. 如权利要求6所述的装置，其特征在于，所述确定模块，具体用于根据所述BPNN模型中每个输入层的第一权重和每个输出层的第二权重，以及输入的每个第一初始负荷预测值，确定目标负荷预测值。
  9. 如权利要求6所述的装置，其特征在于，所述装置还包括：
    训练模块，用于针对训练集中每组第二初始负荷预测值，将该组第二初始负荷预测值，和该组第二初始负荷预测值对应的负荷真实值输入BPNN模型，对所述BPNN模型进行训练。
  10. 如权利要求9所述的装置，其特征在于，所述训练模块，还用于针对测试集中预设数量的每组第三初始负荷预测值，基于所述BPNN模型确定该组第三初始负荷预测值对应的测试负荷预测值；根据每个测试负荷预测值和每个测试负荷预测值对应的负荷真实值，确定所述BPNN模型的误差评价值；判断所述误差评价值是否小于预设的阈值，如果是，确定所述BPNN模型训练完成。
PCT/CN2018/120747 2018-12-12 2018-12-12 一种负荷预测方法及装置 WO2020118587A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/120747 WO2020118587A1 (zh) 2018-12-12 2018-12-12 一种负荷预测方法及装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/120747 WO2020118587A1 (zh) 2018-12-12 2018-12-12 一种负荷预测方法及装置

Publications (1)

Publication Number Publication Date
WO2020118587A1 true WO2020118587A1 (zh) 2020-06-18

Family

ID=71075850

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/120747 WO2020118587A1 (zh) 2018-12-12 2018-12-12 一种负荷预测方法及装置

Country Status (1)

Country Link
WO (1) WO2020118587A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090327173A1 (en) * 2008-06-25 2009-12-31 Inotera Memories, Inc. Method for predicting cycle time
CN102102626A (zh) * 2011-01-30 2011-06-22 华北电力大学 风电场短期功率预测方法
CN103049798A (zh) * 2012-12-05 2013-04-17 浙江大学城市学院 一种应用于光伏发电系统的短期发电功率预测方法
CN104376389A (zh) * 2014-12-10 2015-02-25 国电南京自动化股份有限公司 基于负载均衡的主从式微电网功率负荷预测系统及其方法
CN105719002A (zh) * 2016-01-18 2016-06-29 重庆大学 一种基于组合预测的风电机组状态参数异常辨识方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090327173A1 (en) * 2008-06-25 2009-12-31 Inotera Memories, Inc. Method for predicting cycle time
CN102102626A (zh) * 2011-01-30 2011-06-22 华北电力大学 风电场短期功率预测方法
CN103049798A (zh) * 2012-12-05 2013-04-17 浙江大学城市学院 一种应用于光伏发电系统的短期发电功率预测方法
CN104376389A (zh) * 2014-12-10 2015-02-25 国电南京自动化股份有限公司 基于负载均衡的主从式微电网功率负荷预测系统及其方法
CN105719002A (zh) * 2016-01-18 2016-06-29 重庆大学 一种基于组合预测的风电机组状态参数异常辨识方法

Similar Documents

Publication Publication Date Title
CN106874581B (zh) 一种基于bp神经网络模型的建筑空调能耗预测方法
CN110298501B (zh) 基于长短时记忆神经网络的电负荷预测方法
US20190265768A1 (en) Method, system and storage medium for predicting power load probability density based on deep learning
Kankal et al. Modeling and forecasting of Turkey’s energy consumption using socio-economic and demographic variables
Chang et al. A hybrid ANFIS model based on AR and volatility for TAIEX forecasting
Shao et al. Nickel price forecast based on the LSTM neural network optimized by the improved PSO algorithm
CN104217258B (zh) 一种电力负荷条件密度预测方法
Wang et al. A novel decomposition-ensemble forecasting system for dynamic dispatching of smart grid with sub-model selection and intelligent optimization
Minli et al. Research on the application of artificial neural networks in tender offer for construction projects
WO2020118586A1 (zh) 一种能源消耗预测方法及装置
Zeng et al. Short-term load forecasting of smart grid systems by combination of general regression neural network and least squares-support vector machine algorithm optimized by harmony search algorithm method
Dong et al. A multifactorial short-term load forecasting model combined with periodic and non-periodic features-a case study of qingdao, China
Chen et al. A multivariate grey RBF hybrid model for residual useful life prediction of industrial equipment based on state data
Li et al. Aircraft failure rate prediction method based on CEEMD and combined model
CN106067075B (zh) 一种建筑用能负荷预测模型建立、负荷预测方法及其装置
Chen et al. Fuzzy-neural approaches with example post-classification for estimating job cycle time in a wafer fab
Wei et al. Energy financial risk early warning model based on Bayesian network
CN112686693A (zh) 电力现货市场边际电价预测方法、系统、设备及存储介质
WO2020118587A1 (zh) 一种负荷预测方法及装置
CN111310964B (zh) 一种负荷预测方法及装置
Kim et al. Projection-aware deep neural network for dc optimal power flow without constraint violations
Wang et al. Rolling forecast nature gas spot price with back propagation neural network
Jia et al. Line loss rate prediction method based on deep learning with long short term memory
JP2000305606A (ja) 配水需要量予測方法
Li et al. A new BP neural network model for the prediction problem of equally spaced time sequences and its application

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18943212

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18943212

Country of ref document: EP

Kind code of ref document: A1