A speed control system for a permanent magnet motor

Abstract

The present invention relates to the technical field of permanent magnet motor speed regulation, and specifically to a speed regulation control system for a permanent magnet motor. In the present invention, by identifying the changes in the magnetic properties of a permanent magnet under different thermal conditions and combining the characteristics of the changes in the motor winding structure, a high-dimensional and accurate characterization of the motor operating state is achieved, and the response consistency and error convergence performance of the control gain adjustment are significantly improved. By adjusting the interval gain sequence to obtain a proportional integral factor value group, orderly switching path planning between operating states is achieved, abnormal states are discovered in advance and the source of excitation fluctuations is accurately located. By constructing a voltage excitation adjustment set, the flexible adjustment capability and abnormal response rate of the control strategy are effectively improved, and multi-time domain, multi-stage, fast response and stable speed regulation of the permanent magnet motor under complex working conditions are achieved, the dynamic adaptability and output control accuracy are significantly enhanced, and energy consumption and response lag are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of permanent magnet motor speed regulation, and in particular to a speed regulation control system of a permanent magnet motor. Background Art

[0002] The field of permanent magnet motor speed regulation technology aims to regulate the speed of electric motors that use permanent magnets as excitation sources in electric drive and automation control systems. It has the advantages of compact structure, high efficiency and high power density. It can dynamically adjust the output speed and torque of the motor according to actual working conditions, and can operate efficiently and stably to meet the system requirements of energy saving, safety and response speed.

[0003] The speed control system of a permanent magnet motor aims to solve the problems of unstable speed regulation performance, high energy consumption and delayed response in load fluctuation, starting response, low-speed operation and high-precision positioning scenarios. Through reasonable control strategies, it can achieve the effects of rapid motor start-up, stable operation, high-efficiency output and wide-area speed regulation, thereby improving the overall operating performance and intelligence level of the system.

[0004] During operation, existing technologies generally rely on preset parameters and linear control strategies based on static models to adjust the speed. They do not respond sufficiently to nonlinear factors such as thermal changes in the motor and electromagnetic feedback disturbances. When the load suddenly changes and there are frequent start-stop conditions, the control strategy is difficult to achieve instant perception and fine-grained adjustment of voltage and current disturbances, which can easily lead to control lag and speed overshoot, affecting the overall operational stability of the system. Existing technologies are difficult to accurately identify and respond to abnormal excitation segments. When control instability and overshoot occur, they cannot be corrected immediately through refined gain partitioning and output amplitude regulation, which seriously restricts the performance release of the speed regulation system in intelligent manufacturing and precise motion control scenarios. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a speed control system for a permanent magnet motor.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A speed control system of a permanent magnet motor includes:

[0007] State recognition module: Establishes an operating state configuration set based on the thermal behavior of the permanent magnet, the structural change characteristics of the motor winding, the flux feedback voltage amplitude, and the switching instruction flag;

[0008] Gain adjustment module: Based on the operating state configuration set, a deep feedforward neural network is used to extract the synchronization interval of speed increase and error direction, construct a continuous error change interval, and compare it with the change trend between the jump amplitude of the current cycle gain factor to determine the relationship with the interval matching trend and generate an adjustment interval gain sequence;

[0009] Path scheduling module: Based on the adjustment interval gain sequence, the proportional integral factor value group and the state node jump sequence are obtained, the temperature rise trend fitting and the path segment average jump difference are calculated, and the control response delay sequence is constructed in combination with the continuous cycle error signal response time. The state node sequence is extracted according to the gain jump number, and a state transformation group diagram is constructed with the control response delay sequence to obtain the operation segment switching queue;

[0010] Risk monitoring module: Based on the operating segment switching queue, obtain the voltage amplitude between node segments and the stator current disturbance difference, extract the cross-cycle voltage amplitude mutation point and calculate the current drift peak deviation, and establish the excitation abnormal segment identification set;

[0011] Strategy switching module: Based on the abnormal excitation segment identification set, using the state segment number and the control segment excitation direction, adjusting the interval corresponding to the control factor segment and selecting the segment with the output amplitude peak difference, constructs a voltage excitation adjustment set.

[0012] As a further solution of the present invention, the state recognition module includes:

[0013] Thermal Sequencing Submodule: Based on the thermal behavior of permanent magnets and the structural variation characteristics of motor windings, the module samples and sorts the surface temperature points of the windings. It then segments the temperature difference extreme point spacing into hot zone distribution groups. It extracts the extreme values ​​of the response voltage amplitude within each hot zone to obtain temperature difference variation amplitude data. This data is then sorted to construct a thermal fluctuation positioning index and generate a temperature zone mutation sequence.

[0014] The magnetic flux response submodule: Based on the magnetic flux feedback voltage amplitude, the corresponding speed change rate in the current transient response curve in the reverse jump section is constructed as a rotor response trend curve. The temperature zone mutation sequence generated in the thermal sequencing submodule is called to extract the temperature zone number corresponding to the magnetic deviation section and align the response time window to obtain the thermal-magnetic joint mapping index information. At the same time, the magnetic deviation section and the thermal mutation section are recombined and matched to generate a magnetic deviation structure set.

[0015] Operation mapping submodule: Based on the temperature zone mutation sequence, the magnetic offset structure set and the switching instruction flag bit, the voltage excitation command sequence in each offset segment is extracted and mapped to the control segment mark, the speed mutation amplitude difference before and after the extreme point is calculated and matched with the current jump trajectory segment, and then the speed jump point and the excitation segment are classified and merged to obtain the operation state matching mapping and establish the operation state configuration set.

[0016] As a further solution of the present invention, the gain adjustment module includes:

[0017] Synchronous extraction submodule: Based on the operating state configuration set, the speed difference on both sides of the speed mutation point is calculated and a continuous change vector is generated. The direction change is calculated as a direction change identifier through the speed error signal sequence extracted from the operating state configuration set. The matching sequence is screened, and the phase difference value sequence is calculated based on the time series difference between the speed signal and the excitation signal in the matching sequence. The periodic segments with similar phase difference characteristics are divided into phase groups, and the phase groups are aligned. The excitation signal mutation point is extracted using a deep feedforward neural network, and the intersection extraction is performed with the overlapping area of ​​the speed direction switching point. The offset jump segment area is then excluded to generate a synchronous response segment set;

[0018] Factor generation submodule: Based on the synchronous response segment set, the gradient of the error amplitude sequence in the same direction segment is generated and the jump ratio peak is extracted. The fitting factor structure table is constructed by interpolating the initial values ​​of the proportional and integral factors. The interpolation results are aligned according to the jump segment numbers in the synchronous response segment. The ratio of the factor change before and after the jump to the cycle error increase is calculated to obtain the jump segment interpolation ratio. At the same time, the control error peak of the corresponding cycle is extracted and the difference cross analysis is performed with the parameter amplitude under the same cycle index in the fitting factor structure table. The effective parameters that meet the error constraint conditions are screened to generate the excitation control factor set.

[0019] Interval construction submodule: Based on the excitation control factor set, the direction consistency of the continuous control period factor is judged and the fluctuation reversal samples are eliminated. The difference between the error mutation segment and the fitting factor boundary is calculated to locate the segment boundary. The fitting bandwidth of the continuous segment of the factor is screened and merged with the overlapping segment verification to establish the adjustment interval gain sequence.

[0020] As a further solution of the present invention, the path scheduling module includes:

[0021] Node extraction submodule: Based on the adjustment interval gain sequence, extract the proportional factor sequence, calculate and locate the extreme value points of the jump amplitude difference in the proportional factor sequence, extract the control cycle segment index according to the corresponding number of the integral factor, remove duplicates from the number sequence, accumulate the jump segment length and mark the continuous segment number range, and generate a state jump index set;

[0022] Trend fitting submodule: Based on the state jump index set, the corresponding segment number range in the winding surface temperature measurement point data is matched, and the temperature difference sequence in the corresponding segment is extracted. The fitting difference map is established by calculating the average change speed of the jump segment and aligning it with the temperature increase trend. The integration of the fitting errors and the mean change gradient between each segment are extracted to generate the thermal delay change result;

[0023] Structure construction submodule: Based on the thermal delay change results, compare the excitation order between the control segment nodes and establish a path connection list, extract the peak of the overlapping segment through the temperature rise change direction and time overlap ratio, rearrange the path combination between the nodes and construct a jump control mapping structure to obtain the running segment switching queue.

[0024] As a further solution of the present invention, the rearrangement of the path combination between nodes refers to obtaining the connection paths of multiple control segment nodes, and then performing priority evaluation according to the degree of overlap between the temperature rise change slope corresponding to each path and the excitation time window. By rearranging the order of nodes in the path from high to low according to priority, the path segments that do not meet the time continuity and inconsistent temperature rise trend direction are eliminated, and a control path sequence sorted by excitation consistency and thermal stability is generated.

[0025] As a further solution of the present invention, the risk monitoring module includes:

[0026] Amplitude detection submodule: Based on the operating segment switching queue, extract the voltage values ​​between adjacent nodes and construct a voltage change sequence. It calculates the voltage difference between adjacent cycles to identify the mutation jump point and locate the segment index. It sorts the voltage increase ratios of each segment and extracts the extreme values ​​of the mutation site to generate a voltage mutation index group.

[0027] Disturbance identification submodule: Based on the voltage mutation index group, the stator current waveform in each mutation segment is extracted and a period difference sequence is constructed. The current amplitude peak point is extracted by the fluctuation range. The fluctuation trend slope and the drift value of the mutation zone are extracted by comparing the historical average difference sequence to generate a current drift feature set.

[0028] Abnormal marking submodule: Based on the current drift feature set, cross-screen the current drift peak position and the voltage mutation point segment number and extract the overlapping number sequence, judge the excitation persistence abnormal interval through the synchronous fluctuation of continuous segments, mark and integrate the state number association and abnormal area, and establish the excitation abnormal segment identification set.

[0029] As a further solution of the present invention, the strategy switching module includes:

[0030] Segment number extraction submodule: Based on the excitation abnormal segment identification set, the excitation numbers in the segment number sequence are extracted and classified according to the excitation direction symbol. The segments with opposite directions are excluded and the segments with the same direction are retained to generate an excitation vector group. After merging the consecutive numbers, the difference between the adjacent segment lengths is calculated and the boundary segments are screened to calibrate the jump demarcation points to generate an excitation segment number index set;

[0031] Factor screening submodule: Based on the excitation segment number index set, extract the control factor fragments of the adjustment interval corresponding to each segment number and construct the factor amplitude difference sequence, extract the jump point peak value by calculating the difference between the output control instruction change value and the factor mean value within the segment, summarize the adjacent segments and generate a structured parameter matrix to generate an amplitude difference alignment structure group;

[0032] Instruction construction submodule: Based on the amplitude difference alignment structure group, the corresponding index of the control segment amplitude jump position is converted and a voltage control segment sequence is generated. The non-jump segment is rewritten and filtered out through the control instruction value of each segment. The corresponding position in the original control sequence is replaced while retaining the valid control segment. The complete control instruction set is constructed in chronological order and the voltage excitation adjustment set is constructed.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are:

[0034] 1. In the present invention, by identifying the changes in the magnetic properties of permanent magnets under different thermal conditions and combining the changing characteristics of the motor winding structure, a high-dimensional and accurate characterization of the motor operating state is achieved, which significantly improves the response consistency and error convergence performance of the control gain adjustment.

[0035] 2. In the present invention, by adjusting the interval gain sequence to obtain the proportional integral factor value group, and combining the state node jump sequence to calculate the temperature rise change trend and the path segment jump difference, it is possible to dynamically construct a state transformation diagram and realize orderly switching path planning between operating states, thereby discovering abnormal states in advance and accurately locating the source of excitation fluctuations.

[0036] 3. In the present invention, by constructing a voltage excitation adjustment set, the flexible adjustment capability and abnormal response rate of the control strategy are effectively improved, and multi-time domain, multi-stage, rapid response and stable speed regulation of the permanent magnet motor under complex working conditions are realized, which significantly enhances the dynamic adaptability and output control accuracy, and reduces energy consumption and response lag. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system flow chart of the present invention;

[0038] Figure 2 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] Example:

[0041] See also Figure 1The present invention provides a technical solution: a speed control system of a permanent magnet motor includes:

[0042] State recognition module: Establishes an operating state configuration set based on the thermal behavior of the permanent magnet, the structural change characteristics of the motor winding, the flux feedback voltage amplitude, and the switching instruction flag;

[0043] Gain adjustment module: Based on the operating state configuration set, a deep feedforward neural network is used to extract the synchronization interval between the speed increase and the error direction, construct a continuous error change interval, and compare it with the change trend between the jump amplitudes of the gain factor in the current cycle. The relationship with the interval matching trend is determined and a gain sequence of the adjustment interval is generated;

[0044] Path scheduling module: Based on the adjustment interval gain sequence, the proportional integral factor value group and the state node jump sequence are obtained. The temperature rise trend fitting and the path segment average jump difference are calculated. The control response delay sequence is constructed by combining the continuous cycle error signal response time. The state node sequence is extracted according to the gain jump number. At the same time, a state transformation group diagram is constructed with the control response delay sequence to obtain the operation segment switching queue.

[0045] Risk monitoring module: Based on the operating segment switching queue, it obtains the difference between the voltage amplitude and stator current disturbance between node segments, extracts the cross-cycle voltage amplitude mutation points, calculates the current drift peak deviation, and establishes the excitation abnormal segment identification set;

[0046] Strategy switching module: Based on the identification of the abnormal excitation segment, the state segment number and the control segment excitation direction are used to adjust the control factor segment corresponding to the interval and select the peak value of the segment with the output amplitude difference to construct the voltage excitation adjustment set.

[0047] See also Figure 2 , the state recognition module includes:

[0048] Thermal Sequencing Submodule: Based on the thermal behavior of permanent magnets and the structural variation characteristics of motor windings, the module samples and sorts the surface temperature points of the windings. It then segments the temperature difference extreme point spacing into hot zone distribution groups. It extracts the extreme values ​​of the response voltage amplitude within each hot zone to obtain temperature difference variation amplitude data. This data is then sorted to construct a thermal fluctuation positioning index and generate a temperature zone mutation sequence.

[0049] The magnetic flux response submodule: Based on the magnetic flux feedback voltage amplitude, the corresponding speed change rate in the current transient response curve in the reverse jump section is constructed as a rotor response trend curve. The temperature zone mutation sequence generated in the thermal sequencing submodule is called to extract the temperature zone number corresponding to the magnetic deviation section and align the response time window to obtain the thermal-magnetic joint mapping index information. At the same time, the magnetic deviation section and the thermal mutation section are recombined and matched to generate a magnetic deviation structure set.

[0050] Operation mapping submodule: Based on the temperature zone mutation sequence, the magnetic offset structure set, and the switching instruction flag bit, the voltage excitation command sequence in each offset segment is extracted and mapped to the control segment flag. The speed mutation amplitude difference before and after the extreme point is calculated and matched with the current jump trajectory segment. Then, the speed jump point and the excitation segment are classified and merged to obtain the operation state matching map and establish the operation state configuration set.

[0051] Thermal sequencing submodule: Based on the thermal behavior of the permanent magnet, the temperature distribution reconstruction method is adopted. The surface temperature of the winding is sampled every 0.5 seconds using a thermocouple array sensor. The spatial distribution grid is set to a 10×10 matrix and the center point of each cell is used as the sampling point position. The sampled data is spatially interpolated and fitted through the temperature matrix reconstruction function griddata(points, values, method='cubic') to extract the two-dimensional temperature field. The temperature points are sorted in row priority order using the quick sort algorithm quick_sort(array, low, high). Sorting, recording the sorted point numbers and calculating the temperature difference vector ΔT for all point temperature values ​​in sequence, using a sliding extreme value analysis method with a window size of 5 points to extract the temperature difference change amplitude, and then performing pairwise difference operations on the indexes of the extreme value points in ΔT, calculating the extreme value point spacing, setting the spacing threshold to 4 points for segmentation, and constructing a hot zone distribution group. For the voltage response data corresponding to the sampling points in each hot zone, the maximum and minimum functions np.max() and np.min() are called to extract the extreme values ​​of the response voltage amplitude. Then, the argsort() function is called to sort the extreme value point sequence to construct a thermal fluctuation positioning index and generate a temperature zone mutation sequence.

[0052] Magnetic flux response submodule: Based on the temperature zone mutation sequence, the magnetic flux change rate evaluation method is adopted. The differential function diff(V) is used to perform first-order difference on the sampled magnetic flux voltage in each temperature zone to construct a voltage change rate sequence. The logical operator is used to judge the sign change mode to filter the voltage direction reverse jump segment. The conditional screening logic np.where(np.diff(np.sign(ΔV))!=0) is called to select the sign switching point, extract the reverse segment boundary index, calculate the current change direction of the current frame and the previous frame of each reverse jump segment, extract the current direction consistency segment, and construct a voltage-current matching sequence through Boolean matrix and logical AND operation. The magnetic field switching direction mapping table is constructed according to the index in the matching sequence. The mapping table contains the segment number, jump direction, and current consistency value fields. Then, each mapping item is associated with the magnetic deviation segment position, and is merged and matched with the temperature mutation group according to the segment number to generate a magnetic response segment matrix containing temperature, voltage, and current state vectors, and generate a magnetic offset structure set;

[0053] Operation mapping submodule: Based on the magnetic offset structure set, the operation state mapping modeling method is adopted to extract the voltage excitation command index sequence in the continuous time series of each magnetic offset segment, use the command ID field and timestamp field in the scheduling record data to build the excitation sequence and sort it, call groupby and np.diff() functions to calculate the interval of each excitation command, map the control segment mark field corresponding to the excitation command into the segment index, calculate the difference in speed measurement values ​​in the 20ms window before and after the excitation command is issued, and use the central difference method Δv=(v[t+1]-v[t-1]) / 2 to calculate the speed mutation amplitude. Where t represents the time index of the current analysis moment in the speed measurement sequence. The signal fitting function curve_fit() is called to fit the mutation trajectory and extract the change trend. Then, the second-order difference function np.diff(I, 2) of the current measurement sequence is called to extract the current jump trajectory segment, where I represents the stator current measurement value sequence continuously recorded according to the sampling period. The speed jump point index and current mutation segment are classified and merged to construct the excitation control segment set. The one-to-one mapping relationship between speed, current, and excitation state is matched through the logical aggregation function to obtain the operating state matching mapping and establish the operating state configuration set.

[0054] See also Figure 2 , the gain adjustment module includes:

[0055] Synchronous extraction submodule: Based on the operating state configuration set, the speed difference on both sides of the speed mutation point is calculated and a continuous change vector is generated. The direction change is calculated as a direction change identifier through the speed error signal sequence extracted from the operating state configuration set. The matching sequence is screened, and the phase difference value sequence is calculated based on the time series difference between the speed signal and the excitation signal in the matching sequence. The periodic segments with similar phase difference characteristics are divided into phase groups, and the phase groups are aligned. The excitation signal mutation point is extracted using a deep feedforward neural network, and the intersection extraction is performed with the overlapping area of ​​the speed direction switching point. The offset jump segment area is then excluded to generate a synchronous response segment set;

[0056] Factor generation submodule: Based on the synchronous response segment set, the gradient of the error amplitude sequence in the same direction segment is generated and the maximum jump ratio is extracted. The fitting factor structure table is constructed by interpolating the initial values ​​of the proportional and integral factors. The interpolation results are aligned according to the jump segment numbers in the synchronous response segment. The ratio of the factor change before and after the jump to the cycle error increase is calculated to obtain the jump segment interpolation ratio. At the same time, the control error peak of the corresponding cycle is extracted and the difference cross analysis is performed with the parameter amplitude under the same cycle index in the fitting factor structure table. The effective parameters that meet the error constraint conditions are screened to generate the excitation control factor set.

[0057] Interval construction sub-module: Based on the excitation control factor set, judge the consistency of the factor direction in consecutive control cycles and eliminate the fluctuation reversal samples, calculate the difference between the error mutation section and the fitting factor boundary to locate the section boundary, screen the fitting bandwidth of the continuous factor segments and check and merge them with the overlapping segments, and establish an adjustment interval gain sequence;

[0058] Synchronous extraction sub-module: Based on the operating state configuration set, use the speed difference method to call np.diff(speed_sequence) to calculate the difference on both sides of the mutation point and generate a continuous change vector, call np.sign(np.diff(error_signal)) to obtain the error signal direction change index and match the speed change sequence, calculate the phase delay through the np.correlate() function and reconstruct the alignment sequence, and use a deep feedforward neural network to extract the overlapping region. The network structure includes three hidden layers, with 128 nodes in each layer, the activation function is ReLU, the output layer is a linear structure, the input features are the instantaneous speed, load estimation value, first derivative of the error signal, and the change amplitude of the previous cycle control signal. Use model.predict([ω_t, T_est, de_dt, Δu]) to output the PI control factor and use it as an auxiliary criterion for the mutation point, where Δu is the change amplitude of the voltage control signal in two adjacent control cycles. Use np.intersect1d(trigger_index, switch_index) to extract the overlapping region of the excitation signal mutation point and the speed direction switching point, introduce v as the time alignment error tolerance threshold, filter the abnormal matching pairs of offsets, call np.where(abs(offset)<v) to exclude the offset jump section area, and generate a synchronous response section set;

[0059] Factor generation submodule: Based on the synchronous response segment set, the error amplitude sequence is calculated using the segmented gradient extraction method. The difference function np.diff(error_vector) is called to perform first-order difference on the data in the same direction segment and the np.gradient() function is used to extract the local gradient value. The maximum jump ratio in each segment is extracted by the max(Δe) function, where Δe is the first-order difference sequence of the control error signal in the same direction segment. The initial value interval of the proportional and integral factors is constructed, and the proportional factor interval is defined as Kp∈[0.1, 1.0] and the integral factor interval is defined as Ki∈[0.01, 0.5], the bilinear interpolation method interp2d(Kp, Ki, Z) is used to construct the factor structure table, where Z is the control error response result matrix corresponding to different combinations of proportional factor Kp and integral factor Ki parameters. The interpolation ratio is matched using the principle of minimizing the difference between the fitting error in the jump segment and the interpolation output value in the structure table. The matching result is selected using np.argmin(abs(fitted-actual)), and the np.where() function is called to screen the valid parameter segment corresponding to the intersection of the control error peak and the interpolation factor structure to generate the excitation control factor set.

[0060] Interval construction submodule: Based on the excitation control factor set, the direction consistency judgment logic is adopted to calculate the factor direction difference within the continuous control cycle, call np.where(Δθ==0) to screen the direction consistent segments and eliminate the direction reversal samples, where Δθ represents the control factor direction change sequence within the continuous control cycle, use the mutation boundary detection algorithm to calculate the absolute value of the difference between the error mutation segment and the fitting factor boundary, call abs(error_edge-factor_edge) and set the threshold 0.05 to locate the segment corresponding to the fitting boundary, use the bandwidth aggregation analysis method to calculate the length of the factor continuous segment, call np.bincount() to count the continuous length and extract segments greater than the specified value (such as N=3), use the overlapping segment time series index cross-comparison to call np.intersect1d(segment_1, segment_2) to verify and merge the factor bandwidth segments, and establish the adjustment interval gain sequence.

[0061] Deep feedforward neural network follows the formula:

[0062] ;

[0063] in: For the The composite excitation mutation index value of the sampling period, is the voltage mutation weight coefficient, For the The voltage control signal value within a control cycle, For the The voltage control signal value within a control cycle, is the absolute difference between adjacent cycles of the voltage control signal, is the error signal derivative weight coefficient, For the The first derivative of the error signal within a control period, For the The error signal increment, is the sampling period length, is the absolute value of the derivative of the error signal, is the weight coefficient of the proportional factor change value, For the The change value of the proportional factor within a period, is the absolute value of the proportional factor change, is the load change rate weight coefficient, For the The rate of change of the load estimate within a cycle, For the Load estimation within a cycle, is the absolute value of the load change rate, if is greater than the set threshold, then the decision point is the mutation point of the excitation signal;

[0064] Execution process: First obtain the voltage control signal value of the current cycle Compared with the previous period value , calculate the absolute difference , as the basic measure of voltage variation, extract the first The error value of the cycle and the error value of the previous cycle are used to calculate the derivative , that is, the error change rate, and then the neural network structure outputs the proportional factor in the current cycle PI controller And compare it with the output value of the previous cycle to get the change of the proportional factor , evaluate the controller self-regulation strength, obtain the first Cycle load Calculate the derivative with the previous period value , indicating the changing trend of external working conditions, multiply the four measures by the set weight coefficients , weighted summation to obtain the composite excitation mutation index , compare the value with the set threshold, when When it is greater than the set threshold, it is determined that there is a sudden change in the excitation signal in the current cycle, and the corresponding time point is included as a sudden change point in the control parameter update logic and the synchronous matching sequence.

[0065] See also Figure 2 , the path scheduling module includes:

[0066] Node extraction submodule: Based on the adjustment interval gain sequence, extract the proportional factor sequence, calculate and locate the extreme value points of the jump amplitude difference in the proportional factor sequence, extract the control cycle segment index according to the corresponding number of the integral factor, remove duplicates from the number sequence, accumulate the jump segment length and mark the continuous segment number range, and generate a state jump index set;

[0067] Trend fitting submodule: Based on the state jump index set, the corresponding segment number range in the winding surface temperature measurement point data is matched, and the temperature difference sequence in the corresponding segment is extracted. The fitting difference map is established by calculating the average change speed of the jump segment and aligning it with the temperature increase trend. The integration of the fitting errors and the mean change gradient between each segment are extracted to generate the thermal delay change result;

[0068] Structure construction submodule: Based on the thermal delay change results, compare the activation order between control segment nodes and establish a path connection list, extract the maximum overlap segment based on the temperature rise change direction and time overlap ratio, rearrange the path combinations between nodes and construct a jump control mapping structure to obtain the operation segment switching queue;

[0069] Node extraction submodule: Based on the adjustment interval gain sequence, calculate the extreme points of the scale factor sequence, call the np.diff(Kp_sequence) function to calculate the difference sequence of adjacent scale factor values, calculate the absolute value of the difference through np.abs(), and then use the argrelextrema(array, np.greater) function to extract the first-order maximum point index as the extreme point index position of the jump amplitude difference. According to the corresponding number of the integral factor, use np.where(Ki_index==current_index) to extract the control cycle segment index set, call the np.unique() function to perform deduplication and sort the number sequence in the set, use a loop structure to accumulate and judge the difference between consecutive numbers. If the number difference is 1, it is classified into the same segment, accumulate the jump segment length and define the segment number range, combine all segment numbers and extreme value indexes to build an index dictionary, and generate a state jump index set;

[0070] Trend fitting submodule: Based on the state jump index set, the window interval positioning method is used to match the temperature rise data number range, extract the temperature value sequence in the corresponding segment, calculate the temperature difference of each segment, and construct the temperature difference change vector. Through the time series of each segment in the jump segment, the ΔT / Δt formula is called to calculate the average temperature change rate. Using the linear trend matching algorithm, the scipy.stats.linregress() function is called to fit the temperature change trend to obtain the fitting slope and intercept. The difference between the output value of the fitting function and the actual temperature difference sequence is used as the fitting error. Then, np.mean(abs(fitted-actual)) is called to calculate the mean of the change gradient, construct the difference map structure array, and generate the thermal delay change result;

[0071] Structure construction submodule: Based on the results of thermal delay change, the path connection matrix construction method is adopted. The excitation command sequence between control segment nodes is compared to extract the excitation timestamp sequence between each pair of control segments. After sorting, np.argsort(command_time) is called to construct the excitation sequence list. A two-dimensional path connection array is established and initialized to a zero matrix. The excitation sequence pairs are marked as the starting and ending points of the path, and the corresponding positions in the matrix are assigned 1. The temperature rise change direction vector is cross-compared with the excitation time series. np.where(direction[i:j]==sign(ΔT[i:j])) is called to extract the overlapping direction segments. Then, the time cross-ratio function np.intersect1d(time_list1, time_list2) is used to extract the maximum overlapping segment index. The order between path nodes is rearranged. The node arrangement order is regenerated through the topological sorting algorithm topological_sort(path_list). The node jump connection dictionary is constructed. Then, the path jump structure is assembled based on the index order and the jump control mapping structure is constructed to obtain the running segment switching queue.

[0072] Rearranging the path combination between nodes means obtaining the connection paths of multiple control segment nodes, and then performing priority evaluation according to the degree of overlap between the temperature rise change slope corresponding to each path and the excitation time window. By rearranging the order of nodes in the path from high to low according to priority, path segments that do not meet the time continuity and inconsistent temperature rise trend direction are eliminated, and a control path sequence sorted by excitation consistency and thermal stability is generated.

[0073] See also Figure 2 , the risk monitoring module includes:

[0074] Amplitude detection submodule: Based on the operating segment switching queue, extract the voltage values ​​between adjacent nodes and construct a voltage change sequence. It calculates the voltage difference between adjacent cycles to identify the mutation jump point and locate the segment index. It sorts the voltage increase ratios of each segment and extracts the extreme values ​​of the mutation site to generate a voltage mutation index group.

[0075] Disturbance identification submodule: Based on the voltage mutation index group, the stator current waveform in each mutation segment is extracted and a period difference sequence is constructed. The current amplitude peak point is extracted by the fluctuation range. The fluctuation trend slope and the drift value of the mutation zone are extracted by comparing the historical average difference sequence to generate a current drift feature set.

[0076] Abnormal marking submodule: Based on the current drift feature set, the current drift peak position and the voltage mutation point segment number are cross-screened and the overlapping number sequence is extracted. The excitation persistence abnormal interval is determined by the synchronous fluctuation of the continuous segment. The state number association and abnormal area are marked and integrated to establish the excitation abnormal segment identification set.

[0077] Amplitude detection submodule: Based on the running segment switching queue, the differential analysis method is used to extract the voltage values ​​between adjacent nodes. A loop structure is called to traverse the node number list. U[i] and U[i+1] are called respectively to extract the corresponding voltage values ​​and construct the sequence ΔU. The periodic voltage difference is calculated using np.diff(U_seq). The logical judgment np.where(abs(ΔU)>threshold) is called to identify the mutation jump point. The mutation judgment threshold is set to 20V. The corresponding jump point number is extracted according to the index as the positioning index segment. The sorting function np.argsort(abs(ΔU)) is called to sort the jump segment index from large to small according to the amplification ratio. The point with the largest absolute value among the first N jump sites is extracted. N is set to 5. The extreme jump position number is stored in the index list and a voltage mutation index group is generated.

[0078] Disturbance identification submodule: Based on the voltage mutation index group, the current differential fluctuation detection method is used to extract the stator current waveform data within the mutation segment. The sliding window method is called to extract the current value in each time series. The window width is set to 1 cycle, and a periodic difference sequence is generated. The np.ptp(I_diff) function is called to calculate the range of the difference sequence, extract the current amplitude peak point, and compare each difference sequence with the historical average difference by calling np.mean(H_diff). For each sequence, np.polyfit(x, y, deg=1) is called to perform a first-order fit to obtain the fluctuation trend slope, where x is the time index and y is the difference. The difference between the beginning and end values ​​of the current mutation segment is extracted to calculate the drift value of the mutation area drift=I[end]-I[start]. A vector set containing the slope, drift value, and current range is constructed to generate a current drift feature set.

[0079] Abnormal labeling submodule: Based on the current drift feature set, the index cross-screening method is used to match the current drift peak position with the voltage jump segment number, and np.intersect1d(current_peak_index, voltage_jump_index) is called to extract the overlapping number sequence. The current and voltage sequences are extracted for the continuous time intervals in the overlapping segment. np.sign(diff(I))==np.sign(diff(U)) is called to judge the consistency of the fluctuation direction in the continuous segment. The continuous segments whose length exceeds the specified threshold are counted and marked with segment numbers. The abnormal number of each segment is combined with the original state number by calling zip(). The number-abnormal state dictionary structure is constructed and integrated and classified to establish the excitation abnormal segment identification set.

[0080] See also Figure 2 , the strategy switching module includes:

[0081] Segment number extraction submodule: Based on the excitation abnormal segment identification set, the excitation numbers in the segment number sequence are extracted and classified according to the excitation direction symbol. The segments with opposite directions are excluded and the segments with the same direction are retained to generate an excitation vector group. After merging the consecutive numbers, the difference between the adjacent segment lengths is calculated and the boundary segments are screened to calibrate the jump demarcation points to generate an excitation segment number index set;

[0082] Factor screening submodule: Based on the excitation segment number index set, extract the control factor fragments of the adjustment interval corresponding to each segment number and construct the factor amplitude difference sequence, extract the maximum jump point by calculating the difference between the output control instruction change value and the factor mean within the segment, summarize the adjacent segments and generate a structured parameter matrix to generate an amplitude difference alignment structure group;

[0083] Instruction construction submodule: Based on the amplitude difference alignment structure group, the corresponding index of the control segment amplitude jump position is converted and a voltage control segment sequence is generated. The non-jump segment is rewritten and filtered out by the control instruction value of each segment. The corresponding position in the original control sequence is replaced while retaining the valid control segment. The complete control instruction set is constructed in chronological order to construct the voltage excitation adjustment set.

[0084] Segment number extraction submodule: Based on the excitation abnormal segment identification set, the classification label screening method is used to process the segment number sequence. np.array() is called to convert the segment number sequence into an array structure. The excitation direction value corresponding to each segment number is used to call np.sign(direction_sequence) to generate a sign vector. np.where() is used to extract the segment numbers with positive direction signs as the retained objects. After calling Boolean negation to exclude the reverse sign segments, the excitation vector group is constructed and the direction mark is recorded. np.diff() is called for the numbers in the excitation vector group to calculate the difference between consecutive segment numbers. The conditional judgment logic diff==1 is used to extract the consecutive number set and call the aggregation function to group and merge. np.diff(length_vector) is called for each group of segment numbers to calculate the segment length difference and np.argmax(abs(Δlength)) is used to screen the boundary segment with the largest change amplitude. The number and the index of the point with the largest difference are integrated to generate the excitation segment number index set.

[0085] Factor screening submodule: Based on the excitation segment number index set, the factor difference construction method is used to extract the adjustment interval control factor fragments, and the index screening function segment_K=K_matrix[segment_index_list] is called to extract the corresponding factor fragments from the original control factor matrix. The np.max() and np.min() functions are called to calculate the difference of each factor fragment and construct the amplitude difference sequence ΔK. The output control instruction change value ΔC=|C_out[i]-mean(K_segment)| is calculated and the maximum jump point is determined using np.argmax(ΔC). The set of adjacent segments is concatenated by calling np.concatenate(). The merged segments are statistically constructed by calling np.mean() and np.std() functions to construct parameter matrix fields including mean, range, variance and position index, integrate the structured data of all segments and generate an amplitude alignment structure group;

[0086] Instruction construction submodule: Based on the amplitude difference alignment structure group, the index mapping update method is used to perform conversion operations on the amplitude jump position of the control segment, extract the amplitude jump index list in the structure group, call the position mapping function index_map[i]=control_index[i] in the original control segment array for each index to build an index table, extract the voltage instruction corresponding to the segment number according to the index of the control segment instruction sequence, call the instruction update function V_command[i]=new_value to rewrite the instruction value, use the jump identification vector to filter the voltage value of the non-jump segment, reconstruct the full interval segment number sequence and call np.append() to splice and merge each control segment, organize it into a standard control instruction structure array and construct a voltage excitation adjustment set.

[0087] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A speed control system for a permanent magnet motor, characterized in that: The system comprises: State recognition module: Establishes an operating state configuration set based on the thermal behavior of the permanent magnet, the structural change characteristics of the motor winding, the flux feedback voltage amplitude, and the switching instruction flag; Gain adjustment module: Based on the operating state configuration set, a deep feedforward neural network is used to extract the synchronization interval of speed increase and error direction, and a continuous error change interval based on the speed error is constructed. The interval is compared with the change trend between the jump amplitudes of the gain factor in the current cycle, and the relationship with the matching trend of the continuous error change interval is determined to generate an adjustment interval gain sequence; Path scheduling module: Based on the adjustment interval gain sequence, a proportional integral factor value group and a state node jump sequence are obtained, the temperature rise trend fitting and the control segment average jump difference are calculated, and a control response delay sequence is constructed in combination with the continuous cycle error signal response time. The state node sequence is extracted according to the gain jump number, and a state transformation group diagram is constructed with the control response delay sequence to obtain an operation segment switching queue composed of state nodes; Risk monitoring module: Based on the operation segment switching queue, obtain the difference between the control voltage amplitude and the stator current disturbance between nodes, extract the cross-cycle control voltage amplitude mutation point and calculate the current drift peak deviation, and establish the excitation abnormal segment identification set; Strategy switching module: Based on the excitation abnormal segment identification set, using the control segment number and the excitation direction of the control segment, adjusts the proportional integral factor segment corresponding to the interval, and selects the segment with the peak point of the difference between the output voltage control command value and the factor mean to construct a voltage excitation adjustment set.

2. The speed control system of the permanent magnet motor according to claim 1, characterized in that: The state recognition module includes: Thermal Sequencing Submodule: Based on the thermal behavior of permanent magnets and the structural variation characteristics of motor windings, the module samples and sorts the surface temperature points of the windings. It then segments the temperature difference extreme point spacing into hot zone distribution groups. It extracts the extreme values ​​of the response voltage amplitude within each hot zone to obtain temperature difference variation amplitude data. This data is then sorted to construct a thermal fluctuation positioning index and generate a temperature zone mutation sequence. The flux response submodule: Based on the flux feedback voltage amplitude, the speed change rate corresponding to the stator current transient response curve in the reverse jump section is constructed as a rotor response trend curve. The temperature zone mutation sequence generated in the thermal sequencing submodule is called to extract the temperature zone number corresponding to the magnetic offset section and align the response time window to obtain the thermal-magnetic joint mapping index information. At the same time, the magnetic offset section and the thermal mutation section are recombined and matched to generate a magnetic offset structure set. Operation mapping submodule: Based on the temperature zone mutation sequence, the magnetic offset structure set and the switching instruction flag bit, the voltage excitation command sequence in each magnetic offset segment is extracted and mapped to the control segment mark, the speed mutation amplitude difference before and after the voltage excitation command is issued is calculated and the current jump trajectory segment is matched, and then the speed mutation points and the excitation segments are classified and merged to obtain the operation state matching mapping and establish the operation state configuration set.

3. The speed control system of the permanent magnet motor according to claim 1, characterized in that: The gain adjustment module includes: Synchronous extraction submodule: Based on the operating state configuration set, the speed difference on both sides of the speed mutation point is calculated and a continuous change vector is generated. The speed error signal sequence is calculated based on the speed information in the operating state configuration set. The direction change is calculated as a direction change identifier. According to the direction change identifier, the speed signal period that matches the corresponding excitation signal change is screened out to generate a matching sequence. The phase difference value sequence is calculated based on the time series difference between the speed signal and the excitation signal in the matching sequence. The periodic segments with similar phase difference characteristics are divided into phase groups, and the phase groups are aligned. The excitation signal mutation point is extracted using a deep feedforward neural network, and the intersection extraction is performed with the overlapping area of ​​the speed direction switching point. The magnetic offset segment is then excluded to generate a synchronous response segment set. Factor generation submodule: Based on the synchronous response segment set, the gradient of the error amplitude sequence in the same direction segment is generated and the maximum jump ratio is extracted. The fitting proportional integral factor structure table is constructed by interpolating the initial values ​​of the proportional and integral factors. The interpolation results are aligned according to the jump segment numbers in the synchronous response segment. The ratio of the proportional integral factor change before and after the jump to the period error increase is calculated to obtain the jump segment interpolation ratio. At the same time, the control error peak of the corresponding period is extracted and the difference cross analysis is performed with the parameter amplitude under the same period index in the fitting proportional integral factor structure table. The effective parameters that meet the error constraint conditions are screened to generate the excitation control proportional integral factor set. Interval construction submodule: Based on the excitation control proportional integral factor set, the consistency of the direction of the continuous control period proportional integral factor is determined and the fluctuation reversal samples are eliminated. The difference between the error mutation segment and the fitting proportional integral factor boundary is calculated to locate the segment boundary. The fitting bandwidth of the continuous segment of the proportional integral factor is screened and merged with the overlapping segment verification to establish the adjustment interval gain sequence.

4. The speed control system of the permanent magnet motor according to claim 1, characterized in that: The path scheduling module includes: Node extraction submodule: Based on the adjustment interval gain sequence, extract the proportional factor sequence, calculate and locate the extreme value points of the jump amplitude difference in the proportional factor sequence, extract the control cycle segment index according to the corresponding number of the integral factor, remove duplicates from the number sequence, accumulate the jump segment length and mark the continuous segment number range, and generate a state jump index set; Trend fitting submodule: Based on the state jump index set, the corresponding segment number range in the winding surface temperature measurement point data is matched, and the temperature difference sequence in the corresponding segment is extracted. The fitting difference map is established by calculating the average change speed of the jump segment and aligning it with the temperature increase trend. The integration of the fitting errors and the mean change gradient between each segment are extracted to generate the thermal delay change result; Structure construction submodule: Based on the thermal delay change results, compare the excitation order between the control segment nodes and establish a path connection list, extract the peak overlap section through the temperature rise change direction and time overlap ratio, rearrange the path combination between the nodes and construct a jump control mapping structure to obtain the operation segment switching queue.

5. The speed control system of the permanent magnet motor according to claim 4, characterized in that: The rearrangement of the inter-node path combination refers to obtaining the connection paths of multiple control segment nodes, and then performing priority evaluation according to the degree of overlap between the temperature rise change slope corresponding to each path and the excitation time window. By rearranging the order of nodes in the path from high to low according to priority, path segments that do not meet time continuity and inconsistent temperature rise trend directions are eliminated, and a control path sequence sorted by excitation consistency and thermal stability is generated.

6. The speed control system of the permanent magnet motor according to claim 1, characterized in that: The risk monitoring module includes: Amplitude detection submodule: Based on the operating segment switching queue, it extracts the control voltage values ​​between adjacent nodes and constructs a voltage change sequence. It identifies the mutation jump point and locates the segment index by calculating the voltage difference between adjacent cycles. It sorts the voltage increase ratios of each segment and extracts the extreme values ​​of the mutation site to generate a voltage mutation index group. Disturbance identification submodule: Based on the voltage mutation index group, the stator current waveform in each mutation segment is extracted and a period difference sequence is constructed. The current amplitude peak point is extracted by the fluctuation range. The fluctuation trend slope and the drift value of the mutation zone are extracted by comparing the historical average difference sequence to generate a current drift feature set. Abnormal marking submodule: Based on the current drift feature set, cross-screen the current drift peak position and the voltage mutation point segment number and extract the overlapping number sequence, judge the excitation persistence abnormal interval through the synchronous fluctuation of continuous segments, mark and integrate the state number association and abnormal area, and establish the excitation abnormal segment identification set.

7. The speed control system of the permanent magnet motor according to claim 1, characterized in that: The strategy switching module includes: Segment number extraction submodule: Based on the excitation abnormal segment identification set, the excitation numbers in the segment number sequence are extracted and classified according to the excitation direction symbol. The segments with opposite directions are excluded and the segments with the same direction are retained to generate an excitation vector group. After merging the consecutive numbers, the difference between the adjacent segment lengths is calculated and the boundary segments are screened to calibrate the jump demarcation points to generate an excitation segment number index set; Factor screening submodule: Based on the excitation segment number index set, extract the proportional integral factor fragments corresponding to the adjustment interval of each segment number and construct a proportional integral factor amplitude difference sequence. By calculating the difference between the output voltage control command value and the mean value of the proportional integral factor in the segment, the peak point of the difference is extracted as the jump point, and the adjacent segments are aggregated to generate a structured parameter matrix to generate an amplitude difference alignment structure group. Instruction construction submodule: Based on the amplitude difference alignment structure group, the control segment index corresponding to the jump point is converted, and a voltage control segment sequence is generated. The non-jump segment is rewritten and filtered out through the control instruction value of each segment. The corresponding position in the original control sequence is replaced on the basis of retaining the valid control segment. The complete control instruction set is constructed in chronological order to generate a voltage excitation adjustment set.

Images