A dynamic optimization control method and system for a multi-modal pump group

Abstract

The present invention relates to the field of pump group control, and specifically to a multi-modal pump group dynamic optimization control method and system. The method includes collecting the current waveform, pressure pulsation spectrum and flow time series data of each unit of the pump group through a distributed sensor array, and generating an operating state tensor through time-frequency domain fusion processing. Input a multi-modal recognition neural network, and output a dominant mode identifier and a modal confidence vector. According to the modal confidence, activate the basic optimization strategy library or execute cross-modal strategy collaboration to generate optimization instructions. Then dynamically select the instruction conversion path to obtain the inverter parameter set, load it into the programmable logic controller, and drive the actuator to converge the operating trajectory. The present invention can effectively cope with changes in multi-modal operating states and complex working conditions, and realize dynamic optimization of the pump group.

Description

Technical Field

[0001] The present invention relates to the field of pump group control, and in particular to a multi-modal pump group dynamic optimization control method and system. Background Art

[0002] In modern industrial and civilian fields, pump group systems are widely used in various fluid transportation scenarios, and their operating efficiency and stability have an important impact on the production process and quality of life. Traditional pump group control methods mainly rely on single sensor data and fixed control strategies, which are difficult to adapt to complex working condition changes and multi-modal operating states. For example, the invention patent application with patent number CN113464412A discloses a multi-pump parallel coordinated control method. This method pre-sets the output expected water pressure value, selects one of the multiple intelligent control water pumps in the multi-pump parallel system as the host, and the remaining intelligent control water pumps are slaves. The actual output water pressure value of the multi-pump parallel system is collected, and the actual output water pressure value is compared with the output expected water pressure value. According to the comparison result, the operating state of all intelligent control water pumps is regulated to achieve the effect of pressure stabilization. However, this method only focuses on the stable control of water pressure, lacks comprehensive monitoring and dynamic optimization of the multi-modal operating state of the pump group, and is difficult to meet the efficient operation requirements of the pump group system under complex working conditions. Summary of the Invention

[0003] The purpose of the present invention is to provide a multi-modal pump group dynamic optimization control method and system to solve the problem that the pump group control method in the prior art cannot effectively cope with multi-modal operating states and complex working condition changes.

[0004] To achieve the above objectives, the following technical solutions are adopted.

[0005] A multi-mode pump group dynamic optimization control method includes the following steps:

[0006] S1. Use a distributed sensor array to collect real-time current waveform data, pressure pulsation spectrum, and flow rate time series of each pump unit in the pump system, and perform time-frequency fusion processing on the three types of data to generate a pump system operation status tensor;

[0007] S2. Input the pump group operation state tensor into a pre-built multimodal recognition neural network, and output the current dominant mode identifier and modal confidence vector. The dominant mode identifier includes the high-efficiency zone operation mode, overload warning mode, low-frequency oscillation mode, flow mismatch mode and mixed conflict mode;

[0008] S3. When the maximum value in the modal confidence vector exceeds a preset threshold, the basic optimization strategy library corresponding to the dominant modal identifier is activated and a basic optimization instruction is output; when a mixed conflicting mode is identified or the maximum value in the modal confidence vector does not exceed the preset threshold, the cross-modal strategy coordination step is executed:

[0009] Extract the high-dimensional feature vector of the current running state tensor;

[0010] Calculate the compatibility weight of each basic optimization strategy;

[0011] Generate hybrid optimization instructions based on weight fusion;

[0012] S4. Dynamically select an instruction conversion path based on the optimized instruction type output in step S3 to obtain an inverter parameter set;

[0013] S5. Load the inverter parameter set into the programmable logic controller of the target pump unit, and drive the actuator of the target pump unit to make the pump group operation trajectory converge to the target parameter envelope.

[0014] Optionally, also include,

[0015] S6. Collect the system response matrix after the instruction is executed, calculate the strategy effectiveness evaluation index, and trigger any of the following based on the index deviation:

[0016] When the indicator deviation is ≤ the preset value, fine-tune the convolution kernel parameters of the multimodal recognition neural network;

[0017] When the indicator deviation is greater than the preset value, the decision tree topology of the basic optimization strategy library is reconstructed.

[0018] Optionally, the specific steps of step S1 include:

[0019] S11, synchronously collecting current waveform data, pressure pulsation spectrum and flow rate time series of each pump unit through a distributed sensor array;

[0020] S12, performing fundamental wave separation processing on the current waveform data, and outputting the current effective value component and harmonic distortion component;

[0021] S13, performing frequency band energy integration on the pressure pulsation spectrum, and outputting a set of main frequency band amplitudes and a sub-frequency band energy distribution ratio;

[0022] S14. Perform sliding window statistical analysis on the traffic time series and output the traffic mean series and fluctuation variance series;

[0023] S15. Integrate the current effective value component and harmonic distortion component output from step S12, the main frequency band amplitude set and sub-frequency band energy distribution ratio output from step S13, and the flow mean value sequence and fluctuation variance sequence output from step S14 according to the following structure:

[0024] The first physical dimension: the spatial location identification of each pump unit;

[0025] The second physical dimension: physical characteristic categories of the three parameters of current / pressure / flow;

[0026] The third physical dimension: the number of consecutive sampling time windows;

[0027] S16. Perform feature dimensionality reduction compression on the integrated data structure to generate the pump group operation status tensor.

[0028] Optionally, the specific steps of step S2 include:

[0029] S21, inputting the pump group operation status tensor into the first convolutional layer of a pre-built multimodal recognition neural network, extracting spatial dimension features, and outputting a primary feature map;

[0030] S22, inputting the primary feature map into the time dimension convolution layer, extracting time series correlation features, and outputting a spatiotemporal fusion feature tensor;

[0031] S23, inputting the spatiotemporal fusion feature tensor into a modal classifier, calculating the matching probability value of each preset modality, and generating a modal confidence vector;

[0032] S24. Perform a maximum value screening operation on the modal confidence criterion vector:

[0033] If the maximum value is greater than or equal to the first threshold and the standard deviation of the probability distribution is less than or equal to the second threshold, the dominant mode identifier corresponding to the single mode is output;

[0034] If the maximum value is less than the first threshold or the standard deviation of the probability distribution is greater than the second threshold, output a mixed conflict mode identifier;

[0035] The preset modes include: high-efficiency zone operation mode, overload warning mode, low-frequency oscillation mode, and flow mismatch mode.

[0036] Optionally, the construction and calling of the basic optimization strategy library includes:

[0037] receiving the dominant modality identifier as a strategy selection signal;

[0038] When the strategy selection signal is the high-efficiency zone operation mode: obtain the current pump group total efficiency benchmark value, calculate the efficiency deviation coefficient of each pump unit, generate the pump unit start and stop priority queue based on the efficiency deviation coefficient, and output the basic optimization instruction containing the target start and stop pump number;

[0039] When the strategy selection signal is the overload warning mode: extract the real-time load rate of each pump unit, identify the overloaded pump unit and the underloaded pump unit set, calculate the load transfer matrix, and output the basic optimization instructions including the load redistribution parameters;

[0040] When the strategy selection signal is a low-frequency oscillation mode: obtain the pressure pulsation main frequency phase angle set, calculate the phase compensation amount and damping injection amount, and output the basic optimization instruction including the phase offset instruction;

[0041] When the strategy selection signal is the flow mismatch mode: analyze the target flow-actual flow difference distribution, generate a flow correction gain coefficient vector, and output a basic optimization instruction including the gain coefficient.

[0042] Optionally, the cross-modal strategy collaboration step specifically includes:

[0043] Extracting the time domain feature vector, the frequency domain feature vector, and the time series correlation feature vector from the pump group operation state tensor outputted in step S2, and combining them to generate a high-dimensional feature vector;

[0044] Calculate the matching degree between the high-dimensional feature vector and the constraints of each basic optimization strategy: calculate the energy efficiency adaptability of the high-efficiency zone strategy, calculate the load balancing degree of the overload strategy, calculate the phase compatibility of the oscillation strategy, calculate the flow responsiveness of the mismatch strategy, and output the strategy compatibility measurement set;

[0045] Normalized weighting is performed based on the policy compatibility metric set: when a single metric value exists ≥ 0.8, the policy weight is set to 0.7. When all metric values ​​are < 0.8, the weight coefficients are proportionally distributed to generate a policy weight vector.

[0046] The basic optimization strategy instructions are linearly superimposed according to the strategy weight vector: the start and stop priority queue of the high-efficiency zone strategy is extracted, the load transfer matrix of the overload strategy is extracted, the phase compensation amount of the oscillation strategy is extracted, and the flow correction gain of the mismatch strategy is extracted. The hybrid optimization instructions are generated by fusing them according to the weight coefficients.

[0047] Optionally, the step S4 specifically includes:

[0048] S41, determine the type of optimization instruction output in step S3:

[0049] If the basic optimization instruction is executed, a linear conversion path is performed:

[0050] Extract the start-stop priority queue, load transfer matrix, phase compensation and flow correction gain from the basic optimization instructions;

[0051] Each parameter is converted into frequency modulation amount, phase offset amount and power compensation amount through a linear mapping function;

[0052] Combining and generating a first inverter parameter set;

[0053] If it is a hybrid optimization instruction, a nonlinear conversion path is executed:

[0054] Parse the strategy weight vector and basic strategy parameter set in the hybrid optimization instruction;

[0055] Feed the policy weight vector into the pre-trained neural network transformer;

[0056] Generate frequency modulation waveform, phase compensation curve and power response surface through nonlinear transformation;

[0057] Integrate and generate a second inverter parameter set;

[0058] S42. Perform physical dimension alignment on the converted inverter parameter set: extract the spatial position identifier of the pump group, and reorganize the parameter set into a three-dimensional parameter matrix according to the spatial position identifier, where:

[0059] The first dimension: pump unit spatial index,

[0060] The second dimension: time window sequence,

[0061] The third dimension: parameter type channel;

[0062] S43 , outputting the reorganized three-dimensional parameter matrix as the inverter parameter set for loading and executing in step S5 .

[0063] Optionally, step S5 specifically includes:

[0064] S51, parsing the inverter parameter set outputted in step S4, and extracting a parameter subset corresponding to the spatial position identifier of the target pump unit;

[0065] S52. Based on the parameter type channel identifier of the parameter subset: if it is a frequency modulation amount, generate a PWM waveform loading instruction; if it is a phase offset amount, generate a time delay compensation instruction; if it is a power compensation amount, generate an IGBT trigger pulse sequence; and combine them into a programmable logic controller executable code block;

[0066] S53, injecting the executable code block into the programmable logic controller runtime environment of the target pump unit to drive the actuator to generate a mechanical adjustment action;

[0067] S54. Real-time collection of feedback signals from the actuator: actual change in motor speed, valve opening adjustment value, and pressure transient response value, to construct an execution response triplet.

[0068] S55. Dynamically compare the execution response triplet with the boundary threshold of the target parameter envelope:

[0069] If all parameters ∈ [lower threshold, upper threshold], it is determined to be in convergence state.

[0070] If any parameter exceeds [lower threshold, upper threshold], perform trajectory correction: calculate the parameter deviation gradient value, generate the parameter compensation value based on the gradient value, add the parameter compensation value to the parameter subset in step S51, and trigger a new round of execution loading;

[0071] S56: When the convergence state lasts for more than three control cycles, a closed-loop confirmation signal is sent to step S6.

[0072] Optionally, step S6 specifically includes:

[0073] S61, receiving the execution response triplet outputted in step S54, and generating a system response matrix by aggregating the time window sequence, wherein the row dimension of the matrix corresponds to the pump unit spatial index, and the column dimension corresponds to the response parameter type;

[0074] S62. Extract the following strategic effectiveness evaluation factors from the system response matrix: deviation of the actual energy efficiency improvement rate from the target value, load balancing standard deviation, pressure oscillation attenuation rate, flow tracking response delay, and calculate the strategic effectiveness evaluation index based on the weighted factors;

[0075] S63, calculating the deviation between the strategy effectiveness evaluation index and the target reference value, wherein the target reference value is set based on the historical optimal operation data of the pump group;

[0076] S64, when the indicator deviation is less than or equal to the preset value, performing model fine-tuning: extracting the convolution kernel parameter set of the multimodal recognition neural network in step S2, calculating the parameter gradient based on the indicator deviation, and updating the convolution kernel parameters in the gradient direction to generate a fine-tuned model;

[0077] S65. When the indicator deviation is greater than a preset value, perform strategy reconstruction: obtain the decision tree topology of the basic optimization strategy library, analyze the failure path nodes in the system response matrix, reconstruct the decision tree branch structure based on the node failure frequency, and output the reconstructed decision tree topology;

[0078] S66. Load the fine-tuned model or the reconstructed decision tree topology into the operating environment for use in the next control cycle.

[0079] A multi-mode pump group dynamic optimization control system, comprising:

[0080] Multi-source sensing acquisition module, used to collect the current waveform data, pressure pulsation spectrum and flow time series of each pump unit in the pump group system in real time through a distributed sensor array;

[0081] A time-frequency fusion module is used to perform time-frequency fusion processing on the current waveform data, pressure pulsation spectrum and flow time series to generate a pump group operation status tensor;

[0082] The modal identification module is used to input the pump group operation status tensor into the pre-built multi-modal identification neural network and output the current dominant mode identifier and modal confidence vector;

[0083] A strategy selection module is used to activate the basic optimization strategy library to output basic optimization instructions when the maximum value in the modal confidence vector exceeds a preset threshold, and to execute cross-modal strategies to collaboratively generate hybrid optimization instructions when mixed conflicting modes are identified or the maximum value in the modal confidence vector does not exceed the preset threshold;

[0084] An instruction conversion module is used to dynamically select a conversion path to generate an inverter parameter set according to the optimized instruction type;

[0085] An execution control module is used to load the inverter parameter set into the programmable logic controller of the target pump unit and drive the actuator to make the pump group operation trajectory converge to the target parameter envelope;

[0086] The adaptive optimization module is used to collect the system response matrix calculation strategy effectiveness evaluation indicators, and trigger the convolution kernel parameter fine-tuning of the modal recognition module or the decision tree topology reconstruction of the basic optimization strategy library according to the indicator deviation.

[0087] Compared with the prior art, the present invention has the following beneficial effects:

[0088] This application addresses the problem that the existing pump group control methods are unable to effectively cope with multi-modal operating states and complex working condition changes, and proposes a multi-modal pump group dynamic optimization control method. This method uses a distributed sensor array to collect the current waveform data, pressure pulsation spectrum and flow time series of each pump unit in the pump group system in real time, and performs time-frequency domain fusion processing on these data to generate a pump group operating state tensor. Then, a pre-built multi-modal recognition neural network is used to identify the current dominant mode, and the corresponding optimization strategy is selected according to the modal confidence vector. Finally, the optimization instruction is converted into a frequency converter parameter set and loaded into the programmable logic controller of the target pump unit to achieve dynamic optimization of the pump group operation trajectory.

[0089] First, by integrating multimodal data and using a neural network for modal identification, the current operating mode of the pump set can be accurately identified, thereby providing targeted optimization strategies for different operating states. This effectively addresses the existing technology's inability to adapt to multimodal operating states. Second, this method dynamically selects the conversion path for optimization instructions and generates inverter parameter sets adapted to different operating conditions, further improving the pump set's operating efficiency and stability. Furthermore, by collecting the system response matrix after instruction execution and calculating strategy effectiveness evaluation indicators, the effectiveness of the optimization strategy can be monitored in real time. Model fine-tuning or strategy reconstruction can be triggered based on indicator deviations, ensuring long-term stable operation of the system. Other technical benefits include: improved data accuracy and reliability through detailed data collection and processing steps; enhanced modal identification accuracy through a specific modal identification step; refined optimization for different modalities through the construction and invocation of a basic optimization strategy library; improved adaptability and flexibility of the optimization strategy through cross-modal strategy coordination steps; enhanced dynamic response capabilities of the system through dynamic selection of instruction conversion paths; improved control accuracy and response speed of the actuator through specific execution control steps; and adaptive optimization and long-term stable operation of the system achieved through the adaptive optimization module. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 The figure is a flowchart of an embodiment of a multi-modal pump group dynamic optimization control method according to the present invention.

[0091] Figure 2 It is a schematic diagram of the module structure of an embodiment of a multi-modal pump group dynamic optimization control system according to the present invention. DETAILED DESCRIPTION

[0092] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0093] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise indicated, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present invention.

[0094] Example 1

[0095] like Figure 1As shown, this embodiment provides a multimodal pump group dynamic optimization control method. This method uses a distributed sensor array to collect real-time current waveform data, pressure pulsation spectrum, and flow rate time series from each pump unit in the pump group system. This data is then fused in the time and frequency domain to generate a pump group operating state tensor. A pre-built multimodal recognition neural network is then used to identify the current dominant mode. The corresponding optimization strategy is selected based on the modal confidence vector. Finally, the optimization instructions are converted into a frequency converter parameter set and loaded into the programmable logic controller of the target pump unit, achieving dynamic optimization of the pump group's operating trajectory.

[0096] Step S1: Data collection and processing

[0097] In the pump system, a distributed sensor array collects current waveform data, pressure pulsation spectrum, and flow rate time series from each pump unit in real time. Specifically, current waveform data is collected by high-precision current sensors, pressure pulsation spectrum by high-sensitivity pressure sensors, and flow rate time series by high-precision flow sensors. These sensors, distributed across each pump unit in the system, enable real-time and accurate collection of critical data during pump operation.

[0098] The collected current waveform data is first processed by fundamental wave separation to separate the current effective value component and harmonic distortion component. The specific steps are as follows:

[0099] Fast Fourier transform (FFT) is used to perform frequency domain analysis on the current waveform data to extract the fundamental wave and each harmonic.

[0100] Calculate the effective value of the fundamental wave as the effective value component of the current.

[0101] Calculate the total effective value of each harmonic as the harmonic distortion component.

[0102] The pressure pulsation spectrum performs frequency band energy integration and outputs the main frequency band amplitude set and the sub-frequency band energy distribution ratio. The specific steps are as follows:

[0103] The pressure pulsation spectrum is divided into multiple frequency bands, such as 0-10 Hz, 10-20 Hz, 20-30 Hz, etc.

[0104] Integrate the energy of each frequency band and calculate the total energy of the frequency band.

[0105] Extract the amplitude set of the main frequency bands, that is, the amplitudes of the frequency bands with the largest energy.

[0106] Calculate the energy distribution ratio of the sub-bands, that is, the ratio of the energy of each sub-band to the total energy.

[0107] Perform sliding window statistical analysis on the traffic time series and output the traffic mean series and fluctuation variance series. The specific steps are as follows:

[0108] Use a sliding window to segment the traffic time series. The window size can be set according to actual needs, such as 1 second, 5 seconds, etc.

[0109] The mean of the traffic data in each window is calculated to form a traffic mean sequence.

[0110] The variance of the traffic data in each window is calculated to form a traffic fluctuation variance sequence.

[0111] The processed data are integrated into the following structure:

[0112] First physical dimension: the spatial location identification of each pump unit, such as pump unit 1, pump unit 2, etc.

[0113] The second physical dimension: physical characteristic categories of the three parameters of current / pressure / flow.

[0114] The third physical dimension: the consecutive sampling time window numbers, such as window 1, window 2, and so on.

[0115] The integrated data structure is further compressed by feature dimensionality reduction to generate the pump group operation status tensor. The specific steps are as follows:

[0116] Principal component analysis (PCA) was used to reduce the dimension of the integrated data and extract the main features.

[0117] The reduced feature data is constructed into a multi-dimensional tensor, where each dimension corresponds to a different physical meaning, such as spatial position, parameter category, time window, etc.

[0118] Step S2: Modal identification and optimization strategy selection

[0119] The generated pump group operating status tensor is input into a pre-built multimodal recognition neural network. The neural network extracts spatial dimension features through the first convolution layer and outputs a primary feature map. The convolution layer then extracts time series correlation features through the time dimension and outputs a spatiotemporal fusion feature tensor. The specific steps are as follows:

[0120] The first convolutional layer uses multiple convolution kernels to perform convolution operations on the running state tensor to extract features of the spatial dimension.

[0121] The time dimension convolution layer performs convolution operations on the primary feature map, extracts the correlation features of the time series, and generates a spatiotemporal fusion feature tensor.

[0122] The spatiotemporal fusion feature tensor is further input into the modal classifier to calculate the matching probability value of each preset mode and generate the modal confidence vector. The specific steps are as follows:

[0123] Modality classifiers can use classification algorithms such as support vector machines (SVM) and decision trees.

[0124] Classify the spatiotemporal fusion feature tensors and calculate the matching probability value of each preset modality.

[0125] The matching probability values ​​are constructed into a modal confidence vector, for example, [0.8, 0.1, 0.05, 0.05], which means that the probability of the high-efficiency zone operating mode is 0.8, and the probabilities of the other modes are 0.1, 0.05, and 0.05, respectively.

[0126] In step S3, the maximum value in the modal confidence vector is used to determine the current dominant mode. If the maximum value exceeds a preset threshold and the probability distribution standard deviation is within a certain range, the dominant mode identifier corresponding to the single mode is output; otherwise, the mixed conflicting mode identifier is output. The specific steps are as follows:

[0127] Set the preset threshold, such as 0.7, and the probability distribution standard deviation threshold, such as 0.1.

[0128] If the maximum value is ≥ 0.7 and the standard deviation of the probability distribution is ≤ 0.1, the dominant mode identifier corresponding to the single mode is output.

[0129] Otherwise, output the mixed conflict modality identifier.

[0130] The preset modes include high-efficiency zone operation mode, overload warning mode, low-frequency oscillation mode, flow mismatch mode, etc. When a single mode is identified, the basic optimization strategy library corresponding to the dominant mode identifier is activated and the basic optimization instructions are output. The specific steps are as follows:

[0131] High-efficiency zone operation mode: obtain the current pump group total efficiency benchmark value, calculate the efficiency deviation coefficient of each pump unit, generate the pump unit start and stop priority queue, and output the basic optimization instructions including the target start and stop pump number.

[0132] Calculate the efficiency deviation coefficient of each pump unit, for example, efficiency deviation coefficient = current efficiency / reference efficiency.

[0133] A start-stop priority queue is generated according to the efficiency deviation coefficient, and the pump units with high efficiency are started first and the pump units with low efficiency are stopped.

[0134] Overload warning mode: extracts the real-time load rate of each pump unit, identifies overloaded pump units and underloaded pump unit sets, calculates the load transfer matrix, and outputs basic optimization instructions including load redistribution parameters.

[0135] Calculate the real-time load rate of each pump unit, for example, load rate = current load / rated load.

[0136] Identify overloaded pump units and underloaded pump units, calculate a load transfer matrix, and transfer the load of the overloaded pump units to the underloaded pump units.

[0137] Low-frequency oscillation mode: Obtain the phase angle set of the main frequency of pressure pulsation, calculate the phase compensation amount and damping injection amount, and output basic optimization instructions including phase offset instructions.

[0138] Calculate the pressure pulsation main frequency phase angle set, for example, by extracting the main frequency phase angle through Fourier transform.

[0139] The phase compensation amount and damping injection amount are calculated, and a phase offset instruction is generated to adjust the operating phase of the pump unit and reduce oscillation.

[0140] Flow mismatch mode: Analyze the target flow-actual flow difference distribution, generate a flow correction gain coefficient vector, and output basic optimization instructions including the gain coefficient.

[0141] Calculate the difference distribution between the target flow and the actual flow, for example, difference = target flow - actual flow.

[0142] A flow correction gain coefficient vector is generated to adjust the flow rate of the pump unit so that the actual flow rate is close to the target flow rate.

[0143] When mixed conflicting modes are identified or the maximum value in the modal confidence vector does not exceed the preset threshold, the cross-modal strategy coordination step is executed. The specific steps are as follows:

[0144] The time domain feature vector, frequency domain feature vector and time series correlation feature vector are extracted from the pump group operation status tensor and combined to generate a high-dimensional feature vector.

[0145] The matching degree between the high-dimensional feature vector and the constraints of each basic optimization strategy is calculated, and a strategy compatibility measurement set is output.

[0146] Calculate the energy efficiency adaptability of the high-efficiency zone strategy, for example, energy efficiency adaptability = current energy efficiency / target energy efficiency.

[0147] Calculate the load balancing degree of the overload policy, for example, load balancing degree = current load balancing degree / target load balancing degree.

[0148] Calculate the phase compatibility of the oscillation strategy, for example, phase compatibility = current phase compatibility / target phase compatibility.

[0149] The traffic responsiveness of the mismatch policy is calculated, for example, traffic responsiveness = current traffic responsiveness / target traffic responsiveness.

[0150] Normalized weighting is performed based on the policy compatibility metric set to generate a policy weight vector.

[0151] When there is a single metric value ≥ 0.8, the strategy weight is set to 0.7.

[0152] When all metric values ​​are less than 0.8, weight coefficients are allocated proportionally to generate a strategy weight vector.

[0153] The basic optimization strategy instructions are linearly superimposed according to the strategy weight vector to generate hybrid optimization instructions.

[0154] Extract the start and stop priority queues for the efficient zone policy.

[0155] Extract the load transfer matrix of the overload strategy.

[0156] Extract the phase compensation amount of the oscillation strategy.

[0157] Extract the flow correction gain of the mismatch strategy.

[0158] Generate hybrid optimization instructions by fusing weight coefficients.

[0159] Instruction conversion and execution control

[0160] Step S4: Dynamically select the instruction conversion path based on the optimized instruction type. The specific steps are as follows:

[0161] Basic optimization instructions:

[0162] Extract the start-stop priority queue, load transfer matrix, phase compensation and flow correction gain from the basic optimization instructions.

[0163] Each parameter is converted into frequency modulation amount, phase offset amount and power compensation amount through linear mapping function.

[0164] The first inverter parameter set is generated by combining.

[0165] Hybrid optimization instructions:

[0166] Parse the strategy weight vector and basic strategy parameter set in the hybrid optimization instruction.

[0167] Feed the policy weight vector into the pre-trained neural network transformer.

[0168] Frequency modulation waveform, phase compensation curve and power response surface are generated through nonlinear transformation.

[0169] The second inverter parameter set is generated by integration.

[0170] Perform physical dimension alignment on the converted inverter parameter set, extract the pump group spatial location identifier, and reorganize the parameter set into a three-dimensional parameter matrix according to the spatial location identifier. The specific steps are as follows:

[0171] Extract the spatial location identifier of the pump group, such as pump unit 1, pump unit 2, etc.

[0172] The parameter set is reorganized into a three-dimensional parameter matrix according to the spatial position identifier, where:

[0173] First dimension: pump unit spatial index.

[0174] Second dimension: time window sequence.

[0175] The third dimension: parameter type channel.

[0176] The reorganized three-dimensional parameter matrix is ​​output as the inverter parameter set for loading and execution.

[0177] Step S5: Load the inverter parameter set into the programmable logic controller of the target pump unit to drive the actuator to produce mechanical adjustment action. The specific steps are as follows:

[0178] The inverter parameter set is parsed, and a parameter subset corresponding to the spatial position identifier of the target pump unit is extracted.

[0179] Channel identification according to parameter type of parameter subset:

[0180] If it is a frequency modulation amount, a PWM waveform loading instruction is generated.

[0181] If it is a phase offset, a time delay compensation instruction is generated.

[0182] If it is a power compensation amount, an IGBT trigger pulse sequence is generated.

[0183] Combined into programmable logic controller executable code blocks.

[0184] The executable code block is injected into the programmable logic controller runtime environment of the target pump unit to drive the actuator to produce mechanical adjustment action.

[0185] Collect the feedback signals of the actuator in real time, including the actual change in motor speed, valve opening adjustment value, and pressure transient response value, and construct the execution response triplet. The specific steps are as follows:

[0186] The actual change in motor speed is acquired, for example, by a speed sensor.

[0187] The valve opening adjustment value is acquired, for example via a position sensor.

[0188] Acquire the transient pressure response value, for example, using a pressure sensor.

[0189] Construct an execution response triplet, such as (speed change, valve opening adjustment value, pressure response value).

[0190] Dynamically compare the execution response triplet with the boundary threshold of the target parameter envelope. The specific steps are as follows:

[0191] If all parameters are within the threshold range, for example, [lower threshold, upper threshold], it is determined to be in a converged state.

[0192] If any parameter exceeds the threshold range, trajectory correction is performed:

[0193] Calculate the parameter deviation gradient value, for example, by difference calculation.

[0194] A parameter compensation is generated based on the gradient value, such as by a proportional-integral-derivative (PID) controller.

[0195] The parameter compensation amount is added to the parameter subset to trigger a new round of execution loading.

[0196] When the convergence state lasts for more than 3 control cycles, a closed-loop confirmation signal is sent to the subsequent steps.

[0197] Step S6: Adaptive optimization

[0198] Collect the system response matrix after the instruction is executed, calculate the strategy effectiveness evaluation index, and trigger model fine-tuning or strategy reconstruction based on the index deviation. The specific steps are as follows:

[0199] The system response matrix is ​​generated by aggregation according to the time window sequence, where the row dimension of the matrix corresponds to the pump unit spatial index and the column dimension corresponds to the response parameter type.

[0200] Extracting strategic effectiveness evaluation factors from the system response matrix, including:

[0201] The deviation of the actual energy efficiency improvement rate from the target value.

[0202] Load balancing standard deviation.

[0203] Pressure oscillation decay rate.

[0204] Traffic tracking response delay.

[0205] Calculate strategy effectiveness evaluation indicators based on factor weighting, such as weighted sum.

[0206] Calculate the deviation between the strategy effectiveness evaluation index and the target benchmark value, which is set based on the historical optimal operating data of the pump group. The specific steps are as follows:

[0207] Set target benchmark values, such as energy efficiency improvement rate benchmark value, load balancing benchmark value, etc.

[0208] Calculate the deviation between the strategy effectiveness evaluation indicator and the target benchmark value, for example, deviation = current indicator - benchmark value.

[0209] When the indicator deviation is less than or equal to the preset value, perform model fine-tuning. The specific steps are as follows:

[0210] Extract the set of convolution kernel parameters of the multimodal recognition neural network.

[0211] Calculate parameter gradients based on indicator deviations, for example, via a backpropagation algorithm.

[0212] Update the convolution kernel parameters according to the gradient direction to generate the fine-tuned model.

[0213] When the indicator deviation is greater than the preset value, the strategy reconstruction is executed. The specific steps are as follows:

[0214] Get the decision tree topology of the basic optimization strategy library.

[0215] Resolve failure path nodes in the system response matrix.

[0216] The decision tree branch structure is reconstructed based on the node failure frequency, and the reconstructed decision tree topology is output.

[0217] The fine-tuned model or reconstructed decision tree topology is loaded into the running environment for calling in the next control cycle.

[0218] Through the above steps, this embodiment provides a multimodal pump system dynamic optimization control method that can effectively cope with the multimodal operating states and complex operating conditions of the pump system, improving the operating efficiency and stability of the pump system. This method not only monitors and optimizes the operating state of the pump system in real time, but also achieves long-term stable operation of the system through an adaptive optimization module, thus having broad application prospects and significant economic benefits.

[0219] Example 2

[0220] like Figure 2 As shown, this embodiment provides a multimodal pump system dynamic optimization control system. This system is based on the multimodal pump system dynamic optimization control method described in Example 1 above. Through the collaborative operation of multiple modules, the system achieves efficient and stable operation of the pump system. The system includes a multi-source sensing acquisition module, a time-frequency fusion module, a modal identification module, a strategy selection module, an instruction conversion module, an execution control module, and an adaptive optimization module.

[0221] The multi-source sensing acquisition module is responsible for collecting the current waveform data, pressure pulsation spectrum and flow time series of each pump unit in the pump group system in real time through a distributed sensor array. Specifically, this module includes:

[0222] Current sensor: A high-precision current sensor used to collect current waveform data of each pump unit in real time.

[0223] Pressure sensor: High-sensitivity pressure sensor, used to collect the pressure pulsation spectrum of each pump unit in real time.

[0224] Flow sensor: A high-precision flow sensor used to collect the flow timing sequence of each pump unit in real time.

[0225] These sensors are distributed on each pump unit of the pump system, and can collect key data of the pump operation process in real time and accurately, and transmit the collected data to the time-frequency fusion module.

[0226] The time-frequency fusion module receives data transmitted by the multi-source sensor acquisition module, performs time-frequency fusion processing on the current waveform data, pressure pulsation spectrum and flow time series, and generates the pump group operation status tensor. The specific processing steps include:

[0227] Fundamental wave separation processing: Perform fundamental wave separation processing on the current waveform data to extract the current effective value component and harmonic distortion component.

[0228] Band Energy Integration: Performs band energy integration on the pressure pulsation spectrum and outputs the main frequency band amplitude set and the sub-band energy distribution ratio.

[0229] Sliding window statistical analysis: Perform sliding window statistical analysis on the traffic time series and output the traffic mean series and fluctuation variance series.

[0230] Data integration: Integrate the processed data into the following structure:

[0231] The first physical dimension: the spatial location identification of each pump unit.

[0232] The second physical dimension: physical characteristic categories of the three parameters of current / pressure / flow.

[0233] The third physical dimension: the number of consecutive sampling time windows.

[0234] Feature dimensionality reduction and compression: Perform feature dimensionality reduction and compression on the integrated data structure to generate the pump group operation status tensor.

[0235] The generated pump group operating state tensor will be transmitted to the modal identification module.

[0236] The modal identification module receives the pump group operating status tensor generated by the time-frequency fusion module, identifies the current dominant mode through a pre-built multi-modal identification neural network, and outputs a modal confidence vector. The specific steps include:

[0237] Spatial dimension feature extraction: The pump group operating status tensor is input into the first convolutional layer of the multimodal recognition neural network to extract spatial dimension features and output a primary feature map.

[0238] Time series correlation feature extraction: Input the primary feature map into the time dimension convolution layer, extract the time series correlation features, and output the spatiotemporal fusion feature tensor.

[0239] Modal classification: The spatiotemporal fusion feature tensor is input into the modal classifier to calculate the matching probability value of each preset modality and generate a modal confidence vector.

[0240] Modal judgment: The current dominant mode is determined based on the maximum value and standard deviation of the probability distribution in the modal confidence vector. If the maximum value exceeds the preset threshold and the standard deviation of the probability distribution is within a certain range, the dominant mode identifier corresponding to the single mode is output; otherwise, the mixed conflicting mode identifier is output.

[0241] The modal identification module transmits the modal confidence vector and the dominant modal identifier to the strategy selection module.

[0242] The strategy selection module selects the corresponding optimization strategy based on the modal confidence vector and dominant modal identifier output by the modal identification module and outputs the optimization instructions. The specific steps include:

[0243] Activation of the basic optimization strategy library: When the maximum value in the modal confidence vector exceeds the preset threshold, the basic optimization strategy library corresponding to the dominant modal identifier is activated and the basic optimization instruction is output.

[0244] Cross-modal strategy collaboration: When a mixed conflicting mode is identified or the maximum value in the modal confidence vector does not exceed the preset threshold, the cross-modal strategy collaboration step is executed to generate a mixed optimization instruction.

[0245] The specific implementation method of the basic optimization strategy library and the cross-modal strategy collaboration is consistent with the steps described in Example 1. The strategy selection module transmits the optimization instruction to the instruction conversion module.

[0246] The instruction conversion module dynamically selects the instruction conversion path based on the optimized instruction type output by the strategy selection module and generates the inverter parameter set. The specific steps include:

[0247] Basic optimization instruction conversion:

[0248] Extract the start-stop priority queue, load transfer matrix, phase compensation and flow correction gain from the basic optimization instructions.

[0249] Each parameter is converted into frequency modulation amount, phase offset amount and power compensation amount through linear mapping function.

[0250] The first inverter parameter set is generated by combining.

[0251] Hybrid Optimization Instruction Conversion:

[0252] Parse the strategy weight vector and basic strategy parameter set in the hybrid optimization instruction.

[0253] Feed the policy weight vector into the pre-trained neural network transformer.

[0254] Frequency modulation waveform, phase compensation curve and power response surface are generated through nonlinear transformation.

[0255] The second inverter parameter set is generated by integration.

[0256] The instruction conversion module transmits the inverter parameter set to the execution control module.

[0257] The execution control module receives the inverter parameter set generated by the instruction conversion module, loads it into the programmable logic controller of the target pump unit, and drives the actuator to produce mechanical adjustment action. The specific steps include:

[0258] Parameter set parsing: Parse the inverter parameter set and extract the parameter subset corresponding to the spatial position identifier of the target pump unit.

[0259] Instruction generation:

[0260] If it is a frequency modulation amount, a PWM waveform loading instruction is generated.

[0261] If it is a phase offset, a time delay compensation instruction is generated.

[0262] If it is a power compensation amount, an IGBT trigger pulse sequence is generated.

[0263] Code block assembly: Assemble the generated instructions into programmable logic controller executable code blocks.

[0264] Code block injection: Inject the executable code block into the programmable logic controller runtime environment of the target pump unit to drive the actuator to produce mechanical adjustment action.

[0265] Feedback signal acquisition: Real-time collection of actuator feedback signals, including actual motor speed change, valve opening adjustment value, and pressure transient response value, to construct an execution response triplet.

[0266] Dynamic comparison: The execution response triplet is dynamically compared with the boundary threshold of the target parameter envelope. If all parameters are within the threshold range, it is determined to be in a convergence state. If any parameter exceeds the threshold range, trajectory correction is performed, the parameter deviation gradient value is calculated, and the parameter compensation amount is generated. The parameter compensation amount is superimposed on the parameter subset, triggering a new round of execution loading.

[0267] Closed-loop confirmation: When the convergence state lasts for more than 3 control cycles, a closed-loop confirmation signal is sent to the adaptive optimization module.

[0268] The execution control module transmits the system response matrix to the adaptive optimization module.

[0269] The adaptive optimization module receives the system response matrix transmitted by the execution control module, calculates the strategy effectiveness evaluation index, and triggers model fine-tuning or strategy reconstruction based on the index deviation. The specific steps include:

[0270] System response matrix aggregation: Aggregate the system response matrix by time window sequence, where the row dimension of the matrix corresponds to the pump unit spatial index and the column dimension corresponds to the response parameter type.

[0271] Extraction of strategic effectiveness evaluation factors:

[0272] Extract the deviation of the actual energy efficiency improvement rate from the target value.

[0273] Extract the standard deviation of load balancing.

[0274] Extract the pressure oscillation decay rate.

[0275] Extract traffic tracking response latency.

[0276] Calculation of strategy effectiveness evaluation index: Calculate the strategy effectiveness evaluation index based on the weighted extraction factors.

[0277] Index deviation calculation: Calculate the index deviation between the strategy effectiveness evaluation index and the target benchmark value. The target benchmark value is set based on the historical optimal operating data of the pump group.

[0278] Model fine-tuning or strategy reconstruction:

[0279] When the indicator deviation is less than or equal to the preset value, perform model fine-tuning:

[0280] Extract the set of convolution kernel parameters of the multimodal recognition neural network.

[0281] Calculate parameter gradients based on indicator deviations.

[0282] Update the convolution kernel parameters according to the gradient direction to generate the fine-tuned model.

[0283] When the indicator deviation is greater than the preset value, the strategy reconstruction is executed:

[0284] Get the decision tree topology of the basic optimization strategy library.

[0285] Resolve failure path nodes in the system response matrix.

[0286] The decision tree branch structure is reconstructed based on the node failure frequency, and the reconstructed decision tree topology is output.

[0287] Model or topology loading: Load the fine-tuned model or reconstructed decision tree topology into the runtime environment for use in the next control cycle.

[0288] Through the collaborative operation of the above modules, this embodiment provides a multimodal pump system dynamic optimization control system that can effectively cope with the multimodal operating states and complex operating conditions of the pump system, improving the pump system's operating efficiency and stability. This system not only monitors and optimizes the pump system's operating status in real time but also achieves long-term stable operation through adaptive optimization modules, offering broad application prospects and significant economic benefits.

[0289] It is understood from common technical knowledge that the present invention may be implemented by other embodiments that do not depart from its spirit or essential features. Therefore, the embodiments disclosed above are, in all respects, merely illustrative and not exclusive. All modifications within the scope of the present invention or equivalent to the scope of the present invention are intended to be encompassed by the present invention.

Claims

1. A multi-mode pump group dynamic optimization control method, characterized in that: The following steps are included: S1. Use a distributed sensor array to collect real-time current waveform data, pressure pulsation spectrum, and flow rate time series of each pump unit in the pump system, and perform time-frequency fusion processing on the three types of data to generate a pump system operation status tensor; S2. Input the pump group operation state tensor into a pre-built multimodal recognition neural network, and output the current dominant mode identifier and modal confidence vector. The dominant mode identifier includes the high-efficiency zone operation mode, overload warning mode, low-frequency oscillation mode, flow mismatch mode and mixed conflict mode; S3. When the maximum value in the modal confidence vector exceeds a preset threshold, the basic optimization strategy library corresponding to the dominant modal identifier is activated and a basic optimization instruction is output; when a mixed conflicting mode is identified or the maximum value in the modal confidence vector does not exceed the preset threshold, the cross-modal strategy coordination step is executed: Extract the high-dimensional feature vector of the current running state tensor; Calculate the compatibility weight of each basic optimization strategy; Generate hybrid optimization instructions based on weight fusion; S4. Dynamically select an instruction conversion path based on the optimized instruction type output in step S3 to obtain an inverter parameter set; S5. Load the inverter parameter set into the programmable logic controller of the target pump unit, and drive the actuator of the target pump unit to make the pump group operation trajectory converge to the target parameter envelope.

2. A multi-mode pump group dynamic optimization control method according to claim 1, characterized in that: Also includes, S6. Collect the system response matrix after the instruction is executed, calculate the strategy effectiveness evaluation index, and trigger any of the following based on the index deviation: When the indicator deviation is ≤ the preset value, fine-tune the convolution kernel parameters of the multimodal recognition neural network; When the indicator deviation is greater than the preset value, the decision tree topology of the basic optimization strategy library is reconstructed.

3. A multi-mode pump group dynamic optimization control method according to claim 1, characterized in that: The specific steps of step S1 are: include, S11, synchronously collecting current waveform data, pressure pulsation spectrum and flow rate time series of each pump unit through a distributed sensor array; S12, performing fundamental wave separation processing on the current waveform data, and outputting the current effective value component and harmonic distortion component; S13, performing frequency band energy integration on the pressure pulsation spectrum, and outputting a set of main frequency band amplitudes and a sub-frequency band energy distribution ratio; S14. Perform sliding window statistical analysis on the traffic time series and output the traffic mean series and fluctuation variance series; S15. Integrate the current effective value component and harmonic distortion component output from step S12, the main frequency band amplitude set and sub-frequency band energy distribution ratio output from step S13, and the flow mean value sequence and fluctuation variance sequence output from step S14 according to the following structure: The first physical dimension: the spatial location identification of each pump unit; The second physical dimension: physical characteristic categories of the three parameters of current / pressure / flow; The third physical dimension: the number of consecutive sampling time windows; S16. Perform feature dimensionality reduction compression on the integrated data structure to generate the pump group operation status tensor.

4. A multi-mode pump group dynamic optimization control method according to claim 1, characterized in that: The specific steps of step S2 are: include, S21, inputting the pump group operation status tensor into the first convolutional layer of a pre-built multimodal recognition neural network, extracting spatial dimension features, and outputting a primary feature map; S22, inputting the primary feature map into the time dimension convolution layer, extracting time series correlation features, and outputting a spatiotemporal fusion feature tensor; S23, inputting the spatiotemporal fusion feature tensor into a modal classifier, calculating the matching probability value of each preset modality, and generating a modal confidence vector; S24. Perform a maximum value screening operation on the modal confidence criterion vector: If the maximum value is greater than or equal to the first threshold and the standard deviation of the probability distribution is less than or equal to the second threshold, the dominant mode identifier corresponding to the single mode is output; If the maximum value is less than the first threshold or the standard deviation of the probability distribution is greater than the second threshold, output a mixed conflict mode identifier; The preset modes include: high-efficiency zone operation mode, overload warning mode, low-frequency oscillation mode, and flow mismatch mode.

5. A multi-mode pump group dynamic optimization control method according to claim 1, characterized in that: The construction and calling of the basic optimization strategy library includes: receiving the dominant modality identifier as a strategy selection signal; When the strategy selection signal is the high-efficiency zone operation mode: obtain the current pump group total efficiency benchmark value, calculate the efficiency deviation coefficient of each pump unit, generate the pump unit start and stop priority queue based on the efficiency deviation coefficient, and output the basic optimization instruction containing the target start and stop pump number; When the strategy selection signal is the overload warning mode: extract the real-time load rate of each pump unit, identify the overloaded pump unit and the underloaded pump unit set, calculate the load transfer matrix, and output the basic optimization instructions including the load redistribution parameters; When the strategy selection signal is a low-frequency oscillation mode: obtain the pressure pulsation main frequency phase angle set, calculate the phase compensation amount and damping injection amount, and output the basic optimization instruction including the phase offset instruction; When the strategy selection signal is the flow mismatch mode: analyze the target flow-actual flow difference distribution, generate a flow correction gain coefficient vector, and output a basic optimization instruction including the gain coefficient.

6. A multi-mode pump group dynamic optimization control method according to claim 1, characterized in that: The cross-modal strategy collaboration steps specifically include: Extracting the time domain feature vector, the frequency domain feature vector, and the time series correlation feature vector from the pump group operation state tensor outputted in step S2, and combining them to generate a high-dimensional feature vector; Calculate the matching degree between the high-dimensional feature vector and the constraints of each basic optimization strategy: calculate the energy efficiency adaptability of the high-efficiency zone strategy, calculate the load balancing degree of the overload strategy, calculate the phase compatibility of the oscillation strategy, calculate the flow responsiveness of the mismatch strategy, and output the strategy compatibility measurement set; Normalized weighting is performed based on the policy compatibility metric set: when a single metric value exists ≥ 0.8, the policy weight is set to 0.

7. When all metric values ​​are < 0.8, the weight coefficients are proportionally distributed to generate a policy weight vector. The basic optimization strategy instructions are linearly superimposed according to the strategy weight vector: the start and stop priority queue of the high-efficiency zone strategy is extracted, the load transfer matrix of the overload strategy is extracted, the phase compensation amount of the oscillation strategy is extracted, and the flow correction gain of the mismatch strategy is extracted. The hybrid optimization instructions are generated by fusing them according to the weight coefficients.

7. A multi-mode pump group dynamic optimization control method according to claim 1, characterized in that: The step S4 specifically includes: S41, determine the type of optimization instruction output in step S3: If the basic optimization instruction is executed, a linear conversion path is performed: Extract the start-stop priority queue, load transfer matrix, phase compensation and flow correction gain from the basic optimization instructions; Each parameter is converted into frequency modulation amount, phase offset amount and power compensation amount through a linear mapping function; Combining and generating a first inverter parameter set; If it is a hybrid optimization instruction, a nonlinear conversion path is executed: Parse the strategy weight vector and basic strategy parameter set in the hybrid optimization instruction; Feed the policy weight vector into the pre-trained neural network transformer; Generate frequency modulation waveform, phase compensation curve and power response surface through nonlinear transformation; Integrate and generate a second inverter parameter set; S42. Perform physical dimension alignment on the converted inverter parameter set: extract the spatial position identifier of the pump group, and reorganize the parameter set into a three-dimensional parameter matrix according to the spatial position identifier, where: The first dimension: pump unit spatial index, The second dimension: time window sequence, The third dimension: parameter type channel; S43 , outputting the reorganized three-dimensional parameter matrix as the inverter parameter set for loading and executing in step S5 .

8. A multi-mode pump group dynamic optimization control method according to claim 2, characterized in that: Step S5 specifically includes: S51, parsing the inverter parameter set outputted in step S4, and extracting a parameter subset corresponding to the spatial position identifier of the target pump unit; S52. Based on the parameter type channel identifier of the parameter subset: if it is a frequency modulation amount, generate a PWM waveform loading instruction; if it is a phase offset amount, generate a time delay compensation instruction; if it is a power compensation amount, generate an IGBT trigger pulse sequence; and combine them into a programmable logic controller executable code block; S53, injecting the executable code block into the programmable logic controller runtime environment of the target pump unit to drive the actuator to generate a mechanical adjustment action; S54. Real-time collection of feedback signals from the actuator: actual change in motor speed, valve opening adjustment value, and pressure transient response value, to construct an execution response triplet. S55. Dynamically compare the execution response triplet with the boundary threshold of the target parameter envelope: If all parameters ∈ [lower threshold, upper threshold], it is determined to be in convergence state. If any parameter exceeds [lower threshold, upper threshold], perform trajectory correction: calculate the parameter deviation gradient value, generate the parameter compensation value based on the gradient value, add the parameter compensation value to the parameter subset in step S51, and trigger a new round of execution loading; S56: When the convergence state lasts for more than three control cycles, a closed-loop confirmation signal is sent to step S6.

9. A multi-mode pump group dynamic optimization control method according to claim 8, characterized in that: The step S6 specifically includes: S61, receiving the execution response triplet outputted in step S54, and generating a system response matrix by aggregating the time window sequence, wherein the row dimension of the matrix corresponds to the pump unit spatial index, and the column dimension corresponds to the response parameter type; S62. Extract the following strategic effectiveness evaluation factors from the system response matrix: deviation of the actual energy efficiency improvement rate from the target value, load balancing standard deviation, pressure oscillation attenuation rate, flow tracking response delay, and calculate the strategic effectiveness evaluation index based on the weighted factors; S63, calculating the deviation between the strategy effectiveness evaluation index and the target reference value, wherein the target reference value is set according to the historical optimal operation data of the pump group; S64, when the indicator deviation is less than or equal to the preset value, performing model fine-tuning: extracting the convolution kernel parameter set of the multimodal recognition neural network in step S2, calculating the parameter gradient based on the indicator deviation, and updating the convolution kernel parameters in the gradient direction to generate a fine-tuned model; S65. When the indicator deviation is greater than a preset value, perform strategy reconstruction: obtain the decision tree topology of the basic optimization strategy library, analyze the failure path nodes in the system response matrix, reconstruct the decision tree branch structure based on the node failure frequency, and output the reconstructed decision tree topology; S66. Load the fine-tuned model or the reconstructed decision tree topology into the operating environment for use in the next control cycle.

10. A multi-modal pump group dynamic optimization control system, based on a multi-modal pump group dynamic optimization control method according to any one of claims 1 to 9, characterized in that: include, Multi-source sensing acquisition module, used to collect the current waveform data, pressure pulsation spectrum and flow time series of each pump unit in the pump group system in real time through a distributed sensor array; A time-frequency fusion module is used to perform time-frequency fusion processing on the current waveform data, pressure pulsation spectrum and flow time series to generate a pump group operation status tensor; The modal identification module is used to input the pump group operation status tensor into the pre-built multi-modal identification neural network and output the current dominant mode identifier and modal confidence vector; A strategy selection module is used to activate the basic optimization strategy library to output basic optimization instructions when the maximum value in the modal confidence vector exceeds a preset threshold, and to execute cross-modal strategies to collaboratively generate hybrid optimization instructions when mixed conflicting modes are identified or the maximum value in the modal confidence vector does not exceed the preset threshold; An instruction conversion module is used to dynamically select a conversion path to generate an inverter parameter set according to the optimized instruction type; An execution control module is used to load the inverter parameter set into the programmable logic controller of the target pump unit and drive the actuator to make the pump group operation trajectory converge to the target parameter envelope; The adaptive optimization module is used to collect the system response matrix calculation strategy effectiveness evaluation indicators, and trigger the convolution kernel parameter fine-tuning of the modal recognition module or the decision tree topology reconstruction of the basic optimization strategy library according to the indicator deviation.

Images