A distributed energy two-way prediction optimization scheduling method and system

Abstract

The application discloses a kind of distributed energy two-way prediction optimization scheduling method and system, obtains historical data, historical environmental data, historical load data and historical scheduling strategy and carries out processing to data, is extracted and weighted fusion to the data after processing by the prediction model based on global feature and mutation feature, obtain first prediction result, by adaptive prediction model, obtain second prediction result for reverse distributed energy prediction solution, first prediction result and second prediction result are complex weighted fusion, obtain third prediction result, third prediction result is used to construct distributed energy optimization scheduling model and is solved, obtain scheduling result.This method effectively fuses environment, load, scheduling three, adopts two-way prediction to further improve the accuracy of distributed energy prediction, and optimization scheduling is established on the basis of prediction, so that the overall scheduling of massive distributed energy is more optimized.

Description

Technical Field

[0001] This invention relates to the field of power grid operation planning technology, and in particular to a method and system for bidirectional predictive optimization scheduling of distributed energy resources. Background Technology

[0002] With the rapid development of distributed energy sources such as wind power, photovoltaics, and modern bioenergy, their proportion in the power distribution network is gradually increasing. Because new distributed energy generation, such as wind power and photovoltaics, typically exhibits randomness, volatility, and intermittency, traditional power distribution network dispatching methods are becoming increasingly inadequate. Therefore, it is urgent to study the safety and economic dispatching of the power distribution network under the background of high penetration of new distributed energy sources.

[0003] Achieving optimal scheduling relies on the effective prediction of various influencing factors, including power output prediction based on environmental data and load prediction. Related technologies include using meteorological data to predict photovoltaic or wind power, and modeling methods often employ recurrent neural networks (RNNs), long short-term memory (LSTM) networks, and transform neural networks. Both RNNs and LSTM networks use backpropagation algorithms to calculate the loss function for optimization. The longer the sequence, the more difficult it becomes to calculate the gradient, i.e., the harder it is to converge, thus worsening the performance. Transform neural network structures can completely eliminate the vanishing and exploding gradient problems, but they also suffer from slower speeds when predicting long-term sequences. Regarding optimal scheduling strategies, some studies explore the optimal scheduling strategy from the perspective of minimizing power generation costs and environmental governance costs; however, their constraints often use the maximum output of distributed generation as an upper limit, which may prevent the scheduling strategy from being implemented. Summary of the Invention

[0004] This invention provides a bidirectional predictive optimization scheduling method and system for distributed energy resources. By effectively integrating the environment, load, and scheduling, the output of distributed energy resources is predicted, and then the scheduling scheme is determined based on the prediction results, thereby improving the accuracy of the scheduling scheme.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a distributed energy bidirectional predictive optimization scheduling method, comprising:

[0006] Acquire historical power output data, historical environmental data, historical load data, and historical dispatch strategies within the area to be monitored. The historical power output data includes historical wind power output data and historical photovoltaic power output data.

[0007] Orthogonal wavelet transform and reconstruction processing are performed on historical power output data and historical environment data respectively to obtain historical power output reconstruction data matrix and historical environment reconstruction data matrix. The historical power output reconstruction data matrix and historical environment reconstruction data matrix are then subjected to feature extraction and weighted fusion through a prediction model based on global features and abrupt change features to obtain the first power output prediction result.

[0008] After solving the reverse distributed energy output prediction by using an adaptive prediction model based on historical output data, historical load data, and historical scheduling strategies, the second output prediction result is obtained.

[0009] The first and second power output prediction results are combined and weighted to obtain the third power output prediction result.

[0010] A distributed energy optimization scheduling model is constructed using the third power output prediction results, and the distributed energy optimization scheduling model is solved to obtain the scheduling results, so that the distribution network can schedule energy according to the scheduling results. The scheduling results include photovoltaic power generation power value, wind power generation power value, energy storage system operating power value, and load shedding power value.

[0011] This embodiment acquires historical power output data, historical environmental data, historical load data, and historical scheduling strategies within the monitored area. Orthogonal wavelet transform and reconstruction processing are performed on the historical power output data and historical environmental data to obtain historical power output reconstruction data matrices, historical environmental reconstruction data matrices, and wind speed single-branch reconstruction data matrices. These matrices are then used for feature extraction and weighted fusion through a prediction model based on global and abrupt change features to obtain a first power output prediction result. The historical power output data is then used to perform reverse distributed energy power output prediction using an adaptive prediction model to obtain a second power output prediction result. A composite weighted fusion of the first and second power output prediction results yields a third power output prediction result. This third power output prediction result is used to construct a distributed energy optimal scheduling model, which is then solved to obtain the scheduling result. This method effectively integrates environment, load, and scheduling, employs bidirectional prediction to further improve the accuracy of distributed energy power output prediction, and establishes optimal scheduling based on predicted power output, resulting in more optimized overall scheduling of massive distributed energy resources.

[0012] As a preferred approach, orthogonal wavelet transform and reconstruction processing are performed on historical power output data and historical environment data respectively to obtain historical power output reconstruction data matrix and historical environment reconstruction data matrix, specifically:

[0013] The historical power output data and historical environmental data acquired within a preset time period are normalized to obtain normalized historical power output data and normalized historical environmental data. The historical environmental data includes historical ambient light intensity data, historical incoming wind speed data, and historical outgoing wind speed data. The normalization process is as follows:

[0014]

[0015] Among them, PWH This indicates historical wind power output data for renewable energy, P PVH Indicates historical photovoltaic power generation output data, I H Indicates historical ambient light intensity, V ciH Indicates the historical entry wind speed, V coH Indicates historical cut-out wind speed, P WH,min ,P PVH,min ,I H,min V ciH,min and V coH,min and P WH,max ,P PVH,max ,P LH,max ,I H,max V ciH,max and V coH,max Let P′ represent historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed minimum and maximum values, respectively. WH 、P′ PVH 、I' H V′ ciH and V′ coH These are represented as normalized historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed, respectively.

[0016] Each data point in the normalized historical power output data and normalized historical environmental data is decomposed using a pre-defined transformation formula, yielding the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients corresponding to each data point in the normalized historical power output data and normalized historical environmental data. The pre-defined transformation formula is as follows:

[0017]

[0018]

[0019] Where x(t) is the signal to be processed. Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) is the real part of the wavelet function, ψ g (t) is the imaginary part wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0020] Based on the wavelet coefficients from the real-part tree decomposition and the imaginary-part tree decomposition, the reconstructed wavelet coefficient signal and the reconstructed scaling coefficient signal corresponding to each data point are obtained through the reconstruction formula. The reconstruction formula is as follows:

[0021]

[0022]

[0023] Where, d j (t) represents the reconstructed signal of a certain wavelet coefficient, c J (t) represents the reconstructed signal for a certain data scale coefficient, t = 1, 2, L, M, where M is the length of the signal and n is the sequence number of the signal, n = 1, 2, L, M; Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) represents the real part of the wavelet function, ψ g (t) represents the imaginary part of the wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0024] Based on the wavelet coefficients and scaling coefficients of each data point, a matrix is ​​constructed to obtain the historical output reconstruction data matrix and the historical environment reconstruction data matrix.

[0025] As a preferred approach, the historical output reconstruction data matrix and the historical environment reconstruction data matrix are subjected to feature extraction and weighted fusion using a prediction model based on global features and abrupt change features to obtain the first output prediction result, specifically:

[0026] The historical output data matrix and the historical environment data matrix are input into the prediction model based on global features and mutation features. The first matrix is ​​obtained by extracting data from the global trend feature layer, the second matrix is ​​obtained by extracting data from the local mutation feature layer, and the third matrix is ​​obtained by extracting data from the steady-state features. The steady-state feature calculation formula is as follows:

[0027] x t =φ0+φ1x t-1 +φ2x t-2 +L+φ p x t-p +ε t

[0028] Where, x t φ represents the data at time t, i.e., the extracted data. i (i = 1, ..., p) are the coefficients of the autoregressive model, εt This indicates that the mean is 0 and the variance is σ. 2 The white noise, where p represents the data length of the current prediction time value;

[0029] After performing convolution and distillation operations on the first matrix and the second matrix respectively, new first matrix and new second matrix are obtained;

[0030] The new first matrix and the new second matrix are fused to obtain the fourth matrix. Then, the third matrix and the fourth matrix are weighted and fused to obtain the first output prediction result.

[0031] As a preferred approach, after performing convolution and distillation operations on the first matrix and the second matrix respectively, new first matrix and new second matrix are obtained, specifically as follows:

[0032] After converting the first matrix and the second matrix into vector matrices in sequence, the first matrix is ​​multiplied by a number of preset weight matrices to obtain the first vector matrix, the second vector matrix, and the third vector matrix corresponding to the first matrix.

[0033] The approximation evaluation of the first vector matrix is ​​calculated based on a preset number of dot products, resulting in an approximation evaluation criterion for each vector in the first vector matrix. The approximation evaluation criterion is as follows:

[0034]

[0035] Among them, L Q and L K Represented as the first vector matrix Second vector matrix The length of q i and k j These are the i-th and j-th data points of the first vector matrix Q1 and the second vector matrix K1, respectively. The superscript T represents the transpose operation, and d represents the dimension of matrices Q1, K1, and V1.

[0036] Based on the approximation evaluation criteria of each vector in the first vector matrix, the contribution value of each vector to the calculation of the attention value is obtained. The vector with the largest contribution value is selected as the fifth matrix, and the probabilistic sparse attention value is obtained from the fifth matrix using the attention value calculation formula. The probabilistic sparse attention value is:

[0037]

[0038] in, Let represent the probabilistic sparse attention value, Q1 represent the first vector matrix, K1 represent the second vector matrix, V1 represent the third vector matrix, d represent the dimension of matrices Q1, K1, and V1, and the superscript T represents the transpose operation.

[0039] After performing convolution and distillation operations based on the probabilistic sparse attention values ​​and preset weights, a new first matrix and a new second matrix are obtained. The convolution and distillation operations are as follows:

[0040]

[0041] Among them, [g] AB This represents the value obtained after passing through an attention module. Conv1d is a one-dimensional convolution, ELU is the activation function of the activation layer, and MaxPool is max pooling. At time t The data after the (j+1)th distillation layer, i.e., the new first matrix;

[0042] As a preferred approach, the historical output data, historical load data, and historical scheduling strategies are used to solve for the reverse distributed energy output prediction using an adaptive prediction model, resulting in the second output prediction result, specifically:

[0043] The historical scheduling strategy is extended to obtain a new historical scheduling strategy;

[0044] An adaptive prediction model is constructed based on the historical load data, the new historical scheduling strategy, the historical cut-in wind speed, the historical cut-out wind speed, and the historical ambient light intensity. The adaptive prediction model is then solved to obtain a second power output prediction result. The adaptive prediction model includes a wind power adaptive prediction model and a photovoltaic adaptive prediction module. The second power output prediction result includes wind power output prediction results and photovoltaic power output prediction results.

[0045] This embodiment extends the historical scheduling strategy to obtain a new historical scheduling strategy. An adaptive prediction model is constructed based on the historical load data, the new historical scheduling strategy, historical cut-in wind speed, historical cut-out wind speed, and historical ambient light intensity. The adaptive prediction model is then solved to obtain the second output prediction result. This method considers the reverse prediction of scheduling strategy, load, and weather, and also takes into account different historical scheduling needs, effectively integrating environment, load, and scheduling to achieve more accurate predictions.

[0046] As a preferred embodiment, an adaptive prediction model is constructed based on the historical load data, the new historical scheduling strategy, historical cut-in wind speed, historical cut-out wind speed, and historical ambient light intensity. The adaptive prediction model is then solved to obtain the second power output prediction result, specifically:

[0047] Based on the historical load data, the new historical scheduling strategy, the historical cut-in wind speed, and the historical cut-out wind speed, a wind force adaptive prediction model is constructed, wherein the wind force adaptive prediction model is:

[0048] P WH =β Wi X Wi +β W

[0049] Among them, P WH This represents the predicted wind power output, β. Wi (i = 1, 2, 3, 4) represents the regression coefficient between wind power output and related factors, β W X represents the overall error of the wind power output model. Wi X represents the wind power output prediction factor matrix. Wi =[P LH D H V ciH V coH ], where P LH Represents historical load data, D H This indicates a new historical scheduling strategy, V ciH Indicates historical cut-in wind speed and V coH History cuts out the wind speed;

[0050] Solve for P using the least squares method. WH =β Wi X Wi +β W Regression coefficients, obtaining initial values ​​of the regression coefficients. Further introduce weights Then the weight vector It can be represented as:

[0051]

[0052] renew x j It is X Wi Variables in

[0053] For all regularization factors λ n Solve

[0054] calculate Will Substitute P WH =β Wi X Wi +β W The wind power output prediction results were obtained.

[0055] For photovoltaic power output, an adaptive prediction model based on historical load, new historical dispatch strategies, and historical ambient light intensity should be established:

[0056] P PVH =β PVi XPVi +β PV

[0057] Among them, P PVH This represents the photovoltaic output regression model, β PVi (i = 1, 2, 3) represents the regression coefficient between photovoltaic power output and related factors, β PV X represents the overall error of the photovoltaic power output model. PVi X represents the photovoltaic power output prediction factor matrix. PVi =[P LH D H ,I H ], where P LH Represents historical load data, D H This indicates the new historical scheduling strategy and I H Indicates historical ambient light intensity;

[0058] Solve for P using the least squares method. PVH =β PVi X PVi +β PV The regression coefficients were obtained, and the initial values ​​of the regression coefficients were obtained. Further introduce weights Then the weight vector It can be represented as:

[0059]

[0060] renew x j It is X PVi Variables in

[0061] For all regularization factors λ n Solve

[0062] calculate Will Substitute P PVH =β PVi X PVi +β PV The predicted photovoltaic output was obtained.

[0063] As a preferred approach, the first and second power output prediction results are combined and weighted to obtain the third power output prediction result, specifically:

[0064] The first and second power output prediction results are combined using a weighted average method to obtain the third power output prediction result. The specific calculation formula is as follows:

[0065] P WTP =θP WP1 +(1-θ)PWP2

[0066] Among them, P WTP P represents the predicted third output. WP1 P represents the first output prediction result. WP2 This represents the second output prediction result, and θ represents the coefficient of power predicted using the forward prediction method.

[0067] As a preferred approach, a distributed energy optimal scheduling model is constructed using the third-party output prediction results. This model is then used to optimize the scheduling of historical output data and energy storage equipment operation data, yielding the following scheduling results:

[0068] The power generation cost is constructed based on the third prediction result and the power generation cost coefficient, where the power generation cost is either the photovoltaic power generation cost or the wind power generation cost.

[0069] The operating and management cost of energy storage equipment is obtained by constructing the charging operating cost coefficient, discharging operating cost coefficient, charging efficiency, discharging efficiency, charging power, and discharging power.

[0070] The environmental governance cost of energy storage equipment is obtained by constructing the environmental governance cost coefficient, charging power, and discharging power of the energy storage equipment. The load shedding cost is obtained by constructing the load shedding penalty coefficient and load shedding power.

[0071] The objective function and constraints of the distributed energy optimization scheduling model are constructed based on the generation cost, energy storage equipment operation and management cost, energy storage equipment environmental governance cost, and load shedding cost. The objective function is:

[0072] min(C PV +C WT +C SV +C SVE +C loss )

[0073] Among them, C PV For the cost of photovoltaic power generation, C WT For the cost of wind power generation, C SV For the operation and management costs of energy storage equipment, C SVE For the environmental governance costs of energy storage equipment, C loss For load shedding costs;

[0074] The objective function and constraints are solved using the alternating direction multiplier method to obtain the scheduling results, which include photovoltaic power generation, wind power generation, energy storage system operating power, and load shedding power.

[0075] In this embodiment, the power generation cost is constructed based on the third prediction result and the power generation cost coefficient. The power generation cost can be either photovoltaic (PV) or wind power generation cost. The energy storage device operation and management cost is constructed based on the charging operation cost coefficient, discharging operation cost coefficient, charging efficiency, discharging efficiency, charging power, and discharging power. The energy storage device environmental governance cost is constructed based on the energy storage device operating environment governance cost coefficient, the energy storage device charging power, and the energy storage device discharging power. The load shedding cost is constructed based on the load shedding penalty coefficient and the load shedding power. The objective function and constraints of the distributed energy optimization scheduling model are constructed based on the power generation cost, energy storage device operation and management cost, energy storage device environmental governance cost, and load shedding cost. The objective function and constraints are solved using the alternating direction multiplier method to obtain the scheduling results, which include PV power generation value, wind power generation value, energy storage system operating power value, and load shedding power value. By utilizing the optimized scheduling model based on the prediction results and solving for the optimized scheduling results, the accuracy of the overall distributed energy scheduling scheme is improved.

[0076] As a preferred solution, to address the same technical problem, embodiments of the present invention also provide a distributed energy bidirectional predictive optimization scheduling system, comprising:

[0077] The acquisition module is used to acquire historical power output data, historical environmental data, historical load data, and historical dispatch strategies within the area to be monitored. The historical power output data includes historical wind power output data and historical photovoltaic power output data.

[0078] The first prediction module is used to perform orthogonal wavelet transform and reconstruction processing on historical power output data and historical environment data respectively to obtain historical power output reconstruction data matrix and historical environment reconstruction data matrix. The historical power output reconstruction data matrix and historical environment reconstruction data matrix are then used to extract features and perform weighted fusion through a prediction model based on global features and abrupt change features to obtain the first power output prediction result.

[0079] The second prediction module is used to solve the reverse distributed energy output prediction by using an adaptive prediction model to obtain the second output prediction result after taking the historical output data, historical load data and historical scheduling strategy.

[0080] The third prediction module is used to perform a composite weighted fusion of the first and second output prediction results to obtain the third output prediction result.

[0081] The scheduling result module is used to construct a distributed energy optimal scheduling model using the third power output prediction results, and solve the distributed energy optimal scheduling model to obtain the scheduling results, so that the distribution network can schedule energy according to the scheduling results. The scheduling results include photovoltaic power generation power value, wind power generation power value, energy storage system operating power value, and load shedding power value.

[0082] As a preferred embodiment, the first prediction module includes a normalization unit, a decomposition unit, a reconstruction unit, and a matrix unit.

[0083] The normalization unit is used to normalize the various data points in the historical power output data and historical environmental data within the acquired preset time period, resulting in normalized historical power output data and normalized historical environmental data. The historical environmental data includes historical ambient light intensity data, historical cut-in wind speed data, and historical cut-out wind speed data. The normalization process is as follows:

[0084]

[0085] Among them, P WH This indicates historical wind power output data for renewable energy, P PVH Indicates historical photovoltaic power generation output data, I H Indicates historical ambient light intensity, V ciH Indicates the historical entry wind speed, V coH Indicates historical cut-out wind speed, P WH,min ,P PVH,min ,I H,min V ciH,min and V coH,min and P WH,max ,P PVH,max ,P LH,max ,I H,max V ciH,max and V coH,max Let P′ represent historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed minimum and maximum values, respectively. WH 、P′ PVH 、I' H V′ ciH and V′ coH These are represented as normalized historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed, respectively.

[0086] The decomposition unit is used to perform positive wavelet transform decomposition on each data point in the normalized historical power output data and the normalized historical environmental data according to a preset transformation formula, to obtain the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients corresponding to each data point in the normalized historical power output data and the normalized historical environmental data, wherein the preset transformation formula is:

[0087]

[0088]

[0089] Where x(t) is the signal to be processed. Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) is the real part of the wavelet function, ψ g (t) is the imaginary part wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0090] The reconstruction unit is used to calculate the wavelet coefficient reconstruction signal and scaling coefficient reconstruction signal corresponding to each data point using the reconstruction formula based on the wavelet coefficients of the real part tree decomposition and the wavelet coefficients of the imaginary part tree decomposition. The reconstruction formula is as follows:

[0091]

[0092]

[0093] Where, d j (t) represents the reconstructed signal of a certain wavelet coefficient, c J (t) represents the reconstructed signal for a certain data scale coefficient, t = 1, 2, L, M, where M is the length of the signal and n is the sequence number of the signal, n = 1, 2, L, M; Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) represents the real part of the wavelet function, ψ g (t) represents the imaginary part of the wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0094] The matrix unit is used to construct a matrix based on the wavelet coefficients and scaling coefficients of each data point to obtain the historical output reconstruction data matrix and the historical environment reconstruction data matrix.

[0095] This method acquires historical power output data, historical environmental data, historical load data, and historical scheduling strategies within the monitored area. Orthogonal wavelet transform and reconstruction processing are performed on the historical power output data and historical environmental data to obtain historical power output reconstruction data matrices, historical environmental reconstruction data matrices, and wind speed single-branch reconstruction data matrices. These matrices are then used for feature extraction and weighted fusion through a prediction model based on global and abrupt feature characteristics to obtain the first power output prediction result. The historical power output data is then used to perform reverse distributed energy power output prediction through an adaptive prediction model to obtain the second power output prediction result. The first and second power output prediction results are then combined and weighted to obtain the third power output prediction result. A distributed energy optimal scheduling model is constructed using the third power output prediction result and solved to obtain the scheduling result. This method effectively integrates environmental, load, and scheduling factors, employs bidirectional prediction to further improve the accuracy of distributed energy power output prediction, and establishes optimized scheduling based on predicted power output, resulting in more optimized overall scheduling of massive distributed energy resources. Attached Figure Description

[0096] Figure 1 : A flowchart illustrating an embodiment of the distributed energy bidirectional predictive optimization scheduling method provided by the present invention;

[0097] Figure 2 : A schematic diagram of the data calculation process of an embodiment of the distributed energy bidirectional prediction optimization scheduling method provided by the present invention;

[0098] Figure 3 This is a schematic diagram of the first prediction process of an embodiment of the distributed energy bidirectional prediction optimization scheduling method provided by the present invention.

[0099] Figure 4 : A schematic diagram of the second prediction process of an embodiment of the distributed energy bidirectional prediction optimization scheduling method provided by the present invention;

[0100] Figure 5 : A schematic diagram of the system structure of another embodiment of the distributed energy bidirectional predictive optimization scheduling method provided by the present invention; Detailed Implementation

[0101] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0102] Example 1

[0103] Please refer to Figure 1 This invention provides a distributed energy bidirectional predictive optimization scheduling method, which includes steps 101 to 104, each step of which is as follows:

[0104] Step 101: Obtain historical power output data, historical environmental data, historical load data, and historical dispatch strategies within the area to be monitored. The historical power output data includes historical wind power output data and historical photovoltaic power output data.

[0105] In this embodiment, historical renewable energy output data, historical environmental data, and historical load data within a preset time period are acquired in the monitoring area, and historical dispatch strategies are also obtained. The historical output data includes historical wind power output data and historical photovoltaic power output data, while the historical environmental data includes historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed data.

[0106] It should be noted that the time period and time interval for data acquisition can be determined according to actual needs.

[0107] As an example of this embodiment, historical wind power output data, historical photovoltaic power output data, historical load data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed can be obtained within the monitoring area for the 30 days prior to the current date, at 15-minute intervals.

[0108] Step 102: Perform orthogonal wavelet transform and reconstruction processing on the historical power output data and historical environment data respectively to obtain the historical power output reconstruction data matrix and the historical environment reconstruction data matrix. Then, use a prediction model based on global features and abrupt change features to extract features from the historical power output reconstruction data matrix and the historical environment reconstruction data matrix and perform weighted fusion to obtain the first power output prediction result.

[0109] Optionally, orthogonal wavelet transform and reconstruction processing are performed on the historical output data and historical environment data respectively to obtain the historical output reconstruction data matrix and the historical environment reconstruction data matrix, specifically:

[0110] The historical power output data and historical environmental data acquired within a preset time period are normalized to obtain normalized historical power output data and normalized historical environmental data. The historical environmental data includes historical ambient light intensity data, historical incoming wind speed data, and historical outgoing wind speed data. The normalization process is as follows:

[0111]

[0112] Among them, P WH This indicates historical wind power output data for renewable energy, P PVHThis indicates historical photovoltaic power generation output data, P LH Represents historical load data, I H Indicates historical ambient light intensity, V ciH Indicates the historical entry wind speed, V coH Indicates historical cut-out wind speed, P WH,min ,P PVH,min ,I H,min V ciH,min and V coH,min and P WH,max ,P PVH,max ,P LH,max ,I H,max V ciH,max and V coH,max Let P′ represent historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed minimum and maximum values, respectively. WH 、P′ PVH 、I' H V′ ciH and V′ coH These are represented as normalized historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed, respectively.

[0113] Each data point in the normalized historical power output data and normalized historical environmental data is decomposed using a pre-defined transformation formula, yielding the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients corresponding to each data point in the normalized historical power output data and normalized historical environmental data. The pre-defined transformation formula is as follows:

[0114]

[0115]

[0116] Where x(t) is the signal to be processed. Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) is the real part of the wavelet function, ψ g (t) is the imaginary part wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0117] Based on the wavelet coefficients from the real-part tree decomposition and the imaginary-part tree decomposition, the reconstructed wavelet coefficient signal and the reconstructed scaling coefficient signal corresponding to each data point are obtained through the reconstruction formula. The reconstruction formula is as follows:

[0118]

[0119]

[0120] Where, d j (t) represents the reconstructed signal of a certain wavelet coefficient, c J (t) represents the reconstructed signal for a certain data scale coefficient, t = 1, 2, L, M, M is the length of the reconstructed signal, and n is the index of the reconstructed signal, n = 1, 2, L, M; Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) represents the real part of the wavelet function, ψ g (t) represents the imaginary part of the wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0121] Based on the wavelet coefficients and scaling coefficients of each data point, a matrix is ​​constructed to obtain the historical output reconstruction data matrix and the historical environment reconstruction data matrix.

[0122] In this embodiment, as Figure 2 As shown, in order to eliminate the influence of different data units, the acquired historical power output data and historical environmental data are preprocessed by normalization to obtain normalized historical power output data and normalized historical environmental data. The historical environmental data includes historical ambient light intensity data and historical cut-in wind speed data. The normalization process for the historical cut-out wind speed data is as follows:

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] Among them, P WH This indicates historical wind power output data for renewable energy, P PVH This indicates historical photovoltaic power generation output data, P LH Represents historical load data, I H Indicates historical ambient light intensity, V ciHIndicates the historical entry wind speed, V coH Indicates historical cut-out wind speed, P WH,min ,P PVH,min ,P LH,min ,I H,min V ciH,min and V coH,min and P WH,max ,P PVH,max ,P LH,max ,I H,max V ciH,max and V coH,max These represent historical wind power output data, historical photovoltaic power output data, historical load data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed (minimum and maximum values), respectively. P W ' H P P ' VH P L ' H 、I' H V c ' iH and V c ' oH These are represented as normalized historical wind power output data, historical photovoltaic power output data, historical load data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed, respectively.

[0130] After obtaining the normalized historical wind power output data, historical photovoltaic power output data, historical load data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed data, orthogonal wavelet transform and reconstruction are performed on these data. The specific method is as follows:

[0131] The data is decomposed using the following formula to obtain the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients for each data point. The decomposition formula is as follows:

[0132]

[0133]

[0134] Where x(t) is the signal to be processed. Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) is the real part of the wavelet function, ψ g(t) is the imaginary part wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor.

[0135] Then, based on the wavelet coefficients from the real part tree decomposition and the wavelet coefficients from the imaginary part tree decomposition, the wavelet coefficient reconstruction signal and the scaling coefficient reconstruction signal corresponding to each data point are calculated using the reconstruction formula. Finally, matrices are constructed based on the wavelet coefficient reconstruction signals and scaling coefficient reconstruction signals of each data point to obtain the historical output reconstruction data matrix and the historical environment reconstruction data matrix. The reconstruction formula is as follows:

[0136]

[0137]

[0138] Where, d j (t) represents the reconstructed signal of a certain wavelet coefficient, c J (t) represents the reconstructed signal for a certain data scale coefficient, t = 1, 2, L, M, where M is the length of the signal and n is the sequence number of the signal, n = 1, 2, L, M; Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) represents the real part of the wavelet function, ψ g (t) represents the imaginary part of the wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor.

[0139] As an example of this embodiment, J=4 is selected, and 5 sets of data can be obtained after single-branch reconstruction, as shown in the following formula:

[0140] I RCWT =[c I4 ,d I4 ,d I3 ,d I2 ,d I1 ]

[0141] V ciRCWT =[c ci4 ,d ci4 ,d ci3 ,d ci2 ,d ci1 ]

[0142] V coRCWT =[c co4 ,d co4 ,d co3 ,d co2 ,d co1 ]

[0143] Among them, IRCWT Represented as a single-branch reconstructed data matrix of illumination in historical environmental data, c I4 Represented as the illumination data scaling factor reconstructed signal, d I4 ,d I3 ,d I2 ,d I1 V is represented as the wavelet coefficient reconstructed signal of illumination data. ciRCWT Represented as a reconstructed data matrix of wind speed from historical environmental data, c ci4 Represented as the scale coefficient reconstructed signal of the input wind speed data, d ci4 ,d ci3 ,d ci2 ,d ci1 V represents the wavelet coefficient reconstructed signal from the input wind speed data. coRCWT Represented as a reconstructed data matrix of wind speed extracted from historical environmental data, c co4 This is represented as the reconstructed signal based on the scaling factor of the cut-out wind speed data, d co4 ,d co3 ,d co2 ,d co1 This represents the wavelet coefficient reconstructed signal from the cut-out wind speed data.

[0144] Optionally, the historical output reconstruction data matrix and the historical environment reconstruction data matrix are subjected to feature extraction and weighted fusion using a prediction model based on global features and mutation features to obtain the first output prediction result, specifically:

[0145] The historical output data matrix and the historical environment data matrix are input into the prediction model based on global features and abrupt change features. The first matrix is ​​obtained by extracting data from the global trend feature layer, the second matrix is ​​obtained by extracting data from the local abrupt change feature layer, and the third matrix is ​​obtained by extracting data from the steady-state features. The steady-state feature calculation formula is as follows:

[0146] x t =φ0+φ1x t-1 +φ2x t-2 +L+φ p x t-p +ε t

[0147] Where, x t φ represents the data at time t, i.e., the extracted data. i (i = 1, ..., p) are the coefficients of the autoregressive model, ε t This indicates that the mean is 0 and the variance is σ. 2 The white noise, where p represents the data length of the current prediction time value;

[0148] After performing convolution and distillation operations on the first matrix and the second matrix respectively, new first matrix and new second matrix are obtained;

[0149] The new first matrix and the new second matrix are fused to obtain the fourth matrix. Then, the third matrix and the fourth matrix are weighted and fused to obtain the first output prediction result.

[0150] In this embodiment, as Figure 3 As shown, the historical output reconstruction data matrix and the historical environment reconstruction data matrix are input into a prediction model based on global features and mutation features. This model is a prediction network based on global trends and local mutations, mainly including a global trend feature acquisition branch, a local mutation feature acquisition branch, and a steady-state feature acquisition branch. The global trend feature acquisition branch extracts each data to obtain a first matrix. The local mutation feature layer extracts each data to obtain a second matrix. The steady-state feature layer extracts each data to obtain a third matrix. After convolution and distillation operations on the first matrix and the second matrix respectively through the embedding layer and the distillation layer, new first matrices and new second matrices are obtained. Then, the new first matrix and the new second matrix are fused through a fully connected layer to obtain a fourth matrix. Finally, the third matrix and the fourth matrix are weighted and fused to obtain the first output prediction result.

[0151] Optionally, after performing convolution and distillation operations on the first matrix and the second matrix respectively, new first matrix and new second matrix are obtained, as follows:

[0152] After converting the first matrix and the second matrix into vector matrices in sequence, the first matrix is ​​multiplied by a number of preset weight matrices to obtain the first vector matrix, the second vector matrix, and the third vector matrix corresponding to the first matrix.

[0153] The approximation evaluation of the first vector matrix is ​​calculated based on a preset number of dot products to obtain the approximation evaluation criterion for each vector in the first vector matrix. The approximation evaluation criterion is as follows:

[0154]

[0155] Among them, L Q and L K Represented as the first vector matrix Second vector matrix The length of q i and k j These are the i-th and j-th data points of the first vector matrix Q1 and the second vector matrix K1, respectively. The superscript T represents the transpose operation, and d represents the dimension of matrices Q1, K1, and V1.

[0156] Based on the approximation evaluation criteria of each vector in the first vector matrix, the contribution value of each vector to the calculation of the attention value is obtained. The vector with the largest contribution value is selected as the fifth matrix, and the probabilistic sparse attention value is obtained from the fifth matrix using the attention value calculation formula. The probabilistic sparse attention value is:

[0157]

[0158] in, Let represent the probabilistic sparse attention value, Q1 represent the first vector matrix, K1 represent the second vector matrix, V1 represent the third vector matrix, d represent the dimension of matrices Q1, K1, and V1, and the superscript T represents the transpose operation.

[0159] After performing convolution and distillation operations based on the probabilistic sparse attention values ​​and preset weights, a new first matrix and a new second matrix are obtained. The convolution and distillation operations are as follows:

[0160]

[0161] Among them, [g] AB This represents the value obtained after passing through an attention module. Conv1d is a one-dimensional convolution, ELU is the activation function of the activation layer, and MaxPool is max pooling. At time t The data after the (j+1)th distillation layer is the new first matrix.

[0162] In this embodiment, in the global trend feature acquisition branch, after inputting the historical output reconstruction data matrix and the historical environment reconstruction data matrix into the global trend feature acquisition branch, feature extraction is performed on the data to obtain the first matrix. It should be noted that if the input data has D feature variables, the filter scale is N, and the number of filters is m g The dimension of the first matrix obtained is D×m. g After extracting features from each data point using a local mutation feature layer, a second matrix is ​​obtained. If the input data has D feature variables, the filter scale is L, the length of L is less than N, and the number of filters is m. l After obtaining the branch through local mutation features, the second matrix has a dimension of D×m. l .

[0163] The vector matrix obtained after the first matrix is ​​transformed through the embedding layer is used to randomly generate three weight matrices W. q W k W v Multiply the first matrix by each of the three weight matrices to obtain the first vector matrix, the second vector matrix, and the third vector matrix corresponding to the first matrix.

[0164] Q1 = Wq G1

[0165] K1 = W k G1

[0166] V1 = W v G1

[0167] Where Q1 represents the first vector matrix, K1 represents the second vector matrix, V1 represents the third vector matrix, and W... q Let W represent the first weight matrix. k Let W represent the second weight matrix. v G1 represents the weight of the third matrix, and G1 represents the weight of the first matrix.

[0168] Randomly select L Q lnL K The approximate evaluation criterion formula for calculating the sparsity of each vector in the first vector matrix is ​​as follows:

[0169]

[0170] Among them, L Q and L K Represented as the first vector matrix Second vector matrix The length of q i and k j These are the i-th and j-th data points of the first vector matrix Q1 and the second vector matrix K1, respectively. The superscript T represents the transpose operation, and d represents the dimension of matrices Q1, K1, and V1.

[0171] Based on the approximation evaluation criteria for each vector in the first vector matrix, the portion of the first vector matrix that contributes the most to the attention value is selected as the fifth matrix. The probabilistic sparse attention value is then obtained from the fifth matrix using the attention value calculation formula:

[0172]

[0173] in, Let represent the probabilistic sparse attention value, Q1 represent the first vector matrix, K1 represent the second vector matrix, V1 represent the third vector matrix, d represent the dimension of matrices Q1, K1, and V1, and the superscript T represents the transpose operation.

[0174] By setting higher weights for the main features, and performing the following operations on layers j to j+1 of the distillation layer, a new first matrix is ​​obtained:

[0175]

[0176] Among them, [g]AB This represents the value obtained after passing through an attention module. Conv1d is a one-dimensional convolution, ELU is the activation function of the activation layer, and MaxPool is max pooling. At time t The data after the (j+1)th distillation layer is the new first matrix.

[0177] Then, the vector matrix obtained by the second matrix after the embedding layer transformation is used to randomly generate three weight matrices M. q M k M v Multiplying the second matrix by the three weight matrices respectively yields the first vector matrix, the second vector matrix, and the third vector matrix corresponding to the second matrix.

[0178] Q2 = M q G2

[0179] K2 = M k G2

[0180] V2 = M v G2

[0181] Where Q2 represents the first vector matrix, K2 represents the second vector matrix, V2 represents the third vector matrix, and M... q Let M represent the first weight matrix. k M represents the second weight matrix. v This represents the weights of the third matrix.

[0182] Randomly select L Q lnL K The approximate evaluation criterion formula for calculating the sparsity of each vector in the first vector matrix is ​​as follows:

[0183]

[0184] Among them, L Q and L K Represented as the first vector matrix Second vector matrix The length of q i and k j These are the i-th and j-th data points of the first and second vector matrices, respectively. The superscript T represents the transpose operation, and d represents the dimensions of matrices Q2, K2, and V2.

[0185] Based on the approximation evaluation criteria of each vector in the first vector matrix, the part of the first vector matrix that contributes the most to the attention value is selected as the fifth matrix. The probabilistic sparse attention value is then obtained from the fifth matrix using the attention value calculation formula:

[0186]

[0187] in, Let Q2 represent the probabilistic sparse attention value, K2 represent the first vector matrix, V2 represent the second vector matrix, d represent the dimension of matrices Q2, K2, and V2, and the superscript T represents the transpose operation.

[0188] By setting higher weights for the main features, and performing the following operations on layers j to j+1 of the distillation layer, a new second matrix is ​​obtained:

[0189]

[0190] Among them, [g] AB This represents the value obtained after passing through an attention module. Conv1d is a one-dimensional convolution, ELU is the activation function of the activation layer, and MaxPool is max pooling. At time t Data after the (j+1)th distillation layer.

[0191] Steady-state features are used to extract the data from each data point, resulting in the third matrix, calculated using the following formula:

[0192] x t =φ0+φ1x t-1 +φ2x t-2 +L+φ p x t-p +ε t

[0193] Where, x t φ represents the data value at time t. i (i = 1, ..., p) are the coefficients of the autoregressive model, ε t This indicates that the mean is 0 and the variance is σ. 2 The white noise is p, where p represents the data length of the current prediction time value.

[0194] The new first matrix obtained from the global trend feature acquisition branch and the new second matrix obtained from the local mutation feature acquisition branch are input into the fully connected layer to obtain the feature matrix F. t Then, the steady-state characteristic x t With characteristic matrix F t The weighted fusion is used to arrive at the final wind power output or solar power output prediction result.

[0195] Step 103: After solving the historical output data, historical load data and historical scheduling strategies through the adaptive prediction model to obtain the second output prediction result, the second output prediction result is obtained.

[0196] Optionally, the historical output data, historical load data, and historical scheduling strategies are used to solve for the reverse distributed energy output prediction using an adaptive prediction model to obtain the second output prediction result, specifically:

[0197] By extending the historical scheduling strategy, a new historical scheduling strategy is obtained;

[0198] An adaptive prediction model is constructed based on historical load data, new historical scheduling strategies, historical cut-in wind speed, historical cut-out wind speed, and historical ambient light intensity. The adaptive prediction model is then solved to obtain the second power output prediction result. The adaptive prediction model includes a wind power adaptive prediction model and a photovoltaic adaptive prediction module. The second power output prediction result includes the wind power output prediction result and the photovoltaic power output prediction result.

[0199] Optionally, an adaptive prediction model is constructed based on historical load data, new historical scheduling strategies, historical cut-in wind speeds, historical cut-out wind speeds, and historical ambient light intensity. This adaptive prediction model is then solved to obtain the second power output prediction result, specifically:

[0200] An adaptive wind prediction model is constructed based on historical load data, new historical scheduling strategies, historical cut-in wind speeds, and historical cut-out wind speeds. The adaptive wind prediction model is as follows:

[0201] P WH =β Wi X Wi +β W

[0202] Among them, P WH This represents the predicted wind power output, β. Wi (i = 1, 2, 3, 4) represents the regression coefficient between wind power output and related factors, β W X represents the overall error of the wind power output model. Wi X represents the wind power output prediction factor matrix. Wi =[P LH D H V ciH V coH ], where P LH Represents historical load data, D H This indicates a new historical scheduling strategy, V ciH Indicates historical cut-in wind speed and V coH History cuts out the wind speed;

[0203] Solve for P using the least squares method. WH =β Wi X Wi +β W Regression coefficients, obtaining initial values ​​of the regression coefficients. Further introduce weights Then the weight vector It can be represented as:

[0204]

[0205] renew x j It is X Wi Variables in

[0206] For all regularization factors λ n Solve

[0207] calculate Will Substitute P WH =β Wi X Wi +β W The wind power output prediction results were obtained.

[0208] For photovoltaic power output, an adaptive prediction model based on historical load, new historical dispatch strategies, and historical ambient light intensity should be established:

[0209] P PVH =β PVi X PVi +β PV

[0210] Among them, P PVH This represents the photovoltaic output regression model, β PVi (i = 1, 2, 3) represents the regression coefficient between photovoltaic power output and related factors, β PV X represents the overall error of the photovoltaic power output model. PVi X represents the photovoltaic power output prediction factor matrix. PVi =[P LH D H ,I H ], where P LH Represents historical load data, D H This indicates the new historical scheduling strategy and I H Indicates historical ambient light intensity;

[0211] Solve for P using the least squares method. PVH =β PVi X PVi +β PV The regression coefficients were obtained, and the initial values ​​of the regression coefficients were obtained. Further introduce weights Then the weight vector It can be represented as:

[0212]

[0213] renew x j It is X PVi Variables in

[0214] For all regularization factors λ n Solve

[0215] calculate Will Substitute P PVH =β PVi X PVi +β PV The photovoltaic power output prediction results were obtained.

[0216] In this embodiment, as Figure 4 As shown, for the historical dispatch strategy of distributed energy, it is first expanded to have the same dimension as other historical data. Then, after preprocessing the historical output data, historical load data and historical environmental data of distributed energy, a linear regression model of output and load, dispatch strategy and environment is established. The adaptive prediction model includes wind power adaptive prediction model and photovoltaic adaptive prediction model.

[0217] For wind power output, an adaptive prediction model based on historical load, new historical scheduling strategies, historical cut-in wind speed, and historical cut-out wind speed is established:

[0218] P WH =β Wi X Wi +β W

[0219] Among them, P WH This represents the predicted wind power output, β. Wi (i = 1, 2, 3, 4) represents the regression coefficient between wind power output and related factors, β W X represents the overall error of the wind power output model. Wi X represents the wind power output prediction factor matrix. Wi =[P LH D H V ciH V coH ], where P LH Represents historical load data, D H This indicates a new historical scheduling strategy, V ciH Indicates historical cut-in wind speed and V coH History cuts out the wind speed;

[0220] First, the regression coefficients of the wind force adaptive prediction model are solved using the least squares method to obtain the initial values ​​of the regression coefficients. Further introduce weights Then the weight vector It can be represented as:

[0221]

[0222] Next, updates x j It is X Wi Variables in;

[0223] Furthermore, for all regularization factors λ n Solve

[0224] Further calculations:

[0225] Will Substitute P WH =β Wi X Wi +β W This means obtaining a prediction of wind power output;

[0226] For photovoltaic power output, an adaptive prediction model based on historical load, new historical dispatch strategies, and historical ambient light intensity should be established:

[0227] P PVH =β PVi X PVi +β PV

[0228] Among them, P PVH This represents the photovoltaic output regression model, β PVi (i = 1, 2, 3) represents the regression coefficient between photovoltaic power output and related factors, β PV X represents the overall error of the photovoltaic power output model. PVi X represents the photovoltaic power output prediction factor matrix. PVi =[P LH D H ,I H ], where P LH Represents historical load data, D H This indicates the new historical scheduling strategy and I H Indicates historical ambient light intensity;

[0229] First, the regression coefficients of the photovoltaic adaptive prediction model are solved using the least squares method to obtain the initial values ​​of the regression coefficients. Further introduce weights Then the weight vector It can be represented as:

[0230]

[0231] Next, updates x j It is X PVi Variables in;

[0232] Furthermore, for all regularization factors λ n Solve

[0233] Further calculations Will Substitute P PVH =β PVi X PVi +β PV This means obtaining a prediction of photovoltaic power output.

[0234] Step 104: Perform a composite weighted fusion of the first and second output prediction results to obtain the third output prediction result.

[0235] Optionally, the first and second output prediction results are combined and weighted to obtain the third output prediction result, specifically:

[0236] The first and second power output prediction results are combined using a weighted average method to obtain the third power output prediction result. The specific calculation formula is as follows:

[0237] P HP =θP HP1 +(1-θ)P HP2

[0238] Among them, P HP P represents the predicted third output. HP1 P represents the first output prediction result. HP2 This represents the second output prediction result, and θ represents the coefficient of power predicted using the forward prediction method.

[0239] In this embodiment, the first prediction result and the second prediction result are weighted and averaged to fuse the power output prediction. The wind power output prediction result is as follows:

[0240] P WTP =θ1P WP1 +(1-θ1)P WP2

[0241] The photovoltaic power output forecast results are as follows:

[0242] P PVP =θ2P PVP1 +(1-θ2)P PVP2

[0243] Among them, P WTPP represents the predicted value of the combined wind power output. WP1 P represents the first wind force forecast result. WP2 P represents the predicted wind power output. PVP P represents the predicted value of photovoltaic composite power output. PVP1 This represents the predicted output of the first photovoltaic power plant, P. PVP2 The values ​​represent the predicted power output of photovoltaic power, where θ1 is the coefficient of the power output of wind power predicted using the forward prediction method, and θ2 is the coefficient of the power output of photovoltaic power predicted using the forward prediction method.

[0244] Step 105: Construct a distributed energy optimization scheduling model using the third power output prediction results. Use the distributed energy optimization scheduling model to optimize the scheduling of historical power output data and energy storage equipment operation data to obtain scheduling results, so that the distribution network can schedule energy according to the scheduling results. The scheduling results include photovoltaic power generation power value, wind power generation power value, energy storage system operation power value, and load shedding power value.

[0245] Optionally, the power generation cost can be constructed based on the third prediction result and the power generation cost coefficient, where the power generation cost is either the photovoltaic power generation cost or the wind power generation cost.

[0246] The operating and management cost of energy storage equipment is obtained by constructing the charging operating cost coefficient, discharging operating cost coefficient, charging efficiency, discharging efficiency, charging power, and discharging power.

[0247] The environmental governance cost of energy storage equipment is obtained by constructing the environmental governance cost coefficient, charging power, and discharging power of the energy storage equipment. The load shedding cost is obtained by constructing the load shedding penalty coefficient and load shedding power.

[0248] Based on the generation cost, energy storage equipment operation and management cost, energy storage equipment environmental governance cost, and load shedding cost, an objective function and constraints are constructed for the distributed energy optimization scheduling model. The objective function is:

[0249] min(C PV +C WT +C SV +C SVE +C loss )

[0250] Among them, C PV For the cost of photovoltaic power generation, C WT For the cost of wind power generation, C SV For the operation and management costs of energy storage equipment, C SVE For the environmental governance costs of energy storage equipment, C loss For load shedding costs;

[0251] The objective function and constraints are solved using the alternating direction multiplier method to obtain the scheduling results, which include photovoltaic power generation, wind power generation, energy storage system operating power, and load shedding power.

[0252] In this embodiment, the third prediction result, namely the third predicted power, is used to calculate the power generation cost:

[0253] The cost of photovoltaic power generation is:

[0254] C PV =k PV P PVT

[0255] Where, k PV It is the cost coefficient of photovoltaic power generation, P PVT It is photovoltaic power generation.

[0256] Constraints: P PVmin ≤P PVT ≤P PVP ;

[0257] The cost of wind power generation is:

[0258] C WT =k WT P WTT

[0259] Where, k WT It is the cost coefficient for wind power generation, P WTT It is the amount of wind power generated;

[0260] The constraint is: P WTmin ≤P WTT ≤P WTP ;

[0261] The calculation of energy storage equipment operation and management costs and environmental governance costs includes energy storage equipment operation and management costs and energy storage equipment environmental governance costs.

[0262] The operation and management cost of energy storage equipment is:

[0263] C SV =k ch η ch P ch +k dis η dis P dis

[0264] Where, k ch and k dis These are the cost coefficients for charging and discharging operations, η and η, respectively. ch and η dis These are the charging and discharging efficiencies, P. chand P dis The power of charging and discharging respectively;

[0265] The environmental remediation cost of energy storage equipment is:

[0266] C SVE =k SVE (P ch +P dis )

[0267] Where, k SVE P is the environmental governance cost coefficient for energy storage equipment operation. ch P represents the charging power of the device. dis This is expressed as the discharge power of the energy storage device;

[0268] The constraint is: P ch,min ≤P ch ≤P ch,max and P dis,min ≤P dis ≤P dis,max ;

[0269] Among them, P ch,min and P ch,max P represents the minimum and maximum power required to charge the energy storage device, respectively. dis,min and P dis,max These represent the minimum and maximum discharge power of the energy storage device, respectively.

[0270] The load shedding cost is:

[0271] C loss =k loss P loss

[0272] Where, k loss It is the load shedding penalty factor, P loss It is the load shedding power;

[0273] Constraint: 0 ≤ P loss ≤P loss,max ;

[0274] Among them, P loss,max It is the maximum allowable load shedding capacity;

[0275] The problem of optimizing the dispatch of massive distributed energy resources is defined as minimizing the overall generation cost, environmental governance cost, and load shedding cost, resulting in the objective function for optimizing the dispatch of massive distributed energy resources:

[0276] min(C PV +C WT +C SV +C SVE +C loss )

[0277] The overall power balance constraints are:

[0278]

[0279] Among them, P L1 As a constraint condition during charging, P L2 These are the constraints during discharge.

[0280] Finally, the optimization problem is solved using the alternating direction multiplier method to obtain the values ​​of photovoltaic power generation, wind power generation, energy storage system charging or discharging power, and load shedding power, thus realizing system optimization scheduling based on distributed energy output prediction.

[0281] This method acquires historical power output data, historical environmental data, historical load data, and historical scheduling strategies within the monitored area. It then performs orthogonal wavelet transform and reconstruction processing on the historical power output data and historical environmental data to obtain historical power output reconstruction data matrices, historical environmental reconstruction data matrices, and wind speed single-branch reconstruction data matrices. These matrices are then used to extract features and weighted fuse using a prediction model based on global and abrupt change features to obtain a first power output prediction result. Next, the historical power output data is used to perform reverse distributed energy power output prediction using an adaptive prediction model to obtain a second power output prediction result. Finally, the first and second power output prediction results are combined and weighted to obtain a third power output prediction result. This third power output prediction result is then used to construct a distributed energy optimal scheduling model, which solves for the historical power output data, historical environmental data, and historical load data to obtain the scheduling result. By effectively integrating the environment, load, and scheduling, and adopting bidirectional forecasting to further improve the accuracy of distributed energy output forecasting, the optimization of scheduling is based on the forecasted output, making the overall scheduling of massive distributed energy resources more optimized.

[0282] Example 2

[0283] Accordingly, see Figure 5 , Figure 5 This is a schematic diagram of the distributed energy bidirectional predictive optimization scheduling system provided by the present invention. As shown in the figure, the distributed energy bidirectional predictive optimization scheduling system includes an acquisition module 501, a first prediction module 502, a second prediction module 503, a third prediction module 504, and a scheduling result module 505. The specific units of each module are as follows:

[0284] The acquisition module 501 is used to acquire historical power output data, historical environmental data, historical load data, and historical dispatch strategies within the area to be monitored. The historical power output data includes historical wind power output data and historical photovoltaic power output data.

[0285] The first prediction module 502 is used to perform orthogonal wavelet transform and reconstruction processing on historical power output data and historical environment data respectively to obtain historical power output reconstruction data matrix and historical environment reconstruction data matrix. The historical power output reconstruction data matrix and the historical environment reconstruction data matrix are then used to extract features and perform weighted fusion through a prediction model based on global features and abrupt change features to obtain the first power output prediction result.

[0286] The second prediction module 503 is used to solve the historical output data, the historical load data and the historical scheduling strategy through an adaptive prediction model to obtain the second output prediction result.

[0287] The third prediction module 504 is used to perform a composite weighted fusion of the first output prediction result and the second output prediction result to obtain the third output prediction result.

[0288] The scheduling result module 505 is used to construct a distributed energy optimal scheduling model using the third output prediction result, and solve the distributed energy optimal scheduling model to obtain the scheduling result, so that the distribution network can schedule energy according to the scheduling result. The scheduling result includes photovoltaic power generation value, wind power generation value, energy storage system operating power value, and load shedding power value.

[0289] Optionally, the first prediction module 502 includes a normalization unit 5021, a decomposition unit 5022, a reconstruction unit 5023, and a matrix unit 5024.

[0290] The normalization unit 5021 is used to normalize the various data in the historical power output data and historical environmental data within the acquired preset time period to obtain normalized historical power output data and normalized historical environmental data. The historical environmental data includes historical ambient light intensity data, historical cut-in wind speed data, and historical cut-out wind speed data. The normalization process is as follows:

[0291]

[0292] Among them, P WH This indicates historical wind power output data for renewable energy, P PVH Indicates historical photovoltaic power generation output data, I H Indicates historical ambient light intensity, V ciH Indicates the historical entry wind speed, V coH Indicates historical cut-out wind speed, P WH,min ,P PVH,min ,P LH,min ,I H,min V ciH,min and V coH,minand P WH,max ,P PVH,max ,I H,max V ciH,max and V coH,max Let P′ represent the historical wind power output data, historical photovoltaic power output data, historical load data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed minimum and maximum values, respectively. WH 、P′ PVH 、I' H V′ ciH and V′ coH These are represented as normalized historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed, respectively.

[0293] The decomposition unit 5022 is used to perform positive wavelet transform decomposition on each data point in the normalized historical power output data and normalized historical environmental data according to a preset transform formula, to obtain the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients corresponding to each data point in the normalized historical power output data and normalized historical environmental data. The preset transform formula is:

[0294]

[0295]

[0296] Where x(t) is the signal to be processed. Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) is the real part of the wavelet function, ψ g (t) is the imaginary part wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0297] Reconstruction unit 5023 is used to calculate, using the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients, the wavelet coefficient reconstruction signal and scaling coefficient reconstruction signal corresponding to each data point through a reconstruction formula. The reconstruction formula is as follows:

[0298]

[0299]

[0300] Where, d j (t) represents the reconstructed signal of a certain wavelet coefficient, c J(t) represents the reconstructed signal for a certain data scale coefficient, t = 1, 2, L, M, where M is the length of the signal and n is the sequence number of the signal, n = 1, 2, L, M; Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. ψ represents the scaling factor of the imaginary part tree decomposition. h (t) represents the real part of the wavelet function, ψ g (t) represents the imaginary part of the wavelet function, J represents the number of wavelet decomposition levels, and j represents the scaling factor;

[0301] Matrix unit 5024 is used to construct matrices based on the wavelet coefficients and scale coefficients of each data to reconstruct the signal and obtain the historical output reconstruction data matrix and the historical environment reconstruction data matrix.

[0302] Optionally, the first prediction module 502 includes a third matrix unit 5025, a convolution unit 5026, and a prediction unit 5027.

[0303] The third matrix unit 5025 is used to input the historical output reconstruction data matrix and the historical environment reconstruction data matrix into the prediction model based on global features and abrupt change features. It extracts data from each data point using the global trend feature layer to obtain the first matrix, extracts data from each data point using the local abrupt change feature layer to obtain the second matrix, and extracts data from each data point using steady-state features to obtain the third matrix. The steady-state feature calculation formula is as follows:

[0304] x t =φ0+φ1x t-1 +φ2x t-2 +L+φ p x t-p +ε t

[0305] Where, x t φ represents the data at time t, i.e., the extracted data. i (i = 1, ..., p) are the coefficients of the autoregressive model, ε t This indicates that the mean is 0 and the variance is σ. 2 The white noise, where p represents the data length of the current prediction time value;

[0306] The convolution unit 5026 is used to perform convolution and distillation operations on the first matrix and the second matrix respectively to obtain a new first matrix and a new second matrix;

[0307] The prediction unit 5027 is used to fuse the new first matrix and the new second matrix to obtain a fourth matrix, and then to perform a weighted fusion of the third matrix and the fourth matrix to obtain a first output prediction result.

[0308] Optionally, the convolutional unit 5026 includes a transformation subunit 50261, an evaluation subunit 50262, an attention value calculation subunit 50263, and a convolutional subunit 50264.

[0309] The transformation subunit 50261 is used to sequentially transform the first matrix and the second matrix into vector matrices, and then multiply the first matrix by a plurality of preset weight matrices to obtain the first vector matrix, the second vector matrix and the third vector matrix corresponding to the first matrix.

[0310] Evaluation subunit 50262 is used to perform approximation evaluation calculations on the first vector matrix based on a preset number of dot products, to obtain the approximation evaluation criteria for each vector in the first vector matrix, wherein the approximation evaluation criteria are:

[0311]

[0312] Among them, L Q and L K Represented as the first vector matrix Second vector matrix The length of q i and k j These are the i-th and j-th data points of the first vector matrix Q1 and the second vector matrix K1, respectively. The superscript T represents the transpose operation, and d represents the dimension of matrices Q1, K1, and V1.

[0313] The attention value calculation subunit 50263 is used to obtain the contribution value of each vector to the attention value calculation based on the approximation evaluation criteria of each vector in the first vector matrix, select the vector with the largest contribution value as the fifth matrix, and obtain the probabilistic sparse attention value according to the attention value calculation formula based on the fifth matrix, where the probabilistic sparse attention value is:

[0314]

[0315] in, Let represent the probabilistic sparse attention value, Q1 represent the first vector matrix, K1 represent the second vector matrix, V1 represent the third vector matrix, d represent the dimension of matrices Q1, K1, and V1, and the superscript T represents the transpose operation.

[0316] The convolutional subunit 50264 is used to perform convolution and distillation operations based on the probabilistic sparse attention value and preset weights to obtain a new first matrix and a new second matrix. The convolution and distillation operations are as follows:

[0317]

[0318] Among them, [g] AB This represents the value obtained after passing through an attention module. Conv1d is a one-dimensional convolution, ELU is the activation function of the activation layer, and MaxPool is max pooling. At time t The data after the (j+1)th distillation layer is the new first matrix.

[0319] Optionally, the second prediction module 503 includes an extension unit 5031 and a second prediction unit 5032.

[0320] Among them, the extension unit 5031 is used to extend the historical scheduling strategy to obtain a new historical scheduling strategy;

[0321] The second prediction unit 5032 is used to construct an adaptive prediction model based on historical load data, new historical scheduling strategies, historical cut-in wind speed, historical cut-out wind speed and historical ambient light intensity, and solve the adaptive prediction model to obtain the second power output prediction result. The adaptive prediction model includes a wind power adaptive prediction model and a photovoltaic adaptive prediction module, and the second power output prediction result includes wind power output prediction result and photovoltaic power output prediction result.

[0322] Optionally, the second prediction module 503 is further configured to construct an adaptive prediction model based on historical load data, new historical scheduling strategies, historical cut-in wind speeds, historical cut-out wind speeds, and historical ambient light intensity, and solve the adaptive prediction model to obtain the second output prediction result, specifically:

[0323] An adaptive wind prediction model is constructed based on historical load data, new historical scheduling strategies, historical cut-in wind speeds, and historical cut-out wind speeds. The adaptive wind prediction model is as follows:

[0324] P WH =β Wi X Wi +β W

[0325] Among them, P WH This represents the predicted wind power output, β. Wi (i = 1, 2, 3, 4) represents the regression coefficient between wind power output and related factors, β W X represents the overall error of the wind power output model. Wi X represents the wind power output prediction factor matrix. Wi =[P LH D H V ciH V coH ], where PLH Represents historical load data, D H This indicates a new historical scheduling strategy, V ciH Indicates historical cut-in wind speed and V coH History cuts out the wind speed;

[0326] Solve for P using the least squares method. WH =β Wi X Wi +β W Regression coefficients, obtaining initial values ​​of the regression coefficients. Further introduce weights Then the weight vector It can be represented as:

[0327]

[0328] renew x j It is X Wi Variables in

[0329] For all regularization factors λ n Solve

[0330] calculate Will Substitute P WH =β Wi X Wi +β W This yields the predicted wind power output.

[0331] For photovoltaic power output, an adaptive prediction model based on historical load, new historical dispatch strategies, and historical ambient light intensity should be established:

[0332] P PVH =β PVi X PVi +β PV

[0333] Among them, P PVH This represents the photovoltaic output regression model, β PVi (i = 1, 2, 3) represents the regression coefficient between photovoltaic power output and related factors, β PV X represents the overall error of the photovoltaic power output model. PVi X represents the photovoltaic power output prediction factor matrix. PVi =[P LH D H ,I H ], where P LH Represents historical load data, D H This indicates the new historical scheduling strategy and I H Indicates historical ambient light intensity;

[0334] Solve for P using the least squares method. PVH =β PVi X PVi +β PV The regression coefficients were obtained, and the initial values ​​of the regression coefficients were obtained. Further introduce weights Then the weight vector It can be represented as:

[0335]

[0336] renew x j It is X PVi Variables in

[0337] For all regularization factors λ n Solve

[0338] calculate Will Substitute P PVH =β PVi X PVi +β PV The photovoltaic power output prediction results were obtained;

[0339] Optionally, the third prediction module 504 includes a prediction unit 5041.

[0340] Prediction unit 5041 is used to perform a composite weighted fusion of the first output prediction result and the second output prediction result to obtain a third output prediction result, specifically:

[0341] The first power output prediction result and the second power output prediction result are fused using a weighted average method to obtain the third power output prediction result. The specific calculation formula is as follows:

[0342] P HP =θP HP1 +(1-θ)P HP2

[0343] Among them, P HP P represents the predicted third output. HP1 P represents the first output prediction result. HP2 This represents the second output prediction result, and θ represents the coefficient of power predicted using the forward prediction method.

[0344] The scheduling result module 505 includes a power generation cost unit 5051, an energy storage device operation and management cost unit 5052, a load shedding cost unit 5053, a scheduling model construction unit 5054, and a calculation unit 5055.

[0345] The power generation cost unit 5051 is used to construct the power generation cost based on the third prediction result and the power generation cost coefficient, wherein the power generation cost is either the photovoltaic power generation cost or the wind power generation cost;

[0346] The energy storage equipment operation and management cost 5052 is constructed based on the charging operation cost coefficient, discharging operation cost coefficient, charging efficiency, discharging efficiency, charging power, and discharging power to obtain the energy storage equipment operation and management cost;

[0347] The load shedding cost 5053 is used to construct the environmental governance cost of the energy storage device based on the operating environment governance cost coefficient, the charging power of the energy storage device, and the discharging power of the energy storage device, so as to obtain the environmental governance cost of the energy storage device. It is also constructed based on the load shedding penalty coefficient and the load shedding power to obtain the load shedding cost.

[0348] The scheduling model construction unit 5054 is used to construct the objective function and constraints of the distributed energy optimization scheduling model based on the generation cost, energy storage equipment operation and management cost, energy storage equipment environmental governance cost, and load shedding cost. The objective function is:

[0349] min(C PV +C WT +C SV +C SVE +C loss )

[0350] Among them, C PV For the cost of photovoltaic power generation, C WT For the cost of wind power generation, C SV For the operation and management costs of energy storage equipment, C SVE For the environmental governance costs of energy storage equipment, C loss For load shedding costs;

[0351] The computing unit 5055 is used to solve the objective function and constraints using the alternating direction multiplier method to obtain the scheduling results, which include photovoltaic power generation power, wind power generation power, energy storage system operating power, and load shedding power.

[0352] For a more detailed explanation of the working principle and procedures of this embodiment, please refer to the relevant description in Embodiment 1.

[0353] Compared to existing technologies, this method acquires historical power output data, historical environmental data, historical load data, and historical scheduling strategies within the monitored area. Orthogonal wavelet transform and reconstruction processing are then performed on the historical power output data and historical environmental data to obtain historical power output reconstruction data matrices, historical environmental reconstruction data matrices, and wind speed single-branch reconstruction data matrices. These matrices are then used for feature extraction and weighted fusion through a prediction model based on global and abrupt change features to obtain a first power output prediction result. The historical power output data is then used to perform reverse distributed energy power output prediction using an adaptive prediction model to obtain a second power output prediction result. A composite weighted fusion of the first and second power output prediction results yields a third power output prediction result. This third power output prediction result is used to construct a distributed energy optimal scheduling model, which then solves for the historical power output data, historical environmental data, and historical load data to obtain the scheduling result. This method effectively integrates environment, load, and scheduling, and uses bidirectional forecasting to further improve the accuracy of distributed energy output forecasting. It also establishes optimized scheduling based on forecasted output, making the overall scheduling of massive distributed energy resources more optimized.

[0354] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for bidirectional predictive optimization scheduling of distributed energy resources, characterized in that, include: S1: Obtain historical power output data, historical environmental data, historical load data, and historical dispatch strategies within the area to be monitored, wherein the historical power output data includes historical wind power output data and historical photovoltaic power output data; S2: Perform orthogonal wavelet transform and reconstruction processing on the historical power output data and the historical environment data respectively to obtain historical power output reconstruction data matrix and historical environment reconstruction data matrix. Then, perform feature extraction and weighted fusion on the historical power output reconstruction data matrix and the historical environment reconstruction data matrix through a prediction model based on global features and local mutations to obtain the first power output prediction result. S3: After solving the historical output data, the historical load data, and the historical scheduling strategy through the adaptive prediction model to perform reverse distributed energy output prediction, the second output prediction result is obtained. S4: Perform a composite weighted fusion of the first power output prediction result and the second power output prediction result to obtain the third power output prediction result; S5: Construct a distributed energy optimization scheduling model using the third output prediction results, and solve the distributed energy optimization scheduling model to obtain scheduling results, so that the distribution network can schedule energy according to the scheduling results, wherein the scheduling results include photovoltaic power generation power value, wind power generation power value, energy storage system operating power value, and load shedding power value.

2. The distributed energy bidirectional predictive optimization scheduling method as described in claim 1, characterized in that, The process of performing orthogonal wavelet transform and reconstruction on the historical power output data and the historical environment data respectively to obtain the historical power output reconstruction data matrix and the historical environment reconstruction data matrix is ​​as follows: The historical output data and historical environmental data within the acquired preset time period are normalized to obtain normalized historical output data and normalized historical environmental data. The historical environmental data includes historical ambient light intensity data, historical cut-in wind speed data, and historical cut-out wind speed data. The normalization process is as follows: in, This indicates historical wind power output data for renewable energy sources. Indicates historical photovoltaic power generation output data, Indicates historical ambient light intensity, Indicates the wind speed at which history is approached. Indicates historical cut-out wind speed, , , , and as well as , , , and These represent historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed (minimum and maximum values), respectively. , , , as well as These are represented as normalized historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed, respectively. The normalized historical power output data and the normalized historical environmental data are respectively decomposed using a preset transformation formula using positive wavelet transform to obtain the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients corresponding to each data point in the normalized historical power output data and the normalized historical environmental data. The preset transformation formula is: in, This is a signal to be processed. Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. This represents the scaling factor for the virtual part tree decomposition. It is a wavelet function with the real part. It is the imaginary part wavelet function. J This indicates the number of wavelet decomposition levels. j Indicates the scale factor; Based on the wavelet coefficients of the real part tree decomposition and the wavelet coefficients of the imaginary part tree decomposition, the wavelet coefficient reconstruction signal and the scaling coefficient reconstruction signal corresponding to each data point are obtained through the reconstruction formula, wherein the reconstruction formula is: in, This represents a signal reconstructed from wavelet coefficients of a given data set. Represented as a reconstructed signal based on a certain data scaling coefficient. , M To reconstruct the length of the signal, ; Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. This represents the scaling factor for the virtual part tree decomposition. Denotes the real part of the wavelet function. Denotes the imaginary part of the wavelet function. J This indicates the number of wavelet decomposition levels. j Indicates the scale factor; Based on the wavelet coefficient reconstructed signal and the scale coefficient reconstructed signal of each data, a matrix is ​​constructed to obtain the historical output reconstructed data matrix and the historical environment reconstructed data matrix.

3. The distributed energy bidirectional predictive optimization scheduling method as described in claim 1, characterized in that, The process of extracting features and weighting and fusing the historical output reconstruction data matrix and the historical environment reconstruction data matrix using a prediction model based on global features and local mutations to obtain the first output prediction result is as follows: The historical output reconstruction data matrix and the historical environment reconstruction data matrix are input into a prediction model based on global features and local mutations. A global trend feature layer is used to extract data from each data point to obtain a first matrix. A local mutation feature layer is used to extract data from each data point to obtain a second matrix. Steady-state features are used to extract data from each data point to obtain a third matrix. The steady-state feature calculation formula is as follows: in, express t The data at any given time, i.e., the extracted data. These are the coefficients of the autoregressive model. , This indicates that the mean is 0 and the variance is . White noise, p This indicates the length of the data representing the current predicted time value; After performing convolution and distillation operations on the first matrix and the second matrix respectively, a new first matrix and a new second matrix are obtained; The new first matrix and the new second matrix are fused to obtain the fourth matrix. Then, the third matrix and the fourth matrix are weighted and fused to obtain the first output prediction result.

4. The distributed energy bidirectional predictive optimization scheduling method as described in claim 3, characterized in that, After performing convolution and distillation operations on the first matrix and the second matrix respectively, new first matrix and new second matrix are obtained, specifically as follows: After converting the first matrix and the second matrix into vector matrices in sequence, the first matrix is ​​multiplied by a plurality of preset weight matrices to obtain the first vector matrix, the second vector matrix and the third vector matrix corresponding to the first matrix; The approximation evaluation of the first vector matrix is ​​calculated based on a preset number of dot products to obtain the approximation evaluation criteria for each vector in the first vector matrix. Based on the approximation evaluation criteria of each vector in the first vector matrix, the contribution value of each vector to the calculation of the attention value is obtained. The vector with the largest contribution value is selected as the fifth matrix, and the probability sparse attention value is obtained by using the attention value calculation formula based on the fifth matrix. After performing convolution and distillation operations based on the probabilistic sparse attention value and preset weights, a new first matrix and a new second matrix are obtained.

5. The distributed energy bidirectional predictive optimization scheduling method as described in claim 2, characterized in that, The second output prediction result is obtained by solving the historical output data, historical load data, and historical scheduling strategy through an adaptive prediction model to perform reverse distributed energy output prediction. Specifically: The historical scheduling strategy is extended to obtain a new historical scheduling strategy; An adaptive prediction model is constructed based on the historical load data, the new historical scheduling strategy, the historical cut-in wind speed, the historical cut-out wind speed, and the historical ambient light intensity. The adaptive prediction model is then solved to obtain a second power output prediction result. The adaptive prediction model includes a wind power adaptive prediction model and a photovoltaic adaptive prediction module. The second power output prediction result includes wind power output prediction results and photovoltaic power output prediction results.

6. The distributed energy bidirectional predictive optimization scheduling method as described in claim 5, characterized in that, An adaptive prediction model is constructed based on the historical load data, the new historical scheduling strategy, historical cut-in wind speed, historical cut-out wind speed, and historical ambient light intensity. The adaptive prediction model is then solved to obtain the second power output prediction result, specifically: An adaptive wind force prediction model is constructed based on the historical load data, the new historical scheduling strategy, the historical cut-in wind speed, and the historical cut-out wind speed. The regression coefficients of the wind adaptive prediction model are solved using the least squares method to obtain the initial values ​​of the regression coefficients. The wind power output prediction result is calculated using the regularization factor, the initial value of the regression coefficient, and the corresponding weight vector. A photovoltaic adaptive prediction model is constructed based on historical load, new historical scheduling strategies, and historical ambient light intensity. The regression coefficients of the photovoltaic adaptive prediction model are solved using the least squares method to obtain the initial values ​​of the regression coefficients. The photovoltaic power output prediction results are calculated using the regularization factor, the initial value of the regression coefficient, and the corresponding weight vector.

7. The distributed energy bidirectional predictive optimization scheduling method as described in claim 1, characterized in that, The process of performing a composite weighted fusion of the first power output prediction result and the second power output prediction result to obtain the third power output prediction result is as follows: The first power output prediction result and the second power output prediction result are fused using a weighted average method to obtain the third power output prediction result. The specific calculation formula is as follows: in, This indicates the predicted result of the third output. This indicates the first output prediction result. This indicates the second output prediction result. This represents the coefficient used to predict power using the forward prediction method.

8. The distributed energy bidirectional predictive optimization scheduling method as described in claim 1, characterized in that, The process involves constructing a distributed energy optimal scheduling model using the third power output prediction results, and solving the distributed energy optimal scheduling model to obtain the scheduling results. Specifically: The power generation cost is constructed based on the third output prediction result and the power generation cost coefficient, wherein the power generation cost is either the photovoltaic power generation cost or the wind power generation cost. The operating and management cost of energy storage equipment is obtained by constructing the charging operating cost coefficient, discharging operating cost coefficient, charging efficiency, discharging efficiency, charging power, and discharging power. The environmental governance cost of energy storage equipment is obtained by constructing the environmental governance cost coefficient, charging power, and discharging power of the energy storage equipment. The load shedding cost is obtained by constructing the load shedding penalty coefficient and load shedding power. The objective function and constraints of the distributed energy optimization scheduling model are constructed based on the power generation cost, the energy storage device operation and management cost, the energy storage device environmental governance cost, and the load shedding cost. The objective function is: in, For the cost of photovoltaic power generation, For the cost of wind power generation, For the operation and management costs of energy storage equipment, For the environmental governance costs of energy storage equipment, For load shedding costs; The objective function and constraints are solved using the alternating direction multiplier method to obtain the scheduling results, which include photovoltaic power generation, wind power generation, energy storage system operating power, and load shedding power.

9. A distributed energy bidirectional predictive optimization scheduling system, characterized in that, include: The acquisition module is used to acquire historical power output data, historical environmental data, historical load data, and historical dispatch strategies within the area to be monitored. The historical power output data includes historical wind power output data and historical photovoltaic power output data. The first prediction module is used to perform orthogonal wavelet transform and reconstruction processing on the historical power output data and the historical environment data respectively to obtain a historical power output reconstruction data matrix and a historical environment reconstruction data matrix. The historical power output reconstruction data matrix and the historical environment reconstruction data matrix are then used to extract features and perform weighted fusion through a prediction model based on global features and local mutations to obtain a first power output prediction result. The second prediction module is used to solve the historical output data, the historical load data and the historical scheduling strategy through an adaptive prediction model to obtain the second output prediction result. The third prediction module is used to perform a composite weighted fusion of the first power output prediction result and the second power output prediction result to obtain the third power output prediction result. The scheduling result module is used to construct a distributed energy optimization scheduling model using the third output prediction result, and to optimize the scheduling of the historical output data and energy storage equipment operation data using the distributed energy optimization scheduling model to obtain the scheduling result, so that the distribution network can schedule energy according to the scheduling result. The scheduling result includes photovoltaic power generation power value, wind power generation power value, energy storage system operation power value, and load shedding power value.

10. The distributed energy bidirectional predictive optimization scheduling system as described in claim 9, characterized in that, The first prediction module includes a normalization unit, a decomposition unit, a reconstruction unit, and a matrix unit. The normalization unit is used to normalize the historical output data and historical environmental data within a preset time period to obtain normalized historical output data and normalized historical environmental data. The historical environmental data includes historical ambient light intensity data, historical cut-in wind speed data, and historical cut-out wind speed data. The normalization process is as follows: in, This indicates historical wind power output data for renewable energy sources. Indicates historical photovoltaic power generation output data, Indicates historical ambient light intensity, Indicates the wind speed at which history is approached. Indicates historical cut-out wind speed, , , , and as well as , , , , and These represent historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed (minimum and maximum values), respectively. , , , as well as These are represented as normalized historical wind power output data, historical photovoltaic power output data, historical ambient light intensity, historical cut-in wind speed, and historical cut-out wind speed, respectively. The decomposition unit is used to perform positive wavelet transform decomposition on each data point in the normalized historical power output data and the normalized historical environmental data according to a preset transformation formula, to obtain the real part tree decomposition wavelet coefficients and imaginary part tree decomposition wavelet coefficients corresponding to each data point in the normalized historical power output data and the normalized historical environmental data, wherein the preset transformation formula is: in, This is a signal to be processed. Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. This represents the scaling factor for the virtual part tree decomposition. It is a wavelet function with the real part. It is the imaginary part wavelet function. J The value represents the number of wavelet decomposition layers, and j represents the scale factor. The reconstruction unit is used to calculate, using the real part tree decomposition wavelet coefficients and the imaginary part tree decomposition wavelet coefficients, the wavelet coefficient reconstruction signal and the scaling coefficient reconstruction signal corresponding to each data point through a reconstruction formula, wherein the reconstruction formula is: in, This represents a signal reconstructed from wavelet coefficients of a given data set. Represented as a reconstructed signal based on a certain data scaling coefficient. , M To reconstruct the length of the signal, ; Represents the wavelet coefficients of the real part tree decomposition. This represents the scaling factor for the real part tree decomposition. Represents the wavelet coefficients of the imaginary part tree decomposition. This represents the scaling factor for the virtual part tree decomposition. Denotes the real part of the wavelet function. Denotes the imaginary part of the wavelet function. J This indicates the number of wavelet decomposition levels. j Indicates the scale factor; The matrix unit is used to construct a matrix based on the wavelet coefficients reconstructing the signal and the scaling coefficients reconstructing the signal of each data, so as to obtain the historical output reconstructed data matrix and the historical environment reconstructed data matrix.

Images