Power regulation method and system for off-grid micro-grid

Abstract

The invention discloses a power regulation method for an off-grid micro-grid, and the method comprises the steps: predicting the power data of a plurality of distributed power supplies and a plurality of loads at a next moment according to the historical power data; performing power matching on the plurality of distributed power supplies and the plurality of loads according to the predicted power data of the distributed power supplies and the loads; and the power difference between the distributed power supply and the load is compensated through the energy storage system. The output and the load of the distributed power supply are predicted, on-line adjustment is realized, the prediction precision is improved, optimal source-load matching is realized according to a prediction result, relatively high power vacancy and power excess are avoided, the system stability is improved, and rapid power adjustment is realized by adopting coordinated control between the source, the storage and the load.

Description

Technical Field

[0001] The present invention relates to power regulation of a microgrid, and in particular to a power regulation method and system for an off-grid microgrid. Background Art

[0002] Due to its isolation from the mainland, islands have long faced shortages and restrictions on resources such as electricity and transportation. Off-grid microgrid technology provides a feasible solution for the power supply in islands. An off-grid microgrid is an independent power system built on distributed energy resources and energy storage equipment, which can operate independently in islands. By utilizing the abundant renewable energy resources such as wind and solar energy in islands, combined with off-grid microgrid technology, a clean, stable and sustainable power supply system can be built. Through off-grid microgrid technology, islands can achieve power self-sufficiency and reduce power supply costs.

[0003] However, the off-grid microgrid system is composed of multiple distributed energy resources and power loads. However, when faced with load fluctuations and changes in distributed energy output, traditional systems often have problems such as slow power regulation response and poor system stability. Summary of the invention

[0004] Purpose of the invention: In view of the above shortcomings, the present invention provides a power regulation method for an off-grid microgrid, which has a fast power regulation speed and improves the stability of the microgrid system.

[0005] The present invention also provides a power regulation system for an off-grid microgrid.

[0006] Technical solution: To solve the above problems, the present invention adopts a power regulation method for an off-grid microgrid, comprising:

[0007] For several distributed power sources and several loads, the power data at the next moment is predicted based on the historical power data;

[0008] Power matching is performed between several distributed power sources and several loads according to the predicted power data of distributed power sources and loads; and the power difference between the distributed power sources and the loads is compensated by the energy storage system.

[0009] Furthermore, for the prediction of distributed power generation and load power data, a benchmark prediction model of a long short-term memory neural network combined with a recurrent unit neural network is adopted. Predictions are made separately by the long short-term memory neural network and the recurrent unit neural network, and the prediction results of the two are weighted and summed to form the final prediction output of the benchmark prediction model.

[0010] Furthermore, before predicting the power data of distributed power sources and loads, the error margin between the power data collected by the distributed power sources and loads at the current moment and the power data predicted by the benchmark prediction model is determined. When one or more error margins in the distributed power source and load power data are greater than their corresponding error thresholds, the benchmark prediction model of the corresponding distributed power source or load whose error margin is greater than the error threshold is updated.

[0011] Furthermore, updating the benchmark prediction model includes: using a hedging algorithm and online learning to update the output weights of the long short-term memory neural network and the recurrent unit neural network in real time and back-propagating to update the benchmark prediction model of the distributed power supply or load; when the accuracy of the long short-term memory neural network is higher, increasing the output weight corresponding to the long short-term memory neural network, and when the accuracy of the recurrent unit neural network is higher, increasing the output weight corresponding to the recurrent unit neural network; forward propagating each module in the long short-term memory neural network and the recurrent unit neural network, the expression is:

[0012]

[0013] in, and They represent the initial values ​​of the long short-term memory neural network and the recurrent unit neural network, and They represent the input values ​​of the nth layer of the long short-term memory neural network and the recurrent unit neural network, Represent the output values ​​of the n+1th layer of the long short-term memory neural network and the recurrent unit neural network respectively, FC is the fully connected layer used for residual connection, concat represents vector cascade calculation, n∈(1,2,3,…,L), L is the number of module layers for each channel;

[0014] The output values ​​of the long short-term memory neural network and the recurrent unit neural network are calculated through the fully connected layer to obtain the corresponding prediction results. The weighted sum of the prediction results forms the final prediction output of the benchmark prediction model:

[0015]

[0016] Where: Represents the prediction result of the nth layer of the long short-term memory neural network; represents the output weight of the long short-term memory neural network, Represents the prediction result of the nth layer of the recurrent unit neural network; represents the output weight of the recurrent unit neural network, y F Represents the final prediction output.

[0017] Furthermore, the power matching of several distributed power sources and several loads according to the predicted power data of distributed power sources and loads specifically includes: obtaining the predicted power data of distributed power sources and loads, calculating the correlation coefficient between each distributed power source and each load, generating a corresponding source-load correlation coefficient matrix, sorting the established coefficient matrix from large to small according to the size of the correlation coefficient, and power matching of distributed power sources and loads according to the sorting result; if after power matching, there are multiple distributed power sources and loads with correlation coefficients lower than 0.5, then reallocating the weights of the long short-term memory neural network and the recurrent unit neural network, and re-predicting the power data.

[0018] Furthermore, the energy storage system includes a plurality of bidirectional energy storage converters, and the bidirectional energy storage converters include a current source working state and a voltage source working state; when the bidirectional energy storage converter is in the current source working state, the bidirectional energy storage converter emits or absorbs reactive power or active power; when the bidirectional energy storage converter is in the voltage source working state, the bidirectional energy storage converter acts as a balancing node of the microgrid to compensate for the power difference between the distributed power source and the load;

[0019] The method of compensating the power difference between the distributed power source and the load by the energy storage system is as follows: the bidirectional energy storage converter in the voltage source working state provides power output in real time according to the power difference between the distributed power source and the load, and the calculation formula is:

[0020]

[0021] Among them, P U (i1,k) is the active power output by the i1th bidirectional energy storage converter in voltage source working state at time k, Q U (i1,k) is the reactive power output by the i1th bidirectional energy storage converter in voltage source working state at time k, P I (i2,k) is the active power output by the i2th bidirectional energy storage converter in the current source working state at time k, Q I (i2,k) is the reactive power output by the i2th bidirectional energy storage converter in the current source working state at time k, P L is the active power of the load, P D is the active power of distributed generation, Q L is the reactive power of the load, Q D is the reactive power of distributed generation, i1=1,…,G1, i2=0,1,…G2, G=G1+G2.

[0022] Furthermore, it also includes system control of microgrids, including the following situations:

[0023] Case 1: When the remaining energy SOC in the energy storage system is at the lower limit of the ideal power and the distributed power source cannot increase its output, the output power of the energy storage system is limited and the power supply to some non-important loads is cut off;

[0024] Case 2: When the remaining energy SOC in the energy storage system is within the lower limit of the ideal power and the distributed power source can increase its output, the output power of the energy storage system is limited to quickly charge the energy storage system;

[0025] Case 3: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power generation cannot increase its output, the status quo is maintained;

[0026] Case 4: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load and safely charges the energy storage system at the same time;

[0027] Case 5: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed generation cannot increase its output, the energy storage system provides power to the load;

[0028] Case 6: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load, and the energy storage system limits power to slowly charge the energy storage system.

[0029] The present invention also adopts a power regulation system for an off-grid microgrid, comprising:

[0030] The prediction control layer is used to predict the power data of a number of distributed power sources and a number of loads at the next moment based on the historical power data;

[0031] The source-load matching layer is used to perform power matching between a plurality of distributed power sources and a plurality of loads according to the predicted power data of the distributed power sources and loads;

[0032] The coordination control layer is used to compensate the power difference between distributed generation and load through the energy storage system.

[0033] Furthermore, the prediction control layer uses a baseline prediction model that combines a long short-term memory neural network with a recurrent unit neural network to predict distributed power source and load power data. Predictions are made separately by the long short-term memory neural network and the recurrent unit neural network, and the prediction results of the two are weighted and summed to form the final prediction output of the baseline prediction model.

[0034] Furthermore, the prediction control layer is also used to determine the error amplitude between the power data collected by the distributed power sources and loads at the current moment and the power data predicted by the benchmark prediction model before predicting the power data of the distributed power sources and loads. When one or more error amplitudes in the distributed power source and load power data are greater than their corresponding error thresholds, the benchmark prediction model of the corresponding distributed power source or load whose error amplitude is greater than the error threshold is updated.

[0035] Furthermore, updating the benchmark prediction model includes: using a hedging algorithm and online learning to update the output weights of the long short-term memory neural network and the recurrent unit neural network in real time and back-propagating to update the benchmark prediction model of the distributed power supply or load; when the accuracy of the long short-term memory neural network is higher, increasing the output weight corresponding to the long short-term memory neural network, and when the accuracy of the recurrent unit neural network is higher, increasing the output weight corresponding to the recurrent unit neural network; forward propagating each module in the long short-term memory neural network and the recurrent unit neural network, the expression is:

[0036]

[0037] in, and They represent the initial values ​​of the long short-term memory neural network and the recurrent unit neural network, and They represent the input values ​​of the nth layer of the long short-term memory neural network and the recurrent unit neural network, Represent the output values ​​of the n+1th layer of the long short-term memory neural network and the recurrent unit neural network respectively, FC is the fully connected layer used for residual connection, concat represents vector cascade calculation, n∈(1,2,3,…,L), L is the number of module layers for each channel;

[0038] The output values ​​of the long short-term memory neural network and the recurrent unit neural network are calculated through the fully connected layer to obtain the corresponding prediction results. The weighted sum of the prediction results forms the final prediction output of the benchmark prediction model:

[0039]

[0040] Where: Represents the prediction result of the nth layer of the long short-term memory neural network; represents the output weight of the long short-term memory neural network, Represents the prediction result of the nth layer of the recurrent unit neural network; represents the output weight of the recurrent unit neural network, y F Represents the final prediction output.

[0041] Furthermore, the power matching of several distributed power sources and several loads according to the predicted power data of distributed power sources and loads specifically includes: obtaining the predicted power data of distributed power sources and loads, calculating the correlation coefficient between each distributed power source and each load, generating a corresponding source-load correlation coefficient matrix, sorting the established coefficient matrix from large to small according to the size of the correlation coefficient, and power matching of distributed power sources and loads according to the sorting result; if after power matching, there are multiple distributed power sources and loads with correlation coefficients lower than 0.5, then reallocating the weights of the long short-term memory neural network and the recurrent unit neural network, and re-predicting the power data.

[0042] Furthermore, the energy storage system includes a plurality of bidirectional energy storage converters, and the bidirectional energy storage converters include a current source working state and a voltage source working state; when the bidirectional energy storage converter is in the current source working state, the bidirectional energy storage converter emits or absorbs reactive power or active power; when the bidirectional energy storage converter is in the voltage source working state, the bidirectional energy storage converter acts as a balancing node of the microgrid to compensate for the power difference between the distributed power source and the load;

[0043] The method of compensating the power difference between the distributed power source and the load by the energy storage system is as follows: the bidirectional energy storage converter in the voltage source working state provides power output in real time according to the power difference between the distributed power source and the load, and the calculation formula is:

[0044]

[0045] Among them, P U (i1,k) is the active power output by the i1th bidirectional energy storage converter in voltage source working state at time k, Q U (i1,k) is the reactive power output by the i1th bidirectional energy storage converter in voltage source working state at time k, P I (i2,k) is the active power output by the i2th bidirectional energy storage converter in the current source working state at time k, Q I (i2,k) is the reactive power output by the i2th bidirectional energy storage converter in the current source working state at time k, P L is the active power of the load, P D is the active power of distributed generation, Q L is the reactive power of the load, Q D is the reactive power of distributed generation, i1=1,…,G1, i2=0,1,…G2, G=G1+G2.

[0046] Furthermore, the coordination control layer also includes a microgrid coordination control module for system control of the microgrid, including the following situations:

[0047] Case 1: When the remaining energy SOC in the energy storage system is at the lower limit of the ideal power and the distributed power source cannot increase its output, the output power of the energy storage system is limited and the power supply to some non-important loads is cut off;

[0048] Case 2: When the remaining energy SOC in the energy storage system is within the lower limit of the ideal power and the distributed power source can increase its output, the output power of the energy storage system is limited to quickly charge the energy storage system;

[0049] Case 3: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power generation cannot increase its output, the status quo is maintained;

[0050] Case 4: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load and safely charges the energy storage system at the same time;

[0051] Case 5: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed generation cannot increase its output, the energy storage system provides power to the load;

[0052] Case 6: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load, and the energy storage system limits power to slowly charge the energy storage system.

[0053] The present invention also adopts a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the above method when executing the computer program.

[0054] The present invention also adopts a computer-readable storage medium on which a computer program is stored, characterized in that the computer program implements the steps of the above method when executed by a processor.

[0055] Beneficial effects: Compared with the prior art, the present invention has the significant advantage of predicting the output and load of distributed power sources, realizing online regulation, improving the prediction accuracy, achieving the optimal matching of source and load according to the prediction results, avoiding large power shortages and power surpluses, improving system stability, and realizing rapid power regulation by coordinated control between source, storage and load. Power regulation is achieved for the output of distributed power sources, the charging and discharging of energy storage systems, and load shedding to avoid large-scale collapse of the power grid. Feedback correction exists in each link, so that power can be controlled with high precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic flow chart of the power regulation method of the off-grid microgrid in the present invention.

[0057] Figure 2 This is a structural principle diagram of the prediction control layer in the present invention.

[0058] Figure 3 Schematic diagram of the structure of the source-charge matching layer in the present invention.

[0059] Figure 4 This is a schematic diagram of the structure of the coordination control layer in the present invention. DETAILED DESCRIPTION

[0060] Example 1

[0061] like Figure 1 As shown, in this embodiment, a power regulation system for an off-grid microgrid includes a prediction control layer, a source-load matching layer, and a coordination control layer. The first layer is the prediction control layer, which predicts the output and load of the distributed power source at the next moment and provides the predicted power data for the second layer; the second layer is the source-load matching layer, which matches the predicted distributed power source with the load power data to achieve the optimal power matching between the source and the load, and avoid multiple transactions between the source and the load (such as excessive output power of the distributed power source or excessive power required by the load); the third layer is the coordination control layer, which includes two modules: the equipment on-site control module and the microgrid coordination control module.

[0062] The prediction control layer predicts the output and load of distributed power sources at the next moment, which specifically includes the following steps:

[0063] Step 1: Offline training. Use the historical power data of the distributed power output and load to perform offline training to obtain a benchmark prediction model for the distributed power output and load. First, divide yesterday's distributed power output and load power data into a training set and a validation set according to a specific ratio. The ratio of the training set to the validation set used in this embodiment is 8:2. The specific process of offline training is: construct a long short-term memory neural network and a gate recurrent unit neural network respectively, train the training set through the offline training part, and then verify it with the validation set.

[0064] The input gate, forget gate and output gate are set in the long short-term memory neural network. The input gate is used to determine which power data should be added to the cell state. It contains two parts: one is to determine the acceptance weight of the information through the sigmoid function. t , and secondly, the candidate states are calculated by the tanh function:

[0065]

[0066] Where: i t represents the input gate, σ is the sigmoid activation function, x t represents the input at time t, ht-1 represents the output at time t-1, b i represents the corresponding bias term of the input gate output, is the unit state transition value, tanh is the tanh activation function, W C represents the weight matrix of the input gate, b C Represents the bias term corresponding to the unit state output.

[0067] Forget gate: determines which power data of the previous moment should be forgotten. The activation value of the forget gate is f t From the current input x t and the previous hidden state h t-1 Calculated through a fully connected layer with a sigmoid activation function:

[0068] f t =σ(W f *[x t ,h t-1 ]+b f ) (2)

[0069] Where: f t represents the forget gate, W f represents the weight matrix of the forget gate, b f Represents the bias term corresponding to the output of the forget gate.

[0070] Cell state update: Combine the results of the input gate and the forget gate to update the cell state C t :

[0071]

[0072] Where: C t Indicates the unit status update, C t-1 Indicates the unit update status at time t-1.

[0073] Output gate: determines which power data of distributed power sources and loads should be passed to the next moment. The activation value of the output gate is o t From the current input x t and the previous moment h t-1 It is calculated through a fully connected layer with a sigmoid activation function, and then the final h is obtained by element-wise multiplication with the value of the unit state after the tanh function. t :

[0074]

[0075] Where: t represents the output gate, W o represents the weight matrix of the output gate, b oIt indicates the bias term corresponding to the output gate output, and ⊙ indicates the Hadamard product of the matrix.

[0076] The recurrent unit neural network includes an update gate and a reset gate, which is one less part compared to the long short-term memory neural network. The formula for the update gate is:

[0077] z t =σ(W z *[x t ,h t-1 ]+b z ) (5)

[0078] Where: z t represents the update gate, W z represents the weight matrix of the update gate, b z Represents the bias term corresponding to the update gate output.

[0079] The formula for resetting the gate is:

[0080] r t =σ(W r *[x t ,h t-1 ]+b r ) (6)

[0081] Where: r t Represents the reset gate, W r represents the weight matrix of the reset gate, b r Represents the bias term corresponding to the reset gate output.

[0082] At the same time, the unit transition state is introduced, which is consistent with the long short-term memory neural network. Similarly, it represents the hidden layer information contained in the unit at that moment, and finally updates the transition value of the goalkeeper unit state With the history output h t-1 Perform weighted summation to obtain the output of the recurrent unit neural network:

[0083]

[0084] Where: represents the transition value, W h A weight matrix representing the transition values.

[0085] Step 2: Change identification. Before predicting the power data of distributed power sources and loads, quickly identify the changes in the power data of distributed power sources and loads, determine the error margin between the power data collected by distributed power sources and loads at the current moment and the power data predicted by the benchmark prediction model, and determine whether the constructed benchmark prediction model needs to be corrected based on the error margin. First, set the distributed power output error threshold and the load error threshold. If the error margin of the two data is less than the corresponding error threshold, continue to use the two benchmark evaluation models at the previous moment. If the error margin of one or more of the two data is greater than the corresponding error threshold, correct the corresponding benchmark prediction model to obtain a new benchmark prediction model. According to the result of change identification, when the error margin is greater than the error threshold, enable the online incremental update in the third step.

[0086] Step 3: Online incremental update. Online incremental update is performed based on the hedging algorithm and the online learning method, which are formed by the two independent pathways of the above-mentioned long short-term memory neural network and the recurrent unit neural network. The concept of the hedging algorithm is introduced on this basis, and the output weights of the long short-term memory neural network and the recurrent unit neural network are assigned and adjusted in real time according to online learning. When the accuracy of the long short-term memory neural network is higher, the output weight corresponding to the long short-term memory neural network is increased, and when the accuracy of the recurrent unit neural network is higher, the output weight corresponding to the recurrent unit neural network is increased. When the output of the distributed power source and the load power data change, if the error amplitude is greater than the error threshold, the output weights of the long short-term memory neural network and the recurrent unit neural network are adjusted, and the overall model is updated by back propagation.

[0087] Forward propagation is performed on each module in the long short-term memory neural network and the recurrent unit neural network, and the expression is:

[0088]

[0089] Where: and represents the initial value of the long short-term memory neural network and the recurrent unit neural network, and Represents the input value of the nth layer of the long short-term memory neural network and the recurrent unit neural network, represents the output value of the n+1th layer of the long short-term memory neural network and the recurrent unit neural network, FC is the fully connected layer used for residual connection, concat represents vector cascade calculation, n∈

[0090] (1,2,3,…,L), L is the number of module layers for each channel;

[0091] The output of each module is calculated through the fully connected layer to obtain the prediction result of each module, and the prediction results are weighted and summed to form the final prediction output of the benchmark prediction model:

[0092]

[0093] Where: Represents the prediction results and output weights of the nth layer of the long short-term memory neural network: Represents the prediction result and output weight of the nth layer of the recurrent unit neural network, y F Represents the final output;

[0094] When the error between the actual power data and the prediction model is greater than the preset threshold, the output weight is adjusted online. And back-propagate to update the power prediction model of distributed power sources and loads to achieve online update.

[0095] The source-load matching layer enables optimal matching of source and load, realizes rapid power regulation, and avoids multiple power transactions. Specifically, it includes:

[0096] First, the power data of the predicted distributed power output and load are obtained, and the correlation coefficient between each distributed power source and each load is calculated according to the power data, including: the power data of N distributed power sources is P D = {P D1 ,P D2 ,P D3 ,…,P DN} and the power data of M loads is P L = {P L1 ,P L2 ,P L3 ,…,P LM}; During the operation cycle, the distributed power source P Dn (n=1, 2, ..., N) The power data of each time period is represented as P Dn =[P Dn1 ,P Dn2 ,P Dn3 ,…,P Dnk ], load P Lm (m=1, 2, ..., M) The power data of each time period is P Lm =[P Lm1 ,P Lm2 ,P Lm3 ,…,P Lmk ]; Based on the predicted power data of distributed power output and load, the Pearson correlation coefficient method is used to calculate the distributed power P Dn and load P Lm The correlation coefficient of .

[0097] Then, according to the calculated correlation coefficient, a set of correlation coefficient matrices is generated, including: obtaining the correlation coefficients of each distributed power source and each load in the microgrid, and forming the corresponding correlation coefficient matrix ρ according to the correlation coefficients between a distributed power source and different loads. N×M .

[0098] Finally, the established coefficient matrix is ​​sorted from large to small according to the size of the correlation coefficient, and source-load matching is performed based on the sorting results. If there are more than two distributed power sources with the same load in a column, the correlation coefficients of different distributed power sources with the load are further compared, and source-load matching is performed according to the size of the correlation coefficient.

[0099] Check whether the correlation between the output of each distributed power source and the load is close. If most of the source-load correlation coefficients are quite different, the source-load matching will result in multiple source-load correlation coefficients with large differences. Return to redistribute the weight factors of the long short-term memory neural network and the recurrent unit neural network to maximize the matching between sources and loads and the correlation between the output of distributed power sources and loads. If most of the source-load correlation coefficients are not very different, there is no need to redistribute the weight factors of the objective function of the first layer. The range of the correlation coefficient is [-1,1], and positive linear correlation means that the correlation coefficient is between [0,1]. The larger the correlation coefficient, the stronger the correlation. The sources and loads with stronger correlations are matched first, that is, the higher the matching priority. Under normal circumstances, the difference in source-load matching coefficients below 0.5 is large.

[0100] Coordination control layer: After source-load matching, coordinated control is required due to the volatility of distributed power output and load, which includes two modules: equipment on-site control module and microgrid coordination control module.

[0101] The equipment on-site control module compensates for the power difference between the distributed power source and the load through the energy storage system to achieve dynamic power balance, including: the power balance in the equipment on-site control module mainly utilizes the existence of a bidirectional energy storage converter in the energy storage system, and each bidirectional energy storage converter includes two working states: current source and voltage source; when the bidirectional energy storage converter is in the current source working state, the bidirectional energy storage converter sends or absorbs reactive or active power according to the microgrid instruction; when the bidirectional energy storage converter is in the voltage source working state, the bidirectional energy storage converter acts as a balancing node of the microgrid to compensate for the power difference between the distributed power source and the load.

[0102] The first case: All bidirectional energy storage converters are in the voltage source working state, that is, all bidirectional energy storage converters adjust the power output in real time according to the difference between the distributed power source and the load. The output power of each bidirectional energy storage converter is:

[0103]

[0104] Where: P B (i) is the active power output of the i-th converter under voltage working state, Q B (i) is the reactive power output by the i-th converter under voltage working state, G is the number of converters, P L is the load active power, P D is the active power of distributed generation, Q L is the load reactive power, Q D It is the reactive power of distributed power source.

[0105] The second case: G1 bidirectional energy storage converter is in voltage source working state, G2 bidirectional energy storage converter is in current source working state, the converter in voltage source working state provides power output in real time according to the difference between distributed power source and load, and the converter output power at current time k is:

[0106]

[0107] Where P I (i2,k) and Q I (i2, k) are the active and reactive powers output by the converter when the i2th current source is in working state at the current k moment. When G2 is equal to zero, it is the first case.

[0108] The microgrid coordination control module uses coordinated control between source, storage and load to achieve rapid power regulation, and adjusts the output of distributed power sources, the charging and discharging of energy storage systems, and load shedding to avoid large-scale grid collapse. Rapid power regulation is achieved through coordinated control of the remaining energy SOC in the energy storage system, the output of distributed power sources, and load demand. As shown in Table 1, it includes the following situations:

[0109] Case 1: The SOC in the energy storage system is at the lower limit of the ideal power. At this time, the power needs to be limited. The energy storage system discharges slowly and the distributed power source is affected by environmental factors and cannot increase its output. At this time, it is necessary to cut off some non-important loads to supply power to important loads and achieve rapid power regulation.

[0110] Case 2: The SOC in the energy storage system is at the lower limit of the ideal power. At this time, the power needs to be limited. The energy storage system discharges slowly and the distributed power source can increase the output. In this case, the load does not need to be removed, and the energy storage system can be quickly charged.

[0111] Case 3: When the SOC in the energy storage system is lower than the upper limit of the ideal power but higher than the lower limit of the ideal power and the distributed power source cannot increase its output, the energy storage system can achieve rapid power regulation without load shedding.

[0112] Case 4: The SOC in the energy storage system is lower than the upper limit of the theoretical power, higher than the lower limit of the ideal power, and the distributed power source can increase its output. In this case, the distributed power source provides power to the load and can safely charge the energy storage system.

[0113] Case 5: When the SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed power source cannot increase its output, the energy storage system provides power for the load.

[0114] Case 6: When the SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load, while the energy storage system limits power and charges slowly.

[0115] Table 1

[0116]

[0117] If power balance cannot be met even after non-essential loads are cut off, and a large power shortage occurs, causing power outages to important loads, then the source and load will need to be re-matched; if power balance is met, the current distributed power output control strategy and energy storage system charging and discharging strategy will be adopted.

[0118] Example 2

[0119] In this embodiment, a power regulation method for an off-grid microgrid includes the following steps:

[0120] Firstly, the distributed power output and load forecasting models are trained respectively by combining the long short-term memory neural network and the recurrent unit neural network. According to the results of online volatility identification, if there is no obvious error, the forecast results are output based on the benchmark forecasting model; if there is a large error in the detection result, or the accuracy of the benchmark model is monitored to decrease, the hedging algorithm and the online learning algorithm are enabled. With reference to the forecast accuracy of each module in the benchmark model, the weight coefficients are updated based on the changes in the power data distribution of distributed power sources and loads in actual situations. According to the iterative model, the distributed power output and load at the next moment are predicted for several distributed power sources and several loads based on the historical power data.

[0121] The predicted power data of distributed power sources and loads are used to calculate the correlation coefficient. According to the correlation coefficient between each power source and each load power, the source-load correlation coefficient is arranged in sequence. The source-load matching is performed based on the arrangement results to achieve the optimal power matching of the source and load, and to avoid multiple transactions of the source and load (such as excessive output power of the distributed power source or excessive power required by the load). If the source-load matching results in large differences in the correlation coefficients of multiple sources and loads, the weight factors are returned to the first layer to be redistributed to maximize the matching between the source and the load.

[0122] The energy storage system is used to provide a power difference to maintain the power balance strategy. The energy storage converter is used to compensate for the difference between the distributed power source and the load power to maintain real-time power balance. The source storage load coordinated control is used to achieve rapid power regulation. By controlling the output of distributed energy, the charging and discharging of the energy storage system and the removal of the load, rapid power regulation is achieved to avoid large-scale collapse of the power grid. When power balance cannot be achieved or even important loads need to be removed, the matching between the source and the load is redistributed.

[0123] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0124] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0125] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0127] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A power regulation method for an off-grid microgrid, It is characterized in that include: For several distributed power sources and several loads, the power data at the next moment is predicted based on the historical power data; Power matching is performed between several distributed power sources and several loads according to the predicted power data of distributed power sources and loads; and the power difference between the distributed power sources and the loads is compensated by the energy storage system.

2. The power regulation method for an off-grid microgrid according to claim 1, It is characterized in that For the prediction of distributed power generation and load power data, a benchmark prediction model of long short-term memory neural network combined with recurrent unit neural network is adopted. Predictions are made by long short-term memory neural network and recurrent unit neural network respectively, and the prediction results of the two are weighted and summed to form the final prediction output of the benchmark prediction model.

3. The power regulation method for an off-grid microgrid according to claim 2, It is characterized in that Before predicting the power data of distributed power sources and loads, determine the error margin between the power data collected by the distributed power sources and loads at the current moment and the power data predicted by the benchmark prediction model. When one or more error margins in the distributed power source and load power data are greater than their corresponding error thresholds, update the benchmark prediction model of the corresponding distributed power source or load whose error margin is greater than the error threshold.

4. The power regulation method for an off-grid microgrid according to claim 3, It is characterized in that The updating of the benchmark prediction model includes: using the hedging algorithm and online learning to update the output weights of the long short-term memory neural network and the recurrent unit neural network in real time and back-propagating to update the benchmark prediction model of the distributed power supply or load; when the accuracy of the long short-term memory neural network is higher, the output weight corresponding to the long short-term memory neural network is increased, and when the accuracy of the recurrent unit neural network is higher, the output weight corresponding to the recurrent unit neural network is increased; each module in the long short-term memory neural network and the recurrent unit neural network is forward propagated, and the expression is: in, and They represent the initial values ​​of the long short-term memory neural network and the recurrent unit neural network, and They represent the input values ​​of the nth layer of the long short-term memory neural network and the recurrent unit neural network, Represent the output values ​​of the n+1th layer of the long short-term memory neural network and the recurrent unit neural network respectively, FC is the fully connected layer used for residual connection, concat represents vector cascade calculation, n∈(1,2,3,…,L), L is the number of module layers for each channel; The output values ​​of the long short-term memory neural network and the recurrent unit neural network are calculated through the fully connected layer to obtain the corresponding prediction results. The weighted sum of the prediction results forms the final prediction output of the benchmark prediction model: Where: Represents the prediction result of the nth layer of the long short-term memory neural network; represents the output weight of the long short-term memory neural network, Represents the prediction result of the nth layer of the recurrent unit neural network; represents the output weight of the recurrent unit neural network, y F Represents the final prediction output.

5. The power regulation method for an off-grid microgrid according to claim 4, It is characterized in that The power matching of a plurality of distributed power sources and a plurality of loads according to the predicted power data of the distributed power sources and loads specifically includes: obtaining the predicted power data of the distributed power sources and loads, calculating the correlation coefficient between each distributed power source and each load, generating a corresponding source-load correlation coefficient matrix, sorting the established coefficient matrix from large to small according to the size of the correlation coefficient, and power matching of the distributed power sources and loads according to the sorting result; if after power matching, there are multiple distributed power sources and loads with correlation coefficients lower than 0.5, then reallocating the weights of the long short-term memory neural network and the recurrent unit neural network, and re-predicting the power data.

6. The power regulation method for an off-grid microgrid according to claim 1, It is characterized in that The energy storage system includes a plurality of bidirectional energy storage converters, and the bidirectional energy storage converters include a current source working state and a voltage source working state; when the bidirectional energy storage converter is in the current source working state, the bidirectional energy storage converter emits or absorbs reactive power or active power; when the bidirectional energy storage converter is in the voltage source working state, the bidirectional energy storage converter acts as a balancing node of the microgrid to compensate for the power difference between the distributed power source and the load; The method of compensating the power difference between the distributed power source and the load by the energy storage system is as follows: the bidirectional energy storage converter in the voltage source working state provides power output in real time according to the power difference between the distributed power source and the load, and the calculation formula is: Among them, P U (i 1 ,k) is the i-th 1 The active power output of a bidirectional energy storage converter in voltage source working state, Q U (i 1 ,k) is the i-th 1 The reactive power output by a bidirectional energy storage converter in voltage source working state, P I (i 2 ,k) is the i-th 2 The active power output of a bidirectional energy storage converter in current source working state, Q I (i 2 ,k) is the i-th 2 The reactive power output by a bidirectional energy storage converter in current source working state, P L is the active power of the load, P D is the active power of distributed generation, Q L is the reactive power of the load, Q D is the reactive power of distributed generation, i 1 =1, ..., G 1 ,i 2 =0, 1, ...G 2 , G=G 1 +G 2 .

7. The power regulation method for an off-grid microgrid according to claim 1, It is characterized in that It also includes system control of microgrids, including the following situations: Case 1: When the remaining energy SOC in the energy storage system is at the lower limit of the ideal power and the distributed power source cannot increase its output, the output power of the energy storage system is limited and the power supply to some non-important loads is cut off; Case 2: When the remaining energy SOC in the energy storage system is within the lower limit of the ideal power and the distributed power source can increase its output, the output power of the energy storage system is limited to quickly charge the energy storage system; Case 3: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power generation cannot increase its output, the status quo is maintained; Case 4: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load and safely charges the energy storage system at the same time; Case 5: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed generation cannot increase its output, the energy storage system provides power to the load; Case 6: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load, and the energy storage system limits power to slowly charge the energy storage system.

8. A power regulation system for an off-grid microgrid, It is characterized in that include: The prediction control layer is used to predict the power data of the next moment for a number of distributed power sources and a number of loads based on the historical power data; The source-load matching layer is used to perform power matching between a plurality of distributed power sources and a plurality of loads according to the predicted power data of the distributed power sources and loads; The coordination control layer is used to compensate the power difference between distributed generation and load through the energy storage system.

9. The power regulation system for off-grid microgrid according to claim 8, It is characterized in that The prediction control layer uses a benchmark prediction model that combines a long short-term memory neural network with a recurrent unit neural network to predict distributed power source and load power data. Predictions are made using the long short-term memory neural network and the recurrent unit neural network respectively, and the weighted sum of the prediction results of the two forms the final prediction output of the benchmark prediction model.

10. The power regulation system for off-grid microgrid according to claim 9, It is characterized in that The prediction control layer is also used to determine the error amplitude between the power data collected by the distributed power sources and loads at the current moment and the power data predicted by the benchmark prediction model before predicting the power data of the distributed power sources and loads. When one or more error amplitudes in the distributed power source and load power data are greater than their corresponding error thresholds, the benchmark prediction model of the corresponding distributed power source or load whose error amplitude is greater than the error threshold is updated.

11. The power regulation system for off-grid microgrid according to claim 10, It is characterized in that The updating of the benchmark prediction model includes: using the hedging algorithm and online learning to update the output weights of the long short-term memory neural network and the recurrent unit neural network in real time and back-propagating to update the benchmark prediction model of the distributed power supply or load; when the accuracy of the long short-term memory neural network is higher, the output weight corresponding to the long short-term memory neural network is increased, and when the accuracy of the recurrent unit neural network is higher, the output weight corresponding to the recurrent unit neural network is increased; each module in the long short-term memory neural network and the recurrent unit neural network is forward propagated, and the expression is: in, and They represent the initial values ​​of the long short-term memory neural network and the recurrent unit neural network, and They represent the input values ​​of the nth layer of the long short-term memory neural network and the recurrent unit neural network, Represent the output values ​​of the n+1th layer of the long short-term memory neural network and the recurrent unit neural network respectively, FC is the fully connected layer used for residual connection, concat represents vector cascade calculation, n∈(1,2,3,…,L), L is the number of module layers for each channel; The output values ​​of the long short-term memory neural network and the recurrent unit neural network are calculated through the fully connected layer to obtain the corresponding prediction results. The weighted sum of the prediction results forms the final prediction output of the benchmark prediction model: Where: Represents the prediction result of the nth layer of the long short-term memory neural network; represents the output weight of the long short-term memory neural network, Represents the prediction result of the nth layer of the recurrent unit neural network; represents the output weight of the recurrent unit neural network, y F Represents the final prediction output.

12. The power regulation system for off-grid microgrid according to claim 11, It is characterized in that The power matching of a plurality of distributed power sources and a plurality of loads according to the predicted power data of the distributed power sources and loads specifically includes: obtaining the predicted power data of the distributed power sources and loads, calculating the correlation coefficient between each distributed power source and each load, generating a corresponding source-load correlation coefficient matrix, sorting the established coefficient matrix from large to small according to the size of the correlation coefficient, and power matching of the distributed power sources and loads according to the sorting result; if after power matching, there are multiple distributed power sources and loads with correlation coefficients lower than 0.5, then reallocating the weights of the long short-term memory neural network and the recurrent unit neural network, and re-predicting the power data.

13. The power regulation system for off-grid microgrid according to claim 8, It is characterized in that The energy storage system includes a plurality of bidirectional energy storage converters, and the bidirectional energy storage converters include a current source working state and a voltage source working state; when the bidirectional energy storage converter is in the current source working state, the bidirectional energy storage converter emits or absorbs reactive power or active power; when the bidirectional energy storage converter is in the voltage source working state, the bidirectional energy storage converter acts as a balancing node of the microgrid to compensate for the power difference between the distributed power source and the load; The method of compensating the power difference between the distributed power source and the load by the energy storage system is as follows: the bidirectional energy storage converter in the voltage source working state provides power output in real time according to the power difference between the distributed power source and the load, and the calculation formula is: Among them, P U (i 1 ,k) is the i-th 1 The active power output of a bidirectional energy storage converter in voltage source working state, Q U (i 1 ,k) is the i-th 1 The reactive power output by a bidirectional energy storage converter in voltage source working state, P I (i 2 ,k) is the i-th 2 The active power output of a bidirectional energy storage converter in current source working state, Q I (i 2 ,k) is the i-th 2 The reactive power output by a bidirectional energy storage converter in current source working state, P L is the active power of the load, P D is the active power of distributed generation, Q L is the reactive power of the load, Q D is the reactive power of distributed generation, i 1 =1, ..., G 1 ,i 2 =0, 1, ...G 2 , G=G 1 +G 2 .

14. The power regulation system for off-grid microgrid according to claim 1, It is characterized in that The coordination control layer also includes a microgrid coordination control module, which is used to control the microgrid system, including the following situations: Case 1: When the remaining energy SOC in the energy storage system is at the lower limit of the ideal power and the distributed power source cannot increase its output, the output power of the energy storage system is limited and the power supply to some non-important loads is cut off; Case 2: When the remaining energy SOC in the energy storage system is within the lower limit of the ideal power and the distributed power source can increase its output, the output power of the energy storage system is limited to quickly charge the energy storage system; Case 3: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power generation cannot increase its output, the status quo is maintained; Case 4: When the remaining energy SOC in the energy storage system is lower than the upper limit of the ideal power, higher than the lower limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load and safely charges the energy storage system at the same time; Case 5: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed generation cannot increase its output, the energy storage system provides power to the load; Case 6: When the remaining energy SOC in the energy storage system is higher than the upper limit of the ideal power and the distributed power source can increase its output, the distributed power source provides power to the load, and the energy storage system limits power to slowly charge the energy storage system.

15. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

16. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images