Modeling methods and devices for integrated energy systems

Abstract

The present invention provides a modeling method and device for an integrated energy system, the method comprising: determining multiple historical times, and obtaining multiple features of the integrated energy system at each historical time; selecting a target feature and multiple non-target features from the multiple features; performing dimensionality reduction processing on the multiple non-target features corresponding to each historical time; training each neural network model according to the time features corresponding to the multiple historical times, the target features corresponding to the multiple historical times, and the non-target features after dimensionality reduction; and fusing the trained neural network models to obtain a target neural network model. Thus, using the historical time as alignment data, the required model is obtained by training multiple neural network models with multiple features at the historical time, that is, the modeling of the integrated energy system and the power supply equipment in the system can be realized without mathematical principles, and the modeling accuracy can be improved while reducing the modeling cost.

Description

Technical Field

[0001] The present invention relates to the field of integrated energy technology, and in particular to a modeling method for an integrated energy system and a modeling device for an integrated energy system. Background Art

[0002] With the continuous development of energy technology, the integrated energy system faces the problem of access to a variety of new energy sources and coordinated control with traditional fossil energy. Among them, wind power and photovoltaics are the main forms of new energy, and energy storage power stations are usually built synchronously with wind power and photovoltaics as necessary supporting facilities for the integrated energy system. Therefore, it is extremely important to build a model for the integrated energy system.

[0003] Among the related technologies, mathematical principles are usually used to model the mechanism of a certain power supply equipment in the integrated energy system. However, this modeling method has high requirements for mathematics and requires modelers to understand the principles of power supply equipment. It has high technical barriers and large capital investment, which leads to low accuracy of mechanism modeling and high modeling costs. Summary of the invention

[0004] In order to solve one of the above technical problems, the present invention proposes the following technical solution.

[0005] The first aspect of the embodiment of the present invention proposes a modeling method for an integrated energy system, wherein the integrated energy system includes multiple power supply devices, and the method includes the following steps: determining multiple historical times and obtaining multiple features of the integrated energy system at each historical time; determining a modeling object corresponding to the integrated energy system, wherein the modeling object is the integrated energy system or one of the power supply devices; selecting a target feature and multiple non-target features corresponding to the modeling object from the multiple features; performing dimensionality reduction processing on the multiple non-target features corresponding to each of the historical time; constructing multiple different neural network models, and training each of the neural network models according to the time features corresponding to the multiple historical times, the target features corresponding to the multiple historical times, and the non-target features after dimensionality reduction; and fusing multiple trained neural network models to obtain a target neural network model corresponding to the modeling object.

[0006] In addition, the modeling method of the integrated energy system according to the above embodiment of the present invention may also have the following additional technical features.

[0007] According to an embodiment of the present invention, each of the features has a corresponding feature category, and the feature category includes one of a power supply equipment feature category, an integrated energy system feature category, and a meteorological feature category of an area where the integrated energy system is located.

[0008] According to one embodiment of the present invention, a target feature and multiple non-target features corresponding to the modeling object are selected from multiple features, including: when the modeling object is the integrated energy system, one feature is selected from the features included in the integrated energy system feature category as the target feature, and the features other than the target feature in the multiple features corresponding to each historical time are used as the non-target features; when the modeling object is the power supply equipment, one feature is selected from the features included in the power supply equipment feature category as the target feature, and the features other than the target feature in the multiple features corresponding to each historical time are used as the non-target features.

[0009] According to one embodiment of the present invention, the multiple non-target features corresponding to each of the historical times are subjected to dimensionality reduction processing, including: using a principal component analysis algorithm to perform dimensionality reduction processing on the multiple non-target features corresponding to each of the historical times.

[0010] According to one embodiment of the present invention, the structures and parameters of the multiple neural network models are different, and the method further includes: determining the attributes of the modeling object; determining the number of the multiple neural network models, the structure and parameters of each of the neural network models according to the attributes of the modeling object, wherein the number of the multiple neural network models is an odd number.

[0011] According to one embodiment of the present invention, the features in the meteorological feature category include at least one of solar radiation, rainfall, relative humidity and wind speed, the features in the power supply equipment feature category include instrument measurement data and output of each of the power supply equipment, and the features in the integrated energy system feature category include at least one of external power received by the system, external power received by the system, external power transmitted by the system, external power transmitted by the system and energy supply variables.

[0012] According to one embodiment of the present invention, when the modeling object is the integrated energy system, the power transmitted by the system is used as the target feature; when the modeling object is the power supply device, the output of the power supply device is used as the target feature.

[0013] According to one embodiment of the present invention, after obtaining the target neural network model of the modeling object, it also includes: when the integrated energy system changes, determining the non-target features after the change; performing dimensionality reduction processing on the non-target features after the change through a principal component analysis algorithm, and inputting the non-target features after the change into the target neural network model to obtain new target features; comparing the target features with the new target features to obtain the degree of change of the target features; and judging the degree of influence of the changes in the integrated energy system on the modeling object according to the degree of change of the target features.

[0014] According to an embodiment of the present invention, the plurality of power supply devices include one or more of a photovoltaic power station, a wind turbine, a distributed gas turbine and an energy storage power station.

[0015] The second aspect of the present invention proposes a modeling device for an integrated energy system, wherein the integrated energy system includes multiple power supply devices, and the device includes: an acquisition module, used to determine multiple historical times and obtain multiple features of the integrated energy system at each historical time; a determination module, used to determine the modeling object corresponding to the integrated energy system, wherein the modeling object is the integrated energy system or one of the power supply devices; a selection module, used to select a target feature and multiple non-target features corresponding to the modeling object from the multiple features; a dimensionality reduction module, used to perform dimensionality reduction processing on the multiple non-target features corresponding to each of the historical time; a training module, used to construct multiple different neural network models, and train each of the neural network models according to the time features corresponding to the multiple historical times, the target features corresponding to the multiple historical times, and the non-target features after dimensionality reduction; a fusion module, used to fuse multiple trained neural network models to obtain a target neural network model corresponding to the modeling object.

[0016] The technical solution of the embodiment of the present invention first determines multiple historical times, obtains multiple features of the integrated energy system at each historical time, determines the modeling object, then selects the target features and non-target features corresponding to the modeling object from the multiple features, then performs dimensionality reduction processing on the non-target features corresponding to each historical time, constructs multiple different neural network models, trains each neural network model according to the time features corresponding to the multiple historical times, multiple target features and the non-target features after dimensionality reduction, and finally fuses the multiple trained neural network models to obtain the target neural network model of the modeling object. Thus, using the historical time as the alignment data, the required model is obtained by training multiple neural network models with multiple features under the historical time, that is, the modeling of the integrated energy system and the power supply equipment in the system can be realized without mathematical principles, which can improve the modeling accuracy while reducing the modeling cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a flow chart of a method for modeling an integrated energy system according to an embodiment of the present invention.

[0018] Figure 2 The schematic diagram is a principle diagram of modeling a modeling object according to an example of the present invention.

[0019] Figure 3 This is a schematic diagram of the principle of determining new target characteristics according to a model after adding power supply equipment to an integrated energy system of an example of the present invention.

[0020] Figure 4 Schematic diagram of a block diagram of a modeling device for an integrated energy system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0022] An integrated energy system refers to a system that uses advanced physical information technology and innovative management models to integrate multiple energy sources such as coal, oil, natural gas, electricity, and thermal energy in a certain area, and achieves coordinated planning, optimized operation, collaborative management, interactive response, and complementary assistance among multiple heterogeneous energy subsystems.

[0023] In the related art, when modeling an integrated energy system, a mechanism model is usually established by deducing mathematical principles. However, due to the different working principles of various types of power supplies, the workload of mechanism modeling of the integrated energy system (the process of establishing a system model based on the system mechanism) is large during application, such as in the planning process and the simulation process. In addition, since the manufacturers of various power supply equipment are different, it is difficult to ensure the accuracy of the mechanism modeling, and the modeling cost is high.

[0024] To this end, an embodiment of the present invention adopts a method of fusing multiple machine learning models to learn the real data of the integrated energy system, and mines the patterns in the output data of the integrated energy system and the patterns of a power supply device in the system by changing the granularity of the data, and then reversely reconstructs the model of the integrated energy system or a power supply device in the system, thereby ensuring the similarity between the model and the real system or equipment as much as possible for subsequent integrated energy system planning and simulation.

[0025] The present invention uses real data training to reversely reconstruct and establish a comprehensive energy system model. At the same time, the power supply equipment model in the comprehensive energy system is reversely established through the characteristic of variable data granularity.

[0026] Figure 1 The figure is a flow chart of a method for modeling an integrated energy system according to an embodiment of the present invention.

[0027] The integrated energy system of an embodiment of the present invention includes multiple power supply devices, which may include one or more of photovoltaic power stations, wind turbines, distributed gas turbines and energy storage power stations. For example, the integrated energy equipment may include two photovoltaic power stations, two wind turbines, two distributed gas turbines and two energy storage power stations.

[0028] like Figure 1 As shown, the modeling method of the integrated energy system includes the following steps S1 to S6.

[0029] S1, determine multiple historical times and obtain multiple characteristics of the integrated energy system at each historical time.

[0030] It should be noted that among the data types of the integrated energy system, the time feature is a very special feature type, which contains a lot of implicit information. If a single piece of data of a power supply device in the system contains a time feature, it will be unique and non-repeatable. This also creates conditions for mining the patterns in the output data of the integrated energy system and the patterns of a certain power supply in the system through the granularity of changing data, and then reversely reconstructing the model of the integrated energy system or a certain power supply device in the system.

[0031] Therefore, the time feature is a feature that can be referenced in the data features of each power supply in the integrated energy system. In order to better explore the coupling between various types of power supplies and different power supply devices in the system, in the embodiment of the present invention, time data is used as the alignment feature to integrate and process all data features of each power supply in the integrated energy system.

[0032] Specifically, multiple historical times can be determined, and the characteristic categories of historical times are time characteristics, and multiple characteristics of the integrated energy system at each historical time can be obtained, such as the output of each power supply equipment and the power transmitted by the system, among which multiple characteristics are real output data characteristics at the corresponding historical time.

[0033] It should be noted that each feature carries its corresponding value. When multiple features are obtained, the value of each feature can be obtained, which is the real value.

[0034] S2, determining a modeling object corresponding to the integrated energy system, wherein the modeling object is the integrated energy system or a power supply device.

[0035] Specifically, it can be determined based on specific needs whether the modeling object is an integrated energy system or a certain power supply device therein. When the modeling object is an integrated energy system, it is necessary to regard the integrated energy system as a whole to model the system as a whole; when the modeling object is a certain power supply device, the power supply device is modeled as a whole, that is, at this time, the local power supply device of the system is modeled.

[0036] It should be noted that, in actual applications, step S1 may be performed first and then step S2, or step S2 may be performed first and then step S1, or step S1 and step S2 may be performed simultaneously.

[0037] S3, selecting a target feature and multiple non-target features corresponding to the modeling object from the multiple features.

[0038] Among them, the target feature refers to the feature corresponding to the modeling object among multiple features, and the non-target feature refers to the feature other than the target feature among multiple features.

[0039] It should be noted that the multiple features include the features of the integrated energy system and the features of each power supply device, and therefore the multiple features also include the corresponding features of the modeling object.

[0040] Specifically, after obtaining multiple features and determining the modeling object, a target feature corresponding to the modeling object is selected from the multiple features by utilizing the variable data granularity, wherein the modeling object may have one or more features. When there are multiple features, one is selected as the target feature for model training in the modeling process. After determining the target feature, the features other than the target feature among the multiple features are determined as non-target features.

[0041] S4, perform dimensionality reduction processing on multiple non-target features corresponding to each historical time.

[0042] Specifically, after obtaining the non-target features, any feasible method can be used to reduce the dimension of the non-target features for subsequent model training, thereby reducing the number of non-target feature dimensions and reducing the amount of calculation in the model development process.

[0043] S5, construct multiple different neural network models, and train each neural network model according to the time features corresponding to the multiple historical times, the target features corresponding to the multiple historical times, and the non-target features after dimensionality reduction.

[0044] Specifically, multiple different neural network models are constructed, and the differences between the models may be different machine learning algorithms, or different network structures and parameters, which can be determined according to actual needs. After multiple neural network models are constructed, the time features corresponding to the historical time and the non-target features after dimensionality reduction are used as inputs, and the target features are used as outputs. Each neural network model is trained to obtain multiple trained neural network models.

[0045] Among them, the neural network model can be an LSTM (Long Short-Term Memory) model.

[0046] S6, fusing multiple trained neural network models to obtain a target neural network model corresponding to the modeling object.

[0047] Specifically, due to the different principles of various power sources, some power sources such as photovoltaic and wind power are greatly affected by external factors such as weather. Therefore, in order to better reflect the actual situation of the object and distinguish it from other objects, multiple sets of machine learning algorithms are used for modeling and the final target neural network model is formed through the fusion of the output end. The function of the target neural network model is to output target features based on the input features.

[0048] The embodiment of the present invention adopts machine learning to perform reverse reconstruction modeling based on real output data, and adapts to different power types through the characteristics of variable data granularity. It is suitable for integrated energy systems and can model the entire integrated energy system as well as a single power source within the system. The traditional mechanism modeling method only models a certain type of power source, and lacks sufficient coverage for manufacturers of the same type and multiple sources. The embodiment of the present invention targets the output of the device rather than mathematical principles, and expands the coverage of the modeling work by subdividing the granularity of the training data. Compared with traditional technologies, it reduces manual labor while expanding coverage.

[0049] Therefore, the modeling method of the integrated energy system of the embodiment of the present invention uses historical time as alignment data, trains multiple neural network models according to multiple features under historical time, and fuses the trained models to obtain the required model. That is, the modeling of the integrated energy system and the power supply equipment in the system can be realized without mathematical principles, which can improve the modeling accuracy and reduce the modeling cost.

[0050] In one embodiment of the present invention, each feature has a corresponding feature category, and the feature category includes one of a power supply equipment feature category, an integrated energy system feature category, and a meteorological feature category of an area where the integrated energy system is located.

[0051] Since most of the input characteristics of new energy power stations are meteorological characteristics such as wind speed, irradiance, temperature, air pressure, humidity, etc., these characteristics themselves have large volatility and even randomness. In the process of using traditional mechanism-based modeling, it is easy to cause the model accuracy to be low due to incomplete consideration of influencing factors. In the scenario of integrated energy planning and simulation, since integrated energy is a local power system, the smaller system scale will magnify the difference between the model of a single device and the actual situation. At the same time, the complexity of the input characteristics of new energy power stations increases the difficulty of using mathematical principles for mechanism modeling, and the traditional fossil power sources in the integrated energy system further increase the differences in the power supply forms within the local system, which sets a high technical threshold for using simulation technology for integrated energy system planning.

[0052] To this end, each feature in the embodiment of the present invention has a corresponding feature category, and the feature category includes one of a power supply equipment feature category, an integrated energy system feature category, and a meteorological feature category of the area where the integrated energy system is located.

[0053] Among them, the features in the meteorological feature category include at least one of solar radiation, rainfall, relative humidity and wind speed; the features in the power supply equipment feature category include instrument measurement data and output of each power supply equipment; the features in the integrated energy system feature category include at least one of the power received from outside the system, the amount of power received from outside the system, the power transmitted from the system, the amount of power transmitted from the system and energy supply variables.

[0054] That is to say, the embodiment of the present invention uses time data as an alignment feature to integrate all data features of each power source in the integrated energy system. When the integrated energy system includes two photovoltaic power stations, two wind turbines, two energy storage power stations and two distributed gas turbines, multiple features are shown in Table 1:

[0055] Table 1 Multiple characteristics of integrated energy systems

[0056]

[0057] In one example, the above-mentioned step S3, i.e., selecting a target feature and multiple non-target features corresponding to the modeling object from multiple features, may include: when the modeling object is an integrated energy system, selecting a feature from the features included in the integrated energy system feature category as the target feature, and using the features other than the target feature in the multiple features corresponding to each historical time as non-target features; when the modeling object is a power supply device, selecting a feature from the features included in the power supply device feature category as the target feature, and using the features other than the target feature in the multiple features corresponding to each historical time as non-target features.

[0058] Furthermore, when the modeling object is a comprehensive energy system, the power transmitted by the system is taken as the target feature; when the modeling object is a power supply device, the output of the power supply device is taken as the target feature.

[0059] Specifically, when the modeling object is an integrated energy system, a feature is selected from the features included in the feature category of the integrated energy system as the target feature. The target feature can be selected according to the degree of concern for each feature in the integrated energy system. For example, power transmission can be selected as the target feature; when the modeling object is a power supply device such as a photovoltaic power station, the output of the photovoltaic power station can be selected as the target feature.

[0060] In one embodiment of the present invention, the above step S4, i.e., performing dimensionality reduction processing on the multiple non-target features corresponding to each historical time, may include: performing dimensionality reduction processing on the multiple non-target features corresponding to each historical time using a principal component analysis algorithm.

[0061] It can be understood that principal component analysis is a statistical method that converts a set of variables that may be correlated into a set of linearly uncorrelated variables through orthogonal transformation. The converted set of variables is called principal components.

[0062] Specifically, if Figure 2 As shown, the principal component analysis algorithm is used to reduce the dimension of n non-target features corresponding to each historical time to obtain m principal component features with high correlation with the target features, where m is less than n, and the principal component features are used as training data sets for model training.

[0063] Therefore, using principal component analysis as a prerequisite for modeling can fix the dimensionality of the input features in the model derivation process, merge features with high correlation with the target features, and automatically ignore features with low correlation, thereby reducing the amount of calculation in the model derivation process.

[0064] In one embodiment of the present invention, the structures and parameters of the multiple neural network models are different, and the method also includes: determining the attributes of the modeling object; determining the number of the multiple neural network models, the structure and parameters of each neural network model according to the attributes of the modeling object, wherein the number of the multiple neural network models is an odd number.

[0065] The attributes of the modeling object may include the complexity of the modeling object, the sensitivity of the modeling object to external factors, and the characteristics of the modeling object.

[0066] Specifically, the number of neural network models (which can be LSTM models), the number of network layers of each model, and the number of nodes in each layer can be adjusted according to the properties of the integrated energy system or power source. For more complex systematized objects such as the entire integrated energy system or power sources that are greatly affected by the outside world, such as photovoltaics and wind power, the accuracy of the model can be ensured by increasing the number of LSTM models, the number of neural network layers of a single LSTM model, and the number of nodes in each layer. For objects that are not sensitive to external factors, such as distributed gas turbines, fewer LSTM models, the number of neural network layers of a single LSTM model, and the number of nodes in each layer can be used for modeling.

[0067] After principal component analysis and construction of multiple (for example, three) LSTM models, the time features corresponding to multiple historical times, target features and m principal component features are used as training data sets to train each LSTM model. Afterwards, the three trained first LSTM models, second LSTM models and third LSTM models are fused to obtain the target LSTM model of the modeling object.

[0068] In one embodiment of the present invention, the modeling method of the integrated energy system, after obtaining the target neural network model of the modeling object, may also include: when the integrated energy system changes, determining the non-target features after the change; performing dimensionality reduction processing on the changed non-target features through a principal component analysis algorithm, and inputting the non-target features into the target neural network model to obtain new target features; comparing the target features with the new target features to obtain the degree of change of the target features; and judging the degree of influence of the changes in the integrated energy system on the modeling object according to the degree of change of the target features.

[0069] Specifically, since principal component analysis can reconstruct the input features in advance and ignore features with low correlation, after the integrated energy system is modeled, if the integrated energy system changes, such as adding or reducing power equipment, the non-target features will change. Therefore, to determine the non-target features after the change, the non-target features after the change can be input into the target LSTM model after dimensionality reduction through the principal component analysis algorithm, and then the target LSTM model outputs new target features, and the new target features are compared with the above target features to determine the degree of change of the target features after the system changes, and then the degree of influence of the changes in the integrated energy system on the modeling object is determined according to the degree of change. The greater the degree of change in the target degree, the greater the degree of influence of the changes in the integrated energy system on the modeling object; the smaller the degree of change in the target degree, the smaller the degree of influence of the changes in the integrated energy system on the modeling object.

[0070] After the integrated energy system is modeled, if power sources are added or reduced in the system, the non-target features after the changes can be reduced in dimension and input into the model. The output of the model can be used to determine the degree of change of the target features after the system changes. For example, after adding a distributed gas turbine power station, the output of the model can be used to determine the changes in the target features of the integrated energy system, that is, the power transmitted to the outside, to understand the impact of the changes in power sources on the entire integrated energy system. This is suitable for planning or simulation scenarios.

[0071] After the modeling of power supply equipment such as energy storage power stations is completed, if power points are added or reduced in the system, the non-target features after the change can be reduced in dimension and input into the model. The degree of change of the target features after the system changes can be judged based on the output of the model. For example, after adding a photovoltaic power station, the output of the model can be used to determine the change in the output of a certain energy storage power station in the system, and the impact of the change in the power point on the operating conditions of other power points can be understood. This is suitable for planning or simulation scenarios.

[0072] That is to say, Figure 3 As shown in the figure, when power supply equipment is added to the integrated energy system, it means that non-target features increase. Therefore, all non-target features (including the added non-target features and the original n non-target features) are reduced in dimension using the principal component analysis algorithm, and the same number of principal component features, i.e., m principal component features, are obtained. At this time, the principal component features may change. The m principal component features are input into the target neural network model (LSTM model), and then the target neural network model outputs new target features. The new target features are then compared with the original target features to obtain the degree of change of the new target features. If the degree of change is large, it means that the added power supply equipment has a greater impact on the modeling object; if the degree of change is small, it means that the added power supply equipment has a smaller impact on the modeling object; if the new target feature is the same as the original target feature, it means that the added power supply equipment has no impact on the modeling object.

[0073] The technical solution of the embodiment of the present invention uses time data as an alignment feature to collect data in the entire integrated energy system. The principal component analysis algorithm is used in advance to fix the number of features of the input model to ensure that when a new power point is added to the system, the normal operation of the model is not affected by the new features. At the same time, the principal component screening is completed for non-target features, and the features with high correlation with the target features are merged into a fixed number of principal component features. Multiple LSTM models are fused at the output end to form a fusion model.

[0074] In summary, the embodiments of the present invention have the following advantages:

[0075] 1. Use real output data for reverse reconstruction modeling, and adapt to different power types through the characteristics of variable data granularity. It is suitable for integrated energy systems, and can model the entire integrated energy system or a single power source within the system;

[0076] 2. The traditional mechanism modeling method only models a certain type of power supply, and lacks sufficient coverage for manufacturers of the same type and multiple sources. The advantage of the embodiment of the present invention is that it targets the output of the device rather than the mathematical principle, and expands the coverage of the modeling work by subdividing the granularity of the training data;

[0077] 3. Traditional technology requires a lot of labor to independently model a single device in an integrated energy system. The advantage of the embodiment of the present invention is that it uses machine learning to reverse model based on real data, which reduces manual labor while expanding coverage;

[0078] 4. Traditional mechanism modeling has high requirements for mathematics and requires technicians to understand the principles of the equipment, which makes the technical threshold high and the capital investment large when it is used for planning or simulation. The advantage of the embodiment of the present invention is that it uses machine learning technology, which reduces the requirements for mathematics and the threshold for planning or simulation;

[0079] 5. Once the traditional mechanism modeling is completed, it cannot be iterated. However, the reverse modeling of the embodiment of the present invention can be iterated according to the changes in the integrated energy system. The model can be retrained to approach the real modeling object during the iteration process.

[0080] Corresponding to the modeling method of the integrated energy system of the above embodiment, the present invention further proposes a modeling device of the integrated energy system.

[0081] Figure 4 Schematic diagram of a block diagram of a modeling device for an integrated energy system according to an embodiment of the present invention.

[0082] The integrated energy system of the embodiment of the present invention includes a plurality of power supply devices.

[0083] like Figure 4 As shown, the modeling device 100 of the integrated energy system includes: an acquisition module 110, a determination module 120, a selection module 130, a dimensionality reduction module 140, a training module 150 and a fusion module 150.

[0084] Among them, the acquisition module 110 is used to determine multiple historical times and obtain multiple features of the integrated energy system at each historical time; the determination module 120 is used to determine the modeling object corresponding to the integrated energy system, wherein the modeling object is the integrated energy system or one of the power supply devices; the selection module 130 is used to select a target feature and multiple non-target features corresponding to the modeling object from the multiple features; the dimension reduction module 140 is used to reduce the dimension of the multiple non-target features corresponding to each historical time; the training module 150 is used to construct multiple different neural network models, and train each neural network model according to the time features corresponding to the multiple historical times, the target features corresponding to the multiple historical times and the non-target features after dimension reduction; the fusion module 160 is used to fuse multiple trained neural network models to obtain the target neural network model corresponding to the modeling object.

[0085] It should be noted that the specific implementation of the modeling device of the integrated energy system can refer to the specific implementation of the modeling method of the integrated energy system mentioned above. In order to avoid redundancy, it will not be described in detail here.

[0086] The modeling device of the integrated energy system in the embodiment of the present invention uses historical time as alignment data, trains multiple neural network models according to multiple features under historical time, and fuses the trained models to obtain the required model. That is, the modeling of the integrated energy system and the power supply equipment in the system can be achieved without mathematical principles, which can improve the modeling accuracy and reduce the modeling cost.

[0087] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0088] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" 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 representations of the above terms do not necessarily refer to the same embodiment or example. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be construed as limitations on the present invention, and those of ordinary skill in the art can change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A modeling method for an integrated energy system, wherein the integrated energy system includes a plurality of power supply devices, characterized in that: The method comprises the following steps: Determine a plurality of historical times, and obtain a plurality of characteristics of the integrated energy system at each of the historical times; Determining a modeling object corresponding to the integrated energy system, wherein the modeling object is the integrated energy system or one of the power supply devices; Selecting a target feature and a plurality of non-target features corresponding to the modeling object from the plurality of features; Performing dimensionality reduction processing on the plurality of non-target features corresponding to each of the historical times; Constructing a plurality of different neural network models, and training each of the neural network models according to the time features corresponding to the plurality of historical times, the target features corresponding to the plurality of historical times, and the non-target features after dimensionality reduction; The multiple trained neural network models are fused to obtain a target neural network model corresponding to the modeling object. After obtaining the target neural network model of the modeling object, the method further includes: When the integrated energy system changes, determining non-target characteristics after the change; The changed non-target features are subjected to dimensionality reduction processing by a principal component analysis algorithm and then input into the target neural network model to obtain new target features; Comparing the target feature with the new target feature to obtain a change degree of the target feature; The degree of influence of the change in the integrated energy system on the modeling object is determined according to the degree of change in the target feature.

2. The modeling method of the integrated energy system according to claim 1, characterized in that: Each of the features has a corresponding feature category, and the feature category includes one of a power supply equipment feature category, an integrated energy system feature category, and a meteorological feature category of an area where the integrated energy system is located.

3. The modeling method of the integrated energy system according to claim 2, characterized in that: Selecting a target feature and a plurality of non-target features corresponding to the modeling object from the plurality of features includes: When the modeling object is the integrated energy system, a feature is selected from the features included in the integrated energy system feature category as the target feature, and features other than the target feature among the multiple features corresponding to each of the historical times are used as the non-target features; When the modeling object is the power supply device, one feature is selected from the features included in the power supply device feature category as the target feature, and features other than the target feature among the multiple features corresponding to each historical time are used as the non-target features.

4. The modeling method of the integrated energy system according to claim 1, characterized in that: Performing dimensionality reduction processing on the plurality of non-target features corresponding to each of the historical times includes: The principal component analysis algorithm is used to perform dimensionality reduction processing on the multiple non-target features corresponding to each historical time.

5. The modeling method of the integrated energy system according to claim 1, characterized in that: The structures and parameters of the plurality of neural network models are different, and the method further comprises: determining properties of the modeled object; The number of the plurality of neural network models, the structure and parameters of each of the neural network models are determined according to the properties of the modeling object, wherein the number of the plurality of neural network models is an odd number.

6. The modeling method of the integrated energy system according to claim 2, characterized in that: The features in the meteorological feature category include at least one of solar radiation, rainfall, relative humidity and wind speed; the features in the power supply equipment feature category include instrument measurement data and output of each of the power supply equipment; the features in the integrated energy system feature category include at least one of external power received by the system, external power received by the system, external power transmitted by the system, external power transmitted by the system and energy supply variables.

7. The modeling method of the integrated energy system according to claim 6, characterized in that: When the modeling object is the integrated energy system, the power transmitted by the system is used as the target feature; When the modeling object is the power supply device, the output of the power supply device is used as the target feature.

8. The modeling method of the integrated energy system according to claim 1, characterized in that: The multiple power supply devices include one or more of photovoltaic power stations, wind turbines, distributed gas turbines and energy storage power stations.

9. A modeling device for an integrated energy system using the modeling method for an integrated energy system according to claim 1, wherein the integrated energy system comprises a plurality of power supply devices, characterized in that: The device comprises: An acquisition module, used for determining a plurality of historical times, and acquiring a plurality of characteristics of the integrated energy system at each of the historical times; A determination module, used to determine a modeling object corresponding to the integrated energy system, wherein the modeling object is the integrated energy system or one of the power supply devices; A selection module, used for selecting a target feature and a plurality of non-target features corresponding to the modeling object from the plurality of features; A dimension reduction module, used for performing dimension reduction processing on the plurality of non-target features corresponding to each of the historical times; A training module, used to construct a plurality of different neural network models, and train each of the neural network models according to the time features corresponding to the plurality of historical times, the target features corresponding to the plurality of historical times, and the non-target features after dimensionality reduction; The fusion module is used to fuse multiple trained neural network models to obtain a target neural network model corresponding to the modeling object.

Images