A module power redundancy dynamic switching control system

Abstract

The application provides a module power redundancy dynamic switching control system, and relates to the technical field of power system redundancy control, and comprises: a plurality of power modules, a plurality of loads are connected through a switching module; a switching controller is used for calculating power redundancy in real time according to total load power of each load and rated power and peak efficiency of each power module currently put into, and then controlling the switching module to act, so that the power redundancy is greater than zero and not greater than the minimum value of the rated power of each power module currently put into by adjusting the number of the power modules currently put into. The beneficial effect can automatically adjust the number of redundant modules according to the load rate, solve the problem that the traditional non-redundancy scheme is prone to fault diffusion, and avoid the defects of insufficient redundancy or excessive redundancy of the fixed redundancy scheme, so that the module average load rate is optimized to the 60%-80% efficient interval while ensuring reliability.

Description

Technical Field

[0001] This invention relates to the field of power system redundancy control technology, and in particular to a module power redundancy dynamic switching control system. Background Technology

[0002] In recent years, the global new energy power generation industry, data centers, and industrial automation systems have experienced explosive growth. The stable operation of these sectors heavily relies on power electronic systems: in new energy power generation, photovoltaic inverters and wind power converters are responsible for converting the DC power output from photovoltaic arrays and the AC power output from wind turbines into grid-compliant AC power, and for achieving grid-connected control. In data centers, uninterruptible power supplies (UPS), server power modules, and power conversion units in rack-mount cabinets provide stable, low-ripple power to core computing devices, preventing data loss or computing power interruptions due to power outages. In industrial automation scenarios, frequency converters and servo drives precisely control the switching states of power electronic devices such as IGBTs to achieve real-time adjustment of motor speed and torque, directly determining the continuity of production processes and processing accuracy. Therefore, power electronic systems have become the core carrier of energy conversion and transmission in these fields. Their operational reliability is directly related to the continuity of upstream energy output (such as the power generation duration of photovoltaic power plants), the security of midstream data processing (such as the computing power stability of data centers), and the continuity of downstream industrial production (such as the operating rate of production lines), making them a critical infrastructure for ensuring the implementation of core functions in each field.

[0003] However, existing power electronic systems are difficult to meet the industry's requirements for high reliability, high cost-effectiveness, and strong adaptability in practical applications, mainly in the following two aspects:

[0004] 1) Fixed redundancy solutions are costly and lack flexibility.

[0005] To address module failures, some existing systems employ a fixed redundancy design, configuring a fixed number of spare modules. However, this approach has significant drawbacks: first, spare modules remain idle for extended periods, unable to participate in normal power distribution, leading to a significant increase in hardware costs, installation space, and operating energy consumption; second, the redundancy configuration cannot be dynamically adjusted based on load changes. For example, during low-load periods at night in data centers, fixed-redundancy spare modules remain idle, resulting in resource waste; while during peak industrial production periods, the number of fixed-redundancy spare modules may be insufficient to meet temporary power demands, causing the system to operate at reduced capacity.

[0006] 2) Difficulty in adapting to the dual requirements of new energy power grids

[0007] As the penetration rate of renewable energy generation in the power grid increases, the grid places dual demands on grid-connected power electronic systems for high reliability and high economy. On the one hand, high reliability is required to cope with the impact of fluctuations in renewable energy output (such as sudden changes in solar radiation and sudden drops in wind speed), avoiding grid frequency fluctuations or voltage deviations due to system failures. On the other hand, economy must be considered, reducing hardware costs and operating energy consumption while ensuring reliability, and avoiding resource waste caused by excessive redundancy. Existing technologies either prioritize reliability (such as using 3N redundancy), resulting in high costs, or prioritize economy (such as non-redundant designs), resulting in insufficient reliability. Neither can achieve a balance between the two, making it difficult to adapt to the complex operating scenarios of renewable energy grids.

[0008] Therefore, optimizing power distribution has become a critical issue that the industry urgently needs to address. Summary of the Invention

[0009] To address the problems existing in the prior art, the present invention provides a module power redundancy dynamic switching control system, comprising: multiple power modules connected to multiple loads via switching modules; and a switching controller connected to each of the power modules, the switching modules, and the loads, for calculating the power redundancy in real time based on the total load power of each load and the rated power and peak efficiency of each currently engaged power module, thereby controlling the operation of the switching modules to adjust the number of currently engaged power modules so that the power redundancy is greater than zero and not greater than the minimum rated power of each currently engaged power module.

[0010] Preferably, the switching controller includes: a calculation module, configured to calculate power redundancy in real time based on the total load power of each load and the rated power and peak efficiency of each currently engaged power module; a comparison module, connected to the calculation module, configured to generate a cut-off signal when the power redundancy is greater than the minimum rated power, and to generate an engagement signal when the power redundancy is not greater than zero and the current number of engaged power modules is less than the total number of power modules; and a control module, connected to both the calculation module and the comparison module, configured to control the cut-off of the power module with the minimum rated power based on the cut-off signal and generate a recalculation signal, and to control the engagement of an offline power module based on the engagement signal and generate the recalculation signal; the calculation module recalculates the power redundancy based on the recalculation signal.

[0011] Preferably, the control module includes a cut-out control unit, which issues a soft-disconnect command to the power module with the lowest rated power according to the cut-out signal, so as to control the power module to enter the current reduction mode, and at the same time linearly decreases the output current value of the power module to gradually reduce the output current value of the power module until it is zero, and then controls the switching module to disconnect the power module accordingly to achieve smooth cut-out;

[0012] The expression for the linearly decreasing trend is as follows:

[0013]

[0014] in, This indicates the initial output current value when the power module begins executing the soft-disconnect command. This indicates that the power module has executed the soft disconnect command to begin disconnecting. Output current value after time t. This indicates the preset total duration of the soft disconnect.

[0015] Preferably, the control module further includes a status monitoring unit connected to the cut-out control unit, used to generate a first control signal when the power module with the lowest rated power is in normal working condition according to the cut-out signal, and to generate a second control signal when the power module with the lowest rated power is in fault condition; the cut-out control unit issues the soft disconnect command according to the first control signal, and controls the switching module to directly disconnect the corresponding power module according to the second control signal.

[0016] Preferably, the control module includes: a wake-up unit, used to issue an offline module power-on command according to the input signal, so as to control an offline power module to perform pre-synchronization, phase locking and pre-charging to complete the power-on wake-up; and an input control unit, used to configure the droop coefficient of the power module after power-on to an initial value so that its initial output current is zero, and then gradually decrease the initial value and synchronously increase the droop coefficient of the other power modules currently in operation, until the droop coefficient of the power module after power-on reaches the target value, so as to gradually transfer the power of the other power modules currently in operation to the power module after power-on, so as to achieve smooth input.

[0017] Preferably, the expression for the synchronous increase of the droop coefficient of the remaining power modules currently in operation is as follows:

[0018]

[0019] in, For the currently deployed power modules The synchronous increase in quantity, For the currently deployed power modules The original value, The initial value is... The power module after power-on The target value.

[0020] Preferably, the switching controller further includes a power prediction module connected to the calculation module, used to collect real-time operating data of each power module, the switching module and the load, and input the real-time operating data into a pre-trained load prediction model to obtain the total load power of each load in the next hour; the control module is also used to issue an offline module power-on command in advance when it is determined based on the predicted total load power that the offline power module needs to be put into operation, so as to wake up the offline power module that needs to be put into operation.

[0021] Preferably, each of the power modules integrates a self-testing circuit for performing periodic self-tests when the power module is offline, and marking itself as unusable when the periodic self-test results indicate a fault.

[0022] Preferably, the formula for calculating the power redundancy is as follows:

[0023]

[0024] in, The power redundancy, For power modules Rated power, For power modules Peak efficiency, The total load power is denoted as .

[0025] Preferably, the switching controller further includes an instruction response module, configured to respond to an externally input load increase / decrease instruction containing a load power value when the power redundancy is greater than zero and not greater than the minimum rated power of each of the currently engaged power modules; the switching controller is further configured to update the total load power based on the load increase / decrease instruction, and then adjust the number of currently engaged power modules accordingly so that the power redundancy is again in a state greater than zero and not greater than the minimum rated power of each of the currently engaged power modules.

[0026] The above technical solution has the following advantages or beneficial effects: By introducing an N+X elastic redundancy mechanism, the power redundancy is always greater than zero and not greater than the minimum rated power of each power module currently in operation, based on dynamically adjusting the number N of the currently deployed power modules. This enables dynamic adjustment of the number X of offline power modules as backups, allowing the number of redundant modules to be automatically adjusted according to the load rate. This solves the problem of easy fault propagation in traditional non-redundancy solutions and avoids the defects of insufficient or excessive redundancy in fixed redundancy solutions. While ensuring reliability, it optimizes the average load rate of modules to an efficient range of 60%-80%. Attached Figure Description

[0027] Figure 1 A schematic diagram of a module power redundancy dynamic switching control system is shown in a preferred embodiment of the present invention.

[0028] Figure 2 In a preferred embodiment of the present invention, a schematic diagram of a structure in which multiple power modules are connected to multiple loads through a switching module is shown.

[0029] Figure 3 A circuit diagram of the pre-charging circuit is shown in a preferred embodiment of the present invention.

[0030] Figure 4 A circuit diagram of the current detection circuit is shown in a preferred embodiment of the present invention.

[0031] Figure 5 A circuit diagram of a voltage detection circuit is shown in a preferred embodiment of the present invention. Detailed Implementation

[0032] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment; other embodiments that conform to the spirit of the present invention may also fall within the scope of the present invention.

[0033] In a preferred embodiment of the present invention, based on the above-mentioned problems existing in the prior art, a module power redundancy dynamic switching control system is provided, such as... Figure 1 As shown, it includes: multiple power modules 1, which are connected to multiple loads 3 via a switch module 2; and a switching controller 4, which is connected to each power module 1, the switch module 2, and the loads 3 respectively. The controller is used to calculate the power redundancy in real time based on the total load power of each load 3 and the rated power and peak efficiency of each currently engaged power module 1, and then control the operation of the switch module 2 so that the power redundancy is greater than zero and not greater than the minimum rated power of each currently engaged power module 1 by adjusting the number of currently engaged power modules 1.

[0034] Specifically, in this embodiment, as Figure 2As shown, taking an example where there are 5 power modules 1 and 5 loads 3, each power module 1 is connected to multiple loads 3 through a switch module 2. The switch module 2 includes power switches S1-S5, switching switches S6-S9 and load switches S10-S14.

[0035] Specifically, each power module 1 is connected to the ring bus consisting of switching switches S6-S9 via power switches S1-S5, and then connected to each load 3 via load switches S10-S14. Each power switch controls the connection between the power module and the ring bus one-to-one, which can quickly cut off the circuit in the event of a power module failure and prevent the fault from spreading. Each load switch controls the connection between the ring bus and the load 3 one-to-one, so that the entire system can be started, stopped, or maintained independently of a single load without interrupting the operation of the entire system.

[0036] Furthermore, in response to sudden load changes, this embodiment dynamically adjusts the number of currently deployed power modules 1 to ensure that the power redundancy is greater than zero (avoiding no redundancy) and not greater than the minimum rated power of each currently deployed power module 1 (avoiding excessive redundancy), thereby ensuring that the load rate of each online power module 1 remains in the high-efficiency range and effectively reducing losses.

[0037] Furthermore, with Figure 2 For example, suppose the current system has three power modules (such as...) Figure 2 Power modules 1, 2, and 3) and three loads (such as power ...2, 3, and 4). Figure 2 Loads 1, 2, and 3 are connected and the system maintains optimal peak efficiency. At this time, power switches S1-S3 are turned on, power switches S4-S5 are turned off, switching switches S6-S8 are turned on, and load switches S10-S12 are turned on and S13-S14 are turned off.

[0038] During operation, if it is necessary to dynamically adjust the number of loads according to actual needs, such as the need to increase or decrease the load, the present invention can calculate the power redundancy in real time based on the need to increase or decrease the load, and then readjust the number of power modules 1 currently put into operation based on the power redundancy.

[0039] If, based on real-time calculations of power redundancy, it is determined that another power module needs to be added, such as power module 4, then the power switch S4 can be turned on, and the ring bus can be extended by switching switch S9. The entire load increase process does not require power interruption, and the module load rate remains highly efficient.

[0040] If the power redundancy is determined based on real-time calculation, and it is necessary to switch out a power module, such as power module 3, then the power switch S3 can be disconnected. This allows the power module to operate without restrictions after the load is reduced, maintaining a high load rate and reducing the standby power consumption of the power module, thus avoiding the waste of resources due to high redundancy under low load.

[0041] In a preferred embodiment of the present invention, the switching controller 4 includes: a calculation module 41, configured to calculate power redundancy in real time based on the total load power of each load and the rated power and peak efficiency of each currently engaged power module; a comparison module 42, connected to the calculation module 41, configured to generate a cut-out signal when the power redundancy is greater than the minimum rated power, and to generate an engagement signal when the power redundancy is not greater than zero and the current number of engaged power modules is less than the total number of power modules; and a control module 43, connected to the calculation module 41 and the comparison module 42 respectively, configured to control the cut-out of the power module with the minimum rated power according to the cut-out signal and generate a recalculation signal, and to control the engagement of an offline power module according to the engagement signal and generate a recalculation signal; the calculation module 41 recalculates the power redundancy based on the recalculation signal.

[0042] In a preferred embodiment of the present invention, the formula for calculating power redundancy is as follows:

[0043]

[0044] in, For power redundancy, For power modules Rated power, For power modules Peak efficiency, This represents the total load power.

[0045] Specifically, in this embodiment, peak efficiency is introduced when calculating power redundancy. As a correction factor, it avoids the problem of nominal redundancy being sufficient but actual output efficiency being low, which is caused by the traditional method of calculating redundancy only based on rated power. For example, if a module has a rated power of 10kW but a peak efficiency of only 90% (corresponding to a 70% load rate), its actual effective redundancy should be 10*0.9=9kW, not 10kW, to ensure that the redundancy calculation is more in line with the actual output capacity of the module.

[0046] Furthermore, when the power redundancy exceeds the minimum rated power of the currently deployed power modules, it is determined to be excessive redundancy, and a cut-off signal is generated. For example, if the rated powers of the three currently deployed power modules are 10kW, 10kW, and 15kW respectively, and the power redundancy is 12kW, a cut-off signal is generated, triggering the redundant modules to exit.

[0047] When the power redundancy is not greater than zero, that is, the effective output of the currently deployed power module cannot cover the total load, and the current number of deployed power modules is less than the total number of power modules, it is determined to be insufficient redundancy and a deployment signal is generated to trigger the deployment of offline modules.

[0048] If the load decreases, the total load power will change. At this time, the switching controller 4 will mark the currently activated power module, that is, the one with the lowest rated power among the currently online power modules, as [the one with the lowest rated power]. Recalculate and judge Value, if at this time If the highest rated power module is switched out, the module with the lowest rated power will be switched out to avoid a sudden increase in the load rate of the remaining power modules. After switching out, the number of currently deployed power modules, n = n-1, will be re-marked. And update Value; if at this time Then the system will continue to operate normally.

[0049] For example, in a certain scenario, four 10kW power modules are initially deployed, i.e. = 10 kW, total load 30 kW, then power redundancy = 4 * 10 * 0.965 - 30 = 8.6 kW (normal); when the load drops to 25 kW, power redundancy = 4 * 10 * 0.965 - 25 = 13.6 kW. After cutting out one 10kW power module, n = 3, the power redundancy = 3*10*0.965-25 = 3.95kW, the average load rate of the module increases from 25 / 40=62.5% to 25 / 30≈83.3% (close to the upper limit of the high efficiency range), and the efficiency is improved by 2.3%.

[0050] If the load increases, the total load power will also change. In this case, the switching controller 4 will first determine if the total system power can meet the current load power requirement. If it cannot, the system will refuse to connect the load and issue an overload warning to prevent system overload and crash. Here, the total system power refers to the effective power that all power modules can provide when online. If the requirement is met, the system will calculate and determine... Value, if If the value is less than 0, it is determined that an offline module needs to be added to the system. Then, an offline module power-on command is issued to connect the offline module to meet the load requirements. When the number of power modules added is equal to the total number of power modules, a full load alarm is issued and the system is reported that there are no spare redundant modules.

[0051] For example, in a certain scenario, three 10kW power modules are initially deployed, i.e. = 10 kW, total load 32kW, then power redundancy = 3*10*0.965 - 32 = -3.05kW, which is insufficient redundancy; after adding one 10kW power module, n = 4, power redundancy = 4*10*0.965 - 25 = 6.6kW, the average load rate of the module drops from 32 / 30≈106.7% (overload) to 32 / 40=80% (the upper limit of the high efficiency range).

[0052] As can be seen, the dynamic adjustment of the number N of the currently deployed power modules based on the present invention ensures that the power redundancy is always greater than zero and not greater than the minimum rated power of each currently deployed power module, so that the average load rate of the disclosed modules remains stable in the high-efficiency range of 60%-80% for a long time, which is significantly more efficient than the traditional fixed redundancy scheme.

[0053] In a preferred embodiment of the present invention, the control module 43 includes a cut-out control unit 431, which is used to issue a soft disconnect command to the power module with the minimum rated power according to the cut-out signal, so as to control the power module to enter the current reduction mode, and at the same time linearly decrease the output current value of the power module to gradually reduce the output current value of the power module until it is zero, and then control the switch module to disconnect the power module accordingly to achieve smooth cut-out.

[0054] The expression for linearly decreasing is as follows:

[0055]

[0056] in, This indicates the initial output current value when the power module begins executing the soft-shutdown command. This indicates that the power module has executed a soft disconnect command to begin disconnecting. Output current value after time t. This indicates the preset total duration of the soft disconnect.

[0057] Specifically, in this embodiment, when a power module needs to be switched out, the control module 43 first sends a soft disconnect command to the module to be disconnected (i.e., the power module with the smallest rated power). After receiving the soft disconnect command, the module to be disconnected enters the current reduction mode. In the current reduction mode, the module to be disconnected first exits the current sharing loop, stops receiving the current command from the current sharing loop, and no longer participates in current sharing, so as to realize the coordinated control of the module to be disconnected with other modules and prevent its current change from interfering with the overall current sharing accuracy.

[0058] After exiting the current sharing loop, the module to be disconnected initiates a smooth disconnection strategy. This strategy employs an independent current closed-loop control to linearly decrease the output current of the power module until it reaches zero. Subsequently, the control switch module disconnects the corresponding power module, achieving a smooth disconnection and avoiding current spikes and voltage oscillations caused by traditional hard disconnection, thus providing power supply stability. Specifically, disconnecting the corresponding power module's power switch only after the output current reaches zero ensures that no arcing occurs during disconnection, effectively extending the mechanical life of the power switch and reducing maintenance and replacement frequency, as the current across the switch is close to zero at this point.

[0059] In a preferred embodiment of the present invention, the control module 43 further includes a status monitoring unit 432 connected to the cut-out control unit 431. The status monitoring unit 432 generates a first control signal when the power module with the lowest rated power is in normal working condition, and generates a second control signal when the power module with the lowest rated power is in fault condition. The cut-out control unit 431 issues a soft disconnect command based on the first control signal and controls the switching module to directly disconnect the power module based on the second control signal, thus solving the problem of fault propagation or switching delay caused by the one-size-fits-all approach in traditional solutions.

[0060] In a preferred embodiment of the present invention, the control module 43 includes: a wake-up unit 433, configured to issue an offline module power-on command according to the input signal to control an offline power module to perform pre-synchronization, phase locking, and pre-charging to complete the power-on wake-up; and an input control unit 434, configured to configure the droop coefficient of the power module after power-on to an initial value so that its initial output current is zero, and then gradually decrease the initial value while synchronously increasing the droop coefficient of the other currently input power modules until the droop coefficient of the power module after power-on reaches the target value, so as to gradually transfer the power of the other currently input power modules to the power module after power-on, thereby achieving smooth input.

[0061] Specifically, in this embodiment, when a power module needs to be activated, if it is necessary to... Figure 2 Power module 2 is inserted into power module 1 which is running in the middle. First, the power module to be put into operation is pre-synchronized, phase-locked and pre-charged. The rated input current value of the module is uploaded through high-speed communication. After the control module 43 receives it without error, it adds it to the online module list. Then, switch S2 is closed first and then switch S6 is closed to complete the smooth connection process.

[0062] The purpose of pre-synchronization and phase locking is to ensure that the output voltage phase, frequency, and amplitude of each power module are consistent, avoiding circulating current or inrush current caused by phase difference during parallel connection. When power module 2 needs to be connected in parallel with power module 1, power module 1 is regarded as the master module and power module 2 as the slave module. The master module load generates a synchronization signal and broadcasts the synchronization clock signal through high-speed communication. After receiving the signal, power module 2 uses a phase-locked loop to track the phase and frequency of the master module, adjusts its own phase to be the same as the master module, and then closes the pre-charge circuit of power module 2, charging its own bus capacitor close to the voltage value of power module 1 through the current-limiting resistor. When the voltage, phase, and frequency of power module 2 match those of the master module, the master module issues a parallel connection command, closes the switching switch S6, marks power module 2 as online, and formally connects it to the system.

[0063] Preferred methods include Figure 3 The pre-charging circuit shown performs pre-charging. In the diagram, FU1~FU3 are fuses used for short-circuit protection of the main circuit. They blow when the current exceeds the rated value, cutting off the circuit. QF1 is a circuit breaker used for manual control of the main circuit's on / off state. KM1 is a pre-charging contactor. Upon power-up, KM1 is first closed, and the downstream bus capacitor is slowly charged through the current-limiting resistor R. Once the voltage on the bus capacitor approaches the voltage value of power module 1, QF1 is closed, KM1 is opened, and pre-charging is completed.

[0064] More specifically, by quantifying the adjustment process of the droop coefficient, the overall power balance of the system is ensured, while achieving smooth power transfer. Still using... Figure 2 For example, assuming the current online power modules in the system are power module 1, power module 2, and power module 3, and power module 4 is inserted into the system, the target value for the droop coefficient of each power module is... After power module 4 is put into the system, its droop factor is first set to an initial value, which is usually a large value to ensure that its initial output current is close to 0. The initial value can be 100. Then, the initial value is gradually decreased, while the droop coefficients of the remaining power modules are increased simultaneously. The droop coefficient of power module 4 gradually decreases by an amount of... Preferably -5 The droop coefficient of the remaining power modules is increased by a large amount. And it satisfies the following formula:

[0065]

[0066] in, For the currently deployed power modules The synchronous increase in quantity, For the currently deployed power modules The original value, As the initial value, The power module after power-on The target value.

[0067] The initial droop coefficient of power module 4 is set to After that, the rest Original value; gradually decrease to ,in Simultaneously increase the rest to After the system stabilizes, repeat the above steps until the droop coefficient of power module 4 reaches the target value. .

[0068] In a preferred embodiment of the present invention, the switching controller 4 further includes a power prediction module 44 connected to the calculation module 41, which is used to collect real-time operating data of each power module 1, switching module 2 and load 3, and input the real-time operating data into the pre-trained load prediction model to obtain the total load power of each load in the next hour; the control module 43 is also used to issue an offline module power-on command in advance when it is determined that an offline power module needs to be put into operation based on the predicted total load power, so as to wake up the offline power module that needs to be put into operation.

[0069] Specifically, in this embodiment, a Long Short-Term Memory (LSTM) network is preferably used to establish a load prediction model. By predicting the total load power demand of the system in the following hour, offline modules are woken up in advance, achieving seamless power switching for the entire system. For example, if the predicted load arrival time is at 30 minutes, a power-on wake-up command can be sent to the target offline module 5-10 minutes in advance, so that it can directly participate in power distribution when the load arrives, without waiting for pre-synchronization, phase locking, and pre-charging. This solves the problems of module deployment delay and low efficiency when the load suddenly increases in the traditional passive response mode, significantly improving the system's response speed and operating economy to dynamic loads. This mechanism enables the system to adapt not only to normal load fluctuations but also to cope with sudden scenarios, further expanding the scenario adaptability of the technical solution.

[0070] Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network (RNN) that effectively solves the gradient vanishing or exploding problems of traditional RNNs by introducing gating mechanisms and cell states. LSTMs can adaptively decide whether to retain or discard information, learning both short-term fluctuations in sequential data and memorizing long-term key patterns, thus performing excellently in time-series predictive tasks such as power load forecasting.

[0071] The steps for establishing a load forecasting model are as follows:

[0072] Step 1: First, collect hourly data for the first six months of the year, and align features such as load value, temperature, number of online power modules, system output power, and power module running time according to time series. Define the first four months as the training set, the fifth month's data as the validation set, and the sixth month's data as the test set.

[0073] Next, the data is preprocessed:

[0074] 1) Missing time series values ​​caused by sensor malfunctions or data transmission errors will be filled using linear interpolation based on the following formula:

[0075]

[0076] In the formula, P represents the size of the missing value. This indicates the load power value for a period of time prior to the missing value. This indicates the size over a period of time after a missing value. This indicates the time interval between the two.

[0077] 2) Use 3 The principle is to replace outliers such as sudden spikes or drops in load with rolling averages, and finally normalize the data according to the following formula:

[0078]

[0079] In the formula, P is the preprocessed input sample matrix; This is the normalized data matrix; This represents the maximum value of the historical input data. This is the minimum value based on historical data.

[0080] 3) Simultaneously, the results of training on the normalized data must be denormalized to obtain the loading value. The denormalization formula is as follows:

[0081]

[0082] In the formula, This is the normalized load forecast value. The predicted load value is in true dimensions.

[0083] Step 2: After completing the data preprocessing process, construct a sliding window sequence from the input features, using the feature sequence from the past 24 hours ( Predict the load for the next hour After completing the sequence construction, a three-layer LSTM deep neural network architecture is built.

[0084] The specific computational process of LSTM modeling is as follows:

[0085] First, calculate the gate coefficients using three formulas. A gate is a coefficient within the interval [0-1], representing the input gate. Forgotten Gate Output gate :

[0086]

[0087]

[0088]

[0089] In the formula, , , These are the weighting coefficients. , , For bias terms, This represents the concatenated vector of the previous output and the current input.

[0090] Based on the input at the current moment Output at the previous moment Calculate the candidate state value of the current neuron ,

[0091]

[0092] From the Gate of Oblivion and input gate Determine the state value at the previous moment and the candidate state value at the current time In the new state value The proportion it accounts for.

[0093]

[0094] Finally, the output value was calculated.

[0095]

[0096] When the above-described LSTM calculation process is applied to predict load power in this invention, wherein... This represents the load forecast value at the previous moment. Represents the data characteristics at the current moment. Forget gate. This indicates the proportion of historical load information retained. If the current period is defined as a peak period, the weighting coefficient is increased to record more historical load data. (Input gate) Control the impact of the input feature data of the current new time period on the load. Candidate state Calculate candidate trends for load change based on the characteristics of the current input data. New state value. The load status at the current moment is obtained by combining the historical load patterns filtered by the forgetting gate with the current new trends. It is an LSTM model that summarizes the load pattern up to time t based on the feature sequence of the past 24 hours. When the feature of the next time step is input, The load value at the next moment can be predicted. .

[0097] Step 3: Finally, the model is trained and validated. The training set data of the module is input into the input layer in batches of 32 samples to balance training speed and memory usage. An early stopping mechanism is added to the training method to monitor the loss of the validation set. If the loss does not decrease for 5 consecutive rounds, the training is terminated to avoid invalid iterations.

[0098] Step 4: Predict and deploy the model. Plot the loss curves and MAE curves for the training and validation sets based on the training results. When both curves decrease synchronously and then stabilize, it indicates that the model has no systematic bias and can be deployed and used. At the same time, add a continuous learning mechanism to the model and perform incremental training with new data every month.

[0099] In a preferred embodiment of the present invention, each power module integrates a self-testing circuit for performing periodic self-tests when the power module is offline, and marking itself as unusable when the periodic self-test results indicate a fault.

[0100] Specifically, in this embodiment, in a modular power redundancy system, fault pre-diagnosis of offline modules is a key step in ensuring system reliability. Traditional methods only detect faults when modules are online, while this invention proposes an offline module fault pre-diagnosis technology that performs real-time health monitoring when modules are offline, preventing faulty modules from being mistakenly put into operation and improving overall system availability. Specifically, power modules connected to the system and bearing the load are marked as online modules; while standby power modules not connected to the system and in a low-power standby state are marked as offline modules.

[0101] This invention integrates a self-testing circuit within the offline module, which executes a self-testing program every 24 hours to check the module's output capability and sensor effectiveness, ensuring the backup module is always available. The self-testing includes: the self-testing circuit inputting the rated voltage to the module and detecting the deviation between the output voltage and the rated voltage; the self-testing circuit inputting the rated current to the module and detecting the deviation between the output current and the rated current; and the self-testing circuit inputting a standard temperature of 25°C to the temperature sensor and detecting the deviation between the sensor reading and the standard temperature. When any of the above deviations exceeds a system-set threshold, the offline module issues a corresponding fault alarm and removes the module from the system module list, effectively marking itself as unusable.

[0102] Based on the above self-testing content, the self-testing circuit preferably includes a current detection circuit and a voltage detection circuit.

[0103] The schematic diagram of the current detection circuit is as follows: Figure 4 As shown, during the offline module's self-test, a rated current is injected into its output terminal. The circuit converts the rated current into a U_I_SAMPLE signal, which is then acquired by the ADC and calculated by the MCU to obtain the sampled current value. The samples obtained from the MUC Compared with the actual input rated current value If the deviation exceeds 5%, a fault alarm will be triggered.

[0104] The schematic diagram of the voltage detection circuit is as follows: Figure 5 As shown, during the offline module self-test, a 220V AC voltage is input to the module. The circuit converts the rated current into a UV_V_SAMPLE signal, which is then acquired by the ADC and calculated by the MCU. The voltage value obtained by the MCU is compared with the actual input voltage value. If the deviation exceeds 5%, a fault alarm is triggered.

[0105] In a preferred embodiment of the present invention, the switching controller 4 further includes an instruction response module 45, which is used to respond to an externally input load increase / decrease instruction containing a load power value when the power redundancy is greater than zero and not greater than the minimum rated power of each currently engaged power module; the switching controller 4 is also used to update the total load power based on the load increase / decrease instruction, and then adjust the number of currently engaged power modules accordingly so that the power redundancy is again in a state that is greater than zero and not greater than the minimum rated power of each currently engaged power module.

[0106] Specifically, in this embodiment, the aforementioned external input load increase / decrease instructions typically come from a host computer system, such as the energy management platform of a data center or the MES system of an industrial production line. The instruction format includes the operation type (increase / decrease) and the load power value, such as +10kW or -8kW.

[0107] Preferably, the instruction response module 45 does not execute external instructions unconditionally, but only when the current power redundancy is within a safe and efficient range. That is, it is only allowed to respond to external input load increase or decrease instructions when the power redundancy is greater than zero and not greater than the minimum rated power of each power module currently in operation, thus effectively avoiding overload risk and efficiency drop.

[0108] For example, a data center currently has four 10kW modules deployed, with a total load of 32kW. The power redundancy is (4×10×0.965) - 32 = 38.6 - 32 = 6.6kW. At this time, the command response module 45 can respond to the external command to add 5kW of load. If the current power redundancy is 11kW, the module will first reject the command to reduce the load. After the control module switches out one redundant module, it will then allow the response, ensuring that the system remains in a high-efficiency state after the command is executed.

[0109] More preferably, the command response module 45 will first verify whether the load power value in the load increase / decrease command is within the system's allowable range, that is, the total load power after the increase is ≤ the effective power when all modules are in operation, and the total load power after the decrease is ≥ the system's minimum operating power. For example, if the system's maximum output is 50kW, and the current total load is 32kW, and the external command is +20kW, then because it exceeds the maximum output, the command response module 45 will directly reject it and report "overload risk, command invalid".

[0110] The above description is merely a preferred embodiment of the present invention and does not limit the implementation and protection scope of the present invention. Those skilled in the art should realize that any equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included within the protection scope of the present invention.

Claims

1. A modular power redundancy dynamic switching control system, characterized in that, include: Multiple power modules are connected to multiple loads via a switching module. The power switch includes a power switch, a switching switch, and a load switch. Each power module is connected to a ring bus composed of the switching switches via a corresponding power switch, and then connected to each load via a corresponding load switch. Each power switch controls the connection between the power module and the ring bus on a one-to-one basis, and each load switch controls the connection between the ring bus and the load on a one-to-one basis. A switching controller, connected to each of the power modules, the switching module, and the load, is used to calculate the power redundancy in real time based on the total load power of each load and the rated power and peak efficiency of each currently engaged power module. This allows the controller to control the switching module's operation, adjusting the number of currently engaged power modules to ensure that the power redundancy is greater than zero and not greater than the minimum rated power of each currently engaged power module. The formula for calculating the power redundancy is as follows: ； in, The power redundancy, For power modules Rated power, For power modules Peak efficiency, The total load power is denoted as .

2. The module power redundancy dynamic switching control system according to claim 1, characterized in that, The switching controller includes: The calculation module is used to calculate the power redundancy in real time based on the total load power of each load and the rated power and peak efficiency of each power module currently in operation. The comparison module, connected to the calculation module, is used to generate a cut-off signal when the power redundancy is greater than the minimum rated power, and to generate an input signal when the power redundancy is not greater than zero and the current number of power modules in operation is less than the total number of power modules. The control module, which is connected to the calculation module and the comparison module respectively, is used to control the power module with the minimum rated power to be cut off according to the cut-off signal and generate a recalculation signal, and to control the power module to be put into operation according to the input signal and generate the recalculation signal. The calculation module recalculates the power redundancy based on the recalculation signal.

3. The module power redundancy dynamic switching control system according to claim 2, characterized in that, The control module includes a cut-out control unit, which issues a soft disconnect command to the power module with the lowest rated power according to the cut-out signal, so as to control the power module to enter the current reduction mode, and at the same time linearly decreases the output current value of the power module to gradually reduce the output current value of the power module until it is zero, and then controls the switching module to disconnect the power module accordingly to achieve smooth cut-out; The expression for the linearly decreasing trend is as follows: ； in, This indicates the initial output current value when the power module begins executing the soft-disconnect command. This indicates that the power module has executed the soft disconnect command to begin disconnecting. Output current value after time t. This indicates the preset total duration of the soft disconnect.

4. The module power redundancy dynamic switching control system according to claim 3, characterized in that, The control module further includes a status monitoring unit connected to the cut-out control unit, used to generate a first control signal when the power module with the lowest rated power is in normal working condition according to the cut-out signal, and to generate a second control signal when the power module with the lowest rated power is in fault condition. The cut-out control unit issues the soft disconnect command based on the first control signal, and controls the switching module to directly disconnect the power module based on the second control signal.

5. The module power redundancy dynamic switching control system according to claim 2, characterized in that, The control module includes: The wake-up unit is used to send an offline module power-on command according to the input signal, so as to control an offline power module to perform pre-synchronization and phase locking and pre-charging to complete the power-on wake-up; The control unit is configured to set the droop coefficient of the power module after power-on to an initial value so that its initial output current is zero. Then, the initial value is gradually reduced while the droop coefficient of the other power modules currently in operation is increased simultaneously until the droop coefficient of the power module after power-on reaches the target value. This gradually transfers the power of the other power modules currently in operation to the power module after power-on, thereby achieving smooth power-on.

6. The module power redundancy dynamic switching control system according to claim 5, characterized in that, The expression for the synchronous increase of the droop coefficient of the remaining power modules currently in operation is as follows: ； in, For the currently deployed power modules The synchronous increase in quantity, For the currently deployed power modules The original value, The initial value is... The power module after power-on The target value.

7. The module power redundancy dynamic switching control system according to claim 2, characterized in that, The switching controller further includes a power prediction module connected to the calculation module, used to collect real-time operating data of each power module, the switching module and the load, and input the real-time operating data into a pre-trained load prediction model to obtain the total load power of each load in the next hour; The control module is also used to issue an offline module power-on command in advance when it is determined that the offline power module needs to be put into operation based on the predicted total load power, so as to wake up the offline power module that needs to be put into operation.

8. The module power redundancy dynamic switching control system according to claim 1, characterized in that, Each of the power modules has an integrated self-testing circuit, which is used to perform periodic self-tests when the power module is offline, and to mark itself as unusable when the periodic self-test results indicate that a fault exists.

9. The module power redundancy dynamic switching control system according to claim 1, characterized in that, The switching controller further includes an instruction response module, which is used to respond to an externally input load increase or decrease instruction containing the load power value when the power redundancy is greater than zero and not greater than the minimum rated power of each of the currently engaged power modules. The switching controller is also used to update the total load power based on the load increase / decrease command, and then adjust the number of currently engaged power modules accordingly so that the power redundancy is back to a state that is greater than zero and not greater than the minimum rated power of each currently engaged power module.

Images