CN115149529A - Wave energy power generation power prediction method and system - Google Patents
Wave energy power generation power prediction method and system Download PDFInfo
- Publication number
- CN115149529A CN115149529A CN202211075400.XA CN202211075400A CN115149529A CN 115149529 A CN115149529 A CN 115149529A CN 202211075400 A CN202211075400 A CN 202211075400A CN 115149529 A CN115149529 A CN 115149529A
- Authority
- CN
- China
- Prior art keywords
- data
- time
- prediction
- time period
- long
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/381—Dispersed generators
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wave energy power generation power prediction method and system, and relates to the field of power generation power prediction, wherein the method comprises the following steps: acquiring parameter data of a wave energy power generation device at a current water depth within a current time period; parameter data wave height, wave direction, prediction time period and power generation power; the current time interval is [ a, t ], wherein t is the current moment, and a is less than t; the prediction time period is [ t +1, b ], wherein t +1 represents the next moment of the current moment, and b is more than or equal to t +1; according to the parameter data and the power prediction model at the current water depth in the current time period, obtaining the generated power of the wave energy power generation device at the current water depth within the prediction time period; the power prediction model is constructed based on a long-time memory network model. The invention can realize accurate prediction of wave energy generating power.
Description
Technical Field
The invention relates to the field of power generation power prediction, in particular to a wave energy power generation power prediction method and system.
Background
With the continuous increase of human demand for energy and the continuous increase of attention on global climate change in recent years, the search for alternative energy and the development of renewable energy become important supports for the sustainable development of the human society. Ocean renewable energy (OE for short) comprising wave energy, tidal current and ocean current energy, temperature difference energy and salt difference energy is used as inexhaustible clean energy and has huge global reserves and potential development values.
In recent years, the installed scale of renewable energy is continuously enlarged, the proportion of the power generation in the power grid is also continuously increased, and simultaneously, new challenges are generated for the power grid:
firstly, because the renewable energy is low in hours, the electric quantity of the renewable energy in a high proportion needs the installed capacity of the new energy with multiple loads, which brings a series of changes to the planning design, production management and operation control of the system.
Second, because the power of renewable energy fluctuates greatly in the day, the conventional power regulation capability is difficult to cope with the fluctuation of the power of renewable energy, and the consumption of renewable energy faces a great challenge.
Thirdly, the intermittent and random fluctuation of renewable energy causes the output period to be indefinite, the energy storage requirement is increased under a high output for a long time, and the power balance is required to be realized together with other power supplies under a low output, which brings huge challenges to system consumption, safety and energy storage technology. Meanwhile, the peak output power of the renewable energy is large, the electric quantity is small, and huge flexible resources are required to be mobilized for ensuring peak output consumption. The power prediction technology of ocean energy is based on electric field environment data and historical data, and utilizes physical models, scientific statistics and other methods to directly or indirectly predict the short-term and even long-term power generation power output of an electric field, thereby providing bases for power consumption, scheduling, peak regulation and the like. Therefore, reliable prediction of the power generation power of the ocean energy device plays an important role in improving the reliability and stability of the offshore operation of equipment, reducing the operation and maintenance cost and realizing the commercial operation more quickly.
The wave energy industry is still in the development stage in the world, the wave energy power generation device is used for generating power, the prediction of the power generation power of the wave energy power generation device is not mature, and renewable energy power generation prediction modes such as wind power and solar energy are mainly referred and transplanted. Waves and wind have similar physical characteristics, but the environment of the wave power generation device is simultaneously influenced by elements such as weather, hydrology, landform and even temperature and salinity, and has quite complex mutual connection and influence. With the development of prediction technology, the short-term power prediction method of renewable energy gradually changes from a physical model method to a statistical model.
Currently, there are two main types of methods to predict the wave power: physical model methods and data driven methods. The theory of physical models was created in the late 50 s of the 20 th century and was applied by everyone from the 60 s of the 20 th century. The principle of the physical model method is that local modeling is carried out according to the conditions of sea water depth, topography and the like, so that wave weather forecast of a sea area where a wave power generation field is located is obtained, then the wave forecast of the area is converted into the wave energy density of a section position where a prediction unit is located, and finally the power prediction value of the generator set is obtained by combining a power curve (or a power matrix) of the unit. Through application for many years, the storm mode is mature and business operation is realized. Such as the wave forecasting system of the National Oceanic and Atmospheric Administration (NOAA), have great application based on the WAVEWATCH III model, the ECMWF model, the SWAN model and the like. The statistical model has 2 ideas, a data model is often used in the early stage to express the functional relationship between wave forecasting parameters (wave height and wave direction) and the power generation power, and then the power of waves is predicted according to future wave forecasting values. Another is an artificial intelligence technology developed in recent years, and it is considered that various weather factors are already implied in the historical generated power data of the wave electric field, so the historical data is learned and extrapolated by using a mathematical model, and a predicted value of the wave power is obtained. The linear models widely used are an autoregressive moving average (ARMA) method, a Box-Jenkins method, a kalman filter, a markov chain model, and the like. Artificial Neural Networks (ANN) and Support Vector Machines (SVM) are the two most popular non-linear methods for renewable energy prediction.
The physical model has the characteristics that historical data support is not needed, the method is suitable for the wave energy power generation device which is just installed and operated, the prediction can be simultaneously carried out on a single device and a plurality of arrays, and the method has the obvious defects: (1) numerical weather forecast is based on a complex mathematical model, the modeling processes of a wind field and a flow field are complex, the effect is difficult to say ideal, the forecast computation amount is extremely remarkable, the precision requirement is higher and higher, and the real-time performance of forecast is difficult to guarantee; (2) background environmental information such as water depth, terrain, ocean current, tide, thermohaline and the like of the ocean are complex in influence on a prediction result, and collection and description of the information are difficult; (3) the prediction process of the physical model is long, and errors are easy to accumulate in the prediction process.
The statistical model method is based on accurate historical data rule mining, most of the statistical model method can only obtain ideal effect within a very short time, and error accumulation will increase continuously along with the increase of time. The statistical model is suitable for ultra-short term and short term (within 3 days), because most of power prediction mined based on the historical data rule can only obtain ideal effect in very short time, the data demand is more dependent on electric field real-time monitoring data (hydrology, meteorological data, power data and the like), and the relation with the numerical prediction model is not close. On a long time scale, most meteorological hydrological elements which obviously affect the active power of wave power generation have intermittent variation characteristics, the physical constraints of the variation characteristics are complex, and most of the physical constraints are very typical nonlinear relations, and the characteristics cause the increase of error accumulation.
Disclosure of Invention
Based on the above, the embodiment of the invention provides a wave energy power generation power prediction method and system, so as to accurately predict wave energy power generation power.
In order to achieve the purpose, the invention provides the following scheme:
a wave energy power generation power prediction method comprises the following steps:
acquiring parameter data of a wave energy power generation device at a current water depth within a current time period; the parameter data comprises wave height, wave direction, a prediction time period and generated power; the current time interval is represented as [ a, t ], wherein t is the current time, and a is less than t; the prediction time period is [ t +1, b ], wherein t +1 represents the next moment of the current moment, and b is more than or equal to t +1;
according to parameter data and a power prediction model of the current water depth in the current time period, obtaining the generated power of the wave energy power generation device in the current water depth in the prediction time period; the power prediction model is constructed based on a long-time memory network model.
Optionally, the determination method of the power prediction model comprises the following steps:
obtaining sea test data; the sea trial data comprises a plurality of data sets; the data set includes at least one pair of sample data; the sample data comprises parameter sample data and corresponding label data; any parameter sample data comprises parameter data of a prototype of the wave energy power generation device in a water depth during a test; the tag data comprises a time period and corresponding generated power; the time periods corresponding to the label data in different data sets are different; the time period corresponding to the generated power in the parameter sample data is different from the time period corresponding to the generated power in the tag data; the ending time of the time period corresponding to the generated power in the parameter sample data is continuous with the starting time of the time period corresponding to the generated power in the label data;
establishing a long-time and short-time memory network model; the long-time memory network model comprises an input gate, a forgetting gate and an output gate which are connected in sequence;
and taking parameter sample data in the sea test data as input of the long-time and short-time memory network model, taking corresponding label data as output of the long-time and short-time memory network model for training, and determining the trained long-time and short-time memory network model as the power prediction model.
Optionally, the obtaining of the generated power at the current water depth within the wave energy power generation device prediction time period according to the parameter data and the power prediction model at the current water depth within the current time period specifically includes:
preprocessing the parameter data of the current water depth in the current time period to obtain preprocessed parameter data of the current water depth in the current time period; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
and inputting the parameter data preprocessed at the current water depth in the current time period into a power prediction model to obtain the generated power at the current water depth in the prediction time period of the wave energy power generation device.
Optionally, the training, with the parameter sample data in the marine test data as the input of the long-time and short-time memory network model, with the corresponding label data as the output of the long-time and short-time memory network model, and determining the trained long-time and short-time memory network model as the power prediction model specifically includes:
preprocessing each data set in the sea test data to obtain a preprocessed data set; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
dividing the preprocessed data set into a training data group, a proofreading data group and a prediction data group according to a set proportion;
taking parameter sample data in the training data set as input of the long-time and short-time memory network model, taking label data in the training data set as output of the long-time and short-time memory network model, and performing primary training to obtain an initial long-time and short-time memory network model;
adjusting model parameters in the initial long-time and short-time memory network model by adopting the proofreading data group to obtain an adjusted long-time and short-time memory network model;
and testing the adjusted long-time and short-time memory network model by adopting the prediction data set to obtain the power prediction model.
The invention also provides a wave energy power generation power prediction system, which comprises:
the data acquisition module is used for acquiring parameter data of the wave energy power generation device in the current water depth in the current time period; the parameter data comprises wave height, wave direction, a prediction time period and generated power; the current time interval is represented as [ a, t ], wherein t is the current time, and a is less than t; the prediction time period is [ t +1, b ], wherein t +1 represents the next moment of the current moment, and b is more than or equal to t +1;
the generating power prediction module is used for obtaining the generating power at the current water depth in the wave energy generating set prediction time period according to the parameter data and the power prediction model at the current water depth in the current time period; the power prediction model is constructed based on a long-time memory network model.
Optionally, the wave energy generated power prediction system further includes: a prediction model determination module for determining the power prediction model;
the prediction model determining module specifically includes:
the marine test data acquisition unit is used for acquiring marine test data; the sea trial data comprises a plurality of data sets; the data set includes at least one pair of sample data; any sample data comprises parameter sample data and corresponding label data; the parameter sample data comprises parameter data of a prototype of the wave energy power generation device in a water depth during a test; the tag data comprises a time period and corresponding generated power; the time periods corresponding to the label data in different data sets are different; the time period corresponding to the generated power in the parameter sample data is different from the time period corresponding to the generated power in the tag data; the ending time of the time period corresponding to the generated power in the parameter sample data is continuous with the starting time of the time period corresponding to the generated power in the label data;
the model establishing unit is used for establishing a long-time memory network model; the long-time memory network model comprises an input gate, a forgetting gate and an output gate which are connected in sequence;
and the model training unit is used for taking parameter sample data in the sea test data as the input of the long-time and short-time memory network model, taking corresponding label data as the output of the long-time and short-time memory network model for training, and determining the trained long-time and short-time memory network model as the power prediction model.
Optionally, the generated power prediction module specifically includes:
the parameter data preprocessing unit is used for preprocessing parameter data of the current water depth in the current time period to obtain preprocessed parameter data of the current water depth in the current time period; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
and the prediction unit is used for inputting the parameter data preprocessed at the current water depth in the current time period into the power prediction model to obtain the generated power at the current water depth in the prediction time period of the wave energy power generation device.
Optionally, the model training unit specifically includes:
the sea test data preprocessing subunit is used for preprocessing each data set in the sea test data to obtain a preprocessed data set; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
the data group dividing subunit is used for dividing the preprocessed data set into a training data group, a proofreading data group and a prediction data group according to a set proportion;
an initial training subunit, configured to use parameter sample data in the training data set as input of the long-and-short term memory network model, and use label data in the training data set as output of the long-and-short term memory network model to perform initial training, so as to obtain an initial long-and-short term memory network model;
the proofreading subunit is used for adjusting model parameters in the initial long-time and short-time memory network model by adopting the proofreading data set to obtain an adjusted long-time and short-time memory network model;
and the testing subunit is used for testing the adjusted long-time and short-time memory network model by adopting the prediction data set to obtain the power prediction model.
Compared with the prior art, the invention has the beneficial effects that:
the embodiment of the invention provides a wave energy generating power prediction method and system, which can be used for predicting the generating power in the current water depth in the prediction time period of a wave energy generating device by adopting a power prediction model constructed on the basis of a long-time memory network (LSTM) model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flow chart of a wave energy power generation power prediction method provided by an embodiment of the invention;
fig. 2 is a network structure diagram of an RNN according to an embodiment of the present invention;
FIG. 3 is a network architecture diagram of an LSTM provided by an embodiment of the present invention;
FIG. 4 is a data flow diagram of an input gate portion of an LSTM provided by an embodiment of the present invention;
FIG. 5 is a data flow diagram of an output gate portion of an LSTM provided by an embodiment of the present invention;
FIG. 6 is a data flow diagram of a forgetting gate portion of an LSTM provided by an embodiment of the present invention;
fig. 7 is a diagram of a specific implementation process of the wave energy power generation power prediction method provided by the embodiment of the invention;
fig. 8 is a structural diagram of a wave energy generated power prediction system provided in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof.
In order to improve the accuracy of wave energy power generation power prediction, the embodiment provides a prediction method based on a deep learning model. The deep learning method is used as a branch of a statistical model method, obtains an effect exceeding the expectation by virtue of strong calculation and feature extraction capabilities, and is gradually becoming a research hotspot in the field of power prediction. Deep learning models such as a Recurrent Neural Network (RNN), a Convolutional Neural Network (CNN), a long-term memory network (LSTM) and the like are applied in a large quantity, and deep learning technologies are applied to Google translation, an apple voice tool siri, a Cortana personal voice assistant of Microsoft and the like.
The method realizes the prediction of the short-term power generation power of the wave energy device based on the long-term and short-term memory network model, fully utilizes the time sequence and space multi-source information reconstruction technology of the long-term and short-term memory network model, and obtains the characteristics and the correlation of a power signal in time and space scales, thereby discussing the high-precision prediction method suitable for the short-term power generation power of the wave energy device. Referring to fig. 1, the method for predicting wave energy generated power of the present embodiment includes:
step 101: acquiring parameter data of a wave energy power generation device at a current water depth within a current time period; the parameter data comprises wave height, wave direction, a prediction time period and power generation power; the current time interval is represented as [ a, t ], wherein t is the current time, and a is less than t; the prediction time period is [ t +1, b ], wherein t +1 represents the next moment of the current moment, and b is more than or equal to t +1.
Step 102: according to parameter data and a power prediction model of the current water depth in the current time period, obtaining the generated power of the wave energy power generation device in the current water depth in the prediction time period; the power prediction model is constructed based on a long-time memory network model.
The determination method of the power prediction model comprises the following steps:
1) Obtaining sea test data; the sea trial data comprises a plurality of data sets; the data set includes at least one pair of sample data; the sample data comprises parameter sample data and corresponding label data; any parameter sample data comprises parameter data of a prototype of the wave energy power generation device in a water depth during a test; the tag data includes a time period and a corresponding generated power; the time periods corresponding to the label data in different data sets are different; the time period corresponding to the generated power in the parameter sample data is different from the time period corresponding to the generated power in the tag data; generated power in the parametric sample data the end time of the corresponding time period and the start time of the time period corresponding to the generated power in the tag data is continuous.
2) Establishing a long-time and short-time memory network model; the long-time memory network model comprises an input gate, a forgetting gate and an output gate which are connected in sequence. Specifically, the method comprises the following steps:
the long-term memory network (LSTM) model is a special Recurrent Neural Network (RNN) with the capability of learning long-term memory, which adds a memory function (called Cells) on the basis of the RNN, the Cells determine whether the memory of the information should be written or deleted in the neuron, and the long-term history information can be recorded by combining the previous state, the current memory and the currently input information. The core of the LSTM is a cell state which is similar to a conveyor belt and directly runs on the whole chain, only a small amount of linear interaction exists, and the stability of information flowing on the cell state is guaranteed.
The LSTM is formed by adding gates (gates) on the basis of a Recurrent Neural Network (RNN) to control whether signals can pass or not, wherein the gates are in a mode of allowing information to selectively pass, and three gates are forgetting gates (forget gates) respectively and determine the last momenttCell state of-1C t-1 How much to keep to the current timetCell state of (2)C t (ii) a The other is an input gate (input gate) which determines the input of the network at the present momentX t How much to save the state of the cell to the current timeC t . LSTM uses output gates to control cell statesC t How much current output value to LSTMH t . Fig. 2 shows a network structure of RNN, fig. 3 shows a network structure of LSTM, and fig. 4 to 6 detail data flow diagrams of three gates of LSTM.
As shown in FIG. 2, the elements of the Recurrent Neural Network (RNN)A t Its input value isX t Output ofValue isH t . Information is passed from the current state of the network to the next state through the loop. WhereinA t The output of the hidden layer is two: one as an output value and the other to the neuron at the next time.A t According to the last momenttHidden layer state of-1A t-1 And the current timetIs inputtedX t The result of the calculation is that,A t+1 indicating the next moment of timet+1 hidden layer state. In fig. 3, LSTM connects neurons through self-circulation, so that the output of sequence information can be kept independent over time by the memory unit, and is not influenced by the input and output environments. The input gate is used for determining which newly input information is allowed to be updated or stored in the memory unit; the output gate is opposite to the input gate and is used for determining which information in the memory unit is allowed to be output; the forgetting door is used for controlling whether the memory unit remembers or discards the previous state.
The dotted line in FIG. 4 is the input gate portion of LSTM for judging the input data X t Which new information should be added to the last time instanttCell state of-1C t-1 In (1). Where σ denotes an activation function sigmoid. Input ofH t-1 AndX t the new candidate state value can be obtained through the tanh activation functionC’ t However, these new information are not all useful and therefore need to be usedH t-1 AndX t obtaining an output control value of an input gate through a sigmoid functioni t ， i t Indicating what new information is useful. The result of multiplication of two vectors is added toC t-1 In the method, the unit state at the current time t is obtainedC t . The formula is as follows:
in the formula (I), the compound is shown in the specification,W i for inputting control time steps at the input gatetA weight vector of the input sequence data of (a);U i for inputting control time steps at the input gatet-1 a weight vector of input state values;X t indicating the current time of daytThe input of (1);H t-1 indicating the last moment of timet-an output value of 1;b i inputting a controlled bias for the input gate;W c at time step for input gate state candidatetThe weight vector of the input sequence data of (a),U c time step candidates for entry gate statest-1 a weight vector of input state values;b c is the input gate state candidate.
The dotted line in FIG. 5 is the output gate portion of the LSTM for determining which information should be output toH t In (1). Cell state at the present time tC t Obtaining information which can be output through tanh function, and thenH t-1 AndX t obtaining a vector through sigmoid functionO t ，O t Indicating which locations' outputs should be dropped and which ones should be retained. The result of multiplying the two vectors is the final oneH t . The formula is as follows:
in the formula (I), the compound is shown in the specification,H t as the current timetOutput value of (1), output activation value of output gate and current timetThe Tanh function of the unit state is multiplied to finally obtain the output stateH t 。W o Is a candidate for the output gate stateTime of daytThe weight vector of the output sequence data of (a),U o for output gate state candidates at the previous momentt-a weight vector of output state values of 1;b o is the output gate state candidate.
The dotted line in FIG. 6 is the left behind gate portion of LSTM for determining the previous timetCell state of-1C t-1 Which information should be deleted. Input ofH t-1 AndX t obtaining the activation value of the forgetting gate after the sigmoid activation functionf t ，f t The closer to 1 the value of (A) represents the last momenttCell state of-1C t-1 The value of the corresponding position should be remembered;f t the closer to 0, the higher the value of (A) represents the previous timetCell state of-1C t-1 The value of the corresponding position in (a) should be forgotten. Will be provided withf t AndC t-1 the current time after the useless information is forgotten can be obtained by bit-wise multiplicationtCell state ofC t . The formula is as follows:
in the formula, the cell state at the current time tC t The state of the cell at the last time t-1C t-1 And activation value of forgetting gatef t Multiplying as a decision to discard the amount of information in the old state; input gate output control valuei t And candidate state valuesC’ t Multiplication for controlling the change of the new state; and finally adding the two to obtain a new state value.W f Candidate for forgotten door state at the current momenttThe weight vector of the input sequence data of (a),U f to forget the door stateAt the last momentt-a weight vector of forget state values of 1;b f a bias that is a forget gate state candidate.
3) And taking parameter sample data in the sea test data as input of the long-time and short-time memory network model, taking corresponding label data as output of the long-time and short-time memory network model for training, and determining the trained long-time and short-time memory network model as the power prediction model. Specifically, the method comprises the following steps:
preprocessing each data set in the sea test data to obtain a preprocessed data set; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence.
And dividing the preprocessed data set into a training data group, a proofreading data group and a prediction data group according to a set proportion.
And taking parameter sample data in the training data set as the input of the long-short time memory network model, and taking label data in the training data set as the output of the long-short time memory network model to perform primary training to obtain an initial long-short time memory network model.
And adjusting model parameters in the initial long-time and short-time memory network model by adopting the proofreading data group to obtain an adjusted long-time and short-time memory network model.
And testing the adjusted long-time and short-time memory network model by adopting the prediction data set to obtain the power prediction model.
Wherein, step 102 specifically includes:
preprocessing the parameter data of the current water depth in the current time period to obtain preprocessed parameter data of the current water depth in the current time period; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence.
And inputting the parameter data preprocessed in the current water depth in the current time period into a power prediction model to obtain the generated power of the wave energy power generation device in the current water depth in the prediction time period.
The wave energy power generation power prediction method of the embodiment has the following advantages:
(1) When the marine environment problem is processed, the traditional statistical model is difficult to process a large number of marine environment variables such as waves, currents, wind and the like at the same time, and complex and variable environmental factors are taken into consideration, so that an accurate and credible prediction result is difficult to give; the physical model has complex modeling processes for ocean wind fields, flow fields, wave fields and the like, the computation amount is extremely remarkable, the effect is difficult to be ideal, and the real-time performance of forecasting is difficult to be ensured. The wave energy power generation power prediction method based on the LSTM can take a large amount of time and space related environment variables into consideration, not only solves the problem of complex variables, but also overcomes the problem of poor effect of a single physical model.
(2) In principle, the advantages and disadvantages of the prediction model mainly lie in how to effectively eliminate factors which influence too complicated mechanism (tidal/sea current field, warm salt field, terrain and the like) and have unsatisfactory micro prediction effect (cold tide, red tide and the like), reduce modeling interference and error amplification, and are difficult to realize by adopting a traditional method. Through the mutual correlation of the input of a research model in a time sequence and a space dimension, the influence weight and the bias of different factors are researched, so that the influence degree of different types of data on a prediction result is researched.
In practical application, a specific implementation process of the wave energy power generation power prediction method of the embodiment is as follows:
step 1: the wave energy power generation device generally comprises an energy absorption device, an energy conversion device and an energy transmission device. Firstly, sea test data of a prototype of the wave energy power generation device is used as input, and the method mainly comprises the following steps: and during the test, the model machine of the wave energy power generation device has data such as wave height, wave direction, prediction time period, power generation power and the like within a set water depth range, wherein the set water depth range is 1-3 times of the water depth value of the distribution position where the front end of the model machine of the wave energy power generation device is located.
Step 2: preprocessing the sea test data, which mainly comprises the steps of removing noise and abnormal values, filling blank values, transforming data, merging data, normalizing the data and the like.
And step 3: and (3) dividing the data in the step (2) into a training data set, a proofreading data set and a prediction data set, wherein the data proportion is respectively 80%,15% and 5%.
And 4, step 4: and constructing a long-time and short-time memory network model, and training the long-time and short-time memory network model by adopting a training data set, a proofreading data set and a prediction data set to obtain a power prediction model.
And 5: the parameter data of the wave energy power generation device in the current water depth acquired in real time in the current time period are input into a power prediction model, and the generated power in the current water depth in the prediction time period is calculated through the power prediction model. The prediction time period may be t +1, [ t +1, n ] and [ t +1, t + n + k ]. When the prediction time period is t +1, the generated power prediction value at the next moment can be obtained by realizing one-step prediction; when the prediction time period is [ t +1, n ], n-step prediction is realized, which can also be called ultra-short-term prediction; when the prediction time period is [ t +1, t + n + k ], the n + k step prediction is realized, which can also be called as short-term prediction.
Based on the above steps 1 to 5, a specific flow of the wave energy generated power prediction method of the present embodiment is shown in fig. 7. Referring to fig. 7, firstly, sea test data is acquired, then data preparation is performed, parameter sample data in the sea test data is used as input of the long-time and short-time memory network model, corresponding label data is used as output, and the long-time and short-time memory network (LSTM) model is trained to obtain a power prediction model. When the historical values are input by the power prediction model, corresponding predicted values can be obtained, wherein x (1) represents a matrix consisting of wave height and wave direction at the 1 st moment, x (2) represents a matrix consisting of wave height and wave direction at the 2 nd moment, x (t) represents a matrix consisting of wave height and wave direction at the t th moment, y (1) represents the generated power at the 1 st moment, y (2) represents the generated power at the 1 st moment, and y (t) represents the generated power at the t th moment. As can be seen from fig. 7, in this embodiment, the first-step prediction may be implemented to obtain a one-step prediction value y (t + 1), n-step prediction may be implemented to obtain n-step prediction values y (t + 1),....,. Y (t + n-1), and y (t + n), and n + k-step prediction values y (t + 1),..,. Y (t + n + k-1), and y (t + n + k) may also be implemented to obtain n + k-step prediction values y (t + 1). The n-step prediction may be a 15 minute-4 hour ultra-short term prediction, and the n + k-step prediction may be a 0-4 hour short term prediction. The time window set by the prediction can be selected according to the model expression.
Compared with machine learning algorithms such as an artificial Neural Network (NN) and a Support Vector Machine (SVM), the validity of the embodiment is verified, specifically:
compared with machine learning algorithms such as an artificial Neural Network (NN) and a Support Vector Machine (SVM), a prediction result curve of the SVM can roughly reflect the trend of a measured value, but the deviation from the measured value is more, and the model effect is poor. The NN-based model result can better follow the actual measurement trend, and the deviation is obviously reduced. The results of LSTM show that the predicted results have a small deviation from the actual values, and particularly, the predicted curve in the short term substantially matches the actual values, indicating that the predicted results of LSTM have high accuracy in the short term. This is mainly due to the addition of memory to the LSTM for time series signals. LSTM is a special Recurrent Neural Network (RNN) with the ability to learn long-term memory, which adds a "memory" function (called "Cells") to RNN, which determine whether the memory of information should be written or deleted inside neurons, and can combine the previous state, the current memory and the currently input information to record long-term history information. Due to the characteristics, the LSTM is more and more applied to processing the prediction problem of time series, and the prediction effect is greatly improved.
The method is characterized in that a super-short-term and short-term prediction test is carried out on the generated power of the oscillating float type wave power generation device based on an LSTM model, and if appropriate model parameters are selected, a good result can be obtained by adopting the wave power prediction of a statistical method. In the aspect of model accuracy, the LSTM model has better performance in time series wave energy power generation device power prediction than other models. Because the statistical method establishes a mapping relation similar to a black box between input data and output data by analyzing the internal relation and the change rule characteristics of the data without comprehending the physical significance of each type of process variable, the wave energy power generation device power prediction method based on the statistical method is more suitable for carrying out short-term and ultra-short-term prediction, and for medium-term and long-term prediction, the method based on a physical model can generally obtain higher precision.
The invention also provides a wave energy power generation power prediction system, referring to fig. 8, the system comprises:
the data acquisition module 201 is used for acquiring parameter data of the wave energy power generation device at the current water depth in the current time period; the parameter data comprises wave height, wave direction, a prediction time period and generated power; the current time interval is represented as [ a, t ], wherein t is the current time, and a is less than t; the prediction time period is [ t +1, b ], wherein t +1 represents the next moment of the current moment, and b is more than or equal to t +1.
The generated power prediction module 202 is configured to obtain the generated power of the wave energy power generation device at the current water depth within the prediction time period according to the parameter data and the power prediction model at the current water depth within the current time period; the power prediction model is constructed based on a long-time memory network model.
In one example, the wave energy generated power prediction system further comprises: a prediction model determination module for determining the power prediction model.
The prediction model determining module specifically includes:
the marine test data acquisition unit is used for acquiring marine test data; the sea trial data comprises a plurality of data sets; the data set includes at least one pair of sample data; any sample data comprises parameter sample data and corresponding label data; the parameter sample data comprises parameter data of a prototype of the wave energy power generation device in a water depth during a test; the tag data comprises a time period and corresponding generated power; the time periods corresponding to the label data in different data sets are different; the time period corresponding to the generated power in the parameter sample data is different from the time period corresponding to the generated power in the tag data; the end time of the time period corresponding to the generated power in the parameter sample data is continuous with the start time of the time period corresponding to the generated power in the tag data.
The model establishing unit is used for establishing a long-time memory network model; the long-time memory network model comprises an input gate, a forgetting gate and an output gate which are connected in sequence.
And the model training unit is used for taking parameter sample data in the sea test data as the input of the long-time and short-time memory network model, taking corresponding label data as the output of the long-time and short-time memory network model for training, and determining the trained long-time and short-time memory network model as the power prediction model.
In an example, the generated power prediction module 202 specifically includes:
the parameter data preprocessing unit is used for preprocessing parameter data of the current water depth in the current time period to obtain preprocessed parameter data of the current water depth in the current time period; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence.
And the prediction unit is used for inputting the parameter data preprocessed at the current water depth in the current time period into the power prediction model to obtain the generated power at the current water depth in the prediction time period of the wave energy power generation device.
In an example, the model training unit specifically includes:
the sea test data preprocessing subunit is used for preprocessing each data set in the sea test data to obtain a preprocessed data set; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence.
And the data group dividing subunit is used for dividing the preprocessed data set into a training data group, a proofreading data group and a prediction data group according to a set proportion.
And the initial training subunit is used for taking parameter sample data in the training data set as the input of the long-time and short-time memory network model and taking label data in the training data set as the output of the long-time and short-time memory network model to perform initial training to obtain the initial long-time and short-time memory network model.
And the proofreading subunit is used for adjusting the model parameters in the initial long-time and short-time memory network model by using the proofreading data group to obtain the adjusted long-time and short-time memory network model.
And the testing subunit is used for testing the adjusted long-time and short-time memory network model by adopting the prediction data set to obtain the power prediction model.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the description of the method part.
The principle and the embodiment of the present invention are explained by applying specific examples, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A wave energy power generation power prediction method is characterized by comprising the following steps:
acquiring parameter data of a wave energy power generation device at a current water depth within a current time period; the parameter data comprises wave height, wave direction, a prediction time period and generated power; the current time interval is represented as [ a, t ], wherein t is the current time, and a is less than t; the prediction time period is [ t +1, b ], wherein t +1 represents the next moment of the current moment, and b is more than or equal to t +1;
according to parameter data and a power prediction model of the current water depth in the current time period, obtaining the generated power of the wave energy power generation device in the current water depth in the prediction time period; the power prediction model is constructed based on a long-time memory network model.
2. The wave energy generated power prediction method according to claim 1, characterized in that the power prediction model is determined by:
obtaining sea test data; the sea trial data comprises a plurality of data sets; the data set includes at least one pair of sample data; the sample data comprises parameter sample data and corresponding label data; any parameter sample data comprises parameter data of a prototype of the wave energy power generation device in a water depth during a test; the tag data comprises a time period and corresponding generated power; the time periods corresponding to the label data in different data sets are different; the time period corresponding to the generated power in the parameter sample data is different from the time period corresponding to the generated power in the tag data; the ending time of the time period corresponding to the generated power in the parameter sample data is continuous with the starting time of the time period corresponding to the generated power in the label data;
establishing a long-time and short-time memory network model; the long-time memory network model comprises an input gate, a forgetting gate and an output gate which are connected in sequence;
and taking parameter sample data in the sea test data as input of the long-time and short-time memory network model, taking corresponding label data as output of the long-time and short-time memory network model for training, and determining the trained long-time and short-time memory network model as the power prediction model.
3. The method for predicting the power generated by the wave energy according to the claim 1, wherein the step of obtaining the power generated by the wave energy power generation device in the current water depth within the prediction time period according to the parameter data and the power prediction model in the current water depth within the current time period specifically comprises the following steps:
preprocessing the parameter data of the current water depth in the current time period to obtain preprocessed parameter data of the current water depth in the current time period; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
and inputting the parameter data preprocessed at the current water depth in the current time period into a power prediction model to obtain the generated power at the current water depth in the prediction time period of the wave energy power generation device.
4. The wave energy power generation power prediction method according to claim 2, wherein the training is performed by taking parameter sample data in the sea test data as input of the long-time and short-time memory network model, taking corresponding label data as output of the long-time and short-time memory network model, and determining the trained long-time and short-time memory network model as the power prediction model, specifically comprises:
preprocessing each data set in the sea test data to obtain preprocessed data sets; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
dividing the preprocessed data set into a training data group, a proofreading data group and a prediction data group according to a set proportion;
taking parameter sample data in the training data set as input of the long-time and short-time memory network model, and taking label data in the training data set as output of the long-time and short-time memory network model to perform primary training to obtain an initial long-time and short-time memory network model;
adjusting model parameters in the initial long-time and short-time memory network model by adopting the proofreading data group to obtain an adjusted long-time and short-time memory network model;
and testing the adjusted long-time and short-time memory network model by adopting the prediction data set to obtain the power prediction model.
5. A wave energy generated power prediction system, comprising:
the data acquisition module is used for acquiring parameter data of the wave energy power generation device at the current depth of water in the current time period; the parameter data comprises wave height, wave direction, a prediction time period and generated power; the current time interval is represented as [ a, t ], wherein t is the current moment, and a is less than t; the prediction time period is [ t +1, b ], wherein t +1 represents the next moment of the current moment, and b is more than or equal to t +1;
the generating power prediction module is used for obtaining the generating power of the wave energy generating device at the current water depth within the prediction time period according to the parameter data and the power prediction model at the current water depth within the current time period; the power prediction model is constructed based on a long-time and short-time memory network model.
6. The wave energy generated power prediction system of claim 5, further comprising: a prediction model determination module for determining the power prediction model;
the prediction model determining module specifically includes:
the marine test data acquisition unit is used for acquiring marine test data; the sea trial data comprises a plurality of data sets; the data set includes at least one pair of sample data; the sample data comprises parameter sample data and corresponding label data; any parameter sample data comprises parameter data of a prototype of the wave energy power generation device in a water depth during a test; the tag data includes a time period and a corresponding generated power; the time periods corresponding to the label data in different data sets are different; the time period corresponding to the generated power in the parameter sample data is different from the time period corresponding to the generated power in the tag data; the ending time of the time period corresponding to the generated power in the parameter sample data is continuous with the starting time of the time period corresponding to the generated power in the label data;
the model establishing unit is used for establishing a long-time memory network model; the long-time and short-time memory network model comprises an input gate, a forgetting gate and an output gate which are connected in sequence;
and the model training unit is used for taking parameter sample data in the sea test data as the input of the long-time and short-time memory network model, taking corresponding label data as the output of the long-time and short-time memory network model for training, and determining the trained long-time and short-time memory network model as the power prediction model.
7. The wave energy generated power prediction system according to claim 5, characterized in that the generated power prediction module specifically comprises:
the parameter data preprocessing unit is used for preprocessing the parameter data in the current water depth in the current time period to obtain the preprocessed parameter data in the current water depth in the current time period; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
and the prediction unit is used for inputting the parameter data preprocessed at the current water depth in the current time period into the power prediction model to obtain the generated power at the current water depth in the prediction time period of the wave energy power generation device.
8. The wave energy generated power prediction system according to claim 6, characterized in that the model training unit specifically comprises:
the sea test data preprocessing subunit is used for preprocessing each data set in the sea test data to obtain a preprocessed data set; the preprocessing comprises noise elimination, blank value filling, data transformation, data combination and data normalization in sequence;
the data group dividing subunit is used for dividing the preprocessed data set into a training data group, a proofreading data group and a prediction data group according to a set proportion;
the initial training subunit is used for taking parameter sample data in the training data set as the input of the long-time and short-time memory network model, taking label data in the training data set as the output of the long-time and short-time memory network model, and performing initial training to obtain an initial long-time and short-time memory network model;
the proofreading subunit is used for adjusting model parameters in the initial long-time and short-time memory network model by adopting the proofreading data set to obtain an adjusted long-time and short-time memory network model;
and the testing subunit is used for testing the adjusted long-time and short-time memory network model by adopting the prediction data set to obtain the power prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211075400.XA CN115149529A (en) | 2022-09-05 | 2022-09-05 | Wave energy power generation power prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211075400.XA CN115149529A (en) | 2022-09-05 | 2022-09-05 | Wave energy power generation power prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115149529A true CN115149529A (en) | 2022-10-04 |
Family
ID=83416314
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211075400.XA Pending CN115149529A (en) | 2022-09-05 | 2022-09-05 | Wave energy power generation power prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115149529A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109885982A (en) * | 2019-04-10 | 2019-06-14 | 浙江大学 | Wave-activated power generation method based on power prediction and maximum capture width ratio tracking |
| CN114445634A (en) * | 2022-02-28 | 2022-05-06 | 南京信息工程大学 | Sea wave height prediction method and system based on deep learning model |
-
2022
- 2022-09-05 CN CN202211075400.XA patent/CN115149529A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109885982A (en) * | 2019-04-10 | 2019-06-14 | 浙江大学 | Wave-activated power generation method based on power prediction and maximum capture width ratio tracking |
| CN114445634A (en) * | 2022-02-28 | 2022-05-06 | 南京信息工程大学 | Sea wave height prediction method and system based on deep learning model |
Non-Patent Citations (2)
| Title |
|---|
| CHENHUA NI: "Data-driven models for short-term ocean wave power forecasting", 《IET RENEWABLE POWER GENERATION》 * |
| 倪晨华: "基于数据驱动模式的波浪能装置短期发电功率预测方法", 《科技导报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | A wind speed correction method based on modified hidden Markov model for enhancing wind power forecast | |
| Zhang et al. | Data-driven day-ahead PV estimation using autoencoder-LSTM and persistence model | |
| Gensler et al. | Deep Learning for solar power forecasting—An approach using AutoEncoder and LSTM Neural Networks | |
| Miranda et al. | One-hour-ahead wind speed prediction using a Bayesian methodology | |
| CN111222674A (en) | Short-term photovoltaic power generation capacity prediction method based on long-term and short-term memory neural network | |
| Cheng et al. | Augmented convolutional network for wind power prediction: A new recurrent architecture design with spatial-temporal image inputs | |
| Cao et al. | Multi-step wind power forecasting model using LSTM networks, similar time series and LightGBM | |
| Bendali et al. | Deep learning using genetic algorithm optimization for short term solar irradiance forecasting | |
| CN114021836B (en) | Multi-variable reservoir water inflow prediction system based on different angle fusion, training method and application | |
| Sodsong et al. | Short-term solar PV forecasting using gated recurrent unit with a cascade model | |
| Dai et al. | An online-learning-enabled self-attention-based model for ultra-short-term wind power forecasting | |
| Kharlova et al. | Forecasting photovoltaic power production using a deep learning sequence to sequence model with attention | |
| CN115115125A (en) | Photovoltaic power interval probability prediction method based on deep learning fusion model | |
| Yang et al. | A novel method of wind speed prediction by peephole LSTM | |
| Li et al. | Wind power probabilistic forecasting based on wind correction using weather research and forecasting model | |
| Shi et al. | Deep-learning-based wind speed forecasting considering spatial–temporal correlations with adjacent wind turbines | |
| Alharbi et al. | Short-term wind speed and temperature forecasting model based on gated recurrent unit neural networks | |
| CN116581755B (en) | Power prediction method, device, equipment and storage medium | |
| Zhou et al. | Short-term wind power prediction based on EMD-LSTM | |
| Hossain | Application of Gaussian mixture regression model for short-term wind speed forecasting | |
| Phan et al. | Application of a new Transformer-based model and XGBoost to improve one-day-ahead solar power forecasts | |
| Ayele et al. | Adama II wind farm long-term power generation forecasting based on machine learning models | |
| CN115149529A (en) | Wave energy power generation power prediction method and system | |
| Panda et al. | Forecasting temperature anomalies of planet earth: A comparative analysis of ai models | |
| Xu et al. | A multi-step wind power group forecasting seq2seq architecture with spatial–temporal feature fusion and numerical weather prediction correction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20221004 |